The amyloidogenic C-terminal region of TMEM106B modulates lipid membrane biophysical properties: Functional and pathological insights

TL;DR

This study explores how a part of the TMEM106B protein interacts with cell membranes, revealing its role in altering membrane properties and its potential link to amyloid formation in the brain.

Contribution

The study identifies the C-terminal fragment of TMEM106B as a membrane remodeler that affects lipid bilayer properties without deep insertion.

Findings

TST binds peripherally to lipid bilayers and alters their fluidity and elasticity.

Membranes with anionic lipids and no cholesterol show the strongest effects from TST binding.

TST induces changes in lipid headgroup motion and acyl chain order, indicating surface-level membrane perturbation.

Abstract

The lysosomal transmembrane protein 106B (TMEM106B) forms amyloid filaments in the human brain in an age-dependent manner, observed both in neurologically healthy individuals and in patients with neurodegenerative diseases also containing tau, α-synuclein, or TDP-43 inclusions. Despite its pathological and physiological relevance, the biochemical mechanisms governing TMEM106B structural stability and its functional interactions with membranes remain largely unknown. Here, we examined the luminal C-terminal fragment of TMEM106B (called TST, residues 120–254), corresponding to the amyloid fibril core identified by cryo-electron microscopy, to elucidate its functional membrane-binding properties. Using static solid-state 31P and 2H NMR in combination with magic-angle spinning 13C NMR, we characterized TMEM106B(120–254) interaction with multilamellar vesicles of varying lipid composition that mimic lysosomal membranes. TST binds peripherally to lipid bilayers and remodels their fluidity and elasticity in a composition-dependent manner. 31P NMR spectra revealed reduced chemical shift anisotropy and increased asymmetry, accompanied by an isotropic component indicative of enhanced headgroup motion and local curvature. Complementary 2H NMR spectra of POPC-d31 showed decreased quadrupolar splittings and order parameters, demonstrating reduced acyl chain order upon TST binding. These effects were most pronounced in membranes containing anionic lipids and lacking cholesterol, suggesting that electrostatic interactions and lipid motion modulate the balance between random coil mobile TMEM106B and membrane-immobilized β-rich TMEM106B at the bilayer surface. Together, these findings identify TST as a surface-active remodeler that perturbs membrane structure without deep insertion, providing new insights into the membrane coupling mechanisms of TMEM106B and their potential implications for lysosomal physiology and amyloid formation.

Figures

Full text

Neurodegenerative disorders affect 55 million people worldwide and are a leading cause of death and disability (1). Misfolding of intrinsically disordered and functional proteins into β-sheet aggregates is associated with cellular toxicity. Among the most common misfolding proteins, the microtubule-associated protein tau aggregates in Alzheimer’s disease (AD) and ∼20 other disorders; α-synuclein aggregates in Parkinson's disease; and TDP-43 aggregates in frontotemporal lobar degeneration (FTLD-TDP). Groundbreaking cryo-electron microscopy (cryo-EM) analyses has very recently allowed to discover the presence of TMEM106B amyloid fibrils isolated from sarkosyl-insoluble fractions of post mortem brain tissue from individuals with a range of conditions, including AD, amyotrophic lateral sclerosis, progressive supranuclear palsy, corticobasal degeneration, sporadic and inherited Parkinson’s disease, inherited and sporadic FTLD with TDP-43 inclusions (FTLD-TDP), parkinsonism linked to chromosome 17 (FTDP-17T), and so on, as well as neurologically normal controls (2, 3, 4, 5). These discoveries, mostly made in 2022, now place TMEM106B at the center neurodegeneration and aging, as it is, to date, the only known amyloid protein that tauopathies, synucleinopathies, and TDP-43opathies have in common (6, 7).

TMEM106B is a type II transmembrane protein localized predominantly to late endosomes and lysosomes, where it contributes to the regulation of lysosomal trafficking, positioning, and function (8). It is a 274-residue protein with a short N-terminal cytosolic tail, a single-pass transmembrane helix, and a large luminal C-terminal domain (9, 10). Although initially identified as a risk modifier for FTLD with TDP-43 pathology (11, 12, 13, 14, 15), TMEM106B has since been implicated more broadly in the biology of aging and age-related neurodegeneration. Genome-wide association studies have linked common TMEM106B variants to altered risk of FTLD, AD, and cognitive decline in the elderly. Mechanistically, TMEM106B is thought to regulate lysosomal size, acidification, and trafficking along microtubules (10, 16, 17, 18, 19, 20, 21, 22), although its precise role has been not identified. The biophysical properties of TMEM106B’s function remain completely unknown. Overexpression or disease-associated variants lead to enlarged lysosomes, altered lysosomal distribution, and impaired clearance of proteins, contributing to lysosomal stress (16, 17, 18, 19, 20, 21, 22). More recently, C-terminal fragments of TMEM106B have been found to assemble into amyloid fibrils in the brains of aged individuals, independent of neurodegenerative disease diagnosis (2, 3, 4, 5). These fibrils, which derive from residues ∼120 to 254 of the luminal domain, accumulate with age and have been proposed as a hallmark of brain aging. Thus, TMEM106B has emerged as both a regulator of lysosomal homeostasis and an age-associated aggregation-prone protein with potential roles in neurodegeneration.

Membrane proteins such as TMEM106B operate in a highly complex lipid environment, and their function and misfolding are intimately linked to protein–lipid interactions (23). Conversely, proteins can remodel membranes, induce curvature, or alter lipid phase behavior. These reciprocal interactions are particularly relevant in the lysosomal system, where the lipid composition is enriched in anionic phospholipids, sphingolipids, and cholesterol at a relatively low pH environment (24, 25). A wide range of amyloidogenic proteins interact with membranes during their aggregation pathways. For example, α-synuclein binds acidic phospholipids and can remodel bilayer curvature (26), while amyloid-β can insert into membranes (27, 28). Tau, though largely cytosolic, can also interact with anionic membranes via its microtubule-binding region, with evidence that these interactions promote conformational changes and aggregation (29, 30, 31). These examples underscore how amyloid-prone proteins can directly perturb membranes, and how lipid composition can modulate protein aggregation pathways.

Lysosomal membranes are compositionally distinct from those of the plasma membrane and other organelles (24, 25). They are enriched in cholesterol, sphingomyelin (SM), and the anionic phospholipid bis(monoacylglycero)phosphate (BMP), which is almost exclusively found in late endosomes and lysosomes (32). In addition, lysosomal membranes contain elevated levels of phosphatidylserine and phosphatidylinositol species, which contribute to their net negative surface charge. This unique lipid environment creates a bilayer that is highly charged and optimized for both enzymatic degradation and vesicular trafficking. Perturbations in lysosomal lipid composition are increasingly recognized as hallmarks of aging and neurodegeneration (33, 34, 35). Age-related accumulation of cholesterol and sphingolipids can stiffen lysosomal membranes, impair fusion, and compromise membrane integrity. Because TMEM106B resides in lysosomes, both its normal regulatory roles and its aggregation propensity must be understood in the context of this lipid milieu.

In this study, we investigated how the C-terminal fragment of TMEM106B(120–254) (TST) (Fig. 1, A and B) interacts with model lipid membranes of varying composition, combining ^31^P and ^2^H static solid-state nuclear magnetic resonance (NMR) with magic-angle spinning (MAS) solid-state NMR ^13^C experiments. TMEM106B(120–254) is the amyloidogenic fibrils core of TMEM106B as found in cryo-electron microscopy studies (2, 3, 4, 5, 36). These filaments can be isolated both from neurologically healthy individuals and from patients with neurodegenerative diseases, including those characterized by amyloid-β, tau, α-synuclein, or TDP-43 inclusions. So far, little is known about the biochemical and biophysical pathways underlying TMEM106B aggregation, or about its functional mechanisms within the cellular context. To address this gap, we focused on characterizing the membrane-binding properties of the luminal domain of TMEM106B under conditions that closely mimic the cellular environment. We made three types of multilamellar vesicles (MLVs) with lipid composition close to lysosomal membranes: 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), negatively charged lipids, SM, and cholesterol. Although lysosomes uniquely contain the uncommon negatively charge lipid BMP, we used 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS) to add negative charges to the otherwise neutral membranes. POPS is an anionic glycerophospholipid bearing a single phosphomonoester headgroup linked to serine, whereas BMP is an atypical anionic phospholipid composed of two monoacylglycerol units symmetrically esterified to a central phosphate group (Fig. S1). Owing to their comparable net negative charge, and similar acyl-chain compositions, the use of POPS in this work allows for a direct assessment of the effects of key electrostatic interactions driving TST’s effects on lysosome-like lipid bilayers, without accounting for the different packing properties of atypical BMP-containing membranes (37).Figure 1TMEM106B(120–254) increases the mobility of MLV membranes as observed by static ^31^P NMR. A, schematic representation of TMEM106B sequence. The C-terminal moiety of TMEM106B has been found to form amyloid fibrils in aging brains and brains with neurodegenerative diseases. B, in this study, the constructs TMEM106B(120–254) is called TST, which stands for TMEM106B-Short-T185. C, static ^31^P NMR spectra, at 298 K, of freshly prepared MLVs without (gray) and with (black, blue, and red) TMEM106B(120–254) with three lipid compositions: (right) POPC only membrane; (middle) POPC:SM:chol membrane; (left) POPC:POPS:SM:chol. TMEM106B C-terminal moiety binding reduces the ^31^P CSA of the three membranes, as shown by the emergence of an isotropic peak as measured at 298 K as well as significant lineshape changes observed upon TMEM106B(120–254) binding. Spectral linewidths and parameters are shown in Table S1. MLV, multilamellar vesicle; CSA, chemical shift anisotropy; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin; TST, TMEM106B-Short-T185.

In order to assess the role and effects of POPS, we made intact MLVs in the presence or absence of the TST construct containing POPC:POPS:SM:chol, POPC:SM:chol, and POPC, at ratios of 55 : 10 : 25 : 10, 60 : 30 : 10, or 1, respectively. Overall, our findings reveal that TST associates peripherally with lipid bilayers, without inserting deeply into the hydrophobic core, and perturbs both headgroup and acyl chain dynamics in a lipid-dependent manner. Together, the data support a model in which anionic lipids regulate the balance between dynamic and immobilized states of TST at the bilayer surface or through protein-protein interactions, with additional effects for both membrane biophysics and the conformational properties of the protein.

Results

To mimic TMEM106B C-terminal’s interaction with lysosomal lipid membranes, we produced MLV membranes with three types of lipid compositions in the presence or absence of TMEM106B(120–254)-T185 (Figs. 1, A and B, S2). For clarity, the TMEM106B(120–254)-T185 construct will be called TST in this manuscript, which stands for TMEM106B-Short-T185. TMEM106B carries a common variant at position 185 (T185S) that separates the risk isoform (T185) from the protective isoform (S185). This residue lies next to the N-linked glycosylation site N183, and the S185 form is thought to be degraded faster than T185, resulting in lower TMEM106B levels at lysosomes (5). We used three lipid compositions to assess the roles and effects of the different lipids, namely, POPC:POPS:SM:chol, POPC:SM:chol, and POPC. After mixing TST and the lipid mixtures in aqueous solutions, the proteolipid mixtures were lyophilized and rehydrated in ^2^H-depleted water, then subjected to three freeze/thaw cycles. All NMR measurements were performed immediately after MLV formations.

We first investigated the impact of TST on the membrane integrity by measuring static ^31^P and ^2^H NMR spectra before the protein–lipid mixtures were subjected to high-g MAS (Figs. 1C and 2). These intact proteo-MLVs were immediately packed into 4 mm MAS rotors after preparation without any centrifugation cycles. We measured 1D ^31^P static NMR spectra at 283 K, 298 K, and 310 K (Fig. S3) of the three MLV compositions in the presence and absence of TST. In the absence of TST, the membranes exhibit a uniaxial ^31^P NMR powder pattern, typical of the lamellar organization of lipid bilayers (Figs. 1C, S3). The observed changes in the ^31^P chemical shift anisotropy (CSA) parameters upon protein incorporation into lipid membranes made of POPC:POPS:SM:chol, POPC:SM:chol, and POPC indicate TST induces perturbation of the lipid headgroup environment (Fig. 1C). To study these effects in detail, we used the Topspin tool Sola to generate automated simulations of the ^31^P spectra (Fig. S4) across all samples and temperatures. These CSA changes in the static ^31^P NMR spectra were observed across all membrane types (Fig. S4), especially when the temperatures were above phase transition values. In almost every case, the CSA (δ) indeed decreased (Table S1) in the presence of TST indicating enhanced motional averaging in all three membranes. Interestingly, the asymmetry parameter (η) only significantly increased in temperatures well above the phase transition temperatures (i.e., 298 K and 310 K) and in the POPC:POPS:SM:chol and POPC only MLVs, suggesting a departure from axial symmetry of the phosphate groups in POPS-containing and POPC only MLVs (Table S1). In addition, the appearance of an isotropic component in many of the conditions upon TST addition further suggests the generation of fast-tumbling lipid species or highly curved defects, pointing to a direct impact of TST on bilayer morphology (Fig. S4). These results highlight that membrane composition, and particularly the presence of anionic lipids, modulates the extent of TST-induced perturbation (Fig. 1C). Interestingly, in this experimental setup, at pH 4.5, TST has an approximate net charge of +5. Together, these spectral changes suggest that the protein perturbs the lipid headgroup region, likely inducing local curvature, defects, or dynamic reorganization of the membrane.Figure 2Carbon-deuterium order parameters SCD of ^2^H_31_-POPC MLV in the absence and presence of TMEM106B(120–254) at 298 K and 310K. A–C, experimental ^2^H echo spectra of three reconstructed hydrated MLV membranes made of POPC:POPS:SM:chol (A), POPC:SM:chol membrane (B), POPC only membrane (C) in the absence (gray) and presence (black, blue, and red, respectively) of TMEM106B (120–254) at a molar protein-to-lipid ratio of 1:30. D–F, plot of SCD order parameters at 298K as a function of the labeled carbon position, k, along the palmitoyl acyl chain (k = 16, chain end, k = 2, membrane interface). Error bars represent the propagated uncertainty in the order parameter, calculated from the signal-to-noise ratio of the quadrupolar coupling measurement. MLV, multilamellar vesicle; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin.

Static ^2^H NMR of MLVs containing deuterated POPC-d_31_ (^2^H labeled on the palmitoyl chain) was used to assess how TST affects the acyl chain organization in different membrane compositions. We measured 1D static ^2^H NMR echoes at different temperatures (Fig. 2, A–C,), which overall show typical quadrupolar splittings for lamellar phases of lipid bilayers of 20 to 25 kHz. Temperature-dependent ^2^H NMR spectra showed the expected narrowing of quadrupolar splittings with increasing temperature, reflecting enhanced lipid chain dynamics, while the presence of TST consistently reduced splittings across all temperatures, indicating that its disordering effect persists throughout the thermal range examined (Fig. S5). In the absence of TST, the spectra displayed the expected broad quadrupolar splittings characteristic of well-ordered lipid chains, with slightly narrower splittings at 310 K than at 298 K due to increased thermal motion (Fig. 2, A–C). Upon addition of TST, the quadrupolar splittings were consistently reduced, indicating decreased acyl chain order and enhanced motional averaging. This disordering effect was most evident in pure POPC MLVs and in the complex POPS-containing mixture (POPC:POPS:SM:chol), where significant narrowing of the spectra was observed at both temperatures (Fig. 2, A–C). In contrast, in membranes lacking POPS (POPC:SM:chol), the spectral changes upon TST addition were modest, suggesting a weaker perturbation. We simulated the ^2^H spectral fingerprints of the six samples at 298 K as it is the only temperature at which all spectra show quadrupolar splitting across the entire acyl chain (Fig. S6). Order parameter profiles derived from the simulations of quadrupolar splittings along the POPC-d_31_ palmitoyl acyl chain provided site-specific insights of these effects (Figs. 2, D–F, S6). In POPC MLVs and complex POPC:POPS:SM:chol bilayers, TST lowered order parameters across nearly the entire chain, with the largest decreases observed in the upper and midchain carbons, consistent with disruption of chain packing and increased fluidity (Fig. 2, D–F). In membranes lacking POPS, by contrast, TST induces a larger overall change in the order parameters at 298 K, but this effect is more superficial along the acyl chain, with the last five carbons of the lipid tails remaining essentially unaffected (Fig. 2E). Together, these data demonstrate that TST reduces acyl chain order and promotes chain disorder in a composition-dependent fashion, with the strongest effects in zwitterionic POPC and heterogeneous mixtures containing POPS, and the weakest impact in membranes lacking anionic lipids. In addition, upon addition of TST, the POPC and heterogeneous mixture containing POPS display an isotropic component in the ^2^H spectra, similarly to the ^31^P fingerprints (Figs. 1C and 2, A–C). When the sample was heated from 298 K to 310 K and then cooled back to 298 K, the ^2^H lineshape did not fully recover its initial appearance (Fig. S5), indicating a modest hysteresis on the ∼20 min equilibration timescale used at each temperature step. The spectrum returned to its original state after a longer incubation time at room temperature (data not shown).

Further analysis of spectral moments provided a quantitative measure of TST-induced perturbations in both acyl chain and headgroup dynamics. The first moment (M1) of the ^2^H spectra decreased with increasing temperature and was systematically reduced in the presence of TST, demonstrating a clear loss of chain order (Figs. 2, A–C, S7), specially in temperatures above the phase transitions. Similarly, the second moment (M2) of the ^31^P spectra decreased upon TST addition, consistent with a narrowing of the CSA distribution and enhanced headgroup motional averaging across all samples and temperatures (Figs. 1C, S7). These effects were interestingly most pronounced in POPC only bilayers (Fig. S7) as were already observed on the raw ^31^P and ^2^H spectra, in which the POPC membranes showed the most pronounced isotropic contributions overall (Figs. S3–S5). Interestingly, the POPS-containing and the POPS-free cholesterol-containing membranes showed intermediate effects upon TST addition: with the complex POPC:POPS:SM:chol mixture showing stronger effects on the overall ^2^H-reported membrane fluidity, and the POPS-free cholesterol-containing membranes showing stronger effects on the ^31^P-reported headgroup mobility (Fig. S7). These results confirm that the magnitude of TST-induced disorder strongly depends on bilayer composition. Assuming ideal mixing of the phospholipids and neglecting the direct contribution of cholesterol, the main transition temperatures of the individual components (POPC ≈ −2 °C, POPS ≈ 14 °C, brain SM ≈ 37 °C) predict effective mixture Tm values of ∼11 °C for both the POPC:POPS:SM:chol 55:10:25:10 and POPC:SM:chol 60:30:10 compositions, whereas a POPC-only membrane would remain predominantly fluid at physiological temperatures (Tm ≈ −2 °C). These values should be regarded as theoretical estimates, since cholesterol broadens and suppresses cooperative transitions, yielding gradual changes in bilayer order rather than a sharp phase boundary. Consistent with an additional softening of the bilayer, TMEM106B reduces the first and second spectral moments (M_1_ and M_2_) of the ^2^H and ^31^P NMR powder patterns across almost the entire temperature range examined, indicating enhanced motional averaging of the lipid acyl chains and polar heads, suggesting that TMEM106B further shifts the apparent phase transition toward lower temperatures (Fig. S7).

Analysis of the vesicle shape anisotropy parameter (c/a) from static ^2^H NMR spectral simulations further highlighted the impact of TST on bilayer organization (Fig. S6). In the absence of TST, the complex (POPC:POPS:SM:chol) and POPS-lacking (POPC:SM:chol) mixtures displayed elongated, prolate geometries (c/a = ∼2.2 and ∼1.5, respectively) as deformed in the strong magnetic field, whereas POPC vesicles were essentially spherical (c/a = ∼1). Upon addition of TST, the c/a values for both complex POPC:POPS:SM:chol membranes and ones lacking POPS decreased to ∼0.9, indicating a loss of magnetic-induced deformation and a shift toward near-spherical or slightly oblate morphologies. In contrast, POPC vesicles’ shapes remained spherical regardless of TST (Fig. S6). Interestingly, this shape anisotropy is also observed in the ^31^P experiments. The mismatched signal intensities between simulated and experimental spectra on the left side of the ^31^P static spectra indeed reflect the intrinsic asymmetry of the vesicles (Fig. S4). At temperatures well above the phase transition, TST appears to partially correct this asymmetry in both the POPS-containing and POPC-only vesicles (Fig. S4, A and C). In contrast, for the POPS-lacking vesicles, the simulated ^31^P spectra do not reproduce the well-defined asymmetry correction observed in the ^2^H spectra, suggesting that the 2 to 3 spectral components used in the ^31^P simulations do not fully capture the complexity of the underlying membrane dynamics of POPS-lacking MLVs as the shape anisotropy is not taken into account in the Sola suite (Fig. S4B). Overall, these results demonstrate that TST reduces vesicle shape anisotropy and corrects for vesicle field-induced deformation in heterogeneous bilayers, driving them toward more spherical and isotropic morphologies within the strong magnetic field. To investigate whether these changes in lipid dynamics were caused by the insertion of TST into lipid bilayers, we measured 2D ^1^H-^13^C heteronuclear correlation (HETCOR) spectra that correlate the ^1^H signals of water and lipid chains with the protein ^13^C signals. Even with a long ^1^H mixing time of 100 ms, the spectra displayed water-protein, or residual urea-protein correlations but no lipid-protein cross peaks (Fig. 3). All signals in the lipid ^1^H slice could be assigned to intramolecular lipid correlations. This indicates that TST remains at the membrane-water interface without penetrating the hydrophobic interior of the bilayer. The same result was obtained for all three membrane compositions, indicating that TST does not insert into the hydrophobic core regardless of bilayer lipid compositions under the current conditions (Fig. S8).Figure 32D ^1^H-^13^C HETCOR spectra of MLV-bound TMEM106B(120–254)-T185 monomers. A, spectra of MLV-bound TMEM106B (120–254)-T185 monomers measured with ^1^H spin diffusion mixing times of 100 ms using the POPC:POPS:SM:chol lipid composition. B, ^13^C cross sections at the lipid CH_2_, residual urea, and water ^1^H chemical shifts are shown. This spectrum shows no protein ^13^C cross peaks with the lipid acyl chain protons, indicating that the protein does not significantly insert into the hydrophobic interior of the membrane in the current conditions. The annotation “ajd. db.” stands for “adjacent to double bond”. MLV, multilamellar vesicle; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin; HETCOR, heteronuclear correlation.

To probe the conformational dynamics of TST in different bilayer compositions, we compared ^13^C spectra recorded with direct polarization (DP), cross polarization (CP), and insensitive nuclei enhancement by polarization transfer (INEPT). DP detects all carbons regardless of mobility and provides a baseline that scales with sample amount. CP selectively enhances signals from rigid, immobilized sites with strong ^1^H-^13^ C dipolar couplings, and thus reports on protein segments that are relatively static or tightly associated with the membrane. INEPT, in contrast, relies on scalar couplings and is efficient only for highly mobile, solution-like regions of the protein. As expected, DP spectra showed similar overall intensities across samples, reflecting comparable sample amounts (Fig. 4A). In the complex POPC:POPS:SM:chol mixture MLVs, ^13^C INEPT signals dominated while CP was relatively weak, indicating that TST remains largely dynamic in the presence of anionic lipids (Fig. 4, A–C, black spectra). By contrast, in membranes lacking POPS (POPC:SM:chol and POPC MLVs), CP signals were stronger and INEPT weaker, consistent with a greater fraction of TST adopting immobilized conformations at the bilayer surface (Fig. 4, A–C, blue and red spectra). To further quantify this effect, we integrated the aromatic ^13^C signals (mainly Tyr and Phe, ∼127–133 ppm) relative to the protein CO region (∼166–180 ppm) on the water slice of the 2D ^1^H-^13^C HETCOR spectra. The resulting aromatic/CO ratios are 0.15 for POPC MLVs, 0.17 for POPC:SM:chol MLVs, and 0.10 for POPC:POPS:SM:chol MLVs, indicating that POPS-containing bilayers maintain a smaller fraction of ordered aromatic residues compared to POPS-free membranes. Together, these results demonstrate that POPS promotes a more dynamic and flexible state of TST, whereas its absence favors immobilization of the protein.Figure 4**^13^C MAS spectra of hydrated MLV-bound TMEM106B(120–254) reveal TST is most dynamic in POPS-containing membranes and more immobilized in membranes lacking POPS**. 1D ^13^ C MAS spectra of TST bound to MLVs with three lipid compositions: (i) POPC:POPS:SM:chol; (ii) POPC:SM:chol membrane; (iii) POPC only membrane. A, direct polarization (DP) spectrum of membrane-bound TST detects the signals of both immobilized and dynamic species. B, cross-polarization (CP) spectrum of membrane-bound TST preferentially detects immobilized residues and lipid segments. C, INEPT spectrum of membrane-bound TST in which only highly mobile segments with long ^1^H and ^13^C T2 relaxation times are detected due to magnetization transfer by weak ^1^H-^13^ C scalar couplings. All spectra were measured at 15 to 18 °C. MAS, magic-angle spinning; MLV, multilamellar vesicle; INEPT, insensitive nuclei enhancement by polarization transfer; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin; TST, TMEM106B-Short-T185.

To further investigate residue-specific conformational changes of TST in different membrane environments, we recorded 2D ^13^ C-^13^C CORD MAS NMR spectra using 50 ms CORD mixing time. In the complex POPC:POPS:SM:chol mixture, only a limited set of correlations were observed, primarily from Ala, Thr/Ser, Ile, and Pro residues, consistent with a largely dynamic protein in which only a subset of sites become immobilized (Fig. 5A, Table S2). These few residues mostly display β- and random coil conformations (Fig. S9A). By contrast, in POPC:SM:chol and POPC bilayers, the spectra displayed a richer set of cross-peaks including Ala, Ile, Val, Leu, Lys, and Thr, indicating that a larger fraction of residues adopts immobilized conformations at the membrane surface (Fig. 5, B and C). The POPC:SM:chol MLV sample shows several Ala, Lys, Ile, Val, Thr, and Ser spin systems that could be type-assigned, allowing calculation of their secondary chemical shifts (Fig. S9B). TST adopts predominantly β-like conformations, while a few residues, such as one Ala, one Val, and one Thr, display α-like or random-coil conformations (Figs. S9B and S10). In addition, TST exhibits very similar structural properties in the presence of POPC MLVs and POPC:SM:chol MLVs (Figs. S9, B and C and S10), with several identified amino acids such as Ser, Thr, Ala, Ile, displaying almost identical chemical shifts consistent with β-rich secondary structure (Fig. S9, Table S2). Mapping these residue types onto the TST sequence highlights multiple Ala-, Ile-, and Thr-rich segments that become detectable in the rigid state. Notably the absence or low intensities of Pro signals in the dipolar spectra suggests the central segment of TST, around residues 184 to 201, is not immobilized in the presence of the lipid membranes, while the strong Ala signals points to the N-terminal moiety being highly rigid and ordered in the presence of neutral membranes (Fig. 5D). To assess the assembly state of TST in the presence of membranes under MAS conditions, we performed negative-stain TEM on TST-bound POPC:SM:chol MLVs taken from the same preparation used for MAS experiments (Fig. S11). After ∼1 week of MAS data acquisition and several additional weeks of storage at 4 °C, the samples still contained large MLVs and SUVs which were not present at first but no obvious extended amyloid-like fibrils (Fig. S11). These observations indicate that the immobilized, β-rich fraction detected by CP/INEPT balance and 2D CORD might predominantly correspond to membrane-associated TST assemblies or small oligomers rather than mature fibrils, although the presence of short fibrillar segments below the detection limit of TEM cannot be excluded. Together, these data indicate that POPS-containing bilayers keep TST in a more dynamic conformational state, whereas POPS-free membranes promote more extensive immobilization and structural ordering of the protein, likely through stronger membrane binding and/or self-aggregation/oligomerization (Figs. S10 and S11).Figure 52D ^13^ C-^13^C correlation spectra indicate MLV-bound TMEM106B(120–254) conformation is lipid composition dependent and adopts β-sheet rich conformation. A, 2D CC CORD spectrum of TMEM106B(120–254) bound to POPC:POPS:SM:chol MLVs, measured with CORD mixing times of 50 ms. In the negative charge-containing vesicles, almost no cross peaks are present. B, 2D CC CORD spectrum of TMEM106B(120–254) bound to POPC:SM:chol MLVs, measured with CORD mixing times of 50 ms. β-sheet, α-helical, or random coil Ala, Ser, Thr, Val, Ile, and Lys intensities are detected at high intensities. C, 2D CC CORD spectrum of TMEM106B(120–254) bound to POPC MLVs, measured with CORD mixing times of 50 ms. In the simple POPC only vesicles, β-sheet, α-helical or random coil Ala, Ser, Thr, Val, Ile, and Lys intensities are detected but their intensities are lower than neutral and cholesterol containing POPC:SM:chol MLVs. D, amino acid sequence of TMEM106B(120–254)-T185. Alanines and prolines are highlighted on the sequence. The strong intensities of Ala residues and the absence or low intensities of Pro signals suggest the N-terminal moiety of the constructs is more rigid than the C terminus in the neutral MLVs samples. MLV, multilamellar vesicle; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin.

To characterize the mobile regions of TST, we recorded 2D ^1^H-^13^C INEPT spectra, which selectively detect highly dynamic sites. In the complex POPC:POPS:SM:chol bilayers, numerous protein amino acid correlations as well as lipid-associated signals were observed (Fig. 6A), consistent with TST remaining largely dynamic at the membrane surface. Intensity ratio analysis between the neutral and anionic membranes revealed clear differences between the three bilayer systems. In POPS-free membranes, most protein signals were reduced compared to the complex mixture (Figs. 6, B, S12), indicating that the removal of POPS decreases protein mobility and promotes partial immobilization. In POPC membranes, the protein signals decreased drastically compared to the complex MLVs sample, while only the lipid resonances, including acyl chain CH_2_, terminal methyl (ω, ω2), and glycerol/PC headgroup carbons, increased substantially (Figs. 6C and S12). This shows that in POPC bilayers the lipid matrix itself is more mobile, whereas TST does not gain additional flexibility and in fact is less mobile than in the complex mixture, suggesting the absence of anionic charges promotes protein-protein interactions. Together, these data demonstrate that POPS-containing membranes maintain a balanced dynamic state of the protein.Figure 62D ^1^H-^13^C INEPT spectra of MLV-bound TMEM106B(120–254) reveal that the lipid composition of the vesicles modulates the protein mobility. A, 2D INEPT spectrum of TMEM106B(120–254) bound to POPC:POPS:SM:chol MLVs. Signal assignments of proteins are shown in blue, while that of lipids are shown in orange. The aromatic signals are shown in the excerpts at the bottom right. B, 2D INEPT signal intensities of TMEM106B(120–254) bound to POPC:SM:chol MLVs (I_noPOPS_) are compared to that of TMEM106B(120–254) bound to POPC:POPS:SM:chol MLVs (I_POPX_). C, 2D INEPT signal intensities of TMEM106B(120–254) bound to POPC MLVs (I_POPC_) are compared to that of TMEM106B(120–254) bound to POPC:POPS:SM:chol MLVs (I_POPX_). MLV, multilamellar vesicle; INEPT, insensitive nuclei enhancement by polarization transfer; POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine; POPS, 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine; SM, sphingomyelin

Discussion

A peripheral binding mode that reshapes bilayer morphology

The main finding of this study is that the luminal C-terminal fragment of TMEM106B (TST, residues 120–254) engages peripherally with lipid bilayers and remodels them in a composition-dependent manner. Across all compositions tested, static ^31^P spectra reported a decrease in the CSA span (δ) together with an increase in the asymmetry parameter (η) upon adding TST, and an ^31^P isotropic component emerged (Fig. 1C). These trends indicate enhanced headgroup motional averaging and a departure from axial symmetry consistent with curvature generation and/or fast-tumbling lipid subpopulations. Concordant ^2^H static NMR spectra of POPC-d_31_ revealed smaller quadrupolar splittings and lower order parameters of the lipid acyl chain, establishing a TST-induced reduction of overall order (Fig. 2). Overall, the magnitude of both headgroup and acyl-chain perturbations depends on membrane composition, being largest in POPC and the POPC:POPS:SM:chol mixture and smallest in POPS-free membranes, which show the least changes in CSA, asymmetry parameters (Table S1) and acyl-chain order across all studied temperatures (Figs. S3 and S4). Temperature increases narrowed all ^2^H spectra, as expected, but TST-dependent disorder persisted across the thermal range (Figs. 2, S4). Importantly, 2D ^1^H-^13^C HETCOR experiments failed to detect lipid-protein cross-peaks even at long spin diffusion mixing times, while water-protein correlations were clearly observed. These data provide strong evidence that TST does not insert into the hydrophobic bilayer interior but instead remains confined to the membrane–water interface (Fig. 3). Together these observations position TST as a surface-active remodeler that couples to the headgroup region and transmits disorder into the hydrocarbon core without deep insertion (Fig. 7).Figure 7Model of field-induced membrane deformation correction by TMEM106B and lipid-induced TMEM106B conformation changes.

Geometric analysis of static ^2^H spectra revealed that TST reduced the c/a shape anisotropy of vesicles, shifting them from oblate toward spherical morphologies in complex POPC:POPS:SM:chol and POPS-free mixtures, while leaving POPC vesicles largely unchanged and isotropically shaped even in the absence of TST. This suggests that TST association promotes the formation of more spherical lipid assemblies thus correcting for vesicles magnetic-induced deformation (Fig. S6). TST thus seems to be able to locally tune the lipid membrane’s elasticity and thus curvature/bending under magnetic stress. In MLVs composed of POPC, POPS, SM, and cholesterol, exposure to a strong magnetic field such as that of a 500 MHz NMR magnet induces a pronounced deformation, reflected by an aspect ratio (c/a) of approximately 2.2 (Fig. S6). This elongation arises because the diamagnetic anisotropy of the lipid bilayer generates a magnetic torque that competes with the membrane’s bending elasticity (38, 39). The elasticity, quantified by the bending modulus (κ), determines how easily the membrane can accommodate curvature: lower κ corresponds to a more flexible membrane that deforms readily under external forces, while higher κ implies increased stiffness and resistance to bending (40, 41). When TMEM106B is incorporated into the membrane, the vesicles become significantly less deformable, yielding a nearly spherical shape with c/a ≈ 0.9. This indicates that TMEM106B increases the effective bending modulus, thus stabilizing the bilayer against magnetic torque. In addition to external effects, the elasticity of lipid membranes is governed by several parameters: lipid tail saturation and length (long, saturated chains increase κ), headgroup size and charge (smaller or highly charged headgroups promote curvature and reduce κ), cholesterol content (which orders the lipids and raises κ), temperature (higher temperature lowers κ by increasing fluidity), and protein–lipid interactions (which can rigidify the bilayer by constraining macroscopic motion) (42). Altogether, these factors define the mechanical response of the membrane and its ability to bend or resist deformation under physical perturbations such as magnetic fields. Interestingly, the strong-field deformation reports the membrane’s bending elasticity at mesoscopic scales (hundreds of nm to μm), while by contrast, “fluidifying lipid motions” measured by ^31^P and ^2^H NMR refers to nanoscopic dynamics (ps-μs tail reorientation and ns-ms lateral diffusion) captured by order parameters and correlation times. TMEM106B therefore locally loosens chain packing while globally stiffening the membrane against large-scale bending.

Although the physiological role of the luminal C-terminal domain of TMEM106B remains unknown, our data suggest that it may modulate the mechanical properties of lysosomal membranes. Interestingly TMEM106B is part of the important aggregate clearance system of the cells: the autophagy-lysosome pathway, a modular system that recognizes misfolded material, sequesters it, transports it to degradative organelles and recycles the breakdown products (43). Aggregation-prone proteins first accumulate as diffuse misfolded species and small oligomers. Misfolded proteins form aggregates, which are sorted for selective aggrephagy. Aggrephagy is a selective type of autophagy involved in the lysosomal turnover of cytoplasmic aggregates and condensates (43). Ubiquitylation and phase separation help autophagy receptors gather these aggregates into an “aggresome” (44). The autophagosome then moves to lysosomes, where fusion creates an acidic environment for cathepsins to break down aggregates into reusable peptides and amino acids (45, 46, 47, 48). This process is crucial for normal protein turnover (49, 50, 51, 52). TMEM106B sits at the membrane of late endosomes and lysosomes and might act as a dosage-sensitive gatekeeper of this pipeline: its short cytosolic N terminus influences how lysosomes couple to microtubule motors and where they park within the cell; TMEM106B loss impairs axonal lysosome transport, whereas excess TMEM106B enlarges and immobilizes lysosomes (53, 54, 55, 56). In addition to microtubule-driven trafficking through its N-terminal cytosolic segment, TMEM106B likely stabilizes lipid bilayers and functions as a “fusion guide” or safeguard against premature remodeling through lysosomal fusion making the lipid membranes less elastic and less able to undergo rapid shape fluctuations. Many studies have indeed reported that increased TMEM106B levels cause enlarged, dysfunctional lysosomes and thus may cause a defect in the later stages of late endosome/lysosome fusion or lysosomal degradation (55, 57, 58). One possible interpretation is that C-terminal monomers or small assemblies transiently stiffen the luminal leaflet, thereby tuning membrane curvature, fusion, and resistance to mechanical stress, whereas excessive or irreversible aggregation during brain aging could abnormally rigidify the lysosomal membrane and impair its function. Although these hypotheses remain speculative, our results provide an initial experimental indication that the C-terminal domain of TMEM106B could act as a regulator of lysosomal membrane elasticity.

Lipid composition tunes TST conformational dynamics

Solid-state ^13^C experiments further revealed that lipid composition dictates the balance between rigid and dynamic states of TST (Fig. 4). 1D ^13^ C CP and INEPT spectra showed that in POPC:POPS:SM:chol membranes, INEPT signals were strong and CP relatively weak, indicating that TST remains largely dynamic in the presence of anionic lipids (Fig. 4A). By contrast, in POPC:SM:chol and POPC bilayers, CP signals were enhanced and INEPT signals diminished, consistent with immobilization of a larger fraction of residues at the bilayer surface or increased ordered protein-protein interactions (Fig. 4, B and C). Residue-specific analysis using 2D dipolar-based ^13^C-^13^ C CORD spectra corroborated this trend: in POPS-containing membranes only a limited set of correlations were observed suggesting TST is mostly mobile in the presence of negatively charged vesicles (Fig. 5A), whereas in POPS-free bilayers, numerous cross-peaks from Ala-, Ile-, Val-, Leu-, and Thr-rich regions became visible, and seemed to show random coil, α-helical or β-like secondary structure conformations (Fig. 5, B and C). We therefore infer that removal of anionic lipids broadens immobilization at the surface and allows partial β-structuring of segments, while POPS preserves flexibility. Lysosomal lipid compositions may thus drive TMEM106B amyloid aggregation in aging brains. To directly address the lipid-driven mobility/ordering equilibrium of TMEM106B, we measured 2D ^1^H-^13^C INEPT spectra which provided direct evidence for how protein and lipid mobility are differentially affected by bilayer composition (Fig. 6). In complex membranes, numerous mobile protein correlations were observed (Fig. 6A). In POPS-free membranes, mobile protein signal intensities decreased relative to the complex mixture, consistent with reduced mobility (Fig. 6B), hinting that anionic membranes directly affect TMEM106B’s C-terminal moiety flexibility and conformation. In POPC membranes, protein mobility was again reduced, but lipid resonances, particularly acyl chain and glycerol/PC headgroup signals, were strongly increased, reflecting enhanced lipid flexibility in agreement with increased isotropic signals in both ^31^P and ^2^H static NMR spectra. Thus, anionic lipids appear to stabilize a balanced dynamic state of the protein, while their absence either reduces protein mobility in POPS-free MLVs or shifts the effect primarily toward drastic lipid disordering in POPC only MLVs (Figs. 1C, 2, A–C).

Relationship to membrane-directed amyloid mechanisms

The lipid-programmed behavior we observe for TMEM106B echoes well-established paradigms in which amyloidogenic proteins use membrane interfaces to concentrate, orient, and structurally bias aggregation-prone segments. Amyloid-β (Aβ) interacts strongly with neuronal membranes, a property recognized since early studies demonstrated that synthetic Aβ peptides disrupt lipid bilayers and cause dye leakage from vesicles (59, 60). For example, glycosphingolipid GM1 clusters bind Aβ, stabilize aggregation-competent conformers, and accelerate elongation, yielding the long-described GM1-bound Aβ “seed” (GAβ) (61, 62, 63). In addition, membrane cholesterol further tunes these pathways, promoting or inhibiting Aβ binding and assembly depending on the bilayer context and the stage of the reaction (64, 65, 66). Extensive structural and biophysical work support pore or channel formation by Aβ oligomers and β-barrel-like species, offering a direct path to Ca^2+^ dysregulation (59, 67, 68). These interactions are lipid-dependent, with Aβ showing preferential binding to negatively charged phospholipids, gangliosides, and cholesterol-enriched lipid rafts, highlighting the importance of membrane composition in modulating amyloid toxicity (69). These Aβ precedents reinforce two principles relevant to TMEM106B: (i) specific lipids (GM1, cholesterol, and anionic headgroups) gate interfacial binding and structural transitions; and (ii) substantial membrane remodeling can arise without classical transmembrane helices.

α-Synuclein is perhaps the most studied amyloid protein in terms of membrane interaction. Physiologically, it localizes to presynaptic terminals and binds synaptic vesicle membranes (70). α-Synuclein exemplifies sensitivity to charge and curvature. It is natively disordered in solution but becomes α-helical upon binding anionic, high-curvature vesicles (71). Crucially, lipid chemistry controls whether binding remains helical and dynamic (72) or progresses to β-aggregation (73). Unsaturated, anionic phospholipids can indeed drive nucleation (74). At later stages, lipids can coassemble with α-syn fibrils and be retained within the fibril structure (so-called “lipidic fibrils”), providing direct molecular evidence for persistent protein–lipid contacts in the aggregated state (75, 76). These observations align with our study in two ways. First, anionic lipids (POPS) bias α-syn toward interfacial, dynamic states whereas their absence favors immobilization/ordering at the surface. Second, cholesterol act as higher-order regulators in both systems, shaping whether protein complexes remain mobile membrane remodelers or convert to immobilized β-enriched assemblies (71, 74, 77).

Finally, within the broader amyloid-membrane literature, tau provides a closely related precedent. Tau, traditionally studied as a microtubule-associated protein, has also been shown to interact with cellular membranes. Immunolocalization studies detected tau at the plasma membrane and endosomal compartments (78, 79), while in vitro assays confirmed its strong affinity for lipid vesicles containing negatively charged phospholipids (80). Tau binds membranes primarily through its highly basic microtubule-binding domain, and membrane association induces conformational changes that promote partial β-sheet formation and facilitate aggregation (81, 82). Similarly to TMEM106B, biophysical techniques such as NMR, electron paramagnetic resonance, and surface plasmon resonance have revealed that tau binding can perturb lipid packing, destabilize bilayer integrity, and accelerate nucleation of fibrillar structures (83). In addition, cholesterol-containing high-curvature vesicles can convert soluble tau into interface-immersed amyloid fibrils, whereas lower curvature or cholesterol depletion abolishes fibrillization (30). Thus, membranes act not only as a binding surface but also as cofactors in tau aggregation. Like TMEM106B, tau monomers cannot insert into lipid bilayers (30). Interestingly, some lipid-induced tau amyloid fibrils can penetrate cholesterol-containing and high curvature lipids bilayers as seen by solid-state NMR (29). Whether lipids can nucleate TMEM106B amyloid formation remain to be determined as little to no biochemical data of TMEM106B aggregation is available.

In conclusion, our results suggest the luminal C-terminal fragment of TMEM106B as a surface-active remodeler of lysosome-like membranes. This membrane-coupled activity offers a plausible biophysical pathway through which TMEM106B can alter lysosomal integrity and potentially influence amyloidogenesis. Moving forward, systematic analysis of lipid composition and ionic conditions on TMEM106B amyloid assembly will be essential to define the mechanisms that nucleate pathological TMEM106B aggregates in the human brain.

Experimental procedures

Cloning, expression, and purification of TMEM106B(120–254)-T185 (TST)

The gene encoding TMEM106B(120–254)-T185 contains an N-terminal His_6_ tag, a tobacco etch virus (TEV) cleavage site (6 residues), and residues 120 to 254 of TMEM106B. This gene was cloned into a pET-24a vector and transfected into Escherichia coli BL21(DE3) competent cells (New England Biolabs) or T7 Express Competent E. coli (High Efficiency-New England Biolabs). A starter culture was grown in 50 ml LB medium containing 30 μg/ml kanamycin. After overnight growth at 37 °C with 220 rpm shaking, the 50 ml culture was used to inoculate 1 L of LB medium containing 30 μg/ml kanamycin. Cells were grown at 37 °C and 220 rpm until A600 reached 0.8 to 1.0, then spun down at 1000g and 4 °C for 30 min. The resulting cell pellet was suspended in 1 L minimal media containing M9 salts, 1 g/L ^15^NH_4_Cl, 2 g/L ^13^C-labeled glucose, 1 mM MgSO_4_, 0.1 mM CaCl_2_, 30 μg/ml kanamycin, vitamin, and mineral supplements. Cells were grown in this minimal media at 37 °C under 230 rpm shaking for 2 to 3 h, then protein expression was induced with 1 mM IPTG. Another 1 g/L of ^13^C-labeled glucose was added at this point to the medium to reach 3 g/L ^13^C-labeled glucose. Expression proceeded for 16 to 18 h at 18 °C under 230 rpm shaking.

Bacterial cells were harvested by centrifugation at 6000g for 10 min at 4 °C. The resulting pellet was resuspended in 40 ml of cold lysis buffer A (2 × PBS, 1 × protease inhibitor cocktail, and 5 mM DTT) per liter of culture. Cells were lysed by sonication at 40% amplitude using a cycle of 5 s on and 5 s off for a total of 15 min at 4 °C. The lysate was clarified by centrifugation at 18,000g for 60 min at 4 °C (S0). The resulting pellet, containing inclusion bodies, was resuspended in buffer A supplemented with 2 M urea and 1.5% Triton X-100 and incubated for 30 min at 40 °C. This wash step was followed by centrifugation at 18,000g for 60 min at 4 °C (S1). The resuspension and centrifugation steps were repeated two additional times under identical conditions (S2 and S3).

For solubilization, the final pellet P3 was resuspended in buffer A containing 8 M urea. The suspension was incubated overnight at 40 °C with shaking at 100 rpm. After solubilization, the sample was centrifuged at 18,000g for 60 min to remove insoluble material. The supernatant containing solubilized protein was stored at 4 °C and further used for sample preparation (S′) (Fig. S2).

Preparation of lipid vesicles (MLVs)

Three membrane compositions are used in this work. The first one is a cholesterol (chol)-containing membrane which contains POPC, POPS, brain-extracted SM, and cholesterol at a molar ratio of 55 : 10: 25 : 10 (hereafter mentioned as POPC:POPS:SM:chol MLVs or MLVs). To assess the role and effects of the negatively charged POPS, we also made a similar membrane as the previous one without POPS. It contains POPC, brain-extracted SM, and cholesterol at a molar ratio of 60 : 30: 10 (hereafter mentioned as POPC:SM:chol MLVs or noPOPS MLVs). The last membrane contains only POPC (hereafter mentioned as POPC MLVs). All samples contain 5 mg of deuterated POPC: 16:0-d_31_-18:1 PC. All lipids (Avanti Polar Lipids) were codissolved in a 1:1 chloroform: methanol solution (v/v), then dried with a stream of compressed air gas and lyophilized overnight to a homogenous and dry film.

All membranes were mixed with TMEM106B(120–254) at a protein/lipid molar ratio (P/L) of 1:30, which corresponds to ∼10 mg of ^13^ C/^15^N-labeled protein with ∼14 mg lipids. The current study involved the preparation of intact proteo-MLVs. Overall, we used freeze-thawing and coincubation with the protein to prepare MLVs.

To produce the intact MLV samples, we suspended each air-dried lipid mixture in the 8 M urea (8M urea, 30 mM Tris, pH 7.5) solution containing TMEM106B(120–254) at a concentration of ∼5 mg/ml (3 × 3 ml). In the control MLV only samples, the resuspension buffer did not contain protein monomers but contained only buffer with 8 M urea (8M urea, 30 mM Tris, pH 7.5). The 3 ml lipid-containing mixtures were dialyzed against a pH 4.5 buffer (1 mM Mes, 1 mM NaCl, and 0.033 mM DTT) for 4 h. The 3 ml samples (6 samples) were then lyophilized overnight and then resuspended in 100 μl of deuterium-depleted water thus reaching 30 mM Mes, 30 mM NaCl, and 1 mM DTT buffer concentrations (pH 4.5) in the final sample (accounting for the 30-fold concentration when going from 3 ml to 100 μl through lyophilization). The 100 μl mixtures were vortexed and sonicated for 10 s in an ultrasonic bath (VWR USC100T) at room temperature (no temperature control) to briefly help swiftly suspend the dry lipids in solution. The mixtures were left at room temperature for 1 h with occasional agitation for full hydration of the dry film. The suspension was next subjected to three freeze-thaw cycles between liquid nitrogen and a 38 °C water bath to allow the lipid to reach a temperature above the hypothetical phase transition temperature (Tm) of ∼11 °C for the complex membranes and ∼−2 °C for the POPC only MLVs. Only three freeze-thaw cycles were used to make large MLVs which are better for static ^2^H NMR detection as their signal is not hindered by vesicle tumbling in the highly hydrated suspensions. These MLV samples were indeed not subjected to centrifugation and the full 100 μl solutions were packed into 4 mm Bruker solid-state NMR rotors for immediate ^2^H static NMR analysis. No centrifugation was used so no protein monomers were lost in the packing process. The ^31^P static NMR spectra were also acquired without any centrifugation and immediately after the ^2^H static NMR analysis. Once these acquisitions were finished on intact proteo-MLVs, 4 mm inserts were added in the rotors to avoid leakage, and the samples were spun up to 10.5 kHz for ^13^C detection under MAS.

Solid-state NMR experiments

Static ^2^H solid-state NMR experiments were conducted on an AVANCE III 500 MHz (11.7 T, respectively) spectrometers in the European Institute of Chemistry and Biology (IECB) using a 4 mm CPMAS (^1^H, X:^15^N-^31^P) probe operating in the ^1^H/^2^H double-resonance mode. ^2^H NMR experiments were performed at 77 MHz (^2^H). A phase-cycled quadrupolar echo sequence (90°x-τ-90°y-τ-acq) was used to perform the ^2^H NMR experiments. Acquisition parameters were as follows: spectral window of 500 kHz, π/2 pulse width of 3.25 μs, interpulse delays (τ) of 50 μs, recycle delays of 2 s for ^2^H, and 1024 scans. Samples were equilibrated for 20 min at a given temperature before data acquisition. The temperature was regulated to ± 1 °C (Table S3).

All spectra were processed and analyzed using Bruker Topspin 4.1 software. For the ^2^H NMR spectra, Lorentzian line broadening of 500 Hz was applied before Fourier transformation, starting from the top of the echo. Orientational carbon-deuterium order parameters (S_CD_) were calculated from experimental quadrupolar splittings (Δν_Q_) as described by Seelig in 1977 (84).

Wide-line ^2^H solid-state NMR spectra were simulated in the time domain as free induction decays and Fourier transformed using a FORTRAN code developed internally. Each perdeuterated lipid chain position was described by its experimentally determined quadrupolar splitting (ΔνQ), intrinsic linewidth, and the number of deuterons at the corresponding acyl chain site. Liposome deformation, leading to nonspherical distributions of bilayer normals relative to the magnetic field, was incorporated through the c/a parameter (85). Simulated spectra were iteratively fitted to experimental data until satisfactory agreement was achieved, yielding individual quadrupolar splittings and the corresponding SCD order parameters (86) The uncertainty in the quadrupolar coupling was estimated as σQ = Q/SNR, and propagated to the order parameter using linear error propagation.

Static ^31^P spectra of lipid membranes were measured on a Bruker AVANCE III HD console with a 400 MHz NMR spectrometer using SmartProbe (^15^N-^31^P, ^19^F, ^1^H, ^2^H), 5 mm z-gradients probe. To record static ^31^P ssNMR powder spectra, we applied a Hahn spin echo sequence, 90°-τ-180°-τ-acq at the ^31^P frequency of 162 MHz on a 400 MHz (9.4 T) Bruker Avance III HD spectrometer, with a 90° pulse of 8 μs, an echo delay of 40 μs, a recycle delay of 5 s, a spectral window of 64 kHz, and a number of scans of 320. All spectra were processed and analyzed using Bruker Topspin 4.1 software. Spectra were processed using a Lorentzian line broadening of 200 Hz for ^31^P NMR spectra in the shown plotted experimental spectra. The TopSpin 4.5 tool Sola was used to analyze and simulate the ^31^P spectra (Table S1). For the simulations, the number of CSA components is chosen by the user, while the tool varies all CSA parameters (i.e., intensity, δ(iso), δ(CSA), η(CSA), and LB) to increase the overlap percentage between the experimental and the simulated spectra (area under the peak). Usually, a single CSA component is used with or without an isotropic contribution according to the experimental spectra. In cases where the overlap could not reach at least 85%, a third CSA component was added to obtain a better overlap (Fig. S4, A–C, show in orange). As a practical criterion to evaluate whether changes in the ^31^P CSA asymmetry were meaningful upon addition of TST, we used the linewidth applied during Sola lineshape fitting as an experimental resolution limit. For each spectrum, the Lorentzian line broadening parameter (LB) used in Sola was taken as the full width at half maximum (FWHM, in Hz) of the simulated spectrum and converted to ppm using the ^31^P Larmor frequency (ν0 = 161.9 MHz on a 400 MHz ^1^H instrument). To express the CSA asymmetry in ppm, we converted the fitted Haeberlen parameters (δ, η) into principal components and calculated Δasym = δ22 − δ33 = ηδ, which reports the separation between the two shielded principal components (and is independent of δiso). For each lipid composition and temperature, we then computed ΔΔasym = Δasym^(+TST)^ − Δasym^(−TST)^. A change in asymmetry upon TST addition was considered significant when |ΔΔasym| exceeded the corresponding spectral FWHM (in ppm), i.e., when the inferred change was larger than the linewidth-limited resolution of the fitted spectrum (Table S1).

MAS solid-state NMR experiments were conducted on a Bruker NEO 600 MHz (14.1 T, respectively) spectrometers using a Bruker 4 mm CPMAS (^1^H, ^13^C, ^15^N - E-free) probe. ^1^H chemical shift was internally calibrated using 3-(Trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS) (0 ppm) or to the POPC Hγ chemical shift at 3.264 ppm on the DSS scale. ^13^C chemical shifts were referenced externally to the adamantane CH_2_ chemical shift at 38.48 ppm on the tetramethylsilane (TMS) scale. After the initial calibration, the ^13^C chemical shift was calibrated indirectly through ^1^H by setting the ^13^C the spectral reference value (sr) in the TopSpin software to be equal to sr(^1^H)/4 to 153 Hz. This gave the TMS-referenced ^13^C chemical shift. Typical radiofrequency (rf) field strengths were 50 to 83 kHz for ^1^H and 50 to 62.5 kHz for ^13^C. All reported temperatures are sample temperatures estimated based on the measured water ^1^H chemical shift (33). Samples were spun at 10.5 kHz.

We detected immobilized residues using dipolar-coupling based polarization transfer NMR experiments (Table S3). 2D ^13^ C-^13^C correlation spectra were measured using ^1^H-^13^ C cross polarization (CP) followed by CORD spin diffusion (34). 2D ^13^ C-^13^C dipolar correlation spectra were measured using Combined-Driven (CORD) spin diffusion (87) for ^13^C mixing. The 2D CC spectra were measured under 10.5 kHz MAS. Typical rf field strengths were 70 to 90 kHz for ^1^H and 30 to 50 kHz for ^13^C.

2D ^1^H-^13^C HETCOR experiments (37, 38) were conducted at 295 K under 10.5 kHz MAS. The ^1^H chemical shift evolution period was preceded by a ^1^H T_2_ filter and a mixing time to select the mobile water and lipid ^1^H magnetization and transfer it to the protein protons. The ^1^H T_2_ filter length was 2 × 2.18 ms and the ^1^H mixing time was 100 ms. A pair of ^1^H and ^13^C 180˚ pulses were applied in the middle of the T_2_ filter to refocus the ^1^H chemical shift and recouple the ^1^H-^13^ C J-coupling. In this way, the ^1^H magnetization of ^13^C-labeled protein is suppressed using both ^13^C-^1^H J coupling and the short protein T_2_. After ^1^H chemical shift evolution, the ^1^H magnetization was transferred by CP to ^13^C for detection. No ^13^ C intensities were observed when the ^1^H mixing time was 0 ms, confirming that the protein ^1^H magnetization was suppressed by the T_2_ filter. With 100 ms mixing, the ^13^C spectral envelope is the same as the ^13^C CP spectrum, indicating that the water, lipid and protein magnetization is fully equilibrated.

1D and 2D INEPT ^1^H-^13^ C spectra were measured to detect the signals of dynamic residues. ^1^H-^13^ C CP spectra were measured using CP contact times of 500 μs or 1 ms to preferentially detect rigid and semi-rigid residues. For all 2D spectra, multiple blocks were recorded and added in the time domain before Fourier transformation. Spectra were processed using GM apodization in the Topspin software, and chemical shift assignment was conducted using CCPNMR (88). Typical 2D spectral processing used GM apodization with LB = −30 Hz and GB = 0.03 for HETCOR spectra, and LB = −20 Hz and GB = 0.05 for CORD and INEPT spectra. Typical 1D spectral processing used GM apodization with LB = −20 Hz and GB = 0.05 for ^13^C spectra or LB apodization using LB = 200 to 500 Hz for ^2^H and ^31^P spectra.

Data availability

All data needed to evaluate the conclusions in the article are present in the paper and/or the Supplementary Materials.

Supporting information

This article contains supporting information.

Conflict of interest

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