Molecular substructure of the liquid-ordered phase formed by sphingomyelin and cholesterol: sphingomyelin clusters forming nano-subdomains are a characteristic feature

Abstract

As a model of lipid rafts, the liquid-ordered (Lo) phase formed by sphingomyelin (SM) and cholesterol (Cho) in bilayer membranes has long attracted the attention of biophysics researchers. New approaches and methodologies have led to a better understanding of the molecular basis of the Lo domain structure. This review summarizes studies on model membrane systems consisting of SM/unsaturated phospholipid/Cho implying that the Lo phase contains SM-based nanodomains (or nano-subdomains). Some of the Lo phase properties may be attributed to these nanodomains. Several studies suggest that the nanodomains contain clustered SM molecules packed densely to form gel-phase-like subdomains of single-digit nanometer size at physiological temperatures. Cho and unsaturated lipids located in the Lo phase are likely to be concentrated at the boundaries between the subdomains. These subdomains are not readily detected in the Lo phase formed by saturated phosphatidylcholine (PC) molecules, suggesting that they are strongly stabilized by homophilic interactions specific to SM, e.g., between SM amide groups. This model for the Lo phase is supported by experiments using dihydro-SM, which is thought to have stronger homophilic interactions than SM, as well as by studies using the enantiomer of SM having opposite stereochemistry to SM at the 2 and 3 positions and by some molecular dynamics (MD) simulations of lipid bilayers containing Lo-lipids. Nanosized gel subdomains seem to play an important role in controlling membrane organization and function in biological membranes.

In this review, to describe the physical heterogeneity of model lipid bilayers, we use the conventional terms Ld phase, Lo phase, and gel phase to refer to large-scale domains in which lipids are disordered and lateral diffusion is fast (Ld phase), lipids are ordered and diffusion is fast (Lo phase), or lipids are ordered and their diffusion is much slower than that in Ld (gel phase). When these domains are below the resolution of ordinary light microscopy, we call them nanodomains. Subdomains (or, when small, nano-subdomains) refer to distinct regions within Lo domains or gel domains. Within Lo domains, as discussed in this paper, gel-state subdomains are formed by sphingomyelin (SM) or dhSM, together with interstitial subdomains rich in cholesterol and unsaturated lipids.

As recently confirmed by Cebecauer et al. (2018), small membrane areas (nanodomains) can segregate from the rest of a membrane for various reasons, including raft formation. Rafts in cell membranes generally exist as nanodomains (Fig. 1) and have been identified with liquid-ordered (Lo) phase formed by SM and cholesterol (Cho). Rafts have attracted significant attention in biophysics, biochemistry, and cell biology (Ahmed et al. 1997; Brown and Rose 1992; Lingwood and Simons 2010; Schroeder et al. 1994; Simons and Ikonen 1997).

Fig. 1.

(A) Schematic diagram of the raft structure in a biomembrane. (B) Structure of raft-forming lipids. In biological membranes, strong interactions between raft lipids, such as SM and Cho, induce Lo-like domains. This review also focuses on dihydro-analogue (dhSM). (C) Two types of lipid bilayer models discussed in this review: homogeneous Lo domains (left) and heterogeneous Lo domains composed of relatively stable nano-subdomains (right)

Since tight packing interactions between ordered hydrocarbon chains and Cho are considered imperative in the formation of Lo domains, saturated phospholipids, such as SM and phosphatidylcholine (PC), have been used to investigate the molecular mechanism of Lo phase formation (Ahmed et al. 1997; Grönberg et al. 1991). This tight packing is further promoted by the umbrella effect (Huang and Feigenson 1999), in which non-polar Cho depends on headgroup coverage by nearby polar phospholipids to avoid the unfavorable free energy of Cho when it is in contact with water. In the presence of Cho, SM forms Lo domains more effectively than saturated PC, and its domains have higher thermal stability (Ramstedt and Slotte 2006; Slotte 2016; Veiga et al. 2001; Yasuda et al. 2014); therefore, intermolecular interactions between SM and Cho, or between SM and SM in the presence of Cho, could be a unique structural feature of SM (Grönberg et al. 1991; Slotte 2016; van Blitterswijk et al. 1987). Indeed, blocking the amide moiety and hydroxy group of SM with a methyl group inhibits the formation of Lo domains (Björkbom et al. 2011). Also, threo-SM, a diastereomer of natural SM, shows weaker domain-forming ability than SM, probably due to a difference in the configuration of the C3 position of the sphingosine skeleton (Bruzik and Tsai 1987; Kinoshita et al. 2013; Ramstedt and Slotte 1999). These properties show that not only the packing interactions of saturated chains but also polar interactions via the 2-amide and 3-hydroxyl groups of the sphingosine moiety significantly contribute to the formation of Lo domains. Hence, the structure of SM is suitable for forming a network of interlipid interactions between neighboring SM molecules and Cho. However, few experimental studies have examined the molecular organization of Lo domains due to difficulties in assessing the intermolecular interactions of SM, Cho, and other lipids (Hanashima et al. 2019a, b; Matsumori et al. 2012). Consequently, although these interactions have been investigated using various experimental methods, most studies have focused on SM–Cho interactions rather than on other interactions in Lo domains (Arsov and Quaroni 2008; Bartels et al. 2008; Shirota et al. 2016).

To gain a more detailed understanding of the molecular organization of Lo domains, we investigated the molecular behavior of SM in membranes using solid-state NMR and fluorescence measurements combined with chemical syntheses of fluorescence-labeled and isotope-labeled SM and Cho. These methods are uniquely suitable for studying the behavior of lipid molecules and proteins at a submicrometer resolution. Fluorescence modalities are excellent for visualizing domains and lipid localization under a microscope in the presence of a minute amount of fluorescent probe. Additionally, specialized methods, such as Förster resonance energy transfer (FRET) and fluorescence correlation spectroscopy (FCS), enable the physical properties of lipid domains to be accurately evaluated with nanometer and nanosecond spatiotemporal resolution. This is a significant advantage over NMR methods, as will be discussed. However, in NMR experiments, isotope-labeled lipids usually retain their original properties, and lipid molecules can be feasibly observed without perturbing the dynamic behavior, three-dimensional structure, or intermolecular interaction of individual lipids. Therefore, solid-state NMR has become a common method for investigating the segmental motions of the alkyl chains of fatty acyl groups in membrane lipids; for example, ²H quadrupolar coupling makes it possible to precisely evaluate the local mobility of a labeled segment of the alkyl chain (Leftin and Brown 2011; Matsumori et al. 2012; Molugu et al. 2017; Seelig 1977; Umegawa et al. 2018). Thus, we briefly review the applications of solid-state NMR spectroscopy to hydrated lipid membranes, particularly for site-specific ²H-labeled SM and Cho, to provide insights into Lo domain formation. Regarding NMR studies on the headgroup conformation of membrane phospholipids, Akutsu (2020) recently published a multifaceted review paper. It should be noted that solid-state NMR is sometimes unsuitable for investigating phase transitions between the gel and liquid crystal phases of lipid bilayers. In particular, ²H NMR does not provide clear signals, such as Pake doublets, for lipids residing in the gel phase. Therefore, it is necessary to complement NMR with other analytical methods. Thermal methods, such as differential scanning calorimetry (DSC), can be used to examine the state of phase separation in the gel phase of mixed-lipid bilayers. Also, the phase transition, phase separation, and order of bilayer membranes can be evaluated using fluorescence lifetime and fluorescence anisotropy, as described in a previous review paper (Kinoshita et al. 2018).

In this paper, we mainly review our own and other studies on the atomistic interactions of SM molecules in model lipid bilayers, as summarized in Table 1. Furthermore, we explain in detail how SM molecules form nanosized homogeneous clusters within the Lo phase, which we call nano-subdomains (Fig. 1C). We found that these subdomains are imperative for determining the physical characteristics of the Lo phase, which consists of SM and Cho. Finally, we briefly discuss the possible biological functions of gel-like nano-subdomains.

Table 1.

Recent findings regarding the nano-subdomain formation of sphingomyelin (SM) in the Lo phase

➢ Conformation and orientation of the SM amide group in intermolecular hydrogen-bonding interactions.a−c ➢ In SM-cholesterol (Cho) binary membranes, SM preferentially forms nanosized gel domains, comprising SM nano-subdomains or SM clusters.d ➢ Dihydro-SM, which has even stronger homophilic interactions than SM, possibly forms nano-subdomains without Cho.e–h ➢ SM and its enantiomer separately form homophilic nano-subdomains in the Lo phase.i−k ➢ Molecular dynamic (MD) simulations largely support the presence of intermolecular hydrogen bonding between the amide groups of SM, which is partly responsible for the formation of stable nano-subdomains.l−n ➢ In asymmetric bilayers, SM may form similar nano-subdomains, although experimental verification is needed

References a–n: ^a(Yamaguchi et al. 2012), ^b(Matsumori et al. 2015), ^c(Smith and Klimov 2018), ^d(Yasuda et al. 2015), ^e(Kinoshita et al. 2013), ^f(Kinoshita et al. 2014), ^g(Yasuda et al. 2018), ^h(Kinoshita et al. 2020a), ⁱ(Yano et al. 2018), ^j(Yano et al. 2020), ^k(Hanashima et al. 2020), ^l(Bera and Klauda 2017), ^m(Smith and Klimov 2018), ⁿ(Sodt et al. 2015)

Conformation and orientation of the SM amide group is suitable for intermolecular hydrogen-bonding interactions

SM has several polar functional groups that are potentially involved in intermolecular interactions and possibly responsible for Lo phase formation. Therefore, elucidating their spatial arrangement is key to understanding the intermolecular interactions of SM. The conformation of the SM headgroup in bilayers has been investigated by our group and others (Bittman et al. 2002; Bruzik 1988; Hanashima et al. 2019b; Matsumori et al. 2015; Sankaram and Thompson 2002; Zidar et al. 2009). The conformation around the SM amide group was examined using solution NMR with small bicelles, which reproduced an environment mimicking natural membranes to a certain extent (Liebau et al. 2017). Nuclear Overhauser effect (NOE) and ¹H-¹H spin coupling (³J_H,H), measured using solution NMR techniques, such as NOESY and E.COSY, were employed for conformational analysis (Yamaguchi et al. 2012) (Fig. 2A). Solid-state NMR was used to determine the orientation angles of the SM amide plane to the membrane normal (Matsumori et al. 2015). To determine the orientation angles, three parameters had to be determined: azimuth angle (α angle), orientation angle (β angle), and an order parameter, implying that three kinds of NMR measurements are necessary to obtain at least three experimental values. The values of these parameters were determined by measuring the chemical shift anisotropy of ¹³C and ¹⁵ N nuclei in the amide group and their dipole coupling constants. The amide portion of SM appeared to have a β angle of 31^o (Fig. 2B), which is suitable for intermolecular hydrogen bonding. The importance of this intermolecular hydrogen bonding for SM–SM interactions was confirmed by studies of N-methyl SM in a collaborative study conducted by Slotte and Katsumura’s groups (Björkbom et al. 2011). The N-methyl group interrupted the hydrogen bonding and, as noted above, weakened the formation of the Lo state. Although the difference in the molecular orientation of the SM amide group between the Ld phase without Cho and the Lo phase with Cho was insignificant, the study showed that Cho significantly reduced the wobbling of the acyl chains (Matsumori et al. 2012) and amide group (Matsumori et al. 2015).

Fig. 2.

(A) Conformation of the amide part of SM in bicelles deduced from ¹H NMR data (Yamaguchi et al. 2012). (B) Experimental rotation-axis direction with respect to the amide plane in SM/Cho bilayers and the partial orientations of an SM molecule. As shown in (C), the polar angle β between the amide plane normal and the bilayer normal (the orange rod) was determined to be 31°, which is suitable for a multi-molecular H-bonding network, as indicated by the dotted green lines (Matsumori et al. 2015). (C) An illustration of angle β, denoting the angle between the rotation axis direction and the normal direction of the amide plane

More support for important SM–SM interactions has come from MD studies. As described in detail below, several MD simulations have been used to deduce the lipid–lipid interactions responsible for Lo domain formation (Smith and Klimov 2018; Sodt et al. 2015; Wang and Klauda 2017). Many of these studies revealed that hydrogen bonds are formed between SM and SM and SM and Cho, and a recent study suggested SM–SM hydrogen bonding to be more important than SM–Cho bonding; Smith and Klimov (2018) reported that SM has the higher probability for cluster formation, which probably results in a network of SM-SM interactions, as described in “Impact of lipid asymmetry in bilayers on domain and nano-subdomain formation” section. The fact that this was not found previously may be attributed to recent improvements in force fields, as described below.

Our recent results on depth-dependent chain melting of stearoylsphingomyelin (SSM) further support the intermolecular H-bonding (Tsuchikawa et al. 2022). As shown in Fig. 3, two deuterium atoms were introduced at the upper (C4'), middle (C10'), and lower (C16') positions of SSM, and the same position of the stearoyl group of PSPC. Unitary bilayers consisting of each labeled lipid were subjected to solid-state NMR measurements with precise control of temperature to determine the depth-dependent melting temperature of the acyl chain. The results showed that the terminal 16' position of both lipids melted at a temperature about 2–3 °C below the phase transition temperature, but there was a marked difference between SSM and PSPC in the melting of the 4' and 10' positions (Fig. 3A/B). Both of these positions in PSPC melted near its phase transition temperature. However, for SSM the melting at the 4' positions occurred at a higher temperature than at the 10’ positions. This higher melting point at the 4' position of SSM can be explained by the intermolecular hydrogen bonding by the amide groups of SM, as shown in Fig. 3C. In other words, the SM-SM interaction may retain the gel-like packing structure of the SM alkyl chains to a temperature slightly higher than the main transition temperature.

Fig. 3.

Depth-dependent chain melting of SSM and PSPC. (A, B) Segmental melting points at the 4’, 10’, and 16’ positions of the acyl chains of SSM (A) and PSPC (B). Pale blue bars denote the phase transition temperatures of SSM and PSPC without deuteration. (C) (left) ²H-labeled sites in the stearoyl chain of SSM and intermolecular network via amide-amide hydrogen bonds in the upper part of SSM are shown with the dotted-line box. Headgroups are omitted. (Right) Schematic illustration of depth-dependent chain melting and phase state near the main transition temperature for SSM and PSPC, which shows that the major difference between SSM and PSPC is due to the strong SM-SM hydrogen bonds that act even above the main transition temperature and increase the segment melting temperature in the upper hydrocarbon chains (Tsuchikawa et al. 2022)

Detection of the gel state in SM/Cho binary membranes

Another important aspect of lipid packing in domains involves acyl chain interactions. The chain mobility of SM has been evaluated using model bilayers composed of stearoyl-SM(SSM)/Cho binary systems via the quadrupole splitting width (Fig. 4) of the solid state ²H NMR signal. Also, the fluorescence lifetime of a hydrophobic fluorescent probe (trans-parinaric acid; Fig. 5A–B) was used to measure the order of acyl chains under conditions in which the probe did not affect the phase state (Yasuda et al. 2015). Three important findings regarding the gel-state occurrence in SSM/Cho bilayers were as follows: (a) In the experimental SM/Cho compositions, some of which were assumed to exist entirely in the Lo phase based on reported phase diagrams (Marsh 2010), the NMR results and fluorescence lifetime revealed that the gel state, corresponding to the phase of a unitary SSM bilayer below the melting point (45 °C), appeared to be present (Figs. 4 and 5); (b) these gel domains may be very small; and (c) the gel domains in a membrane lacking Cho and those within the Lo phase have indistinguishable properties.

Fig. 4.

Solid state ²H NMR spectra of 10’,10’-d₂-SSM (stearoyl-SM) in bilayers containing 20 mol% (A), 33 mol% (B), and 50 mol% (C) Cho. Temperature-dependent changes in ²H NMR were recorded in 10 °C steps for the range between a gel-type signal and a typical Pake doublet appearing in a spectrum. The three SM/Cho spectra enclosed in blue squares show mixed Lo-type and gel-type signals (Fig. 5), where a very broad signal due to the gel state (seen at the bottom of lines A and B) overlaps with the Pake doublet (Yasuda et al. 2018). Inset: A dotted line depicts a signal deduced for a gel state overlapping the Pake doublet in the bilayers of 10’,10’-d₂-SSM containing 33 mol% of Cho at 20 °C. Graphs of the temperature-dependent Δν values are shown in Fig. 5C

Fig. 5.

Temperature-dependent changes in apparent percent abundance of the long lifetime fractions (longer than 30 ns) of trans-parinaric acid corresponding to gel state (A, B) and in the Δν values of 10’,10’-d₂-SSM and 10’,10’-d₂-PSPC in ²H NMR spectra (C) in SSM/Cho and PSPC/Cho bilayers containing three different Cho fractions. A large portion of the percent abundance shown above could be attributed to the Cho-poor gel-like domains (CPGLD), since the ratios of CPGLD/cholesterol-rich Lo-like domains were estimated from K_d and fractional amplitudes in SSM and PSPC bilayers for Chol concentrations of 20, 33, and 50 mol % (Yasuda et al. 2018). * Denotes the conditions under which the mixed gel and Lo phases were observed using ²H NMR (blue boxes in Fig. 4). (C) The graphs of Δν show a similar trend to (A) and (B), with the ordering of the SSM acyl chain being less dependent on temperature and Cho content than that of PSPC

Some of the ²H NMR spectra in Fig. 4 (particularly in the blue box) revealed that the gel state coexists with the Lo phase. In the inset spectrum shown in Fig. 4, a broad signal due to the gel state overlapping the Pake doublet is indicated by a dotted line. The Pake signal observations may be due to the fast-diffusing movement of small lipid clusters in the membrane. Considering the NMR timescale, the rotational correlation time of single-digit nanosized SM clusters could undergo lateral motion faster than was necessary for orientation averaging of the NMR signals, thus giving rise to a sharp Pake doublet detected in the Lo phase, e.g., the typical spectra of the Lo phase are those at higher temperatures than in the blue box in each column of Fig. 4. In other words, in SSM/Cho lipid bilayers, large SM domains in the gel state may coexist with smaller SM nanodomains, depending on temperature. We propose that these SM nanodomains, which have the physical properties of lipids in gel phase but are distinguished from them by their faster rotational correlation times, are a signature of the Lo state, and their organizational structure is considered in more detail below. As shown in Fig. 5, the lifetime of trans-parinaric acid in SSM/Cho and palmitoylstearoyl-phosphatidylcholine (PSPC)/Cho bilayers revealed that these coexisting gel and Lo states appeared across a wide range of temperatures and the fraction of SSM in the gel state of SSM/Cho bilayers was less dependent on Cho content than that in PSPC/Cho bilayers (a point that will also be further discussed below).

A previous study using X-ray and EPR showed that egg SM bilayers with a Cho content of over 41 mol% exclusively consisted of the Lo phase with no gel state above 17 °C (Chachaty et al. 2005). However, our findings in the lifetime experiments clearly revealed that the gel state occurred in SM bilayers, even with 50 mol% of Cho. This discrepancy may have been due to the size of the gel domains, which lacked sufficient cooperativity to be detected by X-ray diffraction.

Phase behavior of dihydrosphingomyelin (dhSM) reveals strong homophilic interactions

SM has a propensity to form gel-like domains that are less Cho-dependent than those of saturated PC, as discussed in the previous section. We were also interested in whether Cho is necessary for the Lo-phase formation of SM. To answer this question, we investigated membranes containing dihydrostearoylsphingomyelin (dhSSM), which is known to form more stable Lo domains than SM (Kinoshita et al. 2013; Yasuda et al. 2015). The structure of dhSSM differs from that of SSM only in having a single bond at C4–C5 of the sphingosine base instead of the double bond of SM (Fig. 1), and it is known as a minor component of sphingolipids in mammalian lipidomes. Using dhSM, we conducted microscopic observations, fluorescence lifetime measurements, Förster resonance energy transfer (FRET), isothermal titration calorimetry (ITC), and solid-state NMR.

Figure 6A and B show the phase diagrams of SSM and dhSSM bilayers containing the unsaturated PC, dioleoylphosphatidylcholine (DOPC), respectively. While both mixtures underwent phase separation between the SSM and the dhSM-rich gel phase and DOPC-rich fluid phase in low-temperature regions, the dhSSM-rich gel phase had significantly higher thermal stability than the SSM-rich phase; the dhSSM-rich phase showed an upper limit of T_m higher than the SSM counterpart (compare the gray lines in Fig. 6A and B). Interestingly, there was a region of liquid–liquid immiscibility in which the dhSSM-rich phase was still phase separated from the more DOPC-rich phase above the phase transition temperature. In dhSSM/DOPC 5:5 bilayers (Fig. 6C), a clear phase separation was observed in the temperature range of 44–54 °C under a microscope although our DSC result (Yasuda et al. 2018) did not indicate the presence of large gel domains. In Fig. 6B, this region is shown as “nanodomain/fluid.” These findings indicate that the domain formation ability of dhSSM in the absence of Cho was higher than that of SSM because of the stronger homophilic interactions between dhSSM molecules (Yasuda et al. 2016). As shown in Fig. 6, we also found that at high temperatures, dhSSM/DOPC binary membranes can express two kinds of liquid phases, as is the case with the SM/DOPC/Cho ternary system.

Fig. 6.

Phase diagrams of SSM/DOPC (A) and dhSSM/DOPC (B) binary bilayers. Gray zones roughly indicate the areas where macroscopic gel and fluid phases were observed using microscopy based on ²H NMR and DSC results (Yasuda et al. 2018). The large + nanodomain/fluid regions above the gel/fluid zones denote regions with gel-like nanodomains with a high diffusion rate of 10’,10’-SSM (and dhSSM), in which the lipid gave rise to a typical Pake doublet pattern in ²H NMR spectra. Triangles and circles indicate the phase boundaries, as determined by DSC measurements and confocal observations, respectively. The gray curves in Panels A and B indicate the upper limit of T_m in the SSM-rich and dhSSM-rich gel states, respectively. The phase boundary between the homogeneous gel phase and the gel/fluid phase coexistence regions (dashed lines) could not be determined accurately because, in their DSC thermograms, the superposition of the pre-transition peak on the main transition obscured the lower temperature limit of the main transition. The other phase boundaries were determined by the phase rule. * Nanodomains of dhSSM in this region were considerably larger than those of SSM. This figure was redrawn based on Kinoshita et al. (2014). (C) Confocal images of giant unilamellar vesicles (GUVs) composed of dhSSM-DOPC 1:1 in the temperature range of 34–64.^oC, as indicated by the dotted blue line in Panel B. Ld area was visualized by 0.2 mol% bodipy-PC

Among the ²H NMR spectra shown in Fig. 7, those in Panel A for SSM and Panels F–H for dhSSM contained both a doublet peak and a broad peak, which resembled the SSM-Cho system shown in Fig. 4, except for splitting width Δν. The Δν values of SSM in the SSM/DOPC bilayers became smaller as the temperature rose above 30 °C, whereas those of dhSSM showed little temperature dependence (Fig. 7). This observation indicated that as the temperature increased, the number of molecules exchanged between the gel-like nanodomains formed in the Ld phase and the surrounding liquid area and/or the amount of SM in the Ld phase increased at a higher rate for SSM than for dhSSM. Besides the usual thermal perturbation of lipid order, this exchange influenced the SM Δν values in the Ld phase versus temperature because the large splitting width of the SMs in the gel-like nanodomains was reflected by those in the surrounding area. During the NMR timescale, SSM molecules in these gel-like nanodomains were mostly involved in this exchange. Additionally, the size of the gel-like domain would have to be small to give rise to a clear Pake signal (Yasuda et al. 2015). However, the homophilic interaction of dhSSM was stronger than that of SSM, leading to the low content of dhSSM dissolved in the Ld phase and larger gel-like domains partly residing in the Ld region; the former factor reduced the chain order in the Ld phase, and the latter caused the broad component in ²H signals, as shown in Fig. 7 F–H. Furthermore, the slower exchange rate of dhSSM between the gel-like phase and the liquid phase decreased the influence of the gel-like phase on splitting width. Consequently, the dhSSM Δν values were smaller than those of SSM (Fig. 7).

Fig. 7.

Solid state ²H NMR spectra of 10’,10’-d₂-SSM (A–E) and 10’,10’-d₂-dhSSM (F–I) in binary bilayers containing 50 mol% DOPC without Cho as a function of temperature (the value in kHz for each spectrum denotes the splitting width, Δν, of the Pake doublet). The Δν values and the raised baseline indicate that, in the dhSSM/DOPC bilayers, the gel state partly occurred at 45 °C, whereas the SSM/DOPC bilayers did not show the gel-state pattern above 30 °C, as shown by the blue boxes. The inset in Panel B shows the ²H spectrum (black) and its simulation (red) for 10’,10’-d₂-SSM in SSM/DOPC/Cho 1:1:1 bilayers in comparison with the spectrum without Cho, in which the inner and outer doublet peaks corresponded to the SSM in the Ld and Lo phases, respectively. Overlaid spectra with calculated Pake signals (A–I) are shown in the “Supporting Information” of the original report (Yasuda et al. 2018). ²H NMR experiments have provided further insights into the differences between dhSM and SM physical behavior when mixed with DOPC. These differences manifest themselves in different Pake splitting and different temperature-dependent splitting widths for dhSSM-DOPC vesicles and SSM-DOPC vesicles. In both cases, the signal for which the Pake doublet is visible comes from Ld domains and is influenced by the exchange of.²H-labeled dhSSM or SSM between gel and Ld phase regions (as a function of domain size) and the content of dhSSM or SSM in the Ld phase

The Pake splitting width (Δν = 30 kHz at 30 ℃) of the SSM in the SSM-DOPC binary system was significantly larger than the splitting width (22 kHz at 30 ℃) of the glycerolipid counterpart PSPC under the same conditions (Yasuda 2015). The Δν value of the palmitoyl chain (19 kHz at 27 ℃) of the unitary POPC bilayers (Huber et al. 2001), consisting of equimolar saturated palmitoyl and unsaturated oleoyl chains, which is the same as for the SSM-DOPC 1:1 system in terms of the average number of saturated and unsaturated acyl chains per lipid, was closer to the value of the palmitoyl signal arising from PSPC in Ld domains in PSPC-DOPC bilayers. In the DOPC-rich (Ld) phase of the PSPC/DOPC bilayers at 30 ℃, in which gel and Ld phases coexisted, the saturated and unsaturated chains were thought to be close to the molecularly mixed state, unlike the SSM/DOPC bilayers. These observations indicate that SSM, which gave rise to the Pake doublets in Fig. 7, was not completely molecularly mixed with DOPC, suggesting the formation of small domains, including nanodomains, in the Ld phase of the SSM/DOPC bilayers.

In the ²H NMR spectra measured under the phase-separated conditions according to Fig. 6A–B, the Pake doublets corresponding to the two kinds of SSM (or dhSSM) states were not observed in SSM/DOPC bilayers, whereas two doublet peaks for the Ld and Lo phases were observed in the presence of Cho (inset in Fig. 7B). This difference was attributable to the notion that the Lo and Ld phases were predominant in Cho-containing membranes, while in SSM/DOPC bilayers, the Ld phase occupied the majority, and the macroscopic gel phase was relatively small (Kinoshita et al. 2014). In other words, the macroscopic Ld phase contained a large amount of SSM, while the macroscopic gel phase contained almost no DOPC, resembling the gel phase of SSM pure bilayers. Therefore, the signals in Fig. 7 were considered mostly derived from the Ld phase, which incorporated nanodomains, as described above, since SSM in the gel phase has low mobility and gives almost no (or a very broad) NMR peak. However, in the spectra of the Cho-containing bilayers (inset in Fig. 7B), there was almost no macroscopic gel phase, and about 80% of the SM resided in the fast-diffusing Lo phase; the Cho ordering effect contributed to this large splitting width by inducing the gel-like packing of hydrocarbon chains as shown in the inset spectrum of Fig. 7B.

To summarize, in the presence of a colipid, such as DOPC (which mixes with SM to some extent), SM forms small gel domains. The size of these domains may be slightly larger than that of the nano-subdomains formed in the presence of Cho but much smaller than the macroscopic gel domain. In other words, nanosized domains of SSM can be formed without Cho, while dhSSM provides significantly larger domains under the same conditions.

The origin of the difference between Δν in Cho-containing and Cho-lacking SM-DOPC systems is worth considering. Why does Cho dramatically promote the formation of the highly ordered Lo phase, with its greater order and thus greater Δν? Cho has a remarkable condensing effect on the lipid chains of Lo-forming lipids (Ramstedt and Slotte 2006), involving direct contact of Cho’s rigid tetracyclic core with hydrocarbon chains, making the chains immobile and producing an extended conformation. This is at least partly driven by the minimization of exposure of Cho to water, known as the umbrella model (Huang and Feigenson 1999). The effect proposed in this model is still present even at the melting point (Yasuda et al. 2014), and the Cho-ordering effect promotes the formation of ordered domains above the melting temperature of Lo-forming phospholipids in mixed-phospholipid bilayers in which only one phospholipid favors Lo formation; this temperature is usually lower than the melting point of bilayers composed solely of the Lo-forming phospholipid. It was recently reported for phosphatidylethanolamine (PE) and PC systems that Cho addition caused the formation of Lo microdomains by ordering unsaturated PC at the temperature at which, without Cho, the PC and unsaturated PE completely mixed (Goh and Tero 2021). This observation shows that the selective ordering effects of Cho on the PC, regardless of its melting point (and even if it has unsaturated acyl chains), can efficiently promote Lo phase formation.

To further demonstrate the homophilicity of dhSM, we prepared a fluorescent derivative of dhSM (488neg-dhSM) and compared its behavior to that of fluorescent SM (488neg-SM; Kinoshita et al. 2020a). These headgroup labeled probes contain a polyethyleneglycol (peg) anchor and preserves many properties of unlabeled SM and dhSM (Kinoshita et al. 2017). We gradually added dhSM to SM/Cho/DOPC GUVs and evaluated the dependence of the diffusion coefficients of 488neg-SM and 488neg-dhSM on dhSM content. Because a diffusion coefficient is inversely correlated with cluster size in accordance with the equation by Saffman and Delbrück in 1975, the results were expected to reflect differences in the size of the clusters into which neg-SM and neg-dhSM were recruited. Fluorescence correlation spectroscopy experiments showed that the diffusion coefficients of dhSM in the SM-rich Lo domain decreased significantly as the dhSM content increased, while those of SM did not change significantly (Table 2), indicating that dhSM molecules form homophilic clusters inside the SM-rich Lo domain of the SM/DOPC/Cho system even when the content of dhSM is only 1–3.3 mol% of total lipids (Fig. 8). The inhomogeneous distribution of dhSM in the SM-dominant Lo domains further demonstrates the stronger homophilicity of dhSM-dhSM relative to the heterophilicity of dhSM-SM.

Table 2.

Diffusion coefficients of SM and dhSM in SM/DOPC/Cho GUVs containing minor amounts of dhSM.^a (Kinoshita et al. 2020a)

Molar ratio of SM/dhSM/DOPC/Cho in GUV	Diffusion coefficienta within the Lo domains (µm2/s)b
	488neg-SM	488neg-dhSM
100:0: 100:100	0.72 ± 0.06	0.61 ± 0.07
97:3: 100:100	0.71 ± 0.09	0.57 ± 0.06
94:6: 100:100	0.74 ± 0.08	0.46 ± 0.07
90:10: 100:100	0.65 ± 0.09	0.41 ± 0.06

^aDiffusion coefficients are presented as means ± SEs

^bThe content of fluorescent lipids was 0.002 mol% of the total lipids

Fig. 8.

Schematic diagram of the formation of dhSM clusters in an SM-rich Lo domain. The blue and red headgroups correspond to those of SM and dhSM, respectively. Orange membrane regions indicate the dhSM-rich areas. This figure was redrawn from Kinoshita et al. (2020a)

To summarize, although SSM and dhSSM, which have analogous structures, and can be considered to have the same modes of intermolecular interaction, such as hydrogen bonding, their interaction strengths differ. The ²H NMR and diffusion results showed that a strong dhSSM–dhSSM interaction occurred even above the temperature at which the macroscopic gel phase disappeared, implying that factors other than acyl chain packing contributed to the interaction. In other words, the homophilic interaction via hydrogen bonds facilitates the formation of dhSSM aggregates. We believe that this hydrogen-bonding interaction is the main driving force for the formation of SSM nano-subdomains, which are particularly prominent in Cho-containing membranes.

Phase behavior of SM and its enantiomer examined using calorimetry and.²H NMR

To investigate the stereochemical specificity of SM–SM homophilicity, we compared the interaction between Lo lipids (SM and Cho) and ent-SM (the enantiomer of natural SM) to that between Lo lipids and natural SM (Yano et al. 2018). Ent-SM was chosen because the configurational difference at the amide position (C2) causes steric repulsion between SM and ent-SM, while an intermolecular hydrogen bond is assumed to be formed between SM and SM, as well as between ent-SM and ent-SM via the amide groups (Fig. 2B). Thus, the hydrogen-bond forming ability of SM–ent-SM interactions should be significantly lower than that involved in SM–SM or ent-SM–ent-SM interactions. In the DSC thermograms shown in Fig. 9, two melting peaks were clearly observed for palmitoyl-SM(PSM)/ent-SSM bilayers, implying that ent-SSM molecules aggregated in a homophilic manner and did not mix well with palmitoyl-SM. Subsequently, we examined the ordering effects of Cho on SM and ent-SM acyl chains using ²H NMR, as shown in Fig. 10A (Yano et al. 2018). The 10’ position of the a stearoyl chain of SSM and ent-SSM showed a similar splitting width, indicating that an intermolecular hydrogen bond between SM and Cho, if any, did not influence the Cho-ordering effects on SM in a stereospecific manner. Additionally, Cho methyl ether (CME), in which the 3-OH group of Cho was protected by a methyl group, showed ordering effects on SM that resembled those of Cho, indicating that an intermolecular SM–Cho hydrogen bond is not important for the ordering effects of Cho (Fig. 10B), even though the ordering effect of Cho on hydrocarbon chains of SM and other lipids is vital, as described in “Phase behavior of dihydrosphingomyelin (dhSM) reveals strong homophilic interactions” section. These findings further imply that the SM–SM interaction between amide groups plays an essential role in the formation of subdomains in the presence of Cho.

Fig. 9.

DSC phase transition thermograms showing the main gel–Ld transition of SSM/PSM (A) and ent-SSM/PSM (B). From the top to the bottom of the thermograms, the molar ratios of SSM/PSM and ent-SSM /PSM are 1:0, 3:1, 2:1, 1:1, 1:2, 1:3, and 0:1 (Yano et al. 2018)

Fig. 10.

Acyl chain orders for SSM and ent-SSM in binary systems with Cho or cholesterol methyl ether (CME) as determined by solid-state.²H NMR in the temperature range of 25–50 °C. (A) The Δν values of 10’,10’-d₂-SSM/Cho (1:1), 10’,10’-ent-SSM/Cho (1:1), and their one-to-one mixture. (B) The Δν values of 10’,10’-d₂-SSM/CME (1:1) and 10’,10’-d₂-ent-SSM/CME (1:1) compared with 10’,10’-d₂-SSM/Cho (1:1) (Yano et al. 2018)

Domains formed by saturated PC bilayers, such as dipalmitoylphosphatidylcholine (DPPC), PSPC, and distearoylphosphatidylcholine (DSPC), are thought to differ significantly in size and thermal behavior from those of SM (Yasuda et al. 2014). For example, DSC thermographs of bilayers with a mixture of PSPC and ent-DPPC, under conditions like those shown in Fig. 9, gave rise to a single peak at any mixing ratio (Hanashima et al. 2020), suggesting that the homophilicity of PC in headgroups is weaker than that of SM. Also, gel-phase formation by saturated PC is significantly more dependent on Cho content than that of SM (Fig. 5), implying that as the Chol concentration increased, it reduced gel-phase formation by PC acyl chains to a greater degree than for SM chains. PC forms an Lo phase containing a similar fraction of gel subdomains to SM at 50 mol% Cho, while at lower Cho content (e.g., 20 mol%), PC forms larger and more abundant gel domains (Fig. 5A–B) than SM, as described below (Yasuda et al. 2015).

The solid-state NMR results showed that SM configuration and Cho methylation had little effect on the order of the acyl chain (Fig. 10). This result clearly reveals that, in the headgroup moiety where asymmetric centers are present, the intermolecular attractive force that promotes the formation of SM gel-like domains involves SM–SM hydrogen bonds, not SM–Cho hydrogen bonds. Thus, the ordering effect of Cho on the SM acyl chain derives from the rigid tetracyclic core (and the umbrella effect), which does not require SM-Cho hydrogen bonds. The lack of a crucial SM–Cho hydrogen bond, and the experimental results for ent-SM and dhSM, showed that Cho merely enhances the order of the acyl chains of SM clusters and is not a critical factor in the formation of SM clusters. Taken together, for Lo domains, the effect of Cho should be considered to work on SM clusters rather than on individual SM molecules.

Nanodomain and nano-subdomain formation by SM examined using FRET and microscopy on ternary lipid mixtures

Förster resonance energy transfer (FRET) between lipidic fluorescent probes is promising for detecting nanodomains (Pathak and London 2011, 2015; Petruzielo et al. 2013). In a homogeneous phase, the probes are dispersed on the bilayer plane, so that FRET efficiency varies little according to lipid composition or temperature. However, FRET dramatically changes when crossing into phase coexistence regions. FRET efficiency increases when the probes reside in the same phase, but it decreases when the probes reside in different phases. Such abrupt changes in FRET efficiency clearly indicate phase separation. This methodology detected the formation of SM-rich domains in brain SM/POPC/Cho ternary mixtures, although these mixtures did not show macroscopic phase separation according to fluorescent microscopy (Enoki et al. 2018). Moreover, a combination of FRET and Monte Carlo simulations disclosed that the Lo-domain size was below 10 nm in brain SM/POPC/Cho ternary bilayers at room temperature (Enoki et al. 2018; Frazier et al. 2007; Pathak and London 2011). Notably, although the FRET-based methodology is useful for detecting sub-micrometer phase separation, it does not detect domains smaller than the Förster length (Ro). Consequently, in extreme cases where domain coexistence involves domains smaller than the Förster distance, domain formation is not detectable by FRET (Heberle et al. 2010; Pathak and London 2011).

For the FRET experiments, we took advantage of the phase-segregating properties of SSM and its enantiomer (ent-SSM) to detect nano-subdomains in SSM/ent-SSM/DOPC/Cho membranes containing coexisting Lo and Ld phases and to estimate the size of the nano-subdomains (Yano et al. 2020). The stereochemical specificity of SM–SM interactions causes SM and ent–SM to form separate domains, as described above. FRET was used to probe the localization of SM and ent-SM attached to nonperturbing fluorescent labels. A FRET pair that had labeled SM probes with the same stereochemistry should colocalize in the same subdomains and did give more FRET than a FRET pair with different stereochemistry (one SM probe and one ent-SM probe), which should tend to localize in different SM subdomains (see boxed region in Fig. 11B). Notice that Fig. 11B also shows the effect of stereochemistry decreasing at a higher temperature and disappearing above 35 °C, indicating a decrease in subdomain size with increasing temperature and a subdomain size close to the effective Förster distance for these probes when they were attached to lipids via peg headgroups in membranes (Ro’ of approximately 3 nm) at around 35 °C. Above 50 °C, Lo domains either disappeared completely or decreased to sizes less than Ro’.

Fig. 11.

FRET is affected by the Lo domain and the subdomain structure. (A) Schematic surface view of the membrane leaflet with Lo domains and subdomains. Arrows represent phospholipids, with directions indicating their orientations to the membrane surface (sn-1 position acyl chain at one end of the arrow and sn-2 acyl chain at the other end). Ld-favoring lipid such as DOPC is shown in thick yellow-orange, SM is shown in red–orange, and ent-SM is shown in blue. Cho is shown as green squares. When domains were present, FRET increased relative to a homogeneous membrane due to the concentration of SM FRET probes in the Lo domains. However, if subdomains were present within the Lo domains when one FRET probe was SM based and the other was ent-SM based, segregation into different subdomains decreased FRET more than when both FRET donors and acceptors were SM based and localized in the same type of subdomain. (B) FRET data from Yano et al. (2020) for SSM/ent-SSM/DOPC/Cho and control homogeneous DOPC/Cho vesicles. F/Fo is the ratio of FRET donor fluorescence in the presence of an acceptor to that in its absence; low F/Fo equals stronger FRET. Note that stronger FRET in SSM-containing vesicles at low temperatures relative to control vesicles indicated domain formation, and differences in FRET for vesicles with FRET probes with the same or different stereochemistry indicating subdomain formation (boxed region in the figure)

Other methods can provide information about domain size. Atomic force microscopy (AFM) enables label-free observation of phase separation because it detects differences in membrane thickness between the Lo and Ld phases. Since the lateral resolution of AFM is considerably higher than that of optical microscopes, AFM is a promising methodology for observing nanodomains. Kahdka et al. observed the composition-dependent phase behavior of egg SM/DOPC/Cho ternary bilayers using AFM. Although macroscopic Lo/Ld phase separation was observed across a broad compositional range at 25 °C, they discovered fluctuating nanodomains near the compositional critical point (Khadka et al. 2015). Interferometric scattering microscopy (iSCAT) is another methodology for the label-free observation of nanodomains. The contrast of iSCAT depends on an 800-fold difference in light scattering between the lipid membrane and the environmental phase (De Wit et al. 2015). De Wit et al. (2015) visualized dynamic nanodomains in droplet interface bilayers consisting of a SM/DOPC/POPC/Cho quaternary mixture at room temperature using iSCAT. They found a broad range of domain sizes, with a mean diameter of 120 nm and a lifetime below 220 ms, depending on the domain size.

While these two methodologies enable label-free imaging with high lateral resolution, the setup of bilayer sample preparations may influence the formation and morphology of nanodomains; for example, AFM and iSCAT require a supported bilayer with the membrane surface interacting somewhat with the substrate. Also, in iSCAT sample preparation, a bilayer is formed by contact between two monolayers in an oil environment (Leptihn et al. 2013), raising the possibility of bilayer preparations causing artifacts such as membrane breaks and/or altering the intrinsic properties of lipid bilayers.

MD simulation of molecular interactions in Lo domains and the detection of domains and subdomains

Many MD simulation studies have been performed on SM-containing phase-separated systems. They have provided important insights into the molecular interactions that control domain and subdomain formation. Many of these MD results support the afore-mentioned experimental results derived from NMR and FRET experiments. Wang and Klauda (2017) performed extensive MD simulations of mixed-membrane systems containing SM and Cho and reported that intermolecular SM–SM hydrogen bonds increased as Cho content increased. In the SM/Cho binary and ternary systems, including unsaturated phosphatidylcholine (or phosphatidylethanolamine), Bera and Klauda (2017) reported a probability of hydrogen bonds forming between SM–SM amide groups that was higher than the probability for Cho and SM (Fig. 12A). Smith and Klimov (2018) performed MD simulations on a ternary system of palmitoyl-SM (PSM) and dimyristoylphosphatidylcholine (DMPC) in the presence of Cho, and they investigated the effect of the structural difference between SM and PC on lipid–lipid interactions. They demonstrated that SM–SM hydrogen bonds (particularly between amide groups) formed more frequently, and their lifetimes outlasted hydrogen bonds for other lipid pairs. They also showed that PSM was more likely to form linear molecular clusters than DMPC in the same ternary system (Fig. 12B), suggesting that intermolecular hydrogen bonds can induce cluster formation (Smith and Klimov 2018).

Fig. 12.

(A) MD simulation snapshots showing intermolecular hydrogen bonds (green line and arrowhead) between two amide groups of PSM molecules (intramolecular ones are shown with gray lines). Modified from Bera and Klauda (2017). (B) MD simulation revealed linear clustering of PSM occurring in the ternary PSM/DMPC/Cho bilayer. Yellow balls depict the position of a phosphate group in the PSM cluster (Smith and Klimov 2018)

Sodt et al. (2015) investigated SM hydrogen bonds, focusing particularly on the conformation of the SM headgroup. The MD results suggested that the amide group orientation changed remarkably depending on the presence or absence of Cho; in the presence of Cho, the formation of SM–SM hydrogen bonds was promoted, and the SM Lo phase stabilized (Sodt et al. 2015). The results for the amide plane orientation were inconsistent with our NMR data (Fig. 2B), possibly partly due to the time averaging of interchangeable orientations during the NMR measurements. However, both sets of results showed that Cho significantly reduced the fluctuation of the SM amide plane orientation in the absence of Cho. Furthermore, they detected cluster formation by SM and DPPC in the Lo phase in ternary lipid mixtures also containing DOPC and Cho (Fig. 13). These structures were consistent with SM and DPPC forming subdomains within the Lo phase. They also detected hexagonally packed hydrocarbon chains appearing in SM clusters, suggestive of gel phase-like properties. Overall, the picture provided by MD was strikingly close to that provided by experimental studies.

Fig. 13.

MD simulations showing lateral acyl chains and Cho packing in Lo and Ld phases. A single leaflet is shown with the center of mass of hydrocarbon chains and Cho rendered as discs. A white border indicates lipid in the Lo phase, and a black border indicates lipid in the Ld phase. Red circles indicate PSM or DPPC, blue circles indicate DOPC (or POPC), and yellow circles indicate Cho. Column order: Lo phases (left), Ld phases (middle), and coexisting systems (right). Notice that some areas of the Lo phase show hexagonal chain packing, as seen in the gel phase. (Adapted from Sodt et al. 2015)

Impact of lipid asymmetry in bilayers on domain and nano-subdomain formation

The studies above involve use of symmetric membranes in which the lipid composition in each leaflet was the same. However, many natural membranes, especially eukaryotic plasma membranes, are asymmetric and have a different lipid composition in each leaflet (Lorent et al. 2020). This raises questions about the formation of Lo domains in nature, because the sphingolipids believed to trigger Lo domain formation are concentrated only in the outer leaflets of plasma membranes. Recent studies showed that Lo domain formation occurred in asymmetric membranes of egg SM/DOPC/Cho with SM located only in the outer leaflet (Lin and London 2015), which compositionally resembled membranes in which the Lo domains contained nano-subdomains. For asymmetric bilayers of egg SM/DOPC/Cho, Lin and London (2015) reported that when the Ld phase was visualized with NBD-DOPE under a microscope, it was registered. “Registered” implies that the Lo (or Ld) domains of the outer leaflet induce the Lo (or Ld) phase in the same lateral area of the inner leaflet. In contrast, a membrane in which the inner leaflet of Lo (Ld) contacts Ld (Lo) is “anti-registered.” Other methods can also detect the formation of Lo domains in asymmetric vesicles with this type of composition. Our preliminary ²H NMR spectra of SSMo/DOPCi/Cho 1:1:1 asymmetric large unilamellar vesicles, which largely contained SSM mixed with a small amount of DOPC in the outside leaflet, DOPC in the inside leaflet, and Cho in both leaflets, revealed that Δν SSM values were close to those in the Lo phase (unpublished results). This observation suggests that nanodomains, like those in symmetric membranes, form in asymmetric ternary systems. Coarse-grained MD simulation of membranes with saturated and unsaturated glycerophospholipids recently showed that asymmetric Lo domains that form only in the leaflet with saturated lipids are only slightly less ordered than symmetric Lo domains, and large Lo domains in asymmetric systems induce a slight chain ordering in the opposite leaflet regions (Mohideen et al. 2020). Another MD study found that the long-chain C_24:0 SM in a DOPC/Cho-containing asymmetric ternary system induced anti-registered Lo/Ld distribution (Seo et al. 2020). Thus, it appears that Lo domain formation, like that in symmetric membranes, occurs in asymmetric membranes. We speculate that the Lo domains in asymmetric membranes contain subdomains like those in symmetric membranes. In support of this notion, studies of cells that have highly asymmetric plasma membranes have hinted that sphingolipids with different stereochemistry segregate into different domains (Singh et al. 2006).

Gel-like subdomain formation within the Lo phase by lipids other than SM

We turned our attention to bilayer lipids with higher levels of cooperativity than SM, such as saturated PC, to investigate how their interactions with Cho influenced properties related to phase transitions from gel to liquid states. Kullberg et al. (2015) reported that PSM and SSM in the presence of unsaturated PC formed a gel phase with less order than DPPC and PSPC, respectively, based on fluorescence lifetime and DSC experiments. Together with the FRET results in Fig. 11, SM is assumed to form nano-subdomains in binary bilayers containing unsaturated lipids, as schematically shown in Fig. 14. To examine how the acyl chains of saturated PC were packed in the gel state, static ²H NMR spectra of PSPC bilayers were measured in Cho-containing binary systems (Fig. 15). At 40 °C, the PSPC/Cho bilayers produced the typical peak of the gel- and Lo-coexisting state in the 20 mol% Cho-containing bilayers. The signal pattern was like that of SSM at a temperature of 30 °C (i.e., gel-like domains with properties like those of SM formed in saturated PC). However, as shown in Fig. 5A–B, the effect of Cho on the abundance of the gel state formed by SSM and PSPC differed significantly. The fraction of the gel state in PSPC/Cho bilayers was strongly dependent on Cho concentration, but that in SSM/Cho was much less dependent on Cho. This also suggests that SM (i.e., SM alone) tends to form nano-subdomains in the absence of Cho and already has the ability to form small gel domains, probably due to the hydrogen bonding interaction between the amide groups, which should prefer a different molecular arrangement from the hexagonal packing of saturated chains. However, in the absence of Cho, PSPC, which lacks a hydrogen-bond donating group, tends to form large gel domains, largely due to chain packing force. Cho located between SM subdomains increases their lateral diffusion and thus forms the Lo state, but Cho both breaks up large PSPC gel domains and induces Lo state formation. This topic will be further discussed in “Unique properties of SM nano-subdomain formation in comparison with PC” section.

Fig. 14.

Hypothetical illustration of how gel-phase like domains may differ for SM and saturated PC (here, palmitoyl stearoyl PC). (A) A hypothetical diagram of a lipid monolayer with a binary lipid mixture with SM (blue) and an unsaturated lipid (gold), showing a gel state composed of small subdomains. (B) A hypothetical diagram of a lipid monolayer with a binary lipid mixture with a saturated PC (blue) and an unsaturated lipid (gold), showing a gel state containing larger domains. The membranes are shown as viewed from above the membrane plane. Arrows represent lipids, with their directions indicating orientations to the membrane surface. (sn-1 position acyl chain at one end of the arrow and sn-2 acyl chain at the other end). Although not shown, the gel state may contain a significant amount of unsaturated lipids

Fig. 15.

Solid state.²H NMR spectra of 10’,10’-d₂-SSM (upper row, the same spectra as those in Fig. 4 (A) and 10’,10’-d₂-PSPC (lower row) bilayers containing 20 mol% Cho at various temperatures (the value in kHz for each spectrum denotes the splitting width, Δν, of the Pake doublet). The difference in Δν values and the raised baseline indicate that, in the PSPC-Cho bilayers, the gel state was still present at 40 °C, and the SSM/Cho bilayers showed a gel state at 30 °C, as indicated by blue squares (Yasuda et al. 2015)

Based on the results above, Kinoshita et al. (2020b) examined the formation of substructures containing saturated PC in a gel-like ordered phase in membranes using electron diffraction (ED). As with wide-angle X-ray diffraction (WAXD), ED can provide structural information on the chain packing of lipids and has some advantages over WAXD. First, ED allows the selective acquisition of a diffraction pattern from a designated area of a few square micrometers using the selector aperture. Second, due to the great diffraction potency of electrons, ED can produce clear diffraction peaks from a small membrane area. Due to these advantages, ED is a promising methodology for examining the substructure inside a membrane domain. In this study, we used DSPC/DOPC binary monolayers, which formed coexisting micrometer-sized DSPC-rich ordered and DOPC-rich disordered domains. We selectively obtained diffraction patterns from several different regions across a particular ordered domain (Fig. 16A–C) and found that the ordered domain contained multiple subdomains with different crystallographic axes (Fig. 16D) and subdomain sizes ranging from 2.8–14 μm. Also, because the crystallographic axes were the same in the domain center (see regions 4 and 5 in Panel D), regions 4 and 5 probably belonged to the same subdomain. We found that the size of the subdomains was larger in the domain center and smaller in the vicinity of the phase boundary. Because the DSPC-rich phase forms the gel-like ordered phase, its diffusion coefficient is negligibly small (Kahya et al. 2004). Also, that DSPC/DOPC monolayer undergoes the order/disorder phase separation both on the quartz substrate and water subphase was confirmed by Kinoshita et al. 2020b. Thus, it was expected that solid support does not significantly alter membrane structures in the case of this sample.

Fig. 16.

Local structure inside an ordered domain in a DSPC/DOPC monolayer. (A) Fluorescent micrographs of the DSPC/DOPC (3:7 molar ratio) monolayer formed on a collodion-coated TEM grid. The darker and brighter regions correspond to the DSPC-rich ordered and DOPC-rich disordered domains, respectively. (B) Magnification of the area indicated by the dashed square in Panel A. Bars indicate 30 μm. (C) The ED patterns obtained at regions 1 − 7, as indicated in Panel B. The selected area for each region was 6.2 μm², corresponding to a 2.8-μm diameter, and the exposure time was 0.5 s. (D) The intensity of the hexagonal spots appearing along the azimuthal direction θ, as shown in Panel C. The corresponding region numbers are indicated directly in the panel. The differences in the azimuthal angles of the hexagonal spots between the center (4 and 5) and annular (3 and 6) regions in the ordered domain are indicated by dashed lines and arrowheads, respectively. Schematic illustrations of the directions of the chain packing lattice in regions 3–6 are also shown in Panel D. This figure was redrawn from Kinoshita et al. (2020b)

There are several reasons for the size of subdomains within the Cho-free ordered domains in these experiments being significantly larger than the size of subdomains in the Lo phase in lipid vesicles, including the effect of the lipid structure, limitations on domain size in vesicles due to small vesicle size, and differences in lipid behavior in monolayers and bilayers. Another important factor was the presence of Cho in the Lo phase; for example, an MD simulation (Sodt et al. 2015) demonstrated that Cho molecules (a main component of the Lo phase) were inserted into the large hexagonal lattice of saturated chains, decreasing the size of the saturated chain lattice (Fig. 13).

Another imaging modality—iSCAT—revealed heterogeneity inside the Lo phase. Using iSCAT, Wu et al. (2016) observed the diffusion behavior of a single gold nanoparticle labeled DPPE (GNP-DPPE), which was partiallyf incorporated in the Lo phase. They used DPPC/Cho/diphytanoyl-phosphatidylcholine (diphytanoyl PC) ternary bilayers, which (like SM/DOPC/Cho bilayers) undergo Lo and Ld phase separation. In the Lo phase, the GNP-DPPE showed subdiffusion rather than simple Brownian diffusion in the microsecond timescale (Fig. 17). Based on this result, they speculated that a nanoscopic heterogeneous molecular arrangement existed in the Lo phase, with GNP-DPPE temporally entrapped by nano-subdomains (Fig. 16). Furthermore, analysis of the trajectories revealed an average residential time of 0.62 ms and an average subdomain diameter of 32 nm. On the other hand, STED-FCS by Honigmann et al. (2013) could not detect anomalous diffusion of fluorescently labeled DSPE in the Lo phase, indicating that the slow diffusion in the Lo phase was caused simply by tight lipid packing rather than by anomalous sub-diffusion over a length > 40 nm. The disagreement with the iSCAT results could be attributed to a difference in spatiotemporal resolution, since iSCAT’s resolution of a few nanometers at 50 kHz is high enough to examine nanosized heterogeneity within Lo domains.

Fig. 17.

Schematic diagram showing domains with nano-subdomains and nanoscopic molecular dynamics in the Lo phase deduced from iSCAT experiments with DPPC/diphytanoyl-PC/Cho ternary bilayers. The molecular arrangement is not homogeneous at the nanoscale; there are multiple subdomains of tens of nanometers—potentially clusters of closely packed DPPC. These subdomains affect the diffusion of gold nanoparticle (GNP)-DPPE. For illustration purposes, an artificial diffusion trajectory is plotted, showing fast and free diffusion in the Ld phase (red) with transient confinement in the nano-subdomains of the Lo phase (blue). This figure was redrawn from a paper by Wu et al. (2016)

Unique properties of SM nano-subdomain formation in comparison with PC

As described above, the understanding of nanodomains in model bilayers has advanced considerably in the last two decades, revealing unique aspects of SM behavior that promote its ability to form domains in the outer leaflets of cell membranes and influence biological functions. Thus, it is important to consider the molecular basis of the differences between the nano-subdomains of SM and those of saturated PC (Yasuda et al. 2014, 2015). These differences include:

• Under the same conditions, the sizes of SM nano-subdomains are generally smaller than those of saturated PC with a similar melting point as SM, and also less dependent on Cho content.

• The fraction of SM nano-subdomains in the Lo phase and the chain order of SM in the nano-subdomains are both less temperature-dependent and less Cho-dependent than those of the PC counterpart.

An effect like that of asymmetric carbon atoms near the SM amide group on domain formation was not observed for saturated PC in DSC experiments (Hanashima et al. 2020). Therefore, the difference between SM and PC in the domain properties listed above should largely be attributable to the intermolecular interactions of the polar functionalities, including those of the amide group. The temperature of the normal phase transition of membrane lipids largely depends on the packing energy of the acyl chains and on the ratio of the cross-sectional areas of the headgroups and acyl chains, whereas the lateral interactions between the headgroups are thought to have little effect. However, in the case of SM, the intermolecular interactions between SM headgroups clearly affect the phase behavior; thus, there is a question of whether the most stable arrangement for chain packing matches the most stable arrangement of the headgroups. If they differ considerably, a phase transition may occur twice during increasing temperature due to rearrangement of the stable structure of chain packing (e.g., gel–ripple transition) and due to both dissociation of the stable interaction of headgroup and chain melting (e.g., ripple–fluid transition; De Vries et al. 2005). Although the formation of the ripple phase is not unique to SM (i.e., it is also observed for saturated PC), it is likely that the headgroup interaction is significantly stronger in SM than in PC. For example, the two transition temperatures observed for the ent-SSM and PSM mixed bilayers (Fig. 9) revealed that intermolecular stereochemical matching significantly influenced the most stable arrangement of the headgroups, supporting the importance of headgroup interactions in controlling SM–SM interaction (Yano et al. 2018). Also, the fact that the main transition peak of SM observed in DSC was always broader than that of PC may imply that chain packing was loosened and cooperativity decreased during the first thermal transition observed when heating (pretransition). In X-ray diffraction experiments on gel-phase bilayers, lattice defects have been observed (Akiyama et al. 1982). The difference in steric requirements between the headgroups and hydrocarbon chains of SM should lead to packing frustration. Thus, the SM gel phase may contain more abundant lattice defects than that of PC, meaning that it is composed of nano-subdomains smaller in size than those of PC. The sizes of SM subdomains are probably narrowly distributed under the conditions of typical model membrane experiments because the chain order of SM and the fraction of gel-like domains are less dependent on temperature and Cho content (Yasuda et al. 2015), both of which should greatly influence the size of subdomains. In contrast, large gel domains consisting of saturated PC, which are largely stabilized by the chain packing force rather than headgroup interactions, are strongly affected by temperature and Cho content. Further investigation of SM behavior in model membranes is necessary to elucidate the temperature-dependent aspects of SM structure and orientation that are responsible for these properties of SM domains.

Hypothetical functions of nano-subdomains in heterogeneous biomembranes

As described herein, recent advances in studies of SM-based nanodomains have deepened our understanding of the properties of SM in model bilayer membranes. Whether similar nanosized heterogeneity occurs in biological membranes is unclear (de la Serna et al. 2016; Sevcsik and Schütz 2016). Given the complex structure of biomembranes and the spatiotemporal effects of non-equilibrium conditions upon membrane organization, a definitive answer cannot be given at this time. Nevertheless, we will briefly consider the potential biological effects that SM nano-subdomains could have in biological membranes in vivo.

Conceptual diagrams schematically illustrating SM nano-subdomains based on the studies discussed in this review are shown in Figs. 18 and 19. This model incorporates the observation that the distribution of an SM analog, obtained by Raman imaging under Lo–Ld phase-separated conditions in the single layer of the SM/DOPC/Cho system (Ando et al. 2015), differs from the conventional homogeneous Lo phase, as shown by a fluorescent image (Fig. 18A). The SM concentration gradient running from the central to the peripheral part of an Lo domain indicates that the distribution of SM in the Lo phase is nonuniform, suggesting that nano-subdomains may be unevenly distributed inside the Lo area.

Fig. 18.

Unnatural SM analogs suggest a Lo domain containing nano-subdomains, with hypothetical illustrations of nano-subdomains consisting of SM molecules. (A) Raman microscopy image (top) of an SM-analog/DOPC/Cho monolayer, demonstrating the heterogeneous distribution of the Raman probe of SM (Ando et al. 2015). However, in the fluorescence image of the same frame (bottom, scale bar 10 μm), the black area did not contain the fluorescence probe, thus appearing as homogeneous domains for the SM-rich Lo phase. The top right shows the hypothetical distribution of SM nano-subdomains within the Lo domain at macroscopic scale. The nano-subdomains are tentatively drawn with a circular shape for clarity but are probably irregular (Panel B), and a linear hydrogen-bonding connection between SM amide groups in the membrane may be responsible for the formation of the nano-subdomains. (B) Line profiles of lipid rafts obtained for the Raman (top, red) and fluorescence (bottom, black) images along the dotted lines of the Raman and fluorescence images in B (red and gray, respectively) in Panel A. It is clear that Raman and fluorescence images have complementary contrasts. In the fluorescence image, the Lo domains are visualized as dark circles in which internal nonuniformity can hardly be visualized

Fig. 19.

Conventional and current proposed models for Lo domains in model bilayers show how the lipids are arranged in a Lo/Ld-phase segregated membrane. (A) Macroscopic segregation of uniformly mixed Lo and Ld domains largely composed of SM (blue)/Cho (teal) and unsaturated PC (orange), respectively, satisfying the phase rules. (B) Lo domains consisting of gel-like nano-subdomains rich in SM or saturated lipids and with small single-digit nanometer sizes, separated by interstitial lipids consisting of Cho and a small amount of Ld lipids, such as unsaturated PC. In Panels A and B, the upper illustrations show a cross-section of domain boundaries and the interior of a Lo domain containing nano-subdomains, respectively, as shown in the lower diagrams. (C) Lipid-raft-recruited membrane proteins (green). These proteins do not reside within a gel-state nano-subdomain but stay in the interstitial area, mainly consisting of Ld lipids. Note that these interstitial regions may be somewhat influenced by nearby SM nano-subdomains in terms of the thickness of the lipid bilayers and the ordering of lipid chains (see text). The figure illustrates the bilayer interaction with interleaflet coupling in the registered mode, with the highly ordered upper-leaflet lipids (dark blue) inducing an ordering of the lower-leaflet lipids (light blue)

The question we consider based on this model is: How would this subdomain structure within Lo domains influence membrane protein structure and function? Many transmembrane proteins are recruited to the raft areas and are involved in functions such as signal transduction, but the protein structure remains largely unchanged in both raft and non-raft regions. For example, rhodopsin becomes raft-philic upon dimerization (Hayashi et al. 2019) and can therefore move between phases with different thicknesses. However, the removal of Cho from rod/disc membranes by cyclodextrin (i.e., reducing the percentage of raft domains) does not greatly change the oligomeric state of rhodopsin (Hayashi et al. 2019). In another experiment in which the equilibrium constants (K_eq) between transition states MI and MII in the photocycle of rhodopsin were measured by changing the Cho concentration in rod/disk membranes from 4 to 38 mol% (Niu et al. 2002), the K_eq values varied linearly and exhibited moderate Cho-dependence; K_eq at 4 mol% Cho content was only 25% higher than that of the control at 15 mol% Cho. These observations suggest that some of the membrane proteins can move between raft-like and non-raft phases with different thicknesses without having a significant influence on their conformation due to the proteins altering the extent of hydrophobic matching and mismatching between the protein hydrophobic segments and the lipid bilayer.

This may reflect the ability of membrane proteins to alter the width of their local lipid environments (Killian 1998). A pioneering study by Caffrey and Feigenson (1981) proposed the hypothesis that calcium ATPase, when incorporated into PC membranes with transmembrane domains shorter than the membrane bilayer thickness, can maintain its active conformation by altering local membrane width to dent the bilayer around the proteins. Additionally, Krishnakumar and London (2007) systematically evaluated the stability of the transmembrane structure of model peptides and proposed that denting of the phospholipid bilayer allows for the relatively stable transmembrane incorporation of short α-helical peptides. This behavior may reduce the effect of different thicknesses of bilayers in raft-like and non-raft phases on the structure and function of integral proteins.

However, it is unlikely that lipids in a highly ordered state would readily stretch or compress to accommodate a mismatch between membrane protein hydrophobic domains and the bilayer. If so, how can proteins retain their structure and function when moving from Ld to Lo domains, given the difference in properties between Lo and Ld domains, and especially the difference in properties between Ld domains and SM gel-type subdomains? The hypothetical protein–lipid interactions shown in Fig. 19C illustrate possible lipid arrangements in lipid rafts that may make this possible; that is, considering that a raft comprises nano-subdomains of outer-leaflet Lo lipids, such as SM, the bilayers of which are in the gel phase at biological temperatures, the transmembrane protein would probably be excluded from the nanosized gel state, in which the acyl chain packing is most compact, and instead reside in the interstitial area. Since interstitial lipids are largely non-raft (Ld) lipids, the composition of a protein’s annular lipids (the lipids in contact with the hydrophobic surface of the protein) will resemble that of the non-raft region, which is a favorable environment for proteins to avoid substantial structural changes upon entering the raft region.

Nevertheless, the physical properties of interstitial Lo lipids are unlikely to be the same as those in the Ld state. It is possible that Ld lipids adjacent to Lo domains are stretched to minimize hydrophobic mismatch between lipids at the Lo–Ld boundary and thus avoid a high free energy (line tension) between the Lo and Ld domains (Cordeiro 2018; Nelson et al. 2010). Thus, it is likely that the interstitial Ld-favoring lipids found within Lo domains, which are adjacent to the gel subdomains, would stretch to approach the matching gel lipid bilayer width. Consequently, proteins in Lo domains would find the portion of the protein in the outer leaflet in an environment of Ld lipids with properties somewhat intermediate between the gel state SM and Ld domains. It is unclear how lipid asymmetry would alter this picture. Inner-leaflet (i.e., cytoplasmic leaflet) lipids are more flexible/disordered than Lo lipids, including SM (Lorent et al. 2020), but an increase in inner leaflet ordering by registered interactions with outer leaflet lipids would not greatly modify the picture from that for symmetric membranes.

Thus, the substructure of Lo domains may be one of the reasons that membrane proteins can reside in both the raft and non-raft phases without greatly altering their structure and function. However, this does not mean that the protein structure or local protein concentration would be exactly the same in the Ld domains and the interstitial lipids of the Lo domains. The differences in membrane width between the Ld and Lo domains may control the partition coefficient of membrane proteins for Lo domains, meaning that a protein prefers the lipid environment with least mismatch (Lin and London 2013; Lorent et al. 2020; Nelson et al. 2010). This implies that differences in bilayer width in the Ld domain and the Ld-lipid rich interstitial region of Lo domains control protein localization. Differences in bilayer width in these regions could also alter protein conformation when a protein with transmembrane helices has two conformations with different helix tilts and activities. If the helix tilt differs for a single helix by itself or when interacting with different helices, then different helix tilt in Ld domains and interstitial regions could also alter the tendency of proteins to interact via their transmembrane helices.

Finally, we consider whether the characteristics of SM nano-subdomains are especially suitable for biological functions. It is known that saturated PC and Cho form similar Lo phases to SM and Cho, but there are differences between them, as described above; the formation of SM nano-subdomains is largely due to SM-specific intermolecular interactions, which are less affected by other lipid molecules such as Cho. The size and fraction of SM nano-subdomains in biomembranes may also be relatively unaffected by local lipid composition, as SM homophilicity reduces lipid exchanges between SM nano-subdomains and the interstitial area (i.e., the Lo regions between gel-state subdomains). Comparing the effect of SM on interstitial lipids with saturated PC under the subdomain formation conditions, the lipid composition of the interstitial portion may also be kept more constant in SM-containing membranes.

Various sphingolipids, including SM, are known to have much longer chains than PCs, such as lignoceroyl or nervonoyl (C₂₄-fatty acyl) moieties. Therefore, by interdigitation of long SM acyl chains into the opposite leaflet, the nano-subdomains formed by SMs with a long acyl chain might specifically transmit cellular signals through interleaflet interactions. Other differences between SM and PC in the formation of nano-subdomains in asymmetric membranes (e.g. due to differences in domain size) might further account for the unique biological function of SM, but this remains to be investigated.

Author contribution

Michio Murata and Erwin London had the idea for the manuscript. The manuscript was written by all authors, who approved the final manuscript.

Funding

Some of the research by authors was supported by KAKENHI (grant numbers 19K22257, 16H06315, 20H00405, and 21H04707) to MM and NM, and by a JST ERATO Lipid Active Structure Project grant (JPMJER1005) to MM. The research was also supported by an NIH grant (GM122493) to EL.

Declarations

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

All authors have given final approval for this manuscript version.

Conflict of interest

The authors declare no competing interests.