Atomic force microscopy and other scanning probe microscopy methods to study nanoscale domains in model lipid membranes

ABSTRACT The cell membrane is a fundamental biological structure, which is only 6–10 nm thick. It is composed of hundreds of lipid types, which form small and dynamic lipid domains or rafts. These rafts are thought to be a major aspect of cell organization, to provide support for various transmembrane proteins and are central to the communication of cells with their environs. Understanding the functions of lipid rafts presents an exciting opportunity to understand the molecular mechanisms of biologically important processes, as well as to uncover fundamental molecular mechanisms of membrane-associated diseases. Due to the high complexity of cell membranes, model membranes composed of synthetic lipids have been developed and are widely used to mimic biomembranes in an effort to study the structure and dynamics of lipid domains and their role in cell function. Atomic force microscopy (AFM), Kelvin probe force microscopy (KPFM) and atomic force spectroscopy (AFS) significantly advanced the study of nanodomains in model lipid membranes and monolayers. We review applications of these methods to the study of model membranes, which are widely used to mimic eukaryotic and bacterial cells, as well as neuronal cellular membranes in Alzheimer’s disease (AD). Graphical Abstract


Introduction
Cell membranes are composed of several hundred lipid types, with many thousands of different lipids identified within complex organisms [1][2][3][4].Lipid composition varies by organism, cell type, and even cell state; thus, lipids must play a fundamental role in regulating cell function.However, the precise reason for such complexity and a full understanding of the specific functions of all these lipids remains elusive [2][3][4].The plasma membrane defines the exterior surface of cells and is a key interface for interaction between cells and the environment.In eukaryotes, lipid membranes also compartmentalize different processes in intracellular organelles.In cell membranes, lipids selfassemble into a bilayer structure, with hydrophilic head groups on either side of a hydrophobic core of acyl tails.The lipid membrane anchors proteins that perform specific tasks, such as transporting ions and activating signal cascades that inform cell behavior, among a whole host of other functions [5].The lipid membrane has often been overlooked as a passive structure to support proteins; however, it is increasingly being recognized as a complex, dynamic structure that facilitates protein and cell function.Several models have been proposed for the structure and function of the lipid membrane in biology: first was the fluid mosaic model by Singer and Nicholson in 1972 [6] later, the lipid raft model was proposed, which is now widely accepted as a more complete membrane model [7][8][9].
One fascinating property of complex lipid membranes is the lateral structural heterogeneities that occur as a result of phase separation forming lipid domains or lipid rafts [5,[10][11][12].The presence of these domains in model systems prompted the raft hypothesis which suggested that membranes are polarized in apical and basolateral domains [8].Lipid clusters, enriched with sphingomyelins (SM), glycolipids, and cholesterol (CHOL), form microdomains that act as sorting centres for proteins [8].Lipid rafts are expected to be ubiquitous in biomembranes, with evidence of their presence found in bacteria [13,14], plants [15,16], and animals [17,18].These rafts have been hypothesized to be major aspects of cellular organization, to compartmentalize membrane receptors, and are central to cell homeostasis with the environment [5,19,20] and cell signalling [17,18,21,22].The central importance of lipid rafts in cell membrane function is highlighted by their dysfunction in many diseases, including hostpathogen interactions (HIV, cholera, SARS-CoV) [23][24][25], cancer [26,27], and neurodegenerative disorders, including Alzheimer's Disease (AD) [3,28,29].Lipid rafts and domains are classified according to their degree of order and disorder, as well as their size: nanoclusters (<10 nm), nanodomains (10 nm − 200 nm), and microdomains (>200 nm) [10,12,[30][31][32][33].Though these heterogeneous membrane microdomains are widely observable in model systems [34,35], due to their small size and rapid dynamics they have rarely been observed in living cells [36,37], only recently one paper has claimed to identify nanodomains using super-resolution microscopy approaches limited to domains of resolution between 50 and 200 nm [38].
The thermodynamics of phase separation into domains depends on many factors, such as lipid-chain lengths, degree of chain saturation, headgroup identities, lipid melting temperature, and membrane lipid composition, as well as environmental conditions like temperature and the aqueous environment.Phase separation into domains occurs when two phases occur simultaneously; thus, the phase transitions within these systems are important for understanding the overall-phase behavior.The interfacial energy at the boundary between domains, the so-called 'line tension', is a key parameter in the theory of domain formation, which is related to hydrophobic mismatch between domains [39,40].Hydrophobic mismatches can occur when lipid membranes are composed of lipids with different chain lengths or when a mismatch between hydrophobic thicknesses within the lipid membrane, such as those between disordered (predominantly phospholipid) and more ordered domains (that contain sterols), as illustrated in Figure 1.This principle of hydrophobic mismatch is also important for protein-lipid interactions and the theory of functional lipid rafts [41].Van der Waals and electrostatic forces in lipid-lipid interactions are also important factors in phase separation; the prime example is the interactions between sterols and sphingolipids.With a single-component lipid system, a transition between the solid or gel phase and liquid phase occurs at a well-defined melting temperature (T m ).In general, T m increases with acyl tail length and degree of saturation.In multicomponent systems, with two or more lipids that have different melting temperatures, liquid-phase coexistence occurs, with the two phases being called liquid disorder (L D ) and liquid order (L O ) phases.This liquid-phase coexistence occurs when the system is between the two melting temperatures.When two different melting temperature lipids are combined with cholesterol, L D and L O domains exhibit liquid-phase coexistence [42].The liquid state is typically characterized by the lateral and rotational dynamics (i.e. the lateral diffusion), which are higher in L D than L O states, than in the solid/gel phase [43].In the case of sterols, like cholesterol, these have a tendency to prevent gel phase transitions at low temperatures but stabilize the ordered domains at higher temperatures [44].
Understanding the structure of lipid rafts in living cells therefore presents a challenge to biophysical techniques, as well as an exciting new opportunity to understand the molecular mechanisms of important cellular functions and processes.Due to the high complexity of natural cell and intracellular membranes, models composed of synthetic lipids have been developed to mimic biomembranes.Model membranes with controlled lipid composition form micro-and nano-domains that can be resolved by microscopy techniques in either bilayer or monolayer models.Imaging studies of supported lipid bilayers (SLBs) and Langmuir-Blodgett (LB) monolayers by atomic force microscopy (AFM) have revolutionized basic and applied scientific research pertaining to lipid membranes [35,45].These studies have shed light in areas of understanding fundamental membrane biophysics and mechanisms of disease impacting important health, social, and economic challenges associated with unsolved problems, including the increasing epidemiological burden of Alzheimer's disease, antimicrobial resistances [46], lung surfactant and respiratory diseases [47], and developing new applications of lipid mimics in nano-and bio-technologies.Model lipid membranes can vary in complexity and composition, with the simplest ranging from a single lipid to binary (2-component) and ternary (3-component) lipid models.Recently, more complex models of four or more lipids have been developed, which can bridge the gap between simple models, used in many biophysical studies, to cellular studies, where lipid composition is very complex and cannot be easily controlled or determined.We review applications of AFM and related techniques to study nanoscale lipid domains in model membranes that span from simple 2-lipid component membranes to more complex membranes that mimic specific cell types and disease states, including neuronal cellular membranes for understanding the molecular mechanisms of AD.

Scanning probe imaging: AFM, KPFM and HS-AFM
Atomic force microscopy (AFM) is a powerful tool for characterizing the nanoscale structure of molecules, molecular systems, and living cells; for measuring the mechanical properties of biological systems from lipid bilayers to living cells; and as a means to measure molecule-molecule interactions, all under physiological conditions [32,[48][49][50][51]. AFM generates topographical images by means of the mechanical interaction of the scanning probe with the surface (Figure 2) [52].Forces between an atomically sharp probe and the sample surface cause a proportional bending of a micrometer-sized cantilever, upon which the tip is attached, according to Hooke's law.In contact mode, the feedback system maintains constant tip-sample forces by adjusting the height of the tip as the probe scans over the surface, providing a topographical map [52].Tapping mode is one operational mode that can reduce surface forces and minimize sample damage.In the tapping mode, the cantilever is oscillated near its resonant frequency, with the root mean square of the amplitude of oscillation held constant.This mode also has the advantage of producing phase images that can provide relative differences in the viscoelastic properties of membrane nanodomains.Standard tapping mode images provide a relative measure of probe oscillation damping that cannot be simply related to adhesion or sample compression [53].Peak force tapping mode functionality on newer AFMs isolates compressibility information on the extend portion of the oscillation and adhesion data on the retract [54].
Similar to AFM, Kelvin probe force microscopy (KPFM) relies on tipsample interactions to measure the changes to electrostatic potential over a sample surface [55,56].The electrostatic surface potential of a conductive sample interacts with the cantilever, that already has an AC voltage applied, to cause oscillations.These oscillations generate feedback against the applied AC voltage to produce the contact potential difference (or electrostatic potential) of the sample surface (Figure 2, right) [55,56].Since its inception, KPFM has been used in tandem with AFM to produce corresponding high-resolution images of sample topography and surface electrostatics [55,56].This is most commonly done using a dual pass setup: the first-pass records the sample topography, the tip is raised by a set height, and the AC voltage is applied for the second pass in order to measure the electrostatic potential of the surface.Originally, KPFM was used for conductive samples (i.e.metals or metallic nanoparticles) to determine the relative work functions of these materials, though the applications of KPFM have broadened to study the electrostatic properties of immobilized biomolecules [57,58] and lipid monolayers [59][60][61][62][63]. KPFM studies on biomembranes include studying the electrostatic properties of lipid-based gene delivery systems [62], ocular thin films [63], simple and complex neuronal models for AD [60,64], and fixed and dried PC12 cells [65].
High-Speed AFM (HS-AFM) has been developed more recently that allows for imaging dynamic processes at video rates and large surface areas; overcoming several shortcomings of traditional AFM, though tradeoffs in image resolution and sample damage can occur [66,67].Lipid bilayers have been successfully imaged with ultra-high frequency short level tapping mode HS-AFMs, first developed by the Toshio Ando group, imaging at 1 image/second or, between 100 and 200 lines/second [68].Other groups have used tapping mode HS-AFM systems, with line rates of 17.6 lines/second, to capture membrane processes such as membrane disruption by nanoparticles [54].

AFM-based force spectroscopy
Atomic force spectroscopy (AFS) is an operational mode of AFM that operates in the z-direction to measure interaction forces between the probe tip and the sample.AFS nanoindentation experiments can measure the elastic modulus and adhesion of biological systems.When applied to lipid bilayers, the AFS measures membrane breakthrough forces (schematic in Figure 3) [69,70] and other nanomechanical properties of membrane domains that can be correlated with membrane topography (example in Figure 4) [71].Membrane breakthrough forces are correlated with membrane order; therefore, lipid rafts have higher breakthrough forces [72].Figure 3 shows a representative force curve.During the approach (extend curve in red) the probe comes into contact with the bilayer, causing the cantilever to begin deflecting and compressing the membrane (point A).Fitting the contact region of the force curve (point A -B) to an appropriate model (green line), such as Hertz/Sneddon model, allows for the estimation  of the membrane elastic modulus, a measure of fluidity/stiffness.The breakthrough force is defined as the peak force following the contact region (point B).The length of the contact region is the indentation depth, while the length between the contact point (point A) and the surface of the mica substrate (point C) is the bilayer thickness.On the retract curve (blue), adhesion between the probe and the membrane is measured by the peak adhesion force (point D) and the work of adhesion (the area under the curve at E). Membrane adhesion is important for understanding binding to the lipid bilayer; here probe properties need to be considered for interpreting the adhesion data.
Broadly, there are two theoretical models that can be used to describe membrane breakthrough forces, continuous and discrete models.In each, there are several general assumptions: that the breakthrough event is instantaneous, that it is a statistical process, and that membrane rupture requires overcoming an energy barrier in order to rupture [73,74].The probability P of observing a breakthrough force F under constant velocity conditions can be described using the continuum nucleation model: where A can be approximated as the resonance frequency of the cantilever, R is the radius of the tip, K is the spring constant of the cantilever, k B is the Boltzmann constant, T is the temperature, S is the effective spreading pressure of the bilayer, and Γ is the line tension associated with the hole created by the tip.All of these factors are constants, or can be approximated as such for a given experiment, with the exception of the line tension Γ and spreading pressure S.These factors can differ as a function of lipid phase, as they are dependent upon molecular packing density and lipid headgroup identity [75]; therefore, a variation of breakthrough force between nanodomains can be observed (Figure 4) [72,76].

Other scanning probe methods for studying lipid membranes and nanodomains
In recent years, the combination of AFM with optical and spectroscopic techniques has allowed for impressive spatial resolution beyond the diffraction limit of light.In tip-enhanced Raman spectroscopy (TERS), a visible laser is focused on the tip-sample junction, and the inelastically scattered light is analyzed to measure the Raman spectrum of the sample at the nanoscale.TERS has been applied to the analysis of segregated lipid domains of a DOPC/ DPPC monolayer [77], and the mapping of streptavidin interactions with a DOPC/POPS/Biotin-PS bilayer [78].Recently, STM-TERS has been employed to map the phases of a DPPC monolayer [79].Similarly, the combination of AFM with infrared spectroscopy (AFM-IR) has enabled nanoscale chemical mapping of infrared absorption over the surface of a sample.AFM-IR has been used to chemically map the distribution of sphingomyelin, cholesterol and cyclosporin A in a Langmuir monolayer [80].Beyond the chemical identification of different lipid phases, further applications include the analysis of protein-containing lipid membranes [81] and the interaction of model lipid membranes with amyloid-β as a function of oligomeric state and lipid phase [82].The integration of AFM with waveguide TIRF microscopy has also been used to study the ion channel structure and function in reconstituted lipid membranes [83] this in principle could be used to study the influence of nanodomains in the future.

Sample preparation of monolayers and SLBs for AFM imaging
Langmuir-Blodgett (LB) troughs are used to produce monolayers for the study of lipid membrane structure [84].Lipid monolayers represent a single membrane leaflet and can exhibit domain structures that depend on the surface pressure, lipid composition and temperature [85].Lipid monolayers can have different biophysical properties (miscibility-phase transitions) than in vesicles of the same composition, depending on surface pressure; often bilayer equivalent pressures near 35 mN/m are chosen to mimic the native bilayer.The supported monolayer for imaging is produced by pulling a mica plate through a lipid film at an air-water interface, with the lipid headgroups solubilized in solution, thus orienting the tails normal to the surface in the air, depositing the lipids with headgroups bound to the mica by hydrophilic interactions.These monolayers are amenable to both AFM and KPFM imaging in air to map both the topographical structures and the electrostatic surface potential of the sample, respectively (Figure 2), which can help to understand the electrostatic properties of lipid membranes [59].
Care must be taken with unsaturated lipids prone to oxidation [85].Lipid monolayers are not as physiologically relevant as lipid bilayer models due to the single leaflet and exposed acyl tails rather than lipid headgroups, which would be on the surface of the lipid bilayer under physiological conditions.Resolving nanodomains in monolayers is more straightforward compared to bilayers because one can eliminate the overlap between the domains that can be present in each bilayer leaflets, which is known to give higher complexity of domains, such as those that are out of register [85].
Lipid bilayers are held together by hydrophobic effects; therefore, they must be maintained under fully hydrated conditions.Production of a supported lipid bilayer (SLB) for AFM imaging is typically done by vesicle fusion (AFM imaging in liquid shown in Figure 2).Unilamellar vesicles are prepared by the sonication or extrusion of a multilamellar vesicle solution [86].For the sonication method, ultrasonic sound waves are passed through the lipid vesicle solution, breaking large multilamellar vesicles into smaller vesicles with less lamellar structure.Typically, sonication produces small unilamellar vesicles between 30 and 50 nm in diameter [86,87], though vesicles of approximately 150 nm have been reported [88,89].Extrusion is another method for producing unilamellar vesicle solutions, where the multilamellar solution is passed repeatedly through a filter under pressure, increasing the monodispersity of vesicles.These vesicles are approximately the size of the filter pore, between 50 and 150 nm.Vesicle fusion itself is a multistep process where vesicles bind, adsorb, rupture, and spread on the surface.This process depends on lipid composition, temperature, pH, solutes in the vesicle media, vesicle size, and concentration of vesicles [46,69].During vesicle fusion, predominantly anionic lipid models are difficult to fuse with the mica surface due to repulsion with the negatively charged mica surface, which can be overcome using divalent cations in solution or chemically modifying the mica surface increasing positive charge [90,91].

Differences between model membranes and live cell membranes
As mentioned previously, cellular membranes are composed of complex, species-specific lipid mixtures, with multiple membrane proteins and other associated molecules dispersed throughout [92][93][94][95].As the exact lipid composition of cellular membranes is challenging to discern, reproducibility of samples and the control of physicochemical membrane properties is extremely difficult to achieve [92].In addition, while commercial lipid extracts from biological cells can be used in studies, these are oftentimes cell-and species-specific, with the possibility of altered lipid composition and lipid biosynthesis due to the specific culturing environment [93][94][95].For this reason, despite the trade-off of reduced complexity, biomimetic lipid model membranes are used in studies to circumvent many of these issues as they are easily accessible, reproducible with known lipid composition and allow more control over physicochemical membrane properties [92].These simplified, consistent lipid membrane models can be easily studied and imaged alone or analyzed for their interactions with other biomolecules, proteins or nanoparticles [96,97].For lipid raft studies in particular, biomimetic lipid models are chosen as the micro-and nano-domains are quite large and can be easily observed under various conditions.

Designing model lipid membranes and choosing lipid composition
Choosing the right model is the first step in a scanning probe microscopy study of model membranes.Lipid composition influences the properties, structure, and function of the cell membrane.The physical properties of lipids, such as length and saturation of acyl tail, headgroup identity, lipid electrostatic properties, and lipid geometry are important factors for phase separation and the structure of membrane lipid rafts [5,10,11].The degree of acyl tail order correlates with tail packing density and the size and number of ordered (gel phase) domains or lipid rafts [12,44,98].The length of the acyl tail also affects membrane structure: longer tails induce more order and are thus enriched in lipid rafts, and increase the average bilayer thickness [39].Sterols are an important class of lipid for regulating the structure of membranes, in general stabilizing lipid rafts by increasing the tail packing density, and ordering parameters of the lipid tails.Cholesterol is the sterol found in ordered lipid rafts of animal cell membranes [31,[99][100][101], while phyto-and ergo-sterols are found in plant and fungi, respectively [102][103][104].The choice of lipid model is important to consider for designing experiments; types of considerations include recapitulating different membrane properties (ordered vs. disordered, charged vs. uncharged), regions of the phospholipid membrane (raft vs. non-raft, inner vs. outer leaflet), and type of membrane (cell membrane vs.mitochondria, mammalian vs. bacterial).

Simple lipid models
Dipalmitoyl-phosphatidylcholine (DPPC) and dioleoyl-phosphatidylcholine (DOPC) are the most common simple models used to mimic many types of membranes, since they contain the two most widely distributed fatty acids found in nature, palmitic and oleic acid [105].Phosphatidylcholine (PC) headgroups are zwitterionic and often enriched in the outer leaflet of eukaryotic cell membranes [106], in contrast to amino-phospholipids like phosphatidylserine (PS), phosphatidylethanolamine (PE), and phosphatidylglycerol (PG), which are enriched in the cytosolic leaflet [107,108].This makes PC lipids the most suitable single lipid model for the extracellular membrane surface.DPPC is a fully saturated phospholipid often used to model ordered domains, while DOPC has two unsaturated fatty acyl tails, thereby modeling disordered regions or fluid-phase domains [32,69].DOPC is not a naturally occurring lipid and thus is less physiological; to improve the relevance of model membranes, DOPC can be substituted for POPC, one of the most common lipids found in nature [82,109,110].However, POPC has one unsaturated and one saturated acyl tail and thus it has a closer miscibility to DPPC and does not induce fluid-phase coexistence with DPPC, even in the presence of cholesterol [42].When DOPC and DPPC are combined into binary lipid systems, they exhibit phase separation into micro-and nano-domains, mimicking lipid rafts.

Lipid models to mimic mammalian cell membranes
Cholesterol is the primary sterol in animal and human cells; therefore, any model lipid membrane containing cholesterol could be considered a mammalian mimetic model.Cholesterol condenses membrane lipids, which in turn increases tail packing density and chain ordering; induces liquid to gel-phase transitions; locally stiffens the membrane; and increases membrane breakthrough forces as measured by AFS (Figure 4) [31,71,99,100].Cholesterol levels correlate with ordered lipid domain and lipid raft size [31,72,111].Sphingolipids interact preferentially with cholesterol to form rafts due to electrostatic interactions with the amide bond of the sphingosine, of which SM, ceramides, and gangliosides are important [31,99,100].Lipid headgroups in general play a role in raft structure; for example, the headgroup of gangliosides, an important glycosphingolipid in animals, exhibit dipole interactions with zwitterionic PC headgroups contributing to lipid raft stability [11,112].Lipid charge influences interactions within the membrane and with molecules in the extracellular space and intracellular compartments [60,113,114].Cell membranes generally have a net negative charge due to the presence of PS and PG in the inner leaflet; and thus, mixtures including these lipids can be used to give the model the surface charge more representative of in situ lipid membranes, or for modeling the intracellular leaflet [115].
Common complex mammalian cell membrane mimics include mixtures of DOPC/DPPC/CHOL, DOPC/SM/CHOL and POPC/SM/CHOL at various ratios [71,101,116,117].DPPC/POPC/CHOL membranes do not exhibit observable domains due to the similar miscibility between POPC and DPPC, so are not as useful for studies on lipid rafts [42].Therefore, POPC/ SM/CHOL or DOPC/SM/CHOL models are more widely used for studying lipid rafts [31,82,117].For example, to model red blood cells (or erythrocytes) POPC alone, POPC/CHOL (at 1:0.4 ratio), SM/POPC (at 1:1 ratio), and SM/POPC/CHOL (at 1:1:1 ratio) were used for studying membrane targeting toxins [118].Other Erythrocyte models have included POPC/ CHOL (7:3), POPC/SM/CHOL (1:1:1), and POPC/POPE/CHOL/SM to study the insertion of antimicrobial peptides in mammalian lipid bilayers [117].Another common lipid model is extracted from erythrocyte membranes, which are used for studying different biological phenomena.These extracted erythrocyte membranes have been used to present evidence of lipid rafts in situ, using AFM [119].To mimic the cell membrane of the eye lens and study the influence of oxidative stress, PLPC/SM models at different ratios were studied by AFM and mass spectrometry [120].

Lipid models to mimic neuronal cell membranes
Neuronal models stand-alone as they are more complex and specific to mimic neuronal cells.They are extensively used in membrane biophysics of brain physiology due to the high relevance of the lipid membrane to many neurodegenerative disorders, such as Alzheimer's disease (AD).This includes AFM and AFS studies of the interaction between different neuroactive chemicals with lipid membranes, such as anesthetics with simple dimyristoylphosphatidylcholine (DMPC) and DPPC membranes [53,121], serotonin with POPC/SM/CHOL and POPC/POPG/CHOL models [122,123], melatonin with DOPC/DPPC/CHOL model (unpublished results) [124], neurosteroids with DOPC/SM/CHOL bilayers [125], and ethanol with DOPC/DPPC/CHOL and DOPC/SM/CHOL bilayers [126].All of these studies demonstrate how small molecules can affect domain structure, generally increasing the proportion of disordered domains and disrupting membrane rafts.AFM studies on the effects of amyloidogenic proteins on lipid bilayers have been performed in the context of neurodegeneration, such as in AD and Huntington's disease.For modeling neurons, single lipid mixtures of DPPC, DOPC or POPC; binary mixtures of DOPC/ CHOL and DPPC/DOPC; ternary mixtures of POPC/SM/CHOL, DOPC/ SM/CHOL, and DPPC/DOPC/CHOL; and more complex 4-and 5-component lipid systems that include these lipids, along with gangliosides, have also been used (POPC/SM/CHOL/GM1 [82], DPPC/POPC/SM/CHOL/ GM1 [64], and DSPE/DPPC/SM/CHOL/GM1 [127]. Not only have specific cell types been modeled, but membrane models that mimic changes in lipid composition due to AD have been developed.In our lab, Drolle et al., developed a complex model of lipid membranes that mimic healthy, early-and late-stage AD using 5-component mixtures of DPPC/POPC/CHOL/SM/GM1 at different ratios (images shown in Figure 5).The healthy neuron model was composed of 37:37:10:10:6 lipid ratio.Progressing to the early AD model GM1 was reduced from 6% to 2% by weight, then the late disease model, SM levels were reduced from 10% to 6% both while maintaining constant cholesterol content [64].These changes in lipid ratio correlate with reductions in ganglioside and SM levels in the AD brain.Even these small changes in these lipid proportions dramatically affect lipid monolayer domain structure and electrostatic properties, as revealed by AFM and KPFM imaging shown in Figure 5, and have been shown to play an important role in membrane interactions with amyloid peptide, to be discussed in more detail later [64].
Finally, brain total lipid extracts (BTLE) are commonly used to study many aspects of molecular neuroscience, including neurodegeneration mechanisms.However, the specific structure of these lipid bilayers, such as the presence of nanodomains, depends on the species-specific lipid composition [95].BTLE extracts from commercial bovine sources often do not exhibit clear nanodomains on AFM images [128,129].Moreover, BTLE membranes from mice did not form observable nanodomains, while BTLE extracts from mole-rat brains did exhibit nanodomains by AFM [95].The major limitation of using BTLEs, or other tissue extracts, is that lipid composition is not known, therefore it is difficult to study domains and rafts, especially to understand the contributions of lipid composition to nanodomain formation and membrane structure.

Mitochondrial, cardiolipin, bacterial, fungal and plant membranes
Model membranes of POPE/CHOL (4:1) were used to mimic the inner mitochondrional membrane where AFM imaging showed that cytochrome c preferential bound to the higher, more ordered cholesterol domains [90].Isolated inner and outer mitochondrial membranes, obtained from rat liver cells using osmotic pressure in hypotonic conditions, have also been studied by in situ AFM [130].Mitochondria, as well as several types of bacteria, contain cardiolipin in their lipid membranes.Cardiolipin is an interesting lipid, as it has four acyl tails, rather than the standard two found in most eukaryotic cell membranes.Including cardiolipin into model lipid bilayers thus increases the biological relevance to both mitochondrial and bacterial lipid membrane studies.Lipid membrane models that include cardiolipin have been studied by AFM imaging, these include a PC/Cardiolipin and PE/ Cardiolipin monolayers, which exhibit domain structural differences that depend on molar fractions of cardiolipin and PC or PE, (i.e. the induction of high curvature membrane regions resulting in multilayers and domain formation) [131,132].
Bacterial membranes are largely composed of PE and PG lipids.Model lipid membranes are used to study antimicrobial peptides, an important area of research due to the growing prevalence of antimicrobial resistance in bacteria.Often, simple models containing phosphocholines are used for AFM SLB studies, even though these lipids are not as highly abundant in bacteria, due to the reduced probability of predominantly ionic (PG) and high curvature (PE) liposomes fusing on mica as part of the vesicle planar lipid bilayer transition, as mentioned earlier.Common bacterial models include binary mixtures for modeling gram-positive bacteria with compositions of PC/PG such as DMPC/DMPG, POPC/POPG, DPPC/DPPG and POPE/ POPG [117,[133][134][135]. Cardiolipin can be added to these binary models to improve the similarity to natural in situ bacterial membranes [134,135].
Fungal and plant lipid membrane models that include ergo-and phytosterols have been used to study membrane structure and properties.Fungal lipid membrane models can also include sphingolipids, as sphingolipids are found universally in eukaryotes.Some recent fungal models for AFM studies, such as POPC/SM/Ergosterol and DOPC/DPPC/Ergosterol, are analogous to systems that are often used to study the animal membrane lipid rafts containing cholesterol [104,117].Comparison of cholesterol and ergosterol in mixed lipid membranes reveals some interesting similarities and stark differences between their structural functions in model lipid systems by AFM [136].The role of ergosterol and lipid membrane properties in yeast ethanol resistance has also been studied by AFM, which is important in industrial yeast fermentation processes [104].Finally, the application of ergosterol-containing lipid membranes for studying pore forming antimicrobial peptides (AMPs) that exhibit antifungal activity have also been explored, which has important implications for human and animal health in pharmaceutical and veterinary industries [117,136].Studies of the effects of phytosterols on plants are important for studying plant membrane biophysics and the effect of plant membranes on plant physiology.Plant model lipid membranes generally include phytosterols, phospholipids, such as DPPC or POPC, and can include plant-specific sphingolipids [137].

AFM of other lipid systems: lung surfactant, myelin, stratum corneum and meibum
Lung surfactant is an essential lipid structure in biology, which reduces the work required for the alveoli to open during inhalation: without this collapsible lipid structure the effort to breath would be substantially increased [47].This lipid structure is composed of largely DPPC (about 80%), other phospholipids and cholesterol, with several surfactant proteins [47].Surfactant proteins and lipid composition work to modulate the phase behavior, electrostatic properties, compressibility, and collapse pressure of lung surfactant with elevated cholesterol reducing the self-assembling properties of lung surfactants as revealed by AFM [138][139][140][141]. DPPC alone is often used as a model lipid for studying pulmonary surfactant: for instance, in the study of environmental aerosol toxins [142].Clinical lung surfactants or bovine lung extract surfactants, which are more physiologically relevant, are often used as surfactant models for studying environmental aerosol toxins and pulmonary drug delivery [143][144][145].
Myelin is a crucial multilamellar lipid membrane structure used for insulating the axons of neurons, which greatly increases the rate of action potential propagation in the central and peripheral nervous system.There are several neurodegenerative disorders (e.g., multiple sclerosis) associated with insufficient or loss of myelin that involve its impaired structural assembly.Myelin basic protein (MBP) is a crucial myelin protein that links together the multilamellar sheets of lipids within the myelin structure [146].An AFM study of MBP in DLPA monolayers shows that MDP increases homogeneity of domains [146].The function of myelin depends on the self-assembly of MBP into the lipid structure and properties, a process that is lipid domain-dependent as revealed by AFM [147,148].More sophisticated models of myelin have been used to study interactions with MBP that include porcine brain derived PS, SM, PC, and PE, all with heterogeneous fatty acid tail distributions, and cholesterol from ovine wool.AFM imaging revealed these five-component myelin models exhibit the coexistence of ordered and disordered membrane domains, with disordered domains taking ~ 14% of the surface area.At low concentration, MBP bound preferentially to the disordered regions, but at high concentration bound non-selectively to both domains [148].
Stratum Corneum (SC) is the outermost layer of the skin, responsible to prevent water loss and the absorption of chemicals and pathogens into the body.It is a multilamellar structure comprising dead keratinocytes, which is embedded with several skin cell types.These multilamellar structures are primarily composed of ceramides, free fatty acids, and cholesterol at approximately 1:1:1 ratio.Several Langmuir Blodgett film studies using AFM of SC and SC mimetic model lipid systems have been explored [149,150].Disease-mimetic short-chain SC models exhibit higher permeability, greater water loss, and lesser condensed lipid phases, which are correlated with a decrease in ordered domain (lipid raft) surface coverage of the Cer-16 vs. the Cer-24 model.The lipid constituents of each model were also inverted, with the Cer-16 being enriched with cholesterol in the lower domains, compared to the Cer-24 being enriched in the higher domains [149].
Meibum is a lipid film on the outermost layer of the eye and is important for maintaining ocular hydration and protection of the eye from exposure to chemicals or infection, similar to the purpose of the SC of the skin.Disrupted meibum can cause eye conditions, such as dry eye, thus the structure and properties of the meibum is important for eye health.AFM has been used to study the meibum properties using extracts from the meibomian gland secretions in LB trough monolayer imaging studies [151].A simplified meibum model lipid system (composed of six different lipid types) has also been developed, and was later studied by AFM/KPFM, confirming that the artificial meibum solution recapitulates some of the structural and electrostatic characteristics of natural meibum [63].

Anti-microbial peptides
Nanodomains in model membranes are important factors in peptidemembrane interactions as they create various target sites for binding for a wide range of peptides.For example, LL-37, a well-known human AMP, binds to bacterial model membranes and causes thinning (permeabilization) in all visualized nanodomains in a 'carpet', which acts as a precursor step to the detergent-like mechanism of the peptide [133].After 15 min of interaction with a supported microbial lipid membrane model, LL-37 uniformly binds to the membrane surface, causing membrane thinning at the edges of nanodomains (Figure 6) [133].As time progresses, these areas of thinning show characteristics of membrane disruption and lipid extraction, thought to be a direct result of the peptide-lipid complex formation [133].Conversely, some AMPs are phase-specific, such as G1OLO-L 2 OL 2 , which targets and degrades higher nanodomains in model membranes from the inside out [152] or malculatin 1.1, which avoids detergent-resistant higher nanodomains and instead binds to the disordered phase of reconstituted E. coli membranes [153].The ability of AMPs to target specific membrane sites is extremely relevant, as there is documented evidence that bacteria can restructure their membranes (i.e.properties and composition) to evade AMP activity [154].

Drug delivery systems and therapeutics targeting lipid domains
Interestingly, this relationship between peptides and nanodomains is being explored in the context of potential therapies for cancer and gene delivery systems.Li et al. (2020) developed a molecular self-assembly that inhibits cellproliferation oncoproteins by binding to lipid rafts in the membranes of cancer cells to trigger cytoskeletal reorganization and block tumor progression pathways, showing active tumor suppression in mice [155].Similarly, peptides capable of binding to and crossing cellular-or organelle-specific membranes are being explored for gene delivery systems.HS-AFM studies show that DNA complexes, combined with either a signal peptide or a cell-penetrating peptide, allow for binding to higher (ordered) nanodomains (with varying affinities) on both plasma and mitochondrial model membranes, thus enabling the creation of pores for the delivery of DNA complexes [156].Not only are nanodomains integral in mediating protein-membrane interactions, but they can also be utilized as the basis for novel therapies for a variety of conditions.

AFM for studying protein nanodomain interactions in neurodegenerative diseases
Nanodomains have attracted a lot of interest as target sites for binding and damage by specific peptides involved in neurodegeneration, such amyloid-β (1-42) in Alzheimer's disease, its counterpart α-synuclein in Parkinson's disease and Huntingtin protein [29,[157][158][159].The neuronal cell membrane is of critical importance in neural physiology [160] and is recognized as a target for amyloid attack: we and others have reported the effect of the membrane, and of lipid rafts, on Aβ binding and toxicity [161][162][163].The molecular mechanism of Aβ-induced cytotoxicity depends on accumulation at the neuronal membranes through non-specific interactions [60,64,[164][165][166][167][168].Studies in model lipid systems suggest that Aβ induces defects in the lipid membrane structure through a multistep process: 1) the binding of Aβ to the membrane; 2) the reorganization and conformational changes of the Aβ-lipid membrane complex; 3) the formation of Aβ ion channels and membrane defects that perforate the membrane.These features then trigger downstream cell dysfunction and cell death [169][170][171].These deleterious effects of Aβ depend on the Aβ isoforms and the structure of higher-order aggregates, as well as lipid membrane composition, structure, and properties, with cholesterol and GM1 being especially important [60,64,145].
Biophysical studies using AFM demonstrate different interaction mechanisms with lipid bilayers.DOPC/DPPC lipid bilayers (at a 1:1 ratio), exhibiting phase separation, show increased binding of Aβ42 to preferentially aggregate on gel-phase domains (Figure 7a, b) [168].In a different study of DOPC/ DPPC mixtures (at a 3:1 ratio), Aβ42 avoided binding to the gel-phase domains, and only interacted with the disordered phase (Figure 7c) [95].This seeming contradiction could be due to differences in buffering conditions, Aβ reconstitution methods and lipid composition.
Cholesterol is a critical membrane component, as such it is no surprise that it dramatically affects amyloid aggregation on model lipid membranes, causing topographical defects, increasing surface roughness and membrane permeability [60,64,172].AFM imaging has shown that nanoscale topographical and electrostatic features of cell membranes caused by cholesterol (in DOPC membranes) may cause preferential binding of charged Aβ residues and alignment of β-sheets for further growth into fibrils [60], which was then correlated with KPFM images of the same lipid composition in LB monolayers (Figure 8).Other studies find differences in oligomer and fibril interactions with POPC/POPS model lipid bilayers, mimicking the net negative charge of neuronal membranes, demonstrating that preformed Aβ40 oligomers cause more damage than fibrils [115].A comparison of Aβ40 monomers on pure DPPC and DOPC with and without cholesterol reveals differing modes of interaction between disordered and ordered model membranes: again, Aβ tends to accumulate on top of saturated lipids in the gel phase (DPPC) and grow laterally on the membrane surface, but Aβ can also incorporate into disordered lipids (DOPC) and solubilize the membrane by removing the lipids from their mica support.In both cases, cholesterol increases the kinetics of these interactions [45].In addition, Aβ pore formation appears to be modulated by cholesterol, with low cholesterol concentration (15%) within the membrane having increased Aβ insertion and Ca 2+ pore formation compared to no cholesterol and high cholesterol (above 30%); this effect is correlated with membrane surface roughness but not with the breakthrough force of the lipid bilayer [173].Ganglioside (GM1) and sphingomyelin also have been shown to be important for oligomeric Aβ binding (but not fibril binding) to the cell membrane, which then acts as a seed for further amyloid binding [166,167,174].As mentioned previously, we (Drolle et al.) developed a complex model lipid membranes that mimic healthy and diseased AD  membranes (images shown above in Figure 5) [64].We demonstrated that these models differ in the organization of nanoscale lipid domains in both in morphology and electrical surface potential (Figure 5), and these differences affect amyloid binding to these membranes.AFM imaging of Aβ on complex neuron and AD-specific lipid bilayers reveals how amyloid insertion and accumulation onto the lipid bilayer is time-and compositiondependent. Healthy model membranes aggregated with amyloid show surface roughness increases with time suggesting amyloid accumulates on top of the lipid bilayer.In contrast, in diseased models there were significant fluctuations in membrane surface roughness with time, indicating accumulation followed by penetration of amyloid into the lipid bilayers [64].These AFM experiments were paired with black lipid membrane (BLM) electrophysiology demonstrating that healthy model membranes are less prone to amyloid binding and Aβ-induced damage than diseased model membranes mimicking AD neurons [64].This indicates changes in lipid composition and organization of nanoscale domains are an important factor in amyloid toxicity in AD.
HS-AFM has been used to study time-dependent Aβ aggregation dynamics, structural arrangements of intermediary amyloid species, as well as Aβ interactions with lipid bilayers [82,[175][176][177][178].Structural studies on early aggregatory intermediates demonstrate that higher molecular weight oligomers tend to crowd together but, depending on buffer salt composition, may switch between different fibril structures (straight vs. spiral) [176,177].Interestingly, HS-AFM has also revealed Aβ aggregation dynamics can be influenced by high-speed scanning, causing increased fibrillization rates; this suggests that the transitory nature of Aβ intermediates promotes fibril formation and that high-speed scanning can affect the dynamics of sample processes [176].In recent years, HS-AFM studies have grown to include Aβ-membrane interaction studies [82,178].These HS-AFM Aβ-membrane interaction studies observed that the detergent-like effect of Aβ against lipid membranes was dependent on the presence of both gangliosides and cholesterol in lipid rafts and was initiated by prefibrillar small oligomers [178].In particular, a study by Feuillie et al. has demonstrated that the structure of pre-fibrillar oligomers is related to their binding strength and the resulting damage to model membranes (Figure 9) [82].Oligomers with the anti-parallel β-sheet orientation can strongly associate with model membranes, causing detergent-like effects, while oligomers that have switched to parallel β-sheet (prior to fibrillization) have a weaker Interaction with the membrane [82].The dynamics of these processes can be captured, but are likely affected by the high-speed scanning; however, to what extent HS-AFM impairs or accelerates dynamic interactions with the membrane is still unclear [82,177,178].
One proposed strategy for treating AD is to use Aβ aggregation inhibitors [51,179].In an AFM study of aggregation inhibitors, a model membrane composed of 50% DSPE, 15% DPPC, 25% CHOL, 8% SM, and 2% GM1 (by weight) was used to determine if inhibitors could prevent membrane damage [127].When Aβ and an inhibitor were incubated on the SLB surface, only shallow defects in the lipid membrane were formed, compared to Aβ alone, which caused large holes and defects in the lipid membrane [127].
AFM studies utilizing BTLE models have been used to reveal insights into the mechanisms of Aβ interactions with lipid membranes [108,164].One study combined AFM with BLM to compare interactions of Aβ with BTLE membranes and with DOPS/POPE membranes, finding different structures and ionic current fluctuations due to the differing lipid composition [108].One way to overcome the shortcomings of using BTLE models associated with unknown lipid quantities is to add additional lipids of known identity, or selectively deplete lipids.This was done by Yip et al., who added exogeneous cholesterol or removed cholesterol from the BTLE models using methyl-β-cyclodextrin [164].AFM imaging revealed the association of Aβ molecules with brain lipid extracts containing 10% exogeneous cholesterol that formed ring-like structures alongside oligomers and short fibrils.At 30% exogeneous cholesterol, no ring-like structures or large fibrillar structures were observed; however, small sparse oligomers were detected.This contrasts with normal BTLEs, where large aggregates and fibrils were detected on the surface.AFM-combined fluorescence microscopy studies in BTLE models revealed that Aβ42 increased the extent of the lipid gel phase [129].Interactions of acetylated-Aβ with BTLEs revealed structural changes to Aβ aggregates [180].With increasing acetylation there was a decrease in aggregation, revealing the importance of electrostatic interactions of lysine in fibrillization on brain lipid membranes [180].Acetylated-Aβ added to BTLE SLBs formed annular aggregates (donut or ring-like structures) that resemble pores identified elsewhere [180].
Parkinson's disease is another amyloidogenic-like neurodegenerative disease that is analogous to AD but involves the accumulation and misfolding of a different protein: α-synuclein (rather than Aβ).The accumulation of αsynuclein occurs in dopaminergic neurons, which affect movement and coordination, in contrast to the accumulation of Aβ in cholinergic neurons, which affect memory and learning in AD [158,159,181,182].Interestingly, α-synuclein has been associated with vesicle release during synaptic neurotransmission, a connection that has been proposed to be related to the high curvature nature of vesicle membranes [159].It is believed that α-synuclein affects many intracellular signaling pathways but can also be transported extracellularly to affect neighbouring cells [182].In both instances, αsynuclein interactions with the lipid bilayers, including the permeabilization and disruption of lipid membranes, may be of fundamental importance, similar to the role of Aβ in AD [159,181,183,184].AFM imaging has been used to study the aggregation of α-synuclein on model lipid membranes, including lipid raft models containing nanodomains [181,184].In DOPC/ SM/CHOL (2:1 DOPC to SM, with 5% cholesterol) containing membranes, α-synuclein lipid membrane interactions were shown to be dependent on the presence of iron ions [181], a corollary to the dependence of amyloid aggregation on cations, such as copper [185].Without iron present, the αsynuclein remained largely monomeric, binding to non-raft disordered membrane regions, thinning the membrane.Monomeric α-synuclein interactions destabilized the membrane, producing fractal-shaped defects in the lipid bilayers that were approximately one leaflet (2.5 nm) deep.It is unclear whether these defects involve the removal of lipids from the surface, if so it would suggest there was solubilization of the disordered membrane [181].In contrast, α-synuclein oligomers, which were produced in the presence of iron, accumulated on the raft-like nanodomains [181].These iron-containing oligomers were initially seeded on the ordered nanodomains (i.e.lipid raft models), which then proceeded to grow over time [181].This demonstrates the importance of the aggregation state of α-synuclein and the composition of the lipid membrane in mediating these disruptive protein-lipid interactions.
Huntington's disease is a genetic neurodegenerative disease involving aggregation of an amyloidogenic protein similar to Aβ, which can bind to and disrupt membrane structure and function.When the Huntingtin protein was incubated on BTLE SLB in the presence of H 2 O 2 , it favoured oligomerization on the lipid bilayer compared to non-oxidized conditions, which favoured fibrillization.Similar to Aβ, Huntingtin oligomers are more toxic than fibrils, indicating that an oxidative environment may promote the formation of more toxic aggregate species [186].
Another total BTLE membrane model, that was recently used for AFM study of AD, was formed from rodent brain lipid extracts (mouse and naked mole-rat) [95].Naked mole-rats are very long-lived for rodents (nearly ten times longer than mice and rats), and despite having high levels of Aβ do not accumulate plaques, making them a fascinating model for studying AD mechanisms.Naked mole-rat brain lipid composition is higher in cholesterol and lower in SM (with shorter SM chain lengths), containing similar phospholipid quantities but longer phospholipid acyl tails than mouse brain lipid extracts.Interestingly, brain lipid extracts from naked mole-rats exhibit phase separation with two separate ordered domains, whereas mouse brain lipid extracts do not exhibit phase separation behavior.Moreover, breakthrough forces were recorded for each rodent brain lipid extract model, with naked mole-rat breakthrough forces nearly 8 times higher than mouse bilayers.Aβ exposure (8 µM for 2 h) on mouse brain lipids caused defects in the membrane.These defects were cavities, about 2 nm deep, which are distinct from holes, as they do not extend the complete bilayer thickness.In comparison, the same Aβ exposure to naked mole-rat membranes results in disruption of the lipid bilayer via fragmentation, revealing the mica surface below, implying that naked mole-rat brain lipids are more susceptible to damage than mouse, despite naked mole-rats having resistance to Aβ toxicity.The authors suggest that naked mole-rats have evolved a means of protection against Aβ toxicity, not through protection of the mechanical properties of the membrane but through resistance to oxidative processes induced by Aβ [95].This is consistent with the observation that naked mole-rats have high resistance to cancer, which are known to be caused by oxidative damage to DNA [187].

Conclusion
AFM and related methods (e.g.KPFM, AFS, HS-AFM and other SPM methods) are powerful tools for visualizing the nanoscale structure and for studying the biophysical properties of domains and rafts in lipid bilayers.Lipid rafts are now considered to be an essential organizing principle of membranes that can regulate protein interactions, cell signaling, and function.Due to the small size of these domains, AFM techniques are critical for understanding membrane structure and function in normal physiology and various diseases.Biomimetic lipid membrane models have been developed to study the effect of lipid composition on nanodomain structure and function for a wide range of species, cell types and different disease states including fungi, plants, bacteria, animals, neurons, red blood cells and for AD.AFM-related methods have been used successfully to elucidate how changes induced by small molecules and proteins (e.g., in microbiology and neuroscience applications), influence membrane and lipid raft properties and structure, contributing to a better understanding of many biological processes and membrane-associated diseases.

Figure 1 .
Figure 1.Schematic of domain formation is a thermodynamic process.When the lipid membrane is below the melting temperature (Tm), domain condensation occurs.The hydrophobic thickness mismatch between the phases results in a domain height difference.This can be caused by differences in tail length and/or in degree of chain desaturation and is a result of a different Tm for different lipid types in the mixed membranes.

Figure 2 .
Figure 2. Schematic of Left: AFM imaging of lipid bilayers in aqueous medium.Right: KPFM imaging of lipid monolayers reveals the topography and electrostatic surface potential over two passes.The contact potential difference is the difference between the grounded sample and the AC voltage applied to the AFM cantilever.

Figure 3 .
Figure 3. Atomic force spectroscopy schematic for lipid bilayer breakthrough forces.Left: lipid bilayer breakthrough force schematic.Right: force plot a: contact point; b: breakthrough event; c: mica contact point; d: peak adhesion force; e: area of the retract curve is the work of adhesion.Hertz/Sneddon fit of the indentation region is shown as green dotted line.

Figure 6 .
Figure 6.AFM images of the microbial lipid model membrane a) before addition of the antimicrobial peptide LL-37; b) after 15 minutes of incubation with the peptide; c) 65 minutes after incubation with the peptide, with cross-sectional profiles extracted along the lines as shown.Domain edges are marked by blue and green markers.Reproduced from Majewska, M.; Zamlynny, V.; Pieta, I. S.; Nowakowski, R.; Pieta, P. Interaction of LL-37 Human Cathelicidin Peptide with a Model Microbial-like Lipid Membrane.Bioelectrochemistry Amst.Neth.2021, 141, 107842. .

Figure 9 .
Figure 9. HS-AFM image sequence of G37 Aβ binding to and disrupting POPC/SM/CHOL/GM1 model SLB.Following injection, the peptide destabilized and solubilized the lipid bilayer, causing it to be removed from the surface during scanning.The rate of membrane solubilization depended on specific lipids and Aβ peptide identity.Figure reproduced with permission from Feuillie et al.High speed AFM and NanoInfrared Spectroscopy Investigation of Aβ1-42 Peptide Variants and Their Interaction with POPC/SM/CHOL/GM1 Model Membranes.Front.Mol.Biosci.2020, 7, 571696.