Unveiling the nanoscale architectures and dynamics of protein assembly with in situ atomic force microscopy

2.1 Working principle of conventional AFM

AFM was initially invented in 1986 to obtain topographical information on low-conductive and nonconductive materials.^{[58, 72]} It can also image samples in various environments, such as vacuum, solutions, and air, with sub-nanometer spatial resolution for sensitive biological samples.^{[73, 74]} In the last three decades, AFM has been exhibited as a powerful imaging technique for biological samples, from molecules to living cells to better understand their assembly dynamics, mechanics, and functionality.^{[65, 75, 76]}

In brief, AFM is mainly comprised of four components: (1) a mechanical cantilever with a nanoscale sharp probe, (2) a detection system tracking the cantilever deflection, (3) X, Y, Z piezo scanners, and (4) a feedback loop (Figure 1A). When the AFM probe raster scans sample surfaces, the probe‒sample interactions deflect the cantilever. The feedback loop keeps the cantilever deflection (referred to as probe‒sample interaction force and measured by the detection system) constant (Figure 1A). Specifically, the piezoelectric element on the Z-axis either extends or retracts at each scanning point (X, Y) to keep the probe‒sample interaction force (the deflection of the cantilever) as close to the setpoint as possible. The moving distance of the piezoelectric element (ΔZ) on each scanning point is recorded as the relative height. In the end, the recorded matrix (X, Y, ΔZ) generates morphology images. Until now, optical beam deflection, developed by Meyer and Amer, has been the most successful and widely applied detection technique to monitor cantilever deflection.^[77] A laser beam is reflected from the back side of the cantilever onto a photodiode with four quadrants (Figure 1A). Slight movements of the reflected laser beam caused by probe‒sample interactions introduce potential differences in the four quadrants of the photodiode, and the vertical and torsional deflections of the cantilever are monitored and calculated based on the potential differences.

2.2 AFM operation modes

In general, AFM has four widely used operation modes: contact mode (CM), amplitude modulation mode (AM), frequency modulation mode (FM), and PeakForce tapping mode (PeakForce, detailed in Section 2.3).

CM-AFM is the original operation mode, in which the probe is in constant contact with the sample while raster scanning across it (Figure 1B). CM-AFM usually operates in the repulsive regime of the Lennard‒Jones inter-atomic interaction potential. When scanning native membrane proteins, CM-AFM can apply minimal interaction forces ranging from 50 to 100 pN.^{[62, 73]} These gentle forces allow CM-AFM to successfully characterize purple membranes in aqueous solution, such as the water channel aquaporin Z of Escherichia coli bacteria, and the Rhodopseudomonas viridis photosynthetic core complex in native membranes.^{[63, 78-80]} However, the AFM tip may exert a large shearing force when using CM-AFM to scan over isolated biomolecules on a substrate due to the feedback lag. Therefore, soft biomolecules are liable to be damaged and weakly adsorbed objects may be distorted or pushed away by the AFM tip.

AM-AFM mechanically excites the AFM cantilever at a frequency close to its first flexural resonance frequency (Figure 1C).^[81] When the AFM tip approaches sample surface, the cantilever's oscillation amplitude is dampened due to the increasing repulsive interaction force between the sample and the AFM probe. Hence, in AM-AFM, the cantilever's oscillation amplitude, instead of the cantilever deflection, is monitored by optical beam deflection (Figure 1C).^[77] The amplitude setpoint is usually close to 80%‒90% of the free oscillation amplitude to gain a strong signal-to-noise ratio, but keep the tip‒sample interaction minimal. During scanning, the motion (ΔZ) of the Z piezo is considered equal to the relative height at every scanning point. In addition, the phase lag of the cantilever oscillation to the drive signal can be recorded. It qualitatively indicates the mechanical heterogeneity across sample surfaces.^[82] AM-AFM has exhibited great potential for studying various biomolecular assemblies and complex bio-architectures.^[83-85] Möller et al. first demonstrated that native protein surfaces can be imaged with AM-AFM at molecular scale.^[86] Chung et al. demonstrated that AM-AFM can provide unique insights into protein crystal growth.^[87] In addition to imaging surface morphologies with high spatial resolution, AM-AFM can directly measure the mechanical properties of supported lipid bilayers.^[88]

FM-AFM excites the cantilever close to its resonance frequency but monitors shifts in resonance frequency due to the interaction force between the probe and sample after locking in the oscillation phase (Figure 1D).^{[89, 90]} The feedback mechanism compensates for constant change in the resonant frequency ($$\Delta {f_{\mathrm{r}}}$$). FM-AFM in air and liquid media requires precise control of piezo movement and high-speed feedback; thus, it is widely utilized in ultra-high vacuum AFM. With optimized hardware and operation, FM-AFM has been used efficiently in ambient and liquid media to measure molecular structures at solid‒liquid interfaces with sub-nanometer spatial resolution.^[90-93]

2.3 AFM-based force measurements

As its name states, AFM tries to measure and track the atom/nanoscale interactions between samples and AFM probes. In addition to imaging surface morphologies, AFM can also collect force‒distance curves as force spectroscopy (FS) and generate force‒volume maps.^[94] For example, AFM-FS measures the interaction when an AFM probe approaches a sample surface, “presses into” the sample, and then withdraws from the sample (Figure 1E). This probe movement is controlled via programmed approach-withdraw velocities and dwell time. The cantilever displacement versus probe‒sample distance is recorded and converted to measured interaction force with the Hooke's law. For instance, when an adhesion force occurs between the probe and sample surface, a downward force increases upon cantilever retraction until the adhesion bond breaks and the force versus distance curves are recorded. FS can quantify nonspecific and specific inter-/intra-molecular interactions by modifying the AFM probe and substrate surface with target chemicals/biomolecules.^[94]

AFM-FS has been used to elucidate various phenomena in biological systems, ranging from biomolecules to cells, such as the mechanical characteristics of molecular machines, protein affinity, and folding mechanism. For example, the mechanical properties of proteins, such as I band titin and titin-mimicking artificial elastomeric proteins, were determined by single-molecule AFM-FS.^{[95, 96]} In addition, the receptors on living cell surfaces were successfully mapped by probing individual molecular interactions.^[97] During the COVID-19 pandemic, binding between SARS-CoV-2 and cellular angiotensin-coverting enzyme 2 (ACE2) receptors was evaluated by AFM-FS, providing valuable information for assessing the preclinical potential of a new drug.^{[98, 99]}

Another widely used AFM-based force measurement approach is PeakForce tapping mode. It was developed to obtain force‒distance curves at kilo-Hz speed by modulating the Z-scanner movement with a sinusoidal wave.^[100] Within a modulation cycle, the maximum repulsive force, known as the peak force, is reached when the AFM probe approaches the sample surface. The sample morphology was measured by controlling Z-scanner movement with the peak force as the set point. After calibration, this process enables to extract samples’ mechanical information, such as Young's modulus, deformation, adhesion force, and energy dissipation, from the force‒distance curves recorded at each scanning pixel. It is efficient to obtain quantitative nanomechanical maps and correlate them with protein structures.^{[101, 102]}

2.4 High-speed atomic force microscopy

Although conventional AFM can capture high-resolution images of biomolecular hierarchical structures and phase transitions, tracking the dynamics of most biological events that occur at (milli) second time scale is challenging. To break through the temporal resolution limit, HS-AFM was invented around 2008 with the development of various hardware components and control approaches, such as ultra-short cantilevers, optical beam deflection detectors optimized for small cantilevers, fast scanners, fast deflection-to-amplitude converters, active vibration damping techniques, and dynamic feedback controllers.^[103-111] Reviews by Ando et al. have discussed the working principle and design of HS-AFM in detail.^{[103, 112, 113]}

In addition, HS-AFM can integrate multiple bio-functional modules and characterization methods. For example, bio-enhanced HS-AFM was developed by Miyagi et al. with a buffer exchange perfusion system and a pulse UV laser system.^[68] This allows for the exchange of buffer and uncage-caged compounds to mimic various environmental conditions and stimuli, including homeostasis shift, sudden cation flux into the cytoplasm in a membrane lesion, and upregulation of photon influx that stimulates photosynthetic proteins. Combining HS-AFM with temperature control offers opportunities to study the temperature-dependent biological processes, for example, membrane ripple transition.^[114] Furthermore, the successful development of probe-scan HS-AFM has opened new opportunities to combine HS-AFM with various optical techniques, such as optical tweezers and super-resolution fluorescence microscopy techniques.^[115-118] An integrated system of tip-scan HS-AFM and total internal reflection fluorescence microscopy can simultaneously collect the topographic AFM images and single-molecule fluorescence images at 5 fps.^[119]

Three imaging modes are commonly used in HS-AFM: AM mode, FM mode, and off-resonance tapping (ORT).^{[74, 103, 120]} AM/FM-HS-AFM is based on tapping mode, where the AFM tip oscillates vertically and periodically probes a sample surface during scanning.^[121] The basic configuration and principle of the AM-HS-AFM instrument are similar to those of conventional AFM setups (Figure 2A).

ORT HS-AFM based on force‒distance curve imaging modes with cycle-to-cycle feedback was recently established by Nievergelt et al.^{[120, 122]} They presented an advanced ORT imaging mode using photothermal actuation called high-speed photothermal ORT (HS-PORT). In HS-PORT (Figure 2B), the top and bottom layers of the AFM cantilever are generally coated with different materials that have different thermal expansion coefficients.^[123] When exposed to a modulated laser beam, the coating surfaces (top and bottom surfaces) produce differential thermal expansion, which causes the cantilever to bend, referred to as the bimorph effect, enabling the actuation of the AFM probe for imaging or manipulation purposes. In ORT, the achievable imaging speed is restricted by the cantilever oscillation rate in a controlled manner. The oscillation rate of the cantilever moving up and down is generally smaller than its resonance frequency. In conventional ORT instrumentation, the motion is manipulated by a piezo that moves either the sample or the cantilever chip. In HS-PORT, only the cantilever and probe need to drive when utilizing photothermal actuation; hence, the mass is much smaller. This allows force-curve speeds at several hundred kilohertz, limited only by the HS-AFM cantilever resonance.

HS-AFM can capture images with high spatiotemporal resolution at ∼1 nm in lateral resolution, ∼0.1 nm in vertical resolution, and ∼100 ms in temporal resolution. It has emerged as a valuable tool to understand the behaviors and functions of proteins at single-molecule level, where bioprocesses occur on a time scale of hundreds of milliseconds to minutes (Figure 2C).^{[68, 114, 124, 125]} However, numerous molecular processes are still too fast to be resolved by HS-AFM. To investigate molecular dynamics at the millisecond or even microsecond scale, enhanced HS-AFM techniques such as high-speed AFM line scanning (HS-AFM-LS) (Figure 2D) and high-speed AFM height spectroscopy (HS-AFM-HS) (Figure 2E) have proven highly potent.^[126-130] By fixing one scan axis as zero (X- or Y-direction) and repeatedly scanning over a single line [(X, Z, t) or (Y, Z, t)], HS-AFM-LS supplies sub-millisecond to millisecond temporal resolution (Figure 2D). The generated line scan images show the sample height profiles (Y-axis) versus time (X-axis). Thus, line scan kymographs can reveal the biomolecular structural transition dynamics with angstrom precision.^[131-140]

Inspired by fluorescence spectroscopy, Heath and Scheuring developed HS-AFM-HS by holding the AFM probe at a fixed X‒Y position and monitoring the tip height fluctuation in Z-direction with Angstrom accuracy (Z, t) and ∼10 µs of temporal resolution (Figure 2E,F).^[127] The probe sonicating resonance frequency and feedback loop time delay mainly determine the temporal resolution. HS-AFM-HS allows to directly detect the unlabeled molecules motion underneath the AFM probe and simultaneously measures surface diffusion rates, concentrations, and oligomer sizes.^[127]

In addition to image modes, HS-AFM-FS was developed by Rico et al. by using a short and soft cantilever (0.035 pN µm⁻¹ s⁻¹) and a miniature piezoelectric actuator.^[142] This provides a wide range of pulling velocities from 0.0097 to 3870 µm s⁻¹ (six orders of magnitude), and microsecond time resolution.^[142] The pulling speed is ∼2.5 orders of magnitude faster than that of conventional AFM-FS and reaches the limit of molecular dynamics simulations. Thus, HS-AFM-FS bridges the gap between simulation and conventional low-velocity FS experiments; for example, it allows to directly compare experimental and simulated DNA and protein unfolding forces.^{[142, 143]} In addition, improving protein unfolding experiments to the microsecond time range is vital for studies of biomechanical processes such as receptor‒ligand unbinding, lipid membrane dynamics, and mechanical protein folding and unfolding.^[144-147] HS-AFM-FS has revealed the multistep unfolding mechanisms of titin I91 protein: the protein fluctuates between the native and intermediate states under moderate forces. Still, it remains in the intermediate state under increasing forces.^[142] The ability to directly compare experimental observations with simulations is essential for studying biomechanical processes.

3.1 Investigating protein assembly on inorganic surfaces

Protein assembly on inorganic surfaces is highly relevant in design of biodevices and protein‒mineral composites. In addition, the protein's patchiness and tunable intermolecular interactions can be harnessed to engineer assembly dynamic and multiple phases with emergent capabilities akin to those of living systems.^{[153, 154]} Insights from protein assembly can guide the design of biomimetic materials with organizational sophistication and responsiveness for emergent applications. AFM enables the observation of protein assembly and disassembly processes at high spatiotemporal resolution, provides insights into the protein dynamics under varying environmental conditions, and sheds light on the intricate mechanisms governing their assembly into hierarchical structures across scales.^[155-162]

3.1.1 L-rhamnulose-1-phosphate aldolase on mica

Recently, Zhang et al. reported a modified functional protein as a highly patchy building block that assembles into variable 2D lattices via tunable interactions (Figure 3).^[149] The engineered protein, L-rhamnulose-1-phosphate aldolase with Cys residues at its corners (^C98RhuA, Figure 3A), was modulated into four different patterned 2D crystals (Figure 3C–F). AFM imaging and simulation show that physiochemically distinct growth pathways are controlled by the modulated intermolecular interactions (Figure 3G). ^C98RhuA protein self-assembles into 2D p4212 crystals in solution (pathway 1) via the formation of intermolecular disulfide bonds. In addition, because the N-terminus and C-terminus of the ^C98RhuA molecule have opposite charges (Figure 3A), the neighboring ^C98RhuA molecules need to adopt an out-of-plane antiparallel (up‒down) arrangement to avoid the out-of-plane dipole moment accumulation (Figure 3C,G).

On the contrary, ^C98RhuA assembly on negatively charged mica could compensate for the accumulated dipole moment by modulating electrostatic interactions between the anisotropically charged ^C98RhuA and mica at different concentrations of K⁺. ^C98RhuA forms close-packed 2D crystals in p4 symmetry with the positively charged N-terminus facing mica in 5 mM K⁺ solution because the negatively charged mica can compensate for the accumulated dipole moment (Figure 3D). When incubated in solution with >100 mM K⁺, porous ^C98RhuA crystals with the C-terminus down grow on K⁺ occupied mica surface (Figure 3E). Upon further increase in the K⁺ and protein concentrations, a ^C98RhuA bilayer crystalline forms exclusively (Figure 3F). The bilayer height is slightly smaller than two times of the monolayer thickness, indicating that tail-to-tail packing dimers form via interpenetration of the corrugated N-term faces (Figure 3B,G). The bilayer structure is formed by self-templating growth from the underlying p4-symmetric monolayer. The underlying layer subunits undergo a 22.5° rotation to bring the second-layer monomers close enough for disulfide bonding (Figure 3G).

The results show the mechanism of the patchy protein assembly into alternate 2D crystalline structures by modulating protein‒protein and protein‒substrate interactions. The interactions, including electrostatic, entropic, dipole‒dipole, etc., can be manipulated. If properly exploited, the building blocks with unique sensitivity and angular dependence, such as ^C98RhuA, could allow external fields to direct multi-phase assembly in solution, potentially realizing the template-free self-assembly of tunable biomaterials.

3.1.2 Septin assembly on mica

Self-templating protein assembly is observed not only in ^C98RhuA assembly on mica but also in septin assembly, an essential family of guanosine triphosphate (GTP)-binding proteins. Septins are highly conserved among eukaryotic cells, functioning as a scaffold recruiting various proteins in cell polarity, spore formation, and cell division.^[163-166] They are P-loop-NTPase proteins, in which a shared globular core is opposite in the polybasic domain—a highly conserved GTP-binding domain flanked by the C-terminus and N-terminus.^{[163, 167]} The septin rod in Saccharomyces cerevisiae is composed of four septins by forming a linear palindromic hetero-octamer: Cdc11-Cdc12-Cdc3-Cdc10-Cdc10-Cdc3-Cdc12-Cdc11.^[168] The interactions between adjacent septins dominate with alternate GTP interface and the N-terminal and C-terminal helices interfaces. The octamer rod is a basic unit for further assembly into filaments, filament pairs, and other higher order structures.^{[169, 170]} In vivo, septin filaments and further assembled bundles are present at sites of cell division.^{[171, 172]} They act as scaffolds to recruit and interact with the actomyosin network that drives membrane ingression.^{[173, 174]} The fundamental polymerization mechanism of septins is essential for biophysical properties and functions, yet their assembly mechanism is still unclear.^[175] Jiao et al. used HS-AFM and kinetic modeling to explore septin filament assembly mechanisms and formations of other higher order structures.^[150]

In agreement with previous studies, only the subunits and rod fragments (<32 nm) of septins could be observed in the absence of monovalent ions (Figure 4A,B).^{[169, 176, 177]} In contrast, objects smaller than septin rods were rarely found in a salt solution (Figure 4C). After incubating septin on mica surface in high salt concentration, septin rods quickly assemble into long filaments with lengths up to several micrometers, and the filaments are readily paired (Figure 4D). Septin filament assembly kinetics were investigated with HS-AFM. Rapid surface binding, filament elongation, and pairing of septins were observed (Figure 4E). Once the septin filaments started to assemble, additional filaments grew in their proximity, regulated their orientation, and formed pairs. The formation of long, well-aligned filaments was fast and could be completed within ∼3 min. The observed assembly kinetics data were consistent with the molecular diffusion-driven annealing model. Filament diffusive volumes strongly influenced the septin polymerization (Figure 4F). Once a single layer of well-aligned and well-paired septin filaments formed on the surface, the septin filaments of the second layer started to assemble (Figure 4G). The growth of second-layer septin filaments is due to a template-initiating assembly process with the first-layer filaments as a template. The templating correlation of the second- and first-layer filaments was quantitatively assessed with ∼80% on average. After the second layer was formed, the filament assembly of the third layer was initiated with the underlying filaments as a template (Figure 4G, red arrowhead). These experiments indicate that septin multilayered hierarchical structures are assembled with a complex process synergistically affected by septin vertical and lateral interactions. In short, the assembly of septins is a diffusion-driven process, while the higher order structure formation is a complex process related to self-templating.

3.1.3 S-layers lattice on mica

In addition to the classical nucleation process shown in the study of septin filament, multistep assembly involving intermediate states,^[178-181] is also common in protein assembly. In the folding and assembly of native proteins at solid‒liquid interfaces, it is highly possible that one portion of protein molecules reaches the lowest-energy state, while the rest of the molecules get kinetically trapped in a metastable state along the energy landscape. For example, in situ AFM results showed that S-layer protein goes through a multistage assembly pathway at the mica‒liquid interface^[148] (Figure 5): (1) the main portion of S-layer molecules goes through the classical nucleation and formed domains in a stable tall phase (T phase, Figure 5C,D); (2) a small portion of S-layer molecules fold to form domains in a metastable short phase (S phase, ΔE = 1.6 kJ/mol, Figure 5D); (3) the S-phase domains transform to the T-phase domains (Figure 5A,B) in an exponential dependence; and (4) the ratio between S and T phases depends on the incubation temperature. Both states emerge from monomer clusters but follow two pathways—one leads to the kinetically trapped state, and the other directly reaches the final low-energy state. However, the trapped state transforms into a stable state by overcoming an energy barrier, E_s (Figure 5D). The transition initiated at the free edge of the short domain and gradually propagated from one side to the other rather than coinciding across the whole domain (Figure 5B).

These results demonstrate that kinetic traps, which might be triggered by the properties of substrate‒solution interfaces, play an essential role in determining the pathways and phases of protein assembly (Figure 5C,D). Extended protein arrays require conformational changes from a monomeric structure to a long-range order during protein assembly. They could fold into a metastable state (a local energy minimum) and slowly relax into a thermodynamic equilibrium state. The described picture of protein multistage assembly, which is based on in situ AFM observations, could be generally applicable to the phenomenon of native protein folding and assembly.

3.1.4 Cytochrome b₅₆₂ (cyt b₅₆₂-(H63C)) assembly on HOPG

In addition to hydrophilic surfaces, protein assemblies on hydrophobic surfaces have also been investigated by in situ AFM. For example, Kitagishi et al. studies the assembly of the hemoprotein system on HOPG.^[151] Cytochrome b₅₆₂ is a subclass of cytochromes that play a vital role in energy transfer by undergoing reduction and oxidation.^[182] The heme group bonded to the protein can also undergo these reversible reactions. Externally introduced heme analogs, such as heme 1, can bind to the surface of a H63C single mutant of cytochrome b₅₆₂ (cyt b₅₆₂-(H63C)) through a thiol-reactive group after knocking out the native heme of the modified protein. Variable morphologies of synthetic heme analogs on HOPG substrate were characterized by AFM, which have regioisomers regarding the substitution position in the two heme propionate side chains of protoheme IX (Figure 6A).^[151] Compared to analog 1, heme analog 2 has a higher reactivity to the thiol group of cyt b₅₆₂-(H63C), enabling the formation of long linear protein fibers through interprotein heme‒heme pocket interactions on HOPG, 2-apo-cyt b₃₆₂-(H63C) (Figure 6B). In the presence of heme triad 3 having a threefold center, cyt b₅₆₂-(H63C) can form dispersed protein particles. However, the coexistence of heme 2 and 3 at appropriate molar ratios results in the assembly of cyt b₅₆₂-(H63C) into a two-dimensional (2D) protein monolayer network, 3/2-apo-cyt b₃₆₂-(H63C), on HOPG (Figure 6B–D). The heme triad 3 incorporates into the linear fibers as a branching point for the trident hemoprotein assembly. The assembly continues to expand and diverge, eventually forming unique 2D hemoprotein networks on HOPG. This design strategy for controlling 1D and 2D supermolecular architectures on HOPG holds promise in developing protein-based materials.

3.2 Imaging of protein dynamics on lipid membranes

One key aspect of protein pivotal roles in modulating the biological membrane functionality is their ability to oligomerize and self-assemble on lipid membrane surfaces. These protein complexes and supramolecular structures contribute to the membrane organization, stability, and function.^{[19, 22, 23, 28, 32]} Protein‒protein/protein‒lipid interactions and the self-assembly processes are tightly regulated and essential for various cellular processes. Conversely, certain proteins can perturb a membrane integrity and disrupt its structural organization. For instance, specific toxin proteins can interact with membranes, leading to membrane disruption, pore formation, or alterations in membrane permeability.^[183] The interplay between proteins and membranes is a complex and dynamic, underlying fundamental biological processes. AFM provides opportunities to in situ study the dynamics of these proteins on lipid membranes in a near-neutral environment.

3.2.1 Soluble protein on supported lipid membrane

S-layers lattice on lipid membrane

Similar to multistep assembly on mica, in situ AFM can be used to observe the assembly of S-layer proteins on supported lipid bilayers through a multistep process.^{[87, 148]} Unlike the assembly on mica controlled by mica‒liquid interfaces and showing the phase transition between a metastable crystal phase and a thermodynamically stable phase (Figure 5); S-layer nucleation on supported lipid bilayers is mainly controlled by the folding kinetics between condensed liquid-like amorphous clusters and a crystalline structure (Figure 7). Monomers initially from a mobile adsorbed phase on the lipid membrane and transform into amorphous/liquid-like clusters (Figure 7A–C). These clusters then further undergo a phase transition into crystalline arrays composed of compact tetramers through simultaneous folding, and rearranged tetramers occur in the clusters (Figure 7C,D). Following growth occurs through new tetramer formation at the array edges—a template-guiding process with the existing tetramer matrix as a template (Figure 7E–H).^[87] Growth kinetics reveal that tetramer formation of the protein folding is the growth-limiting step (Figure 7I,J). The barrier of the new tetramer formation is significantly reduced by the pre-folded tetramers at the edges of the crystal clusters.

Perforin-2 (PFN2, MPEG1)

Membrane attack complex/perforin-like (MACPF) proteins are a huge family of pore-forming proteins.^[184] The primary role of MACPF proteins is to form transmembrane pores in membranes with various downstream biological effects, including targeted cell destruction by transport of granzymes and solute homeostasis devastation.^[185-190] These proteins play a critical function in immune defense and characteristically undergo a multistep conformational transition from water-soluble monomers into large β-barrel pores.^{[137, 186, 189-206]} The multistep pore formation mechanism commonly includes the following steps: (i) soluble MACPF monomers are recognized and recruited by membrane receptors or lipids onto membrane surface and (ii) MACPFs activate and oligomerize into ring- and/or arc-shaped pre-pore and pore structures upon membrane binding.

As a member of the MACPF family, PFN2 is a crucial effector in killing engulfed bacteria in the phagolysosome by forming pores in the cell membranes.^[207] PFN2 is a critical component of the innate immunity conserved throughout the animal kingdom and is close to the MAC and perforin-1 in evolution.^[208-211] Despite extensive exploration of the MACPF pore formation mechanism and recent structure advances of both pre-pore and pore states of PFN2 and MAC using cryo-electron microscopy (cryo-EM), the pre-pore-to-pore transition mechanisms and related dynamics of MACPF remain primarily unknown.^{[189, 190, 198, 200, 202, 203, 212]}

With the help of HS-AFM and HS-AFM-LS, Jiao et al. investigated the pre-pore-to-pore transition mechanism of PFN2.^[137] Using HS-AFM with sub-second temporal resolution (200 ms) (Figure 8A, top center), PFN2 oligomers were revealed to proceed in a multistep, clockwise conformational transition. In this transition: (i) pre-pore rings broke open upon triggering by acidification, (ii) the subunit located clockwise at the breakage site proceeded a conformational transition from the original pre-pore state (referred to as pre-pore-I) to an intermediate state (referred to as pre-pore-II) with ∼1.6 nm increase in height, (iii) the conformational transition further propagated to adjacent subunit in clockwise direction, (iv) the subunit in initiating site proceeded further conformational transition from the intermediate pre-pore-II state to the pore state by ∼2.2 nm increase in height, and (v) was also propagated clockwise to the neighboring subunit. The pre-pore-to-pore transition occurred within a few seconds. In addition, HS-AFM-LS experiments were performed to explore the transition dynamics of PFN2 (Figure 8A, bottom center). Due to the increased temporal resolution (2 ms) of HS-AFM-LS, state transition dynamics could be quantified with discernable height plateaus. The height increases were in accordance with the HS-AFM data and showed a height increase of 1.6 ± 0.4 nm and 2.2 ± 0.2 nm of pre-pore-I to intermediate pre-pore-II and the pre-pore-II to pore, respectively. These HS-AFM-LS data revealed the conformational change speed by fitting the height profiles (Figure 8A (bottom center), B). Together, the HS-AFM imaging and HS-AFM-LS experiments determine the progressive clockwise transition mechanism and dynamics of the PFN2 pore formation (Figure 8B).

3.2.2 Integral membrane proteins reconstituted in lipid membranes

Aquaporin Z water channels in lipid membranes

Biological membranes are not passive and actively organize their proteins.^{[213, 214]} Long-range membrane-mediated interactions, modulated by membrane properties, drive initial protein assembly.^[215] Rather than direct protein‒protein contacts alone, these interactions yield nonrandom organization and domains.^[216] Fundamental studies of how membranes physically orchestrate protein assembly promise to enhance our understanding of membrane organization. However, directly studying membrane‒protein interactions at single-molecule spatiotemporal resolution is still challenging.

Using HS-AFM, Jiang et al. successfully visualized and quantified the membrane-mediated interactions of E. coli water channel Aquaporin Z (AqpZ) at high spatial and temporal resolution.^[71] Initially, the membrane was reconstituted with AqpZ at a low lipid‒protein ratio and covered only a tiny portion of the surface. A continuous fluid lipid membrane was then formed by adding lipids with defined hydrophobic thicknesses for the membrane proteins diffusion and interaction. Membrane protein association/dissociation (Figure 9A) and unbounding diffusion (Figure 9B) were measured in C14, C16, C18, and C20 phosphatidylcholine lipid membranes by HS-AFM and HS-AFM-LS. Based on these measurements and a mechanical model of the lipid membrane, the researchers revealed that the interaction energies are proportional to the hydrophobic mismatch between the protein and lipid membranes. The oligomerization energetics of AqpZ-W14A were estimated based on statistical observations of non-tetrameric complexes (Figure 9C). The protomer interaction $$\Delta G_{{\mathrm{olig}}}^0$$ is weakest (∼ −2k_BT) in C20 lipids, which matches the protomer interface hydrophobic thickness. AqpZ diffusion speed is inversely proportional to the hydrophobic mismatch of the lipids (thicker and thinner). The most favorable membrane‒protein interaction $$\Delta G_{{\mathrm{asso}}}^0$$ was estimated (∼ −6.5k_BT) in lipids with a good match in the membrane thickness.

Elucidating the principles of membrane assembling and regulating protein organization is crucial to understanding cellular processes. These results demonstrate how membrane properties directly control AqpZ interactions and dynamics at the single-molecule level. By revealing how lipid membranes assemble and regulate model membrane proteins, this work establishes a framework for understanding the active role of membranes in orchestrating protein function. The methodology can be broadly applied to determine membrane contributions to the structure and function of other proteins.