A key process in the crystal growth of both inorganic and macromolecular crystals like proteins, nucleic acids, and virus crystals is the formation and growth of step edges ^[1][14]. It is indeed in the investigation of step-edge growth that a real-space imaging method like AFM can provide the most useful information.

The two primary mechanisms that govern the formation of step edges are growth via screw dislocation and growth via the spontaneous emergence of two-dimensional nuclei on active surfaces. A third mechanism known as normal growth does not result in layer addition. Instead, it relies on intense random nucleation on active crystals that possess an unusually low surface free energy. While this mechanism is infrequent, it has been observed in several macromolecular crystals ^[1][14].

In addition to the previous mechanisms, a fourth mechanism, not observed in conventional crystal growth, comes into play for macromolecular crystals due to the distinctive properties of concentrated macromolecular solutions ^[2][105]. AFM studies focusing on protein crystal growth dating back to the 1990s ^{[2][3][4][5][6][7][8][9][10][11][12][13][14]}[2,3,105,106,107,108,109,110,111,112,113,114,115] revealed the presence of multilayer stacks of monomolecular protein layers. These stacks can form hillocks, with shapes that usually reflect the overall morphology of the crystal. The local hypersaturation of the molecular solution in the droplet stimulates the crystal epitaxial growth, resulting in the formation of multilayer stacks ^[15][16][17][116,117,118]. Each layer provides step edges, allowing for tangential growth and development of new terraces ^[2][105]. Moreover, contaminants in the growth solution (foreign microcrystals, dust, and other contaminating macromolecules) can be incorporated, causing dislocations and morphological defects ^[9][110].

An interesting example of the use of AFM for the analysis of macromolecular crystal growth was reported by McPherson et al. ^[18][119]. As reported by the authors, the dislocation was the starting point that triggered the step-bunching process. Another application of the AFM on a lysozyme crystal was performed by Mollica et al. ^[19][120]. Differently from previous works that investigated the (110) face of lysozyme crystals ^[4][11][20][3,112,121], the novelty of Mollica et al. was the study of the (101) face. The investigation, performed by tapping mode AFM, revealed an anisotropic growth with a two-dimensional nucleation.

Leveraging the AFM capability to measure the z profile of samples, the authors reported a step height histogram which showed a bimodal distribution, with the Gaussian fit indicating step height values of approximately 3.1 nm and 6.7 nm. These values correspond well to the step heights expected for monomolecular and bimolecular steps in the (101) crystal face (3.4 nm and 6.8 nm, respectively) calculated from the parameters of the tetragonal unit cell of lysozyme crystals. The step height distribution suggests that, unlike the (110) face, the terraces of the (101) face are mostly due to single-molecule steps, as was previously observed through electron microscopy ^[21][122] and AFM ^[18][119].

The resolution of an AFM image can be pushed further to reach molecular resolution over a crystal face. The work by Mollica et al. ^[19][120] demonstrated molecular resolution on the (111) face of a ferritin crystal. The 450 nm × 400 nm image allows for the observation of the surface structure at the molecular level, revealing fine details of the terraces. Consistent with the (111) face of a face-centered cubic (fcc) crystal, a hexagonal molecular packing is observed. To determine the lattice symmetry and spacing, the authors performed a two-dimensional Fourier transform analysis on image portions corresponding to single terraces. The value obtained for the lattice constant (13.1 nm) was in good agreement with the 13.0 nm lattice constant calculated from X-ray diffraction experiments.

The importance of AFM in achieving molecular-level details of protein crystals was underlined also by the study by Malkin et al. ^[5][106] investigating the satellite tobacco mosaic virus (STMV). Earlier studies on STMV analyzed the aggregation pathway leading to the formation of 3D critical nuclei using quasi-elastic light scattering (QELS) ^[22][123] and investigated crystallization kinetics using Michelson interferometry ^[23][124]. The authors focused their attention on the growth mechanisms and molecular-scale processes involved in the development of the (111) face of cubic STMV crystals, revealing the individual virus monomers with a center-to-center distance of 18 nm.

Although the structure of the orthorhombic form of STMV had been determined ^[24][125], the packing and position of the virus particles in the unit cell were unknown for the cubic crystal form, which was determined to have space group P23 and unit cell parameters of a = b = c = 25.7 nm through X-ray diffraction experiments ^[25][126]. The hexagonal array and intermolecular spacing revealed by the AFM images of the (111) plane indicate that molecules are packed into an fcc lattice, which was impossible to determine from the X-ray data due to the lack of fourfold symmetry in the STMV molecule itself.

The potential of AFM as a tool for surface molecular structure analysis has been illustrated in more recent works by Guo et al. ^[26][127]. After successful acquisition of the surface lattice structures of L-valine, D-valine, and D-alanine crystals ^[27][128], the authors employed AFM to capture, for the first time, images of the surface-ordered lattices of single crystals of DL-valine and L-alanine ^[26][127]. The molecular-resolution images are in excellent agreement with the simulated models, showing the exceptional capability of AFM in investigating the surface molecular structures of amino acid crystals.

The AFM image acquired of the DL-valine crystal displays a valine molecule as a single protrusion. Despite attempts to obtain more detailed information from a higher-resolution AFM image, submolecular resolution could not be achieved, and an accurate description of the submolecular structure of a valine molecule remains challenging. The AFM image of an L-alanine crystal reveals easily distinguishable parallel molecular rows of different levels of brightness, forming a periodic structure composed of four molecular chains per unit cell (outlined by a white box in the figure). The inset of the image presents the Fourier transform pattern of the data.

The study from Guo et al. showed that AFM is a powerful tool for measuring the surface structure of single crystals.

The capability of AFM to produce molecularly resolved images of protein assemblies, previously discussed in the case of 3D protein crystals, can be advantageously exploited to investigate also 2D supramolecular ordered architectures.

The first molecular level AFM studies were reported by Ohnishi et al. in 1992 ^[28][129] and Furuno et al. in 1998 ^[29][130]. The protein quaternary structure was successfully resolved by Ohnishi et al. in 1993 ^[30][131].

A remarkable example of 2D protein architectures are the so-called S-layers, which are two-dimensional arrays of proteinaceous subunits that form surface layers ^[31][132]. The S-layers, firstly described as “macromolecular monolayers” by Houwink and Le Poole in ^[32][133], have now been identified in hundreds of different species of bacterial and archaeal species ^{[33][34][35][36][37][38][39][40]}[134,135,136,137,138,139,140,141]. S-layers can have different functions, such as protection, adhesion, and cell shape determination. S-layer proteins are excellent model systems for exploring the synthesis, secretion, and assembly of extracellular proteins, and are one of the most abundant biopolymers on Earth ^[41][142]. These protein layers consist of a single molecular species that can assemble on the cell surface into closed regular arrays in a low free-energy arrangement.

S-layer properties can moreover be changed due to environment chemical modifications, emphasizing the versatility and significance of S-layers as a building block for constructing supramolecular complex structures that involve proteins, lipids, glycans, and nucleic acids ^{[42][43][44][45][46]}[143,144,145,146,147]. Considering their unique structural and physicochemical properties, S-layers have shown great potential in various fields of application including nano(bio)technology, design of vaccines, diagnostics, and drug delivery ^{[47][48][49][50][51]}[148,149,150,151,152].

Various electron microscopy studies on S-layers have been reported in the literature; they require sample preparation such as thin-sectioning, freeze-etching, freeze-drying, and negative staining ^{[34][52][53][54][55][56]}[135,153,154,155,156,157]. On the other hand, AFM has proved to be a very versatile analytical tool, with the advantage, over other techniques, of allowing for in situ experiments under controlled conditions in the growth environment ^{[57][58][59][60][61][62][63][64][65][66][67]}[158,159,160,161,162,163,164,165,166,167,168].

An example of the high-resolution capability of AFM applied to S-layers can be found in the work by Scheuring et al. ^[67][168]. The authors studied the S-layer formed by the PS2 protein, responsible for the S-layer formation of the Corynebacterium glutamicum ^[68][69][70][169,170,171]. A hydrophobic amino acid stretch of 21 residues is present at the C-terminus of the protein. When the S-layer is proteolyzed, this C-terminus is removed, which results in detachment of the S-layer from the cell. Bacteria that express the truncated PS2 protein cannot assemble the S-layer and release PS2 into the medium, which confirms the role of the C-terminus in S-layer/cell wall attachment ^[70][171].

In the reported work ^[67][168], both native and trypsin-treated forms were analyzed, revealing differences in their adsorption behavior. AFM acquisitions showed two different surface types, a flower-shaped one and a triangular one, with a thickness of 4.6 nm for the native and 4.1 nm for the proteolyzed sample. Electrostatic balancing was employed to minimize the tip–sample interaction, allowing for high-resolution topography down to 1 nm ^[71][172]. The height information from both surfaces spanned the total height of the S-layer, allowing for a first-time reconstruction of a three-dimensional model from AFM topography.

The study showed evidence of the different behavior of the two sides of the native S-layer in terms of hydrophilicity/hydrophobicity. This property generates the tendency to stack into double-layered assemblies that could be dissected using the AFM tip ^[72][73][74][173,174,175].

Another highly studied system is the SbpA protein from Lysinibacillus sphaericus (CCM2177), whose reassembly in S-layers on mica does not follow the classical pathway of crystal growth. In this case, a kinetic trap may hinder the reassembly into extended matrices ^[75][176], and SbpA proteins can self-assemble into a square lattice symmetry by diffusion from solution to the surface ^[75][76][176,177]. Moreover, the SbpA proteins follow a two-stage nucleation process ^[65][75][166,176], and the protein–substrate and protein–protein interactions may play a role in the assembly pathway ^[77][178].

The role that the hydrophobicity of the surfaces has on the SbpA adsorption kinetics was investigated by Herrera et al. ^[61][162]. Their investigation was intended to fill the gap in knowledge on the recrystallization kinetics of S-proteins on functional thiols. Indeed several studies had already addressed that aspect on solid and soft interfaces ^{[59][65][77][78]}[160,166,178,179].

In their paper ^[61][162], the authors investigated how surface modification affects the adsorption rate of S-layer proteins and the formation of protein crystals. They functionalized gold substrates with various alkanethiols with different hydrophobic properties and surface charges, including five types of self-assembled monolayers with methyl, hydroxyl, carboxylic acid, and mannose terminal functional groups.

Protein adsorption rates were faster on uncharged hydrophobic substrates compared to hydrophilic ones. Small S-layer domains were observed on hydrophobic substrates, whereas on OHC₁₁S substrates, the protein adsorbed without forming crystalline layers. When SbpA interacted with a hydrophilic ManC5S surface, partial S-layer recrystallization occurred, but no specific carbohydrate/protein interaction was observed. However, decreasing the pH from 9 to 5 induced different S-layer recrystallization pathways, demonstrating that electrostatic interactions between SbpA and the COOHC₁₁S substrate can tune the adsorption rate. Another interesting result was the dependence of the structure of the protein crystal domains on concentration and surface chemistry. For hydrophobic substrates, concentration helped to enlarge the crystalline domain area, while for COOHC₁₁S substrates, the protein concentration had no influence on the domain size.

Some years later, the same authors ^[62][163] explored the role played by protein concentration and observation time on the adsorption of SbpA proteins on hydrophobic SiO₂. AFM measurements were performed using four protein concentrations (from 0.08 μM to 0.8 μM) and five different time sequences. The results showed that the rate of adsorption, the nucleation points, and the crystal growth were influenced by the protein–surface interactions, with protein concentration being a crucial factor. Therefore, any alteration in experimental conditions would impact the protein–substrate interactions, highlighting their significance in the formation of crystal layers. The observation time influenced the formation of 2D crystals, with nucleation points and protein domain size increasing with time. The lowest concentration did not form a confluent protein layer, while the highest concentration completely covered the substrate surface.

The AFM phase images confirmed that a more compact protein layer was achieved at higher concentrations, as the gaps between the growing crystalline domains were reduced. The results also showed that the best resolution in AFM imaging is obtained at higher protein concentrations (harder surfaces), confirming previous observations that sample softness affects AFM resolution ^[79][180]. The authors reported that the main limitation to the experimental procedure was the rapid adsorption kinetics of the SbpA protein on these surfaces. The AFM imaging started after the hydrophobic surface had already been exposed to the protein solution for about 15 min, which was enough time for the smallest crystalline domains to form. While the initial adsorption and recrystallization steps were not directly visualized by AFM, previous QCM-D results indicated that the process is primarily governed by protein–substrate interactions, followed by subsequent protein self-assembly.

The AFM capability of surface analysis at the molecular scale and at timescales typical for protein crystallization was exploited also by Vekilov et al. ^[63][164]. The aim of their experiments was to gain insight into the nucleation pathways by imaging clusters for the crystallization of a model protein system. In situ AFM was performed in supersaturated solutions to monitor the growth of clusters on the bottom of the AFM cell. The authors focused on apoferritin, a protein consisting of 24 subunits arranged in pairs along the 12 walls of a quasi-rhombo-dodecahedron ^[80][81][82][181,182,183]. Apoferritin crystals have a face-centered cubic lattice exposing (111) planes with a hexagonal molecular packing ^[83][84][184,185].

The authors observed the adsorption of a disordered monomolecular protein layer covering the entire glass surface. Throughout the experiment, no partial or full second layer of adsorbed molecules was observed, and there were no changes in the molecular arrangements. Therefore, it was concluded that the apoferritin molecules were rigidly attached to the glass substrate, allowing for the imaging of individual molecules and determination of their size, which was found to be approximately 13 nm, as observed in crystals ^[84][185].

Moreover, considering two clusters containing four and nine molecules, according to nucleation theories the smaller adsorbed cluster is expected to disappear quickly, while the larger one should be closer to a labile equilibrium with the solution ^[85][186]. However, the analysis showed that the smaller cluster gains and loses two molecules, ultimately retaining four molecules after about 20 min of observation. Conversely, the larger cluster dissolves progressively over the same period. This apparent contradiction is explained by the fact that nucleation theories describe the evolution of large populations of clusters, and the fate of individual clusters is determined by a random sequence of events. The authors suggested that their findings may have implications for understanding and predicting the behavior of ensembles of small aggregates driven by surface energy, with implications in nanotechnologies ^[86][187].

Another example of molecule exchange dynamics presented by the same group was focalized on the surface of a large apoferritin crystal, with an on-top cluster of apoferritin with the same molecular arrangement of the underlying crystal ^[87][88][188,189]. The molecules in the cluster are organized in rods of four to eight molecules, with the center-to-center distance between adjacent molecules in a rod equal to that along the close-packed (110) direction in the crystal lattice. The subsequent observations on the same cluster showed that molecules attach and detach from the cluster with comparable frequencies. From the low net exchange rate between the cluster and the solution, the authors inferred that the size of the considered cluster was slightly above the critical size for the saturation condition.

Although AFM has been widely recognized as a powerful tool in the biological sciences, its limited imaging speed (tens of seconds to minutes to complete one image) has hindered its widespread use compared to other popular techniques like optical microscopies. In many cases, understanding the assembly process and the molecular dynamics at their very initial step is crucial for understanding the biological processes that occur within living cells, such as molecular interactions and conformational changes. To overcome the above-mentioned slow imaging speed limits, several research groups have worked to increase the imaging speed of AFM ^{[89][90][91][92][93][94][95]}[97,98,99,100,101,102,190], and high-speed atomic force microscopy (HS-AFM) was developed. The use of tapping mode ^[96][191] to reduce lateral forces during scans, miniaturized cantilevers, high-frequency feedback systems, and newly developed amplitude detectors ^[97][98][192,193] allow the current HS-AFM to capture images at a speed in the range of 30–60 ms per frame with a scan size of about 250 nm and 100 scan lines under typical conditions for biomolecular investigations ^[99][100][104,194].

Based on these characteristics, HS-AFM was employed, e.g., to observe the conformational changes of proteins ^{[101][102][103][104][105][106][107][108][109][110][111][112][113][114][115]}[15,195,196,197,198,199,200,201,202,203,204,205,206,207,208], to monitor enzyme reactions ^{[116][117][118][119][120]}[209,210,211,212,213], to study the growth of amyloid-like fibrils ^{[121][122][123]}[214,215,216], to investigate the DNA conformational changes ^{[124][125][126]}[217,218,219], and to follow protein dynamics while interacting with other proteins ^{[127][128][129][130][131]}[220,221,222,223,224] or with membranes ^{[132][133][134][135][136][137][138][139][140]}[225,226,227,228,229,230,231,232,233]. In addition to imaging applications, HS-AFM has been shown to be valuable for high-speed force spectroscopy (HS-FS) and active high-frequency microrheology (HF-MR) ^{[141][142][143][144][145]}[234,235,236,237,238]. These advancements have enabled the exploration of previously unattainable dynamic ranges, enabling experimental protein unfolding studies that can be directly compared to molecular dynamics simulations ^[141][142][234,235].

Furthermore, recent advancements in HS-AFM have provided valuable insights into the molecular mechanisms of cyclic nucleotide-gated (CNG) channels, as investigated in the work from Marchesi et al. ^[146][239]. In their work, the authors studied the bacterial SthK CNG, chosen as a model CNG channel that is important for linking intracellular cyclic nucleotide signaling with electrical signaling at the cell membrane ^{[147][148][149][150]}[240,241,242,243]. These channels are regulated by the binding of cyclic nucleotides (cAMP or cGMP) to a specific intracellular domain called the cyclic nucleotide-binding domain (CNBD) ^[151][152][244,245]. This binding induces conformational changes that open or close the channel pore, controlling the flow of ions across the membrane. These channels, which are crucial for neuronal excitability and sensory pathways ^[153][154][246,247], consist of tetrameric subunit arrangements surrounding a central pore ^[155][248]. HS-AFM has enhanced  understanding of the dynamic behavior and functional properties of CNG channels, revealing their physiological roles and molecular mechanisms. In particular, studies demonstrated that cGMP does not activate SthK channels but instead inhibits the cAMP-induced activity, leading to channel closure ^[156][157][249,250]. In the work of Ruan et al. ^[104][197], the authors utilized HS-AFM to observe the transition of SthK channels from an activated/open state to a resting/closed state by adding an excess amount of cGMP. Starting with an initial concentration of 0.1 mM cAMP, increasing concentrations of cGMP were added until reaching a final concentration of 7 mM. The imaging revealed a noticeable change in the surface topography and in the two-dimensional crystal packing. This change originated from the molecules located at the border of the crystal patch and gradually propagated towards the center with slow kinetics.

Analyzing the morphological changes of the SthK, the authors concluded that upon the transition from a closed to an open state, the four CNBDs that form the channel undergo distinct structural changes. These CNBDs move vertically towards the membrane by approximately 0.6 nm and spread out by approximately 0.4 nm. Additionally, they rotate relative to the pore domain. Structural analysis suggested that this rotation of the CNBDs, triggered by cAMP activation, results in a displacement of approximately 1.3 nm on the periphery of the C-linker. This movement is transmitted through the C-linker to the pore domain, leading to the opening of the channel pore.

These observations suggested that the long-range interactions between adjacent and distant protein domains are responsible for structural transitions not only of cyclic nucleotide-modulated channels but also of other channels controlled by the binding of small intracellular ligands to carboxyl terminal regulatory domains. This indicates a shared mechanism for conformational changes and functional regulation among these channels.

Alongside membrane channels, which passively allow or block the transport of solutes through membranes, a different assisted mechanism of intra/extramembrane communication is represented by the membrane transporters ^[158][251].

Between them, extensive research has been performed on glutamate, a membrane transporter that plays a crucial role in maintaining appropriate glutamate levels in the synaptic cleft to prevent excitotoxicity ^[159][252]. Dysfunctions in these transporters are linked to various neurological disorders including epilepsy, Alzheimer’s disease, and amyotrophic lateral sclerosis ^[160][253]. The elucidation of the crystal structures of the glutamate transporter homolog from the archaebacterium Pyrococcus horikoshii (GltPh) has been a significant milestone in understanding the mechanism of ion-coupled transport. Previous studies on glutamate transporters have provided valuable insights into their localization, function, and structure ^{[161][162][163]}[254,255,256]. However, dynamic studies of the transport mechanism have been limited, making HS-AFM observations a valuable contribution to understanding the functioning of glutamate transporters. In this direction, a recent study from Ruan et al. ^[104][197] exploited HS-AFM to obtain direct insights into the dynamic behavior of glutamate transporters.

The authors observed a membrane-reconstituted GltPh, whose trimers in the membrane were visualized as formed by three protomers arranged in a triangular pattern with a central cavity. Each protomer was imaged as a ∼2 nm diameter and 2 nm height protrusion. These observations suggested that GltPh was exposing the transport domains towards the extracellular side facing the HS-AFM tip. The transport domains exhibited reversible conformational changes between an outward-facing (up) and an inward-facing (down) distinct state. The trimeric structure of the transporter was visible when all three subunits were in the outward-facing state, with the transport domains clearly protruding from the membrane plane. The domains demonstrated vertical motion, moving up and down with an amplitude of approximately a couple of nm. These observations were consistent with crystal structures, single-molecule Förster resonance energy transfer, and electron paramagnetic resonance studies ^{[164][165][166][167][168][169][170]}[257,258,259,260,261,262,263].

To investigate the relationship between molecular motion and function, the GltPh domain movements between the “up” and “down” states were studied under various environmental conditions, including the absence of substrate, different concentrations of Na⁺/Asp, and the presence of an inhibitor (DL-TBOA) ^[171][264]. As expected for a Na⁺/Asp symporter, the HS-AFM revealed that GltPh exhibits greater dynamics in the absence and presence of Na⁺ and Asp compared to when only Na⁺ is present. Contextually, when Asp was replaced with the competitive inhibitor, GltPh remained trapped in the outward-facing state with minimal observed movements. Furthermore, analyzing the correlations between the conformations of the protomers within the trimer revealed that each protomer acted independently and without a specific order in terms of their location within the trimeric structure.

The HS-AFM imaging capability has been used also to investigate the structural dynamics of bacteriorhodopsin (bR), a light-driven proton pump ^{[172][173][174][175]}[265,266,267,268]. Imaging on a slow photocycle bR mutant (D96N), conformational changes in the cytoplasmic domains were observed ^[108][201]. Upon light illumination, HS-AFM allowed the observation of approximately 0.8 nm lateral outward displacement of the E–F loops in each bR. This information was correlated with high-resolution electron microscopy and X-ray crystallography data, demonstrating the powerful combination of these techniques in assigning structural dynamics.

The capability of HS-AFM to catch protein dynamics at the single-molecule level has been recently demonstrated by a study on the interaction between spike variants and ACE2 receptors ^[176][269]. Hinterdorfer and coworkers ^[176][269] could follow the spike trimer dynamics in the presence of ACE2, showing that Delta and Omicron variants enhance viral attachment to the host-cell receptor compared to the early Wuhan-1 isolate. This finding would explain not only the increased rate of viral uptake but also the increased resistance of the variants against host–cell detachment by shear forces.

The above studies highlight the exceptional ability of HS-AFM to investigate dynamic processes involving individual unlabeled proteins. Traditional techniques often struggle to capture conformational changes with high precision due to low signal-to-noise ratios. In contrast, HS-AFM provides a powerful tool to directly observe and characterize these conformational changes, enabling a more detailed understanding of protein dynamic behavior at the single-molecule level.