The actin cytoskeleton: Morphological changes in pre- and fully developed lung cancer

Abstract

Actin, a primary component of the cell cytoskeleton can have multiple isoforms, each of which can have specific properties uniquely suited for their purpose. These monomers are then bound together to form polymeric filaments utilizing adenosine triphosphate hydrolysis as a source of energy. Proteins, such as Arp2/3, VASP, formin, profilin, and cofilin, serve important roles in the polymerization process. These filaments can further be linked to form stress fibers by proteins called actin-binding proteins, such as α-actinin, myosin, fascin, filamin, zyxin, and epsin. These stress fibers are responsible for mechanotransduction, maintaining cell shape, cell motility, and intracellular cargo transport. Cancer metastasis, specifically epithelial mesenchymal transition (EMT), which is one of the key steps of the process, is accompanied by the formation of thick stress fibers through the Rho-associated protein kinase, MAPK/ERK, and Wnt pathways. Recently, with the advent of “field cancerization,” pre-malignant cells have also been demonstrated to possess stress fibers and related cytoskeletal features. Analytical methods ranging from western blot and RNA-sequencing to cryo-EM and fluorescent imaging have been employed to understand the structure and dynamics of actin and related proteins including polymerization/depolymerization. More recent methods involve quantifying properties of the actin cytoskeleton from fluorescent images and utilizing them to study biological processes, such as EMT. These image analysis approaches exploit the fact that filaments have a unique structure (curvilinear) compared to the noise or other artifacts to separate them. Line segments are extracted from these filament images that have assigned lengths and orientations. Coupling such methods with statistical analysis has resulted in development of a new reporter for EMT in lung cancer cells as well as their drug responses.

I. INTRODUCTION

In this review, we first summarize the vast literature on actin filament dynamics in health (Sec. II) and disease (Sec. III). We then review analytical methods that are used to quantify and characterize actin filaments (Sec. IV). In Sec. V, we review our recent findings regarding dynamic morphological changes of actin filaments during epithelial mesenchymal transition (EMT) and in pre- and fully developed lung cancer cell lines. We describe the experimental and analysis tools and demonstrate that the orientational order parameter (OOP) can be used as a marker for EMT/cancer progression. We conclude with outlook for possible future extensions of this line of research. Future prospects of such image quantification techniques are discussed in Sec. VI.

II. ACTIN AND ACTIN DYNAMICS

A. Actin cytoskeleton

The cytoskeleton of the cell is essential for a range of functions in the cell ranging from basic processes, such as maintaining cell shape and aiding in cell movement, to more complicated activities, including intracellular cargo transport and nuclear blebbing. The cell cytoskeleton essentially consists of actin, tubulin,^1,2 and intermediate filaments.^3–5 Actin demonstrates a high degree of conservation throughout the evolutionary tree of eukaryotes starting from the very primitive eukaryote organisms to modern humans.⁶ On the prokaryotic side (specifically in bacteria), filaments of actin or actin homologs show a greater divergence, though the underlying genes are shown to be related both structurally and functionally.⁶ This overall conservation, though not unique, is very uncommon in other types of proteins and is suggestive of actin's importance in organisms.⁷ The different types of actin in plants is responsible for the maintenance of vegetative and reproductive tissues and are expressed from around ten different genes.⁶ On the other hand, humans have six isoforms of actin,⁸ each expressed from one of six different genes and can be broadly classified into three categories α-actin, β-actin, and γ-actin.^9,10 Though these isoforms have varied overall charge of the molecules and hence their isoelectric points,^11,12 they share as much as 95% similarity in their amino acid sequences; and despite their structural similarity, their functions are heavily diverse and they are localized^13–16 at different and often unique organs. This spatially and temporally dissimilar expression of isoforms can be observed throughout development demonstrating not only tissue but sub-cellular specificity as well.^17–21

The first isoform, α-actin is found in most type of tissues at low quantities but is predominantly expressed in vascular smooth muscles,²² such as cardiovascular and skeletal cells. α-actin starts expressing in smooth vascular tissues from early stages of embryonic development, eventually becoming the most abundant protein in adult vascular smooth muscle tissues like the aorta.²³ The primary difference between α-actin and other isoforms is in the N-terminal section of the protein, which is responsible for actin polymerization.^8,24 There are two types of α-actin: (1) cardiac α-actin or smooth muscle actin (SMA), which is predominant in the striated heart tissues,^25,26 and (2) skeletal α-actin, which is primarily found in skeletal muscles.^27–30 This spatially divergent expression of α-actin is due to the fact that all the promoters responsible for the expression of different α-actins contain a conserved region known as the CArG box,²³ which is believed to encode the spatial divergence.¹⁸ Though this spatial divergence is indicative that each of these types have their own specific property and function,^10,31 they can compensate for one another. This is evidenced in mice generated without SMA still have functional cardiovascular system, which suggests that skeletal α-actin can offset the lack of SMA if required.³² α-actin is believed to be involved in mechanotransduction pathways serving as an important factor in how the cell communicates with and perceives the outside.²³ α-actin is upregulated in cells when they are subjected to external mechanical forces implicating them in mechanotransduction pathways.^33–37 They are also involved in wound healing and tissue fibrosis.^38–42 Their role in mechanotransduction is further supported by their increased presence in myofibroblasts, which indicates that α-actin is involved in connecting the cortical actin with the extracellular matrix.^43,44

The second isoform, β-actin is an exclusively cytoskeletal actin that is ubiquitous across most cells and tissues.⁴⁵ The most common actin superstructures, such as stress fibers, circular bundles, and filopodias, all consist primarily of β-actin.¹⁶ Though α-actin is primarily responsible for the contraction of the vascular tissue, β-actin, coupled with myosin II, can aid in cell contraction and adhesion in the normal epithelium where α-actin is absent or negligible in amount.¹⁶ Downregulation of myosin II or inhibition of its interaction with actin causes a decrease in the amount of stress fibers in cells.¹⁶ Knocking out β-actin renders embryos non-viable indicative of its tremendous importance and the fact that other isoforms or proteins cannot compensate for its absence.^46,47 In cells, knocking out β-actin results in drastic decrease in growth rate and motility with a concurrent increase in apoptosis rate.¹⁵ β-actin is required for cell division which is evidenced by the fact that β-actin null cells are also often polynucleated indicating that this actin isoform is specifically involved in formation of contractile rings.^16,31,47 Actin known to be associated with all three RNA polymerases^48,49 and as such actin dynamics affects cellular gene regulation through serum response factor (SRF).^50,51 Of the three isoforms, cells lacking β-actin show a higher variance in gene expression as well as decreased availability of G-actin in the cytoplasm.⁵²

The third isoform γ-actin has two main types, the γ smooth muscle actin and γ cytoplasmic actin. The γ cytoskeletal actin is present in most cells of the epithelium where they form the basis of their cortical actin structures.^16,53 γ smooth muscle actin is usually present in vascular structures, such as walls of blood veins, gastrointestinal and genital tracts.⁵⁴ This isoform is also present in the costamere of skeletal muscles, which is located right under the cell membrane and supports its involvement in cortical actin mesh.⁵⁵ Further evidence of this involvement is found in the role this isoform plays in maintaining the asymmetry in the first meiotic division.⁵³ Interestingly, γ-actin is found in the mechanosensory sterocilia of the hair cells involved in hearing. Knocking out γ-actin resulted in higher mortality and progressive hearing loss in mice; but the mice remained viable.⁵⁶ This indicates that unlike β-actin, the absence of γ-actin can be partially compensated for by other isoforms (most likely β-actin). This theory is further supported by the upregulation (up to 1.5 folds) of β-actin in γ-actin null cells. However, these deletions also initiate neoplastic transformations such as epithelial-mesenchymal transition.¹⁶

B. Actin polymerization

Actin is a biopolymer that is able to form filaments and even larger structures under certain conditions, which is likely why it has been conserved across species.⁶ Polymerization of actin monomers into filaments is the first step toward the development of cell cytoskeleton. In certain instances, the polymerization (and concurrent depolymerization) process is capable of generating force for movement of the cell.^57–60 Actin monomers, also called globular actin (G-actin), have an asymmetric structure; they have a barbed end and a pointed end.^58–61 Usually, these monomers are attached with actin depolymerization factor (ADF) and cofilin,^57–63 which prevents random actin polymerization and filament formation in the cytoskeleton. On certain cellular cues, the ADF/cofilin complex is replaced with profilin (actin-binding protein, ABP), which enables them to polymerize and bind with existing filaments.^57–63 Repetitive addition of monomers to the barbed end of filaments result in elongation of the filaments. Conversely, monomers are detached from the pointed end of filaments when de-polymerization is required (Fig. 1). Adenosine triphosphate (ATP) serves as the source of energy for this polymerization/de-polymerization process.^59,60

FIG. 1.

Actin filament formation. (a) Actin monomers attach to the barbed end of an existing filament. Detachment of profilin and ATP hydrolysis accompanies this process. (b) Actin polymerization mediated via formin, where local concentration of G-actin is enhanced near the barbed end of the filament. (c) Ena/VASP complex-mediated actin polymerization. The Ena/VASP complex enables recruitment of G-actin while preventing filament capping. Reproduced with permission from T. Svitkina, Cold Spring Harbor Perspect. Biol. 10, a018267 (2018). Copyright 2018 Cold Spring Harbor Laboratory Press.⁶⁰

Being built from asymmetric monomers that only bind selectively to one end of another monomer, the actin filaments are also asymmetric, and the elongation process is highly directional. At the barbed end (plus end), new monomers are attached to the filament while on the pointed end (minus end), depolymerization takes place aided by the ADF/cofilin complex.^57,62,63 The depolymerization process replenishes the G-actin pool, and the monomers are recycled for further elongation of the barbed end. This concurrent polymerization and depolymerization on opposite ends essentially result in the movement of the whole filament and plays a key role in cell movement processes such as cytokinesis.^57,60,63 Actin polymerization is energetically favorably leading to frequent nucleation events, but the dimers and trimers are extremely unstable as their depolymerization is entropically favored; thus, very few nucleation events actually result in filament formation (Fig. 2).⁵⁸ Other than the simple polymeric filaments, actin is capable of forming branched networks from shorter filaments. Actin related protein 2/3 (Arp2/3) binds to an actin monomer, which is attached to the barbed end of an existing filament.^57,60 Now, along with normal elongation of the filament, the Arp2/3 serves as a nucleation site for another filament formation, forming a Y-type branched structure.^57,60 This branching very consistently generates an angle of ∼70° (Refs. 64 and 65) and is controlled by the WASP/Scar genetic pathway.⁶⁶ This Y-type branching is primarily observed near the cell membrane where the thicker fibers taper and interact with membrane proteins and receptors.^57,60

FIG. 2.

Rates of nucleation and elongation of different mono-, oligo-, and polymeric actin units. Relative reaction rates demonstrate that for shorter oligomers, the rate of depolymerization is higher than elongation rendering them unstable. Only in longer polymeric structure, the elongation rate is higher that depolymerization rate. Reproduced with permission from T. D. Pollard and J. A. Cooper, Science 326, 1208 (2009). Copyright 2009 The American Association for the Advancement of Science.⁵⁸

Apart from the Arp2/3-mediated polymerization of actin, another class of proteins called formins^67,68 plays a key role in actin polymerization. Formins are characterized by two domains, which are involved in actin nucleation/elongation: formin homology 1 (FH1) and formin homology 2 (FH2).⁶⁷ FH1 attaches to the barbed end of a filament and increases the local concentration of G-actin near the elongation site by binding multiple monomers and preventing capping proteins from binding.^69,70 FH2, on the other hand, forms a homodimer near the barbed end of a filament and serves as a capping protein.⁶⁷

C. Stress fibers and actin binding proteins

Polymerized actin filaments serve as building blocks to a diverse range of larger structures, such as stress fibers and contractile rings. In these larger structures, a family of proteins called actin-binding proteins (ABP)^71–73 serve as the glue that holds the filaments together. Stress fibers, which are the primary contractile structure in most cell types, are formed from 10 to 30 actin filaments bundled together by a myriad of ABPs.^58,59,61 For example, actin filaments held together by myosin is the primary source of the contractile force.^58,59 These fibers, called actomyosin bundles, function similarly in both muscle and non-muscle cells. The actin-binding site of myosin is capable of constantly attaching and detaching form the actin filament and its motor domain provides the energy for force generation at the expense of ATP hydrolysis.⁷⁴ A conformational change in the myosin, caused by ATP hydrolysis, is amplified by the lever arm resulting in a motion called the power stroke; this is the basis of the motor-like function of myosin.^75–77 The movement of the lever arm and the constant attachment/detachment of myosin and actin fibers cause the contraction and movement. In certain cases, a two-headed myosin motor is connected to two actin filaments, which causes a sliding motion of one filament over the other also known as “precessive walking.”^74,78,79

α-actinin is probably the next most important and ubiquitous protein involved in stress fibers. It is a cross-linking protein belonging to the spectrin superfamily of proteins and it holds multiple actin filaments together.^80–82 One end of α-actinin can bind to another α-actinin monomer, thereby forming a dimer with their individual actin binding ends pointing outwards.⁸⁰ This dimeric structure with two actin-binding sites on opposite ends can bind to two separate actin filaments at once. Out of the four isoforms of α-actinin, isoforms 1 and 4 are present in all types of cells except for cardiac cell, which have isoforms 2 and 3.⁸⁰ In contractile stress bundles, myosin and α-actinin forms a periodic alternating pattern on the actin fibers in which force is generated by myosin and the individual filaments are held together by α-actinin.^83,84 Apart from these two, other less ubiquitous cross-linking proteins like fascin,⁸⁵ zyxin,^86,87 filamin,^88,89 and epsin⁹⁰ are also involved in stress fibers.

Stress fibers also serve as a connection to the outer matrix/extra-cellular matrix (ECM);⁶⁰ these junction points are known as focal adhesions (FA)⁹¹ and are often characterized by the presence of actin-binding proteins such as focal adhesion kinase (FAK)^92,93 and paxillin.⁹⁴ Apart from forming the connection between actin and the ECM, FAs are hypothesized to act as a nucleation site of stress fiber formation.⁶⁰ As the junction between the inside and outside of the cell, FAs are predictably responsible for mechanotransduction which is verified in studies where the application of physical force on the cell resulted in formation and development of FAs.^91–93

As we have discussed already, though the main component of stress fibers is actin, stress fibers are still distinct based on their location and the ABPs involved.^{59,61,71,72,95} The three major categories are dorsal stress fibers, transverse arcs, and ventral stress fibers (Fig. 3).^59,61,96 Dorsal stress fibers do not have any myosin and, thus, are non-contractile and have one of their ends capped with FAK which enables them to act as precursors for elongation/fusion into other types of stress fibers.⁵⁹ Sometimes, dorsal fibers can exert contractile force, but that originates from them being connected to another type of contractile stress fibers known as transverse arcs.⁹⁵ Transverse arcs are actin bundles with both myosin and α-actinin, but no FAK capping on either of their ends enabling them to “float” around the cell.^59,61 Their constant contractile motion during cell migration drive them away from the leading edge of the cell toward the center, resulting in a phenomenon known as retrograde flow.^97,98 The third category of stress fibers, the ventral stress fibers, are the most commonly seen type of the three. They are connected to FAs on both ends and is usually the major contractile actomyosine moiety in cells.⁵⁹ It is conjectured that they are formed via the fusion of two dorsal fibers which might have multiple transverse arcs in between.^59,61 Another type of stress fibers, which are compositionally same as ventral fibers, are formed over and around the nucleus encasing it and is known as perinuclear actin cap. These fibers aid in mechanosensation of the nucleus with the rest of the cell as well as controls its shape and size.^59,95

FIG. 3.

Different types of stress fibers. Transverse arcs are fibers without FAK (orange) on either end and contain both myosin (blue) and α-actinin (red). Fibers radiating from transverse arcs with one end FAK capped are dorsal stress fibers and only contain α-actinin. Ventral stress fibers contain both myosin and α-actinin and have FAK on both ends. Fibers that are situated over the nucleus are perinuclear actin caps. Reproduced (without any changes) from Basu et al., Commun. Biol. 5, 407 (2022). Copyright 2022 Authors, licensed under a Creative Commons Attribution (CC BY) License.⁹⁶

A meshwork of genetic pathways and mechanisms controls the formation, recruitment, functioning, and eventual disintegration of actin filaments, ABPs, and stress fibers.^57,59,61,99 The small GTPases—RhoA,⁹⁹ RhoB,¹⁰⁰ and RhoC¹⁰¹—are involved in the most studied and validated genetic pathway involved in actin dynamics. Though RhoA is the major controlling factor in stress fiber assembly in most cell types,⁹⁹ it is hypothesized that in certain cells RhoB¹⁰⁰ and RhoC¹⁰¹ are relevant in the process. A library of proteins required for the formation of stress fibers is controlled by Rho through myocardial-related transcription factor. Rho-associated protein kinase (ROCK), which is a downstream protein of Rho, affects the stress fiber assembly more directly^102,103 by phosphorylating the myosin light chain¹⁰⁴ and increasing the contractility of the stress fibers.¹⁰⁵ The Rho-GTPase pathway also controls the nucleation and branching of actin filaments by controlling Arp2/3¹⁰⁶ and mDia1 (Formin).^107–109 Two GTPases in the Rho-GTPase signaling pathway, named Rac1¹¹⁰ and CDC42,¹¹¹ affect the Arp2/3 complex.

D. Roles in cellular processes

Stress fibers are the primary mechanotransduction machinery in cells, often forming or dissolving as a response to mechanical stress in cells.^112,113 Not only do stress fibers communicate the external stimuli with the aid of FAK, they also respond to it.¹¹⁴ The response of the actin cytoskeleton to external stimuli is demonstrated in studies where cells grown on soft and hard surfaces manifest completely different organization and type of stress fibers.^115,116 In the case of cells grown on soft surfaces, the stress fibers are thin and disorganized consisting of mostly dorsal fibers and transverse arcs. On the other hand, the same cell, when grown on hard surfaces, exhibit thick parallel ventral stress fibers.¹¹⁴ As the primary force generating machinery in the cell, owing to the contractile actomyosin bundles, stress fibers are involved in all forms of elastic motion in and of the cell, ranging from cell migration to shape change.^59,60,117 Cell migration is accompanied by and is often described as a series of protrusion and retraction events at the leading and lagging edge of the cell, respectively; this is achieved by constant assembly of new actin filaments at the leading edge of the cell with simultaneous disassembly at the lagging edge.¹¹⁸ Changing the shape of the cell can also be interpreted as directed growth of stress fibers.^119,120 Stress fiber formation also accompanies cell migration during wound healing.¹²¹ Actomyosin being the main contractile machinery in cells, in contractile tissues, the main purpose of the cells is carried out via the contraction of stress fibers in concert with ABPs.⁵⁹

Intra-cellular cargo transport, though thought of as a feature of the tubulin network, is also carried out by actin stress fibers in certain cases.¹²² Myosin can slide along actin filaments via its precessive motion without detaching and this phenomenon is often utilized to transport cargo (sometimes as big as whole organelles such as lysosomes) through the actin network.¹²³ Actin cytoskeleton is also reported to be involved in endocytosis¹²⁴ and exocytosis¹²⁵ of cells.

III. CANCER AND ACTIN DYNAMICS

A. Lung cancer

As one of the leading causes of death in the world, lung and bronchial cancer resulted in more than 135 000 deaths (22.4% of cancer deaths) in the United States in the year 2020.^126–128 Smoking causes 80% of lung cancer fatalities. Radon, asbestos, long-term, and cumulative air pollution exposure, notably PAH emission exposure, and personal or family lung cancer history are all risk factors.^129,130 As per the World Health Organization (WHO), lung cancer may be broadly classified as non-small cell lung cancer (NSCLC), which accounts for about 80%–85% of lung cancer cases, and small cell lung cancer (SCLC), which accounts for the remaining 15%.^131,132 NSCLC includes adenocarcinoma, squamous cell carcinoma, and giant cell carcinoma (LCC). Each molecular targetable genetic subtype can be subcategorized. The 5-year survival rate for NSCLC and SCLC metastatic lung cancer is about 4%.^126,129 Although surgery, chemotherapy, and irradiation are used in the treatment of NSCLC and SCLC, more novel treatment strategies are needed in order to cure lung cancer, especially the late-stage malignancies.¹³¹ Recent studies have revealed the association between EMT and lung cancer, especially in the context of mechanism, tumor progression, resistance development to drugs, metastasis, tumor vasculature, and cancer stem cells; hence, targeting EMT-associated unique proteins can be an exciting tumor targeting strategy. Cancer is usually identified and characterized by uncontrolled growth of malignant cells in a tumor that starve the neighboring healthy tissue.¹³³ Cells in a healthy tissue possess a property known as contact inhibition, which prevents them from overgrowing thereby maintaining equilibrium in the tissue. Through certain genetic mutations cells lose their “contact inhibition,” which enables them to proliferate uncontrollably and form tumors.¹³⁴ The ability of tumor cells to metastasize and colonize to distant organs depends on tumor cells-induced neovascularization to create vascular linkage to the tumor, a process referred to as angiogenesis,¹³⁵ the metastatic aspect of which will be discussed in Sec. III B.¹³³

B. Metastatic cancer

Like the outer layer of organs of the body, tumors are also primarily comprised of epithelial cells, which adhere to one another¹³⁶ rendering them incapable of migration. Another characteristic of epithelial cells is their apical to basal polarity,^137,138 which essentially translates into them having distinct ends that point to the inside and outside of the tumor (organ). A Ca²⁺-mediated adhesion protein named E-cadherin, which is expressed specifically in epithelial cells, forms the epithelial sheet, which aids in the cell–cell adhesion.¹³⁹ The primary tumor, which forms due to certain mutations such as p53¹³⁴ and uncontrolled cell growth, still consists of epithelial cells; thus, even with all the cancer related mutations, these tumor cells are incapable of leaving the primary tumor site. The spreading of cancer via metastasis requires the migration of these tumor cells beyond their original location. The process through, which these non-migratory epithelial cells gain their motility during metastasis is called epithelial mesenchymal transition (EMT).^{121,140–143} This transition, as indicated by the name, transforms the epithelial cells into a different phenotype known as mesenchymal phenotype. Mesenchymal cells, unlike epithelial ones, do not exhibit cell–cell adhesion which is demonstrated by the drastic decrease is E-cadherin expression level in the mesenchymal cells, rendering them highly motile.¹⁴⁰ Mesenchymal cells also demonstrate a lower level of apoptosis, resulting in a high degree of success in their ability to form secondary tumors.¹²¹ The phenomenological conversion of epithelial cells into mesenchymal type is actually a protocol that healthy organisms use during embryonic development¹⁴⁴ and wound healing, which is later co-opted by cancer cells. However, the genetic protocol involved in each of these EMT types vary slightly from one another.

Apart from maintaining the epithelial sheet, E-cadherin is also responsible for contact inhibition of cells, which is a signaling pathway of the cell which arrests overgrowth of an epithelial layer, thereby suppressing formation and growth of tumors.¹³⁹ During EMT, the loss of E-cadherin is compensated by another form of cadherin known as N-cadherin,¹⁴⁵ the expression of which initiates the MAPK/ERK pathway of EMT.¹⁴⁶ The inhibition of E-cadherin and the concurrent upregulation of N-cadherin^147,148 is affected by proteins such as Zeb,¹⁴⁹ Snail, and Slug,¹⁵⁰ which are also upregulated during EMT. The enhanced motility of the mesenchymal cells^151,152 can be ascribed to the formation of stress fibers,^153,154 which also modifies the elastic properties^155–160 of the cells as is required to navigate through the extremely crowded tumor environment. Apart from the previously described function of stress fibers in cell migration, perinuclear actin caps play a very special role in metastasis as well. When cells enter the blood vessels, they must pass through very small pores, smaller the size of their nucleus. The process of deforming the nucleus to enter through small pores is called nuclear blebbing;^161,162 but cells do not survive unless they can regain the shape of their nucleus after the deformation.¹⁶³ Perinuclear actin caps play a key role here in post-blebbing recovery and eventual maintenance of shape and size of nucleus.^164,165

There are multiple theoretical models for the formation of stress fibers, which have been studied and validated, but the details of the cytoskeletal reorganization during EMT are not well understood. It is known that focal adhesions sites are essential for the nucleation/formation of stress fibers;⁶¹ this is validated by the fact that vimentin, an intermediate filament and another marker protein of EMT,¹⁶⁶ is known to increase the formation and maturation of focal adhesion spots.^167,168 The vimentin intermediate filament (VIF) cytoskeleton forms early in EMT to be eventually replaced by the actin cytoskeleton.¹⁶⁶ It is worth noting that in certain cell lines, not only do actin and vimentin cytoskeletons coexist, they even form an inter-penetrating network.¹⁶⁹

Actin dynamics share an extensive amount of genetic cues with EMT (Fig. 4).¹⁷⁰ Like actin polymerization, the Rho family of proteins play a very integral role in EMT, serving as one of the most upstream proteins in the EMT signaling cascade.^140,170 There is a growing recognition that though often described as a relatively downstream event in EMT, stress fiber formation can also facilitate EMT forming a loop in the signaling pathway.¹⁷⁰ Recent studies have also revealed that EMT is not a binary process where the cells transition from epithelial to mesenchymal directly; but the epithelial cells traverse through multiple intermediates states before reaching the mesenchymal phenotype.^96,171 These intermediates states are predicted and identified by RNA¹⁷¹ sequencing and can have unique cytoskeletal signatures as well, further demonstrating the extensive crosstalk between the actin cytoskeleton and EMT.⁹⁶ MAPK/ERK pathway, which is another pathway that is actively involved in EMT signaling, has a considerable amount of crossover with the Rho-GTPase pathway.^172,173 This is evidenced from the studies that show that a known inducer of EMT in mammalian cells named transforming growth factor β (TGFβ) initiates EMT through the activation of both the Rho-GTPase and MAPK/ERK pathways.¹⁷⁴ An emergent pathway which has recently been reported to be relevant for EMT is the Wnt pathway,¹⁷⁵ which again has crossover with the MAPK/ERK pathway through the Jun-N-terminal kinase (JNK) protein; but both these pathways can affect and initiate EMT independently as well.¹⁷⁶

FIG. 4.

Genetic pathways of actin regulation and cytoskeletal reorganization during EMT. The figure shows the interconnected nature of Rho/ROCK, Rac, and Cdc42 pathways and how they affect actin organization. Cdc42 is primarily responsible for N-WASP-mediated activation of Arp2/3 which is required for actin polymerization and branching. Rac is known to affect cofilin (through LimK), which is another protein involved in actin polymerization. Rho pathway affects the phosphorylation of myosin light chain which controls the contractility of actin stress fibers. Reproduced with permission from Fife et al., Br. J. Pharmacol. 171, 5507 (2014). Copyright 2014 John Wiley and Sons.¹⁷⁰

Although EMT confers increased invasiveness and metastasis to primary solid tumors, secondary tumors that have spread to distant locations often retain the histology of the primary tumor. Mesenchymal-epithelial transition (MET) links the reversibility of EMT in secondary mesenchymal tumors enabling the acquisition of the primary tumor's epithelial characteristics.¹⁷⁷ As with EMT, the MET process is very dynamic; cells exist in numerous hybrid states, and further research is required to comprehend the underlying molecular pathways. Karacosta et al. used a neural-net-based computational method (TRACER) to dissect EMT and MET trajectories in lung cancer cells and created an EMT-MET PHENOtypic STAte MaP (PHENOSTAMP).¹⁷¹ The role of tumor cell heterogeneity on EMT-MET transitions and the role of tumor mutation on MET phenotype also need attention. More studies are needed to understand the genetic and molecular nuances that can be used in the phenotypic–genotypic characterization of clinical samples and help develop EMT-MET-directed therapeutics. It is worth noting that the cytoskeletal aspects of MET are equally fascinating, but unfortunately beyond the scope of this review.

C. Pre-cancer

Lung cancer is considered a stealth disease and develops due to a sequence of pro-tumorigenic genetic and epigenetic alterations that transform the epithelium from normal to pre-malignant lesions to in situ carcinoma. Even with current advances in treatment, over 50% of lung cancer patients die due to metastases and more than half of them die within the first year of diagnosis. While metastatic activity is generally seen as a late stage event, there is strong evidence that tumor cells spread through EMT early in development of lung tumors.^178–182 Mutant lineage originating from mutant clone can expand and populate widespread fields and are predisposed to eventually initiate tumors even before primary tumor has formed¹⁸³ posing a great challenge for conventional treatment approaches. This phenomenon is called “field cancerization,” a paradigm reported in multiple carcinomas.^184,185 Pipinikas et al. provided molecular evidence by studying synchronous preinvasive biopsies that field cancerization of the upper airways may proceed by cell migration rather than local contiguous cell growth as previously believed.¹⁸⁶ Using longitudinal tracking, they demonstrated that a single P53 p.E294fs*51 gene mutation was observed at multiple bronchial lesions in patients. Data suggest that the accumulation of genetic and molecular anomalies after carcinogen exposure culminates in selecting cells with metastatic potential within pulmonary pre-malignant lesions, resulting in geographically discrete and clonally linked lesions.¹⁸⁷ Long-term monitoring of patients with preinvasive endobronchial lesions show that there was a higher likelihood of developing lung cancer in those with high-grade preinvasive endobronchial lesions than low-grade lesions in the lungs,¹⁸⁸ strongly supporting the concept of field cancerization.¹⁸⁹ Another exciting piece of evident is the detection of disseminated tumor cells (DTCs) and circulating tumor cells (CTCs) in blood, lymph nodes and bone marrow confirmed using highly sensitive techniques as early as stage I non-small cell lung cancer (NSCLC) patients.^179–182 CTCs are rare tumor cells that travel via the circulation, undergo mesenchymal-epithelial transition (MET), and metastasize to distant organs.¹⁹⁰ It has been recently demonstrated that certain non-tumor cells also demonstrate high motility and apoptotic resistance like mesenchymal cells.^178,191 Thus, evidence suggests that micrometastic lesions are often present at the time of surgery but are not detectable with existing imaging techniques.¹⁸⁴ Therefore, cancer interception is a proactive transformational strategy for targeting pre-malignant stage disease pathogenesis to stop the progression to aggressive cancer.¹⁹²

The “linear progression models” and the “parallel progression” model explain the metastatic progression and relies on EMT.^193–196 According to the “linear progression” paradigm, EMT occurs only in a small percentage of cells at the leading invasive edge of the advanced tumor, hence accelerating the last stage in tumor propagation, i.e., metastasis. It was recently postulated that EMT enhances lung epithelial cell spread before and concurrent with malignant conversion, a process dubbed the “parallel progression” model.¹⁹⁷ This concept represents a paradigm leap in our knowledge of cancer growth, development, and metastasis. The factors enabling the “parallel progression” in lung cancer are unknown. Hait and Levin hypothesized that pre-malignant lesions would include decipherable interception targets capable of halting the development of malignancy. Mapping genotype and phenotype relation and elucidating the range of intermediate EMT and MET states in clinical samples holds the possibility of proving insight into phenotype dimensions responsible for cancer development and medication resistance.

Pre-malignance and cancer development biology is still poorly understood, especially in terms of EMT in lung cancer. Clonal population amongst a pre-malignant lesion may vary significantly in gene expression and have advantageous cellular biophysical parameters that affect stiffness, motility, and migration. Pagano et al. recently identified a discrete subpopulation of human airway pre-malignant (p53 null, activated Kras-G12D) epithelial cells that differed in terms of biophysical, molecular, and metastatic features from the rest of the population. EMT induces changes in mechanical characteristics, mainly owing to increased cell contractility and the production of actin stress fibers.^178,191 The molecular mechanisms that control actin dynamics during EMT in pre-malignant cells remain unknown. With the help of a novel restricted migration selection method and various physiomic techniques, such as deformability cytometry and atomic force microscopy, the team discovered a highly motile (HM) subset of human bronchial epithelial cells (HBEC) that had improved heritable migrating fitness observed both in vitro and in vivo. Comparative RNA-seq data and confocal live cell imaging demonstrate that HM-HBECs exhibit enhanced migration and expression of critical EMT genes, thus serving as a unique model for studying pre-malignant cell migration.¹⁷⁸ Compared to parental HBECs, HM-HBECs were shown to accumulate robust actin stress fibers. Though these highly migratory cells exhibit stress fibers, they are not entirely mesenchymal. While this study sheds light on molecular underpinnings of pre-malignant change, additional research is necessary to understand better the role of actin epithelial mesenchymal plasticity (EMP), intermediate stages of EMT, and the mechanisms underlying the pre-malignant migration that can aid in its targeting.¹⁹⁸

Recent studies have pointed out that EMT is a non-binary phenomenon and renewed attention is paid to the intermediate phases. Recent studies have characterized the molecular and epigenetic landscape regulating the intermediate “metastable” phases of EMT in different tissue systems.^199–201 Deciphering pro-tumorigenic EMT-actin cytoskeletal rearrangement crosstalk and phenotypic understanding of intermediate EMT plastic states associated with pre-malignant migration can facilitate the designing of novel lung cancer interventions. Methods to study the dynamics of actin polymerization states can provide the clues to quasi-epithelial/mesenchymal phenotypic dimensions.

IV. ANALYTICAL METHODS

Over the years, a myriad of approaches has been used to study the regulation of actin cytoskeleton of cells. For analyzing simple up- or downregulation of actin western blots are the most widely used technique.^202,203 Gown et al.²⁰² and Kinner et al.²⁰³ showed that even the different isoforms of actin can be identified from tissue cultures of animals using western blots. In a similar study, Tsai et al. demonstrated differential expression of one actin isoform, the smooth muscle actin can affect the migration and healing of tendon cells.²⁰⁴ However, change of the overall amount of actin in cells is usually a very extreme response; in most cellular processes, actin gets regulated and recycled into different structures, such as cortical actin, filaments, and stress fibers. In fact, the amount of β-actin is usually so constant that it is used as a loading control in western blots where expression levels of related proteins are compared.⁹⁶ Also, the cell has mechanisms to increase the concentration of actin in certain areas of the cell locally without affecting the overall amount; this is often seen in keratinocyte migration.^57,60 Western blot is performed on cell lysate, so it cannot provide any information regarding sub-cellular concentration changes or the conversion of one form of actin to another. Also, as western blot analyzes a whole cell population in a single experiment, it will also miss out any special cells that behave differently in the population, such as the highly migratory sub-population in premalignant lung cancer cells.¹⁹⁸ Some of these problems can be solved by employing quantitative fluorescence microscopy, such as stress fiber formation or concentration of actin in one side of the cell. However, the actin monomer background limits the quality of the images. Probes like Lifeact,²⁰⁵ which selectively bind to actin filament and not monomers, mitigate the problem of high background to a certain extent. With the advent of multicolor super-resolution microscopy, fluorescence imaging has become an indispensable tool in studying actin dynamics.^206–208 Apart from super-resolution, other techniques such as total internal reflection fluorescence (TIRF), can also decrease the background by limiting the thickness of the sample which is excited; the first visualization of myosin “walking” on actin filaments was carried out using polarized TIRF microscopy.²⁰⁹ However, little to no quantitative information can be extracted from the fluorescent images. Despite the lack of quantitative aspects, some studies have explored the actin polymerization depolymerization processes using fluorescent microscopy.^210,211 A more quantitative technique, fluorescence recovery after photobleaching (FRAP), which can measure the lifetime of the recovery, has been employed to study actin polymerization; specifically, FRAP has been used to verify the active transport and enhanced polymerization rate of actin to the leading edge of keratinocytes during migration.²¹² Fluorescence correlation spectroscopy (FCS) can quantify the motion of fluorescent particles. The is a difference between the free diffusive motion of free actin monomers compared to the motion of actin monomers that are part of a filament and this phenomenon makes FCS a perfect technique to identify and distinguish between different forms of actin.²¹³ Beyond the spatial resolution of optical/fluorescence microscopy lies the domain of cryo-electron microscopy (Cryo-EM), which is used to delve further into the structural properties of actin.^63,75,214 The faster polymerization at the barbed end of actin filaments due to a conformational change brought about by ATP binding was explained by Chou et al.⁶³ using Cryo-EM. Mentes et al.⁷⁵ studied myosin and its interaction with actin using Cryo-EM to better understand the force sensing and response of myosin. In optical microscopy, the proteins that are going to be imaged need to be labeled, so foreknowledge of which proteins to tag is required; thus, it is not useful or efficient in discovering novel proteins (or known proteins in new locations) in actin stress fibers. In such cases, where the proteins involved are not well known, a pull-down assay is carried out followed by mass spectrometry to identify the fragments that were pulled down with the actin filaments indicating that they were likely attached to the filament.²¹⁵ It has also been used to analyze the dynamic interaction between actin filament and ABPs.²¹⁶ A somewhat indirect, but highly quantitative technique is quantitative RNA sequencing that generates information regarding the expression levels of genes that are involved in actin regulation.²¹⁷ With modern techniques like single cell RNA sequencing, Pastushenko et al. studied the expression actin and ABP related genes to identify and characterize intermediate EMT states.²⁰⁰

There is a growing recognition that there is a vast amount of quantifiable information in fluorescent image (especially of actin) which is not extracted or exploited in most biological studies. This led to recent efforts of using image processing techniques to extract more quantifiable information from fluorescent images, such as localization of ABPs or the relative organization of stress fibers.^96,218–221 Rimoli et al.²²⁰ evaluated the orientation of single fluorophores using super-resolution technique STORM, eventually inferring the three-dimensional orientation information of the cytoskeleton. X-ray diffraction has also been used to quantify the relative orientation of myofibrils; Ma et al.²²¹ proposed this relative orientation as a metric for characterizing heart diseases.

V. ACTIN IMAGE QUANTIFICATION METHODS

A. Image analysis methods

As a dynamic network that can react to external and internal stimuli, the actin cytoskeleton often exhibits that response in terms of polymerization or depolymerization with the goal of balancing the stress (stimulus). Thus, it is possible to evaluate the correlation between relative amount of different types of stress fibers, their relative orientation, and the stimulus responsible for bringing about the change. It has already been shown using fluorescence microscopy that external forces like fluid shear stress,^222–226 simulated microgravity,^227–231 cyclic stretching,^232–234 etc., all of which mimic biological scenarios, induce drastic cytoskeletal response in cells. Fluorescence microscopy provides a platform to study biological processes ranging from a few seconds to multiple days and with the development of a library of super-resolution techniques, each with its own time-resolution window, the spatial resolution of these images has improved vastly without losing any temporal resolution.^235,236 However, no matter how accurate, an image or video is subject to interpretation that can lead to bias or other manual errors. So, methods for objectively and reliably quantifying different parameters from cytoskeletal images have garnered immense attention.

There are three broad categories of the basis of extraction in the existing strategies: filament directionality, filament network, and single filament extraction^237–239 (Fig. 5). In analyses that focus of filament directionality, earlier approaches involved Fourier transform followed by filtering the Fourier transform coefficients.²⁴⁰ More sophisticated transforms such as wavelets,^241,242 curvelets,^243,244 shearlets,^245,246 and contourlets²⁴⁷ have been developed and used in recent studies. As the shape and geometry of the features understudy are taken into consideration in these transforms, each of these transforms are uniquely suited for specific applications based on the features developed in the biological process under study. The orientation information is then extracted from the basis function of these transforms. Though these processes have low levels of error due to not separating background and foreground segmentation, but on the downside, that limits their application to only directionality and orientation related analyses.

FIG. 5.

Summary flow chart of the different categories of existing strategies for fiber extraction. The basis for extraction of filaments is shown in red (second row). The information extractable from each of these strategies is shown in blue (third row). Reproduced (with alterations) from Alioscha-Perez et al., PLoS Comput. Biol. 12, e1005063 (2014). Copyright 2014 Authors, licensed under Creative Commons Attribution (CC BY) License²¹⁸ and Fig. 1 (Chap. 2) [reproduced with permission from A. Basu, “Statistical Parameterization of the Cell Cytoskeleton (SPOCC): A noninvasive image quantification tool to identify and track sub-cellular processes,” Ph.D. dissertation (University of California, Los Angeles, 2021)].²⁶⁵

In the case of network based image analysis techniques, the whole filament network is extracted as a whole from the underlying image.²⁴⁸ Further improvement can be achieved by developing templates that contain some information regarding the target and then filtering the image (network) with respect to that template matching.²⁴⁹ These techniques and algorithms have a limitation in detecting network junctions, so they do not perform well in systems where there is a significant amount of overlap between the filaments.²¹⁸ Incorporation of linear models for filaments can solve this problem partly in contour based methods.²¹⁸ Another approach information regarding velocity and density is extracted from these methods instead of orientation, which is especially relevant for understanding polymerization/depolymerization processes.²⁵⁰

One of the most important steps in feature extraction algorithms is the segmentation of an image into multiple parts. Usually in an image, apart from the feature of interest, there is a non-negligible amount of background fluorescence (noise) and it is essential to separate that from the features. As the background noise has a frequency much higher than the signal, simple or directed Gaussian filters²⁵¹ or Laplace filters²⁵² are used to filter out the background in the frequency domain. Though this process gets rid of the background noise, the left over signal can have artifact features, such as aggregates or debris, along the feature of interest which has to be separated as well. If the feature of interest has a flowing pattern like in the case of filaments, then coherent enhancements filters²¹⁸ can distinguish them from artifacts. However, as cells are three dimensional in nature, the out of focus blurring creates a problem for these methods. Another transform that has been used for denoising in some studies is called Radon transform.²⁵³

The separation of filaments from the noise and artifact has to be followed by binarizing the filament network. In this process, pixels that are part of the filament image are separated from the ones that are not. Binarization is usually carried out using thresholding protocols, such as global thresholding, Otsu's brightness thresholding,²⁵⁴ and local thresholding based on neighboring regions.^238,255 The binary image can then be used to extract more information, such as filament length, thickness, and orientation. To obtain the information of individual filaments, they need to be extracted as separate entities and this is done via multiple line detection algorithms. Both the thresholding and line detection steps are computationally intensive and thus require some balance between accuracy and computational load.²³⁹ In certain techniques, the lines extracted from the line detection algorithm are iteratively merged to create longer filaments, which more closely mimic the real system.²¹⁸

B. Orientational order parameter

Physical characteristics, such as size, aspect ratio, and flatness, have been assigned to cells as figures of merit.^114,256 Similarly, properties of the cytoskeleton can also be used to characterize the cells. Especially due to the extensive correlation between the cytoskeletal organization and physical properties, such as elasticity and motility,^60,61 these cytoskeletal figures of merit can be more informative and biologically significant than others. The most intuitive feature of the cytoskeleton that can be quantified and used for characterization is the length of individual filaments or the whole filament network; but a caveat of such an approach is that the extracted length of individual filaments is limited by the line accuracy of the line detection algorithm which often breaks up the larger curved filaments into smaller straight line and the overall length of the network can vary vastly from cell to cell depending on their size.²¹⁸ However, the relative orientation of stress fibers does not suffer from the aforementioned problems and provides a better figure of merit for cellular characterization and provide insight into cytoskeletal properties. Relative orientation also has a direct relation with the type of fibers.⁵⁹ By the virtue of their structure and FAK capping, transverse arcs and dorsal stress fibers form a disorganized pattern, whereas ventral stress fibers are more likely to be situated parallel inside the cell.⁵⁹ Based on the relative number of different types of fibers and their positions, the angular distribution of fibers can be broad or narrow. To further quantify this distribution, they have been fitted to a range of mathematical equations ranging from simple Gaussian or normal distribution²⁵⁷ to more sophisticated von Mises circular distribution²⁵⁸ (wrapped normal distribution). The standard deviation of the distributions, or an equivalent parameter for more complex distributions, is used as a figure of merit to quantify and characterize the relative orientation of the fibers.^257,258 Apart from the cell cytoskeleton, these models have been used in evaluating orientation in other systems such as hydrophobic elastomers²⁵⁹ and cells in a soft biological tissue.²⁶⁰ However, in the absence of a universal empirical model for actin regulation, fitting this distribution to any known mathematical distribution is likely to introduce errors in the analysis.

An alternate mathematical protocol, which does not fit the data to any mathematical distribution model or require any foreknowledge of the actin reorganization, is to calculate the orientational order parameter (OOP) from the angular distribution.²⁶¹ This method has been used to evaluate the angular organization of biological systems, such as bacterial cultures,²⁶² fibroblasts,²⁶³ and cardiac muscles.²⁶⁴

The first step of OOP calculation is to generate an order tensor for every angle; angles are treated as vectors in this analysis

$$\mathrm{Angle}\hspace{0.17em}\stackrel{\mathit{yields}}{\to }\left[\begin{array}{c}{p}_{i,x}\\ p_{i,y}\end{array}\right],$$ (1)

$$\mathrm{Order}\hspace{0.17em}\mathrm{tensor}=\left[\begin{array}{cc}{p}_{i,x}{p}_{i,x}& p_{i,x}{p}_{i,y}\\ p_{i,x}{p}_{i,y}& p_{i,y}{p}_{i,y}\end{array}\right].$$ (2)

Then, the mean order tensor is calculated for the distribution

$$\mathrm{Mean}\hspace{0.17em}\mathrm{order}\hspace{0.17em}\mathrm{tensor}=T=⟨2\left[\begin{array}{cc}{p}_{i,x}{p}_{i,x}& p_{i,x}{p}_{i,y}\\ p_{i,x}{p}_{i,y}& p_{i,y}{p}_{i,y}\end{array}\right]-\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]⟩.$$ (3)

Orientational order parameter is defined as the maximum possible eigenvalue of the mean order tensor

$$\mathit{OOP}=\text{max}\left[\mathrm{eigenvalue}\left(T\right)\right].$$ (4)

If the p_x and p_y components of the angle vectors are replaced with cos(α) and sin(α), respectively, then the equation can be simplified further.

$$\mathit{OOP}=⟨\text{cos}\left[2\left(\alpha -{\alpha }_0\right)\right]⟩,$$ (5)

where α₀ is the director of the distribution.

Theoretically, for completely ordered systems, the OOP value should be 1 and in the case of a completely disordered system (complete random distribution of angles) the OOP value should be 0. However, is real biological cases, the system is neither completely ordered nor completely random, so the OOP value can range from 0 to 1; higher OOP values correspond to higher degree of alignment or order (Fig. 6). This ability of OOP to quantify order in the system has been validated by simulated data with Gaussian and normal distribution.

FIG. 6.

Angular distributions with different levels of order and their corresponding OOP values. Individual arrows correspond to the direction of individual vectors (angles). As the order of the system decreases from the left panel to the right panel, the OOP correspondingly deceases. This image is reproduced from A. Basu, “Statistical parameterization of the cell cytoskeleton (SPOCC): A noninvasive image quantification tool to identify and track sub-cellular processes,” Ph.D. dissertation (University of California, Los Angeles, 2021)²⁶⁵ (Fig. 2, Chap. 2).

C. Filament extraction and quantification methods

Combining the image analysis methods with orientational order parameter, researchers have developed techniques that can reliably quantify the cytoskeleton and its dynamics from fluorescent images.⁹⁶ Here, we will be discussing three such techniques and their possible applications.

Statistical parameterization of cell cytoskeleton (SPOCC) is an image analysis technique developed by Basu et al. which can quantify different aspects of the cytoskeleton, such as amount of stress fibers and their relative organization during dynamic cellular processes.^96,265 In this method, the fluorescent image is assumed as a sum of multiple components, such as filament image, artifact image, and noise. Sparse source separation approach is used to calculate the filaments and artifact images. Filaments demonstrate a curvilinear geometry, which is entirely lacking in artifact structures, and this difference is exploited is their individual evaluation.²⁶⁶ Curvelet transform was used for modeling of the filaments image whereas the artifact image was modeled via undecimated wavelet transform. This image decomposition protocol follows the morphological component analysis lab (MCALab) libraries, using variable iterations.²⁶⁷ The filaments in the filaments image are then enhanced using a sequence of Gaussian, directed Gaussian, and Laplacian filters.

Following the signal enhancement step, the filament network is segmented using a multi-scale line detection algorithm.²⁶⁸ This step iteratively analyzes the neighborhood of every pixel in the image, and each individual pixel in the image is assigned a likelihood of being part of a line of a defined width. The likelihood of every pixel being the center of a line oriented in any particular direction is also calculated. The output likelihood image is then binarized using Wellner's adaptive thresholding,²⁶⁹ which generates an image of only those pixels that are part of a line.

A list of line segments with a fixed (minimum) length is obtained from the binarized image through line segmentation. In this step, a line segment (fixed length) is used to fit the non-zero pixels via least squares fitting. Any two lines that overlap and have the similar directionality (based on a pre-set curvature threshold) are then stitched together to generate longer lines, which better mimic the actual filaments. Orientational order parameter (OOP) is then calculated from the list of angles (protocol described in Sec. V B^261,262,264), which acts as a figure of merit for the relative alignment of fibers (Fig. 7).

FIG. 7.

Flowchart of individual steps involved in statistical parameterization of cell cytoskeleton (SPOCC). Two cells with different degrees of angular order are processed using SPOCC. The individual steps that move the algorithm ahead is shown in different colors (bottom row) and the outputs of individual steps are shown in red (top row). Scale bar: 16 μm. This image is reproduced with permission from A. Basu, “Statistical parameterization of the cell cytoskeleton (SPOCC): A noninvasive image quantification tool to identify and track sub-cellular processes,” Ph.D. dissertation (University of California, Los Angeles, 2021) (Fig. 3, Chap. 2).²⁶⁵ The cells and the corresponding data displayed in this figure are from Basu et al., Commun. Biol. 5, 407 (2022). Copyright 2022 Authors, licensed under a Creative Commons Attribution (CC BY) License.⁹⁶

The second technique uses a very similar pipeline and is named focal adhesion filament cross-correlation kit (FAFCK).²¹⁹ Here, the cytoskeletal filaments are extracted using the method described by Eltzner et al.²³⁸ From the binarized image a width map is created based on the relative clustering of non-zero pixels around a certain pixel. Next, for every non-zero pixel from the width map, the maximum segment length is estimated for the neighboring pixels in every direction. Now, these line segments are discarded if there is a high degree of variation in pixel orientations. The technique eventually extracts the location, length, and individual width of filaments.

They have further quantified and cross-correlated the localization of focal adhesion kinase (FAK) caps with the actin stress fibers. The FAK image is binarized by automated or manual thresholding from which they evaluate the area and aspect ratio of the FAK spots. Depending on the number of FAK cappings on individual stress fibers, they can predict the type of stress fiber as well (Fig. 8).

FIG. 8.

Flowchart of individual steps involved in focal adhesion filaments cross-correlation kit (FAFCK). In this algorithm, the actin fiber and focal adhesion datasets serve as the input. Then for every filament, the algorithm identifies if there are one or more focal adhesions on their ends and are sorted accordingly. Filaments that are shorter than the size of a focal adhesion are rejected and focal adhesions can be screened based on a predetermined user-chosen shape. Reproduced from Hauke et al., PLoS One 16, e0250749 (2021). Copyright 2021 Authors, licensed under Creative Commons Attribution (CC BY) License.²¹⁹

The third study is where an order parameter is used as a quantifier of cytoskeletal alignment.²⁷⁰ In this work, Marcotti et al. developed an algorithm to evaluate the alignment using fast Fourier transforms (FFT). The whole image is broken down into multiple windows of defined (customizable) size. Existence of aligned features in these windows result in the FFT domain being skewed orthogonally to the direction of alignment in the frequency domain. The orientation of FFTs of all such windows are used to calculate the orientational order parameter. The length scale of the cytoskeletal alignment can be quantified by changing the size of the windows used for FFT. They have demonstrated the efficiency of their algorithm by measuring the change in cytoskeletal alignment of cells undergoing Rho-kinase inhibitor treatment.

D. Application of SPOCC to the analysis of actin fibers in EMT and in pre- and fully developed lung cancer

SPOCC has been used to quantify the dynamic actin remodeling in lung cancer EMT⁹⁶ as well as pre-cancer cells.^198,265 This work has identified a novel partial-EMT intermediate with unique cytoskeletal features. They have also demonstrated that there is an increase in the relative alignment of fibers with the progression of EMT, which can be quantified by SPOCC using OOP as a figure of merit. Thus, the group has developed the relative alignment of stress fibers as a reliable marker for EMT progression. The reliability of OOP as a reporter for EMT was verified by protein expression levels and elastic properties of the cells.⁹⁶ It has also been demonstrated that SPOCC can be used to quantify the drug/inhibitor response of these cells. SPOCC can extract the amount of stress fibers present in cells along with their location and orientation. As drug response of cells usually entail a change in the amount of stress fiber or their distribution, SPOCC can be used as an effective screening tool.

Apart from lung cancer cells, SPOCC has also been employed to identify pre-malignancy. A sparse sub-population of highly migratory cells are reported in pre-malignant bronchial cells.^178,265 These cells show unique cytoskeletal features that is consistent with their higher motility. They also respond to drugs/inhibitors differently compared to the rest of the population. The same group has employed SPOCC to quantify the difference in the cytoskeleton of the high migratory cells. They have also tracked the drug response of these cells using SPOCC to identify the differences between the high migratory and low migratory populations.^198,265

VI. FUTURE PROSPECTS

The image quantification technique discussed here could possibly have broad biological applications in the future. The correlation between cytoskeleton and physical properties of the cell, such as motility and elasticity, is well known. Our method quantifies this correlation and hence can serve as benchmark for future studies. Moreover, a library of correlations between several different relevant observables can aid in indirect measurement of motility and elasticity of cells in cases where direct mechanical measurements are not possible (for example, when the biological processes being tracked are too fast to be captured by mechanical observables).

In cancer biology, genetic pathways and specific events are often identified using sequencing or mass-spec analyses, which can destroy the sample. However, as demonstrated here, morphological changes of the cytoskeleton could serve as a reporter of such events. Albeit indirect, using available imaging techniques, cells can be imaged noninvasively and re-used for further studies. This makes image quantification a suitable candidate for screening or understanding of precious samples. More importantly, cytoskeletal imaging can be multiplexed with other imaging-based quantification methods, such as fluorescent in situ hybridization (FISH) or single-molecule fluorescent in situ hybridization (smFISH), to increase the amount of information extracted from the same sample. Also, given that the cytoskeletal response often mimics physical stresses to the cell, quantifying the cytoskeletal response can help build physical/biophysical models for biological events. Further work is required to evaluate the efficiency and applicability our approach to more biologically relevant conditions, such as tissue sections, 3D culture, and organoid culture of cells.

Image quantification techniques together with artificial intelligence/deep learning (AI/DL) could be made to require very limited user input and hence more immune to user-introduced biases. Further progress is required for integrating AI/DL in these methods to eliminate biases. Finally, such techniques could find broader applications in studying other filamentous networks such as vimentin and tubulin.

VII. CONCLUSIONS

In this review, we have summarized the dynamics and importance of actin and its polymeric forms in biological context. We have discussed how they behave in healthy cells as well as how their behavior changes in diseased ones. The properties of actin binding proteins as well as their importance in the cytoskeleton have also been covered. We have briefly discussed the existing analytical methods to quantify and study actin along with their limitations. We have introduced recent techniques that rely on imaging that can quantify actin and its dynamics in real time with sub-cellular resolution. Coupled with statistical analysis, these methods are highly objective and efficient in studying biological processes involving any change in actin. We have shown the application of such techniques in studying cancer metastasis as well as pre-malignancy. We envision that in the future, these techniques can be coupled with other analysis methods such as FISH²⁷¹ or single cell RNA-sequencing²⁷² to obtain a more comprehensive understanding of the systems. There is further scope of development in terms of integrating artificial intelligence in the image quantification methods to make them more objective. These techniques should also be improved to quantify actin from images with lower magnifications, so as to study more cells at a time and then segment them to obtain single cell information. In conclusion, studying and quantifying actin dynamics accompanying biological processes is a potent and efficient tool, but further coupling of methods is likely to make it more universal.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts to disclose.

Author Contributions

Arkaprabha Basu and Shimon Weiss conceptualized and wrote the manuscript. Manash K. Paul wrote the pre-cancer section of the manuscript.

Arkaprabha Basu: Conceptualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Manash K. Paul: Writing – original draft (supporting); Writing – review & editing (supporting). Shimon Weiss: Supervision (lead); Writing – review & editing (supporting).

DATA AVAILABILITY

Data sharing is not applicable to this article as no new data were created or analyzed in this study.