GPI-anchored protein organization and dynamics at the cell surface

The surface of eukaryotic cells is a multi-component fluid bilayer in which glycosylphosphatidylinositol (GPI)-anchored proteins are an abundant constituent. In this review, we discuss the complex nature of the organization and dynamics of GPI-anchored proteins at multiple spatial and temporal scales. Different biophysical techniques have been utilized for understanding this organization, including fluorescence correlation spectroscopy, fluorescence recovery after photobleaching, single particle tracking, and a number of super resolution methods. Major insights into the organization and dynamics have also come from exploring the short-range interactions of GPI-anchored proteins by fluorescence (or Förster) resonance energy transfer microscopy. Based on the nanometer to micron scale organization, at the microsecond to the second time scale dynamics, a picture of the membrane bilayer emerges where the lipid bilayer appears inextricably intertwined with the underlying dynamic cytoskeleton. These observations have prompted a revision of the current models of plasma membrane organization, and suggest an active actin-membrane composite.

The ability of GPI anchor to incorporate proteins into membrane domains helps in sorting these proteins all along the secretory pathway ( 18 ). There exist different mechanisms by which sorting can take place in the secretory pathway ( Fig. 1C ). The sorting takes place at the ER exit site, in the case of yeast, in a sphingolipid-dependent manner; whereas in mammalian cells, this event primarily occurs at the Golgi ( 14,19,20 ). The sorting process is preceded by posttranslational modifi cation of the lipid tail, so that GPI-anchored proteins are incorporated into specifi c membrane domains and can be transported effi ciently to the cell surface . Depending on the cell type, for example, in MDCK or Fisher rat thyroid cells, GPI-anchored proteins are sorted and routed to the apical or basal membrane, respectively. For the apical sorting of GPI-anchored protein oligomerization, N-glycosylation and raft association have been shown to be important ( 13 ); although the underlying mechanism for basolateral sorting of GPI-anchor proteins is not fully understood ( 21 ). Synthetic chemistry approaches have led to the development of modifi ed GPI analogs and, hence, allowed studying the traffi cking of analogs with modifi ed glycan cores, which showed that despite exhibiting a minor difference in the effi ciency of membrane incorporation, the traffi cking and endocytic fates of these analogs with differing glycan core does not show an appreciable change ( 22 ). However, there is compelling evidence that these analogs differ hugely in their diffusion properties, possibly due to their differential engagement with membrane molecules ( 6 ). Further studies utilizing this powerful synthetic tool would provide answers to several questions prevailing in the fi eld.

THE BEHAVIOR OF GPI-ANCHORED PROTEINS AT THE CELL SURFACE
Once at the cell surface, GPI-anchored proteins exhibit a rich diversity of dynamic behaviors in terms of their diffusion, organization, and interactions with other membrane-resident proteins. This has indeed been an active area of research for the past three decades. Studying these behaviors has been instrumental in shaping the modern understanding of how specialized domains in the cell membrane could be organized, maintained, and utilized for signal transduction. In the next sections, we provide a chronological perspective of how the study of GPI-anchored proteins has led to a complex understanding of cell membrane organization and structure .

GPI-ANCHORED PROTEINS AND LATERAL HETEROGENEITIES IN THE CELL MEMBRANE
The plasma membrane has been thought of as a fl uid mosaic, as proposed by Singer and Nicholson ( 23 ) in 1972. The fl uid-mosaic model was a simple, yet elegant, model, rooted in the principles of equilibrium thermodynamics of lipid-protein interactions. Here the cell membrane essentially behaves like a "fl uid," where membrane lipids act as "solvents" for the TM and peripheral membrane proteins.
adhesion. This has implications in health and disease, because GPI-anchored proteins carry out diverse functions in different organisms by acting as surface hydrolases, coat proteins, protozoan antigens, toxin binders, receptors, and so on ( 7 ). Moreover, impairment of GPI anchoring is also implicated in a large number of diseases, such as the formation of the scrapie form of the prion protein, the causative agent for Creutzfeldt-Jakob disease. Paroxysmal nocturnal hemoglobinuria is caused by the absence of GPI on the membrane. The role of GPI anchoring is necessary for embryonic development ( 8 ), and its perturbation is the cause of several neurological disorders ( 9 ), abnormal cell growth in yeast ( 10 ), and the survival of many protozoan parasites ( 11 ). A study of the cell surface dynamics and the organization is crucial to provide an understanding of their many roles.

DELIVERY OF GPI-ANCHORED PROTEINS TO THE PLASMA MEMBRANE
The GPI anchor plays an important role in delivering the attached membrane protein to the plasma membrane, and this has been addressed in a review from Muniz and Reizman also in this thematic review series. The plasma membrane domain to which the GPI-anchored protein localizes also depends on the cell type; for example, in Madin-Darby canine kidney (MDCK) cells, GPI-anchored proteins are present mainly on the apical surface. On the contrary, in Fisher rat thyroid cells, they are sorted to the basolateral domain; the information for targeting the proteins to these distinct locations lies in the GPI anchor sequence ( 12,13 ). The mechanism by which sorting takes place from the Golgi to the appropriate membrane domain also depends on the cell type ( 13 ).
Posttranslational modifi cation of the GPI anchor also affects its sorting at the level of the ER. Unlike TM proteins, GPI-anchored proteins do not have a cytoplasmic domain that can directly interact with coat proteins. Hence, the modifi ed GPI anchor effi ciently interacts with protein complexes, like the p24 family of proteins, which selectively sort them into COPII-containing vesicles for anterograde transport, thereby establishing quality control at the site of protein attachment itself ( 14 ). Knocking down Sec24, an interactor of the p24-p23 complex, was shown to impede the traffi cking of GPI-anchored proteins from the ER to the Golgi, while the other cargo molecules remained unaffected ( 15 ). In yeast, the p24 protein complex acts as an adaptor between GPI-anchored proteins and the CO-PII coat by recognizing the remodeled GPI-anchored proteins. On the contrary, in mammalian cells, the p24 protein complex mediates the concentration of GPI-anchored proteins to ER exit sites, because the lipid remodeling occurs much later at the Golgi ( 14 ). Posttranslational modifi cations of GPI-anchored proteins, which involve remodeling enzymes like post GPI attachment to protein (PGAP)1 and PGAP5, are also known to tightly regulate ER-Golgi transport ( Fig. 1B ) of several GPI-anchored proteins in mammalian cells ( 14,16,17 ).
In the next sections, we focus on the biophysical and microscopy studies on cell-surface dynamics of GPI-anchored proteins, which have given pointers regarding the development of a new model of membrane organization that has been driven by these studies.

DIFFUSIONAL DYNAMICS OF GPI-ANCHORED PROTEINS
Fluorescence recovery after photobleaching (FRAP), single particle tracking (SPT), and fl uorescence correlation spectroscopy (FCS) are three widely used tools to study the diffusion of GPI-anchored proteins . Although all three techniques monitor diffusion, they do so over different spatial-temporal regimes, which are in the order of SPT (10-20 nm, ~1 ms) < FCS ( ‫ف‬ 40 nm-1 m, ‫ف‬ 100 s-100 ms) < FRAP (>1 m, ~1 s ) ( 36 ). This has indeed provided complementary insights on the scale-dependent diffusion behavior of GPI-anchored proteins, as well as the local organization of the membrane.
Pioneering FRAP studies ( 37,38 ) on GPI-anchored proteins from Jacobson's group reported that GPI-anchored proteins, such as Thy-1, placental alkaline phosphatase, and Ly6E, exhibit diffusion coeffi cients comparable to membrane lipids, but ‫ف‬ 2-to 5-fold faster than TM proteins like VSV-G. However, the GPI molecules showed a mobile fraction of only ‫ف‬ 50% across different cell types ( 38 ). The origin of this reduced lateral mobility was not well-understood. It was speculated that the size, glycosylation, and interactions of the ecto-domain of the GPIanchored proteins with the extracellular matrix, other TM receptors, or cortical cytoskeletal corrals could be a potential explanation ( 37,39 ). FRAP-based studies also showed that apically sorted GPIs are immobile just after reaching the cell surface, and their mobility increases as they become long-term residents at the cell surface, suggesting that GPI-anchored proteins are delivered to the membrane in a clustered form ( 40 ). Studying the dynamics of GPI-anchored proteins ( 41 ) provided a scale dependence to GPI-anchored protein diffusion, which suggested that they diffuse freely as individual molecules over a larger length scale, but may dynamically partition into raft domains. However, the poor spatial resolution, the inherent averaging of dynamics of multiple individual molecules, and a possible convolution of diffusion, turnover, and interactions (like trapping or binding) in a typical FRAP measurement precluded further investigation ( 42 ). Hence, studying the dynamics at (or near) the scale of single molecules (or a small group) was necessary to gain a more resolved picture regarding the microscopics of the diffusion process.
The advent of fast (video-rate and higher) and sensitive "low" light detectors in the early 90s coupled with the ability to specifi cally tag membrane molecules led to the emergence of SPT as a popular method in studying membrane dynamics ( 42,43 ). A unique feature about SPT is the ability to classify the observed particle trajectory into several categories/modes of diffusive behavior like free diffusion, The proposed fl uid-like nature of membrane lipids implied the absence of local lateral heterogeneities in the distribution of lipids. However, studies on the lateral heterogeneity of membrane lipid and protein composition (24)(25)(26) and their polarized distribution in epithelia ( 27 ) suggested that the cell surface and internal membranes were not a well-mixed membrane, and this led to a rethinking of the fl uid-mosaic model.
In 1988, Simons and van Meer ( 27 ) argued that polarized membrane composition could be achieved by sorting GPI-anchored proteins and sphingolipids into lipidic microdomains at the level of Golgi membranes, followed by their polarized traffi cking. However, a biochemical correlate of such membrane domains was lacking. In 1992, Brown and Rose ( 2 ) showed that placental alkaline phosphatase, a GPI-anchored protein destined for apical transport in MDCK cells, was enriched in membrane extracts insoluble in cold nonionic detergents, such as Triton X-100 (1% at 4°C ). These detergent-resistant membrane (DRM) fractions had very high lipid content and exhibited an enrichment of sphingolipids and cholesterol along with GPI-anchored proteins. The DRMs hence appeared as a biochemical correlate of the functional membrane domains proposed by Simons and van Meer ( 27 ). This laid the foundations of the "lipid raft" hypothesis ( 28,29 ) in cellular and functional context, as well as from a biochemical perspective, wherein these "rafts" were envisaged as lateral sorting platforms for specifi c lipids (sphingolipids), lipidlinked proteins (like GPI-anchored proteins), and other TM proteins (e.g., CD44), which form distinct domains at the cell surface (schematic in Fig. 2D ). Moreover, it was shown that the partitioning of GPI-anchored proteins and other raft markers into DRMs depended on cholesterol levels. For example, the depletion of cholesterol led to nearly complete solubilization of GPI-anchored proteins by the same detergents ( 30 ). Hence, the ability to partition into DRMs and its sensitivity to cholesterol depletion gained ground as a working defi nition of lipid rafts in cells. Rafts were envisaged as domains formed by the phase segregation of lipids, as observed in artifi cial membranes of composition resembling the plasma membrane ( 31,32 ).
These initial developments gave an impetus to using GPI-anchored proteins as a marker to study the organization and dynamics of the raft domains, using both fl uorescence and electron microscopy in intact cell membranes. In one such early study ( 33 ), fl uorescence microscopy of labeled folate receptor (FR)-GPI in CHO cells showed a diffuse and uniform distribution at the cell membranes. Upon detergent extraction, FR clearly redistributed to detergent-insoluble patches on the membrane. By this time, it was recognized that the process of detergent extraction could induce artifacts by coalescence of microdomains, differential removal of specifi c lipid species, and inducing changes in fl uidity and phase behavior of the native lipids in the membrane ( 34,35 ). Hence, while the direct visualization of native raft domains remained elusive, the focus shifted to other noninvasive or minimally perturbing approaches to study GPI-anchored protein organization under native (-like) conditions. In parallel, SPT data from the Kusumi group on TM proteins, like cadherin, EGF-receptor, and transferrin receptor, showed similar signatures of diffusional trapping for these proteins into compartments of ‫ف‬ 300 nm in NRK cells ( 43,47 ). To explain these fi ndings, Kusumi and Sako ( 48 ) proposed the "membrane-skeleton (MSK) fence" model, wherein the diffusion of TM receptors could be corralled into confi ned zones by the barrier effects of the membrane underlying the cortical matrix, which include the actin cytoskeleton and spectrin meshwork ( Fig. 2E , left). This presented a conundrum about the true nature of TCZs, do they represent MSK fence-bound compartments or cholesterol-sphingolipid-rich "raft domains" ( 49 ). Moreover, it was unclear how MSKs could affect the diffusion anomalous diffusion, directed motion, confi ned diffusion, and trapping/stalling. Early SPT studies ( 44,45 ) on GPIanchored proteins, like Thy-1 and NCAM-125 (a GPI-linked isoform of NCAM), showed that the individual trajec tories of both proteins exhibit all the different modes, like stationary, free, anomalous sub-diffusive, and transiently confined diffusion. The most striking observation was that ‫ف‬ 35% of all tracked molecules showed compartmentalized/confi ned diffusion in compartments of sizes ‫ف‬ 250-300 nm in C3H fi broblasts. These regions/domains were called "transient confi nement zones" (TCZs), and GPI-anchored proteins and other glycosphingolipids, like GM-1, were found to be trapped for ‫ف‬ 5-10 s in these cholesterol-and sphingolipid-sensitive domains ( 44,46 ). The residency time within each compartment is shown and is color coordinated with the respective compartment. The numbers in the square brackets indicate the order of the compartments the molecules entered. When repeated passages across the same compartment took place in these trajectories, the compartment is numbered by two numbers [(A-C) have been reproduced with permission from ( 55 )]. D: Schematic showing lipid rafts in cell membrane, which are envisaged as cholesterol-and sphingolipid-rich liquid-ordered domains enriching GPI-anchored proteins and raft-partitioning TM receptors. Adapted from ( 131 ). E: A contrasting picture of the membrane structure arose from the SPT studies of Kusumi et al. ( 133 ). Schematic of the bottom view (left) of the plasma membrane depicting the MSK fence model and the top view (right) of the membrane representing the anchored TM protein picket model. This model shows that the membrane constituents can undergo short-term confi ned diffusion within the domains and long-term hop diffusion between neighboring compartments. The actin meshworkbased MSK (fence) and the TM proteins anchored and aligned across the actin fences (pickets) create the compartmental boundaries. Reproduced with permission from ( 133 ). dictate the diffusion behavior of TM components, as well as inner and outer leafl et lipids.
FCS is the third major technique to study the dynamics of GPI-anchored proteins on the cell surface. In FCS, a confocal spot is parked on the cell membrane and the emitted fl uorescence is recorded at a very high temporal resolution using avalanche photo diodes over a few seconds (typically 5-10 s). The fl uctuation of fl uorescence intensity is then autocorrelated to generate the autocorrelation decay over time. This can then be fi tted with suitable models describing dynamics (like single component or multi-component diffusion/mobilities/fl ow, etc.) to extract quantitative measures of diffusion timescales (and their amplitudes) and the number of molecules. Unlike FRAP, FCS monitors equilibrium diffusion and, hence, the measurements are made at sparse molecular density ( ‫ف‬ 10-100 molecules per square micrometer) and low illumination intensities. This allows monitoring dynamics without the deleterious effects of photobleaching and overexpression, often associated with FRAP studies on cellular physiology. While FCS can readily measure the local density and the diffusion coeffi cients of membrane molecules, just the value of the diffusion constant is not sufficient to infer on the local membrane environment, which the diffusing probe encounters. However, measuring the diffusion properties across a range of observation areas provided more resolved understanding. In fact, FRAP studies had already exploited the scale-dependence of the mobile fraction and diffusion coeffi cient to infer on the size of membrane domains ( 59 ). Marguet and coworkers implemented this approach in FCS and developed a simulation-based approach to interpret the scale-dependence data. This FCS-based technique, also called spot-variation FCS (sv-FCS), and the scale-dependence of the diffusion timescales ( D ) as a function of the spot size were analyzed using the "FCS Diffusion Law " ( 60 ). According to this framework ( Fig. 3 ), linear fi ts of D versus confocal spot area plots, which pass through the origin (hence have zero y-intercept), were interpreted as simple free diffusion; a positive y-intercept would refl ect transient dynamic association with raft-like microdomains and fi nally a negative y-intercept would signify meshwork compartmentalized diffusion ( 60,61 ). The authors used this to show that Thy1, a GPI-anchored protein, shows a positive y-intercept, suggesting dynamic partitioning into membrane domains of ‫ف‬ 120 nm in size ( Fig. 3 , right panel) ( 62 ). The diffusion behavior of GPI-anchored proteins switched to free diffusion upon cholesterol and sphingomyelin depletion, while actin perturbations did not show any discernible effect ( 62 ). Subsequently, the spatial resolution of FCS measurements was improved using two different approaches. Marguet and colleagues also reported the use of nanometric apertures ( 63 ) for FCS measurements on the cell membrane, while Stephan Hell's group used stimulated emission depletion (STED) beam for excitation ( 64 ), thereby bringing down the resolution of FCS measurements to ‫ف‬ 40 nm. Both approaches confi rmed the cholesterolsensitive positive y-intercept for GPI-anchored proteins. Further STED-FCS studies suggested that GPI-anchored of outer-leafl et lipid-linked proteins, like GPI-anchored proteins, which, unlike TM receptors, cannot directly interact with the underlying cortex, but show distinctly confi ned diffusion.
Two new lines of observation attempted to resolve this. The fi rst approach combined particle tracking with optical trapping, wherein an individual (or a small group of) membrane protein (here GPI-anchor protein probes) could be bound to a bead held by the laser trap. The fl uctuations of the bead position ( 50 ) and the resistance felt by the bead in its movement during stage scanning ( 51 ) provided a measure of local viscous drag ( 50 ) and the nature and abundance of resistive barriers in diffusion ( 51 ), respectively. In conjunction, they showed that the viscous drag and the density of these domains were sensitive to cholesterol levels. The local viscous drag measurements yielded a size of ‫ف‬ 26 nm for individual domains and the force measurements from scanning of beads estimated ‫ف‬ 50-100 domains for every square micron . However, the inherent noise in these measurements precluded any defi nitive conclusion.
The second approach, pioneered by Kusumi et al. ( 52 ), involved enhancing the temporal resolution of SPT measurements to an unprecedented 40,000 frames per second for colloidal gold nanoparticle-labeled species and 10,000 frames per second for fl uorescently labeled molecules. In 2002, Fujiwara et al. ( 53 ) reported that outer-leafl et phospholipids also showed the signature of compartmentalized diffusion. The high temporal resolution of the particle trajectories showed that particles tend to diffuse freely for ‫ف‬ 1-5 ms in a given compartment before "hopping" onto a neighboring compartment . In a series of SPT-based studies ( 54,55 ), the Kusumi group subsequently showed that GPIanchored proteins (like GPI-linked MHC-II) exhibit this distinct hop-diffusion behavior ( Fig. 2A-C ), like other phospholipids and TM proteins ( 56 ). These SPT studies were also able to estimate the membrane compartment sizes in many cell lines ( 55 ), which closely correlate with the sizes of the juxtamembrane cortical actin (CA) meshwork ( 57 ). It was suggested that the diffusion of the outerleafl et components could be corralled due to the presence of TM-protein "pickets," which can directly sense the MSK "fences" underneath ( 52 ). This led to the more generalized form of MSK concept, the "anchored picket fence" model. These SPT studies clearly implicated the CA network as a major regulator of membrane structure.
Despite these studies, the link between these compartments and the lipid raft-like (cholesterol-sphingolipid-rich liquid-ordered) microdomains remained unclear. An explanation for the cholesterol sensitivity of the TCZs (or any microdomains brought about by lipidic interactions) also comes from the fact that the perturbation of cholesterol can lead to the loss of the PI 4,5-bis phosphate (PIP 2 ) pool from the inner leafl et, effectively disrupting both the membrane-cortex coupling and the organization of the actin cortex ( 58 ). This key observation not only linked the TCZs to anchored picket fences, but also hinted at farreaching implications, that the spatial architecture and topology of the membrane cytoskeleton interactions could resolution, its application is limited to fi xed cells only. Moreover, potential artifacts of the fi xation and EM preparation process might alter the native organization of the molecule studied . However, the early electron microscopy studies hinted of the possible nanoscale organization of GPI-anchored proteins ( 33,66,67 ).
Proximity measurements based on fl uorescence (or Förster) resonance energy transfer (FRET) [reviewed in (67)(68)(69)] and, more recently, super-resolution imaging ( 71,72 ) have been utilized to explore the nanoscale organization of the GPI-anchored proteins at the cell surface.

FRET imaging of GPI-anchored protein organization
FRET is a photophysical phenomenon involving the nonradiative energy transfer between a donor fl uorophore in excited state and an acceptor fl uorophore in ground state ( 73,74 ). The effi ciency of energy transfer is inversely proportional to the sixth power of distances between the fl uorophores, rendering FRET an ideal "spectroscopic ruler" ( 75,76 ) to probe distances at the biomolecular scales (1-10 nm). Energy transfer can also occur between like fl uorophores (homo-FRET) with low Stokes shift and, hence, a signifi cant overlap between their absorption and emission spectra ( 69,75 ). Homo-FRET is estimated by measuring the loss of polarization of emitted fl uorescence, which in turn is monitored by determining fl uorescence emission anisotropy. Homo-FRET can be estimated from steady-state anisotropy images or from time-resolved anisotropy decays ( 69 ). These approaches have been extensively used to study the nanoscale organization of multiple cell surface constituents, such as fl uorescent lipid analogs, GPI-anchored proteins, and the EGF receptor ( 5,70,78 ).
Over the years, homo-FRET ( 68 ) has been employed to study the nanoscale clustering of GPI-anchored proteins. Using this technique with steady-state anisotropy as a proteins and sphingolipids could be trapped into lipid shell-like domains of ‫ف‬ 20 nm size, which were likely to be held together by engaging the CA cytoskeleton ( 65 ).
In summary, studying diffusion dynamics of membrane molecules such as GPI-anchored proteins has contributed signifi cantly to the evolving picture of cell membrane structure and organization. However, the picture is not complete without understanding the molecular scale organization of GPI-anchored proteins at the cell surface. In the next section, we summarize the studies focused on the nanoscale organization of GPI-anchored proteins and its functional role in cellular physiology.

CELL-SURFACE ORGANIZATION OF GPI-ANCHORED PROTEINS
Different microscopy approaches have been employed to unravel the organization of GPI-anchored proteins in their native environment. Direct visualization of the membrane domains of these proteins also posed unique challenges on multiple fronts because living cells at (or near) physiological temperatures do not show phase-segregated domains, unlike model membranes. Instead, fl uorescence imaging of GPI-anchored proteins shows a uniform distribution under both native conditions and in DRMs ( 33 ). Moreover, it was only upon antibody-mediated cross-linking of GPI-anchored proteins that optically resolvable patches could be generated ( 66 ), possibly by bringing together the sub-resolution native domains of GPI-anchored proteins. Early studies of GPI-anchored protein organization using electron microscopy also pointed out that GPIanchored proteins primarily show diffused distribution and clustering only upon cross-linking ( 66 ) and in DRMs ( 33 ). Although electron microscopy can provide the ultimate Subsequently, GPI-anchored green fl uorescent protein (GFP) and monomeric-yellow fl uorescent protein were also shown to form cholesterol-sensitive sub-resolution clusters ( 79 ). Time-resolved anisotropy decays of these probes ( 77 ) allowed a more direct estimation of the fraction of molecules undergoing FRET and their intermolecular distances. These measurements clearly showed that at least 10-20% of the cell surface pool of both probes are organized into very high density clusters with intermolecular distances of ‫ف‬ 4 nm ( 79 ). To estimate cluster size at the nanoscale, a methodology that is based on how fl uorescence anisotropy changes when local fl uorophore density is reduced via photobleaching was adopted ( 79,81 ). If, indeed, robust homo-FRET occurs in specifi c arrangements, this should predictably increase distances between fl uorophores, thereby reducing FRET probabilities and leading to a systematic increase in anisotropy ( Fig. 4A, B ). Photobleaching and/or chemical quenching can reduce fl uorophore densities. The change in the fl uorescence anisotropy upon photobleaching was analyzed by Monte-Carlo-based simulations of a number of expected models of organizing the fl uorophores, wherein the expected FRET changes from different nanoscale confi gurations of probes are able to distinguish the nature of the cluster. This analysis suggested about readout, Varma and Mayor ( 3 ) showed that FR-GPI labeled with a fl uorescent analog of folate undergoes homo-FRET, suggesting for the fi rst time that these proteins can form small submicron-sized domains at the surface of the living cell. Subsequently, in collaboration with Madan Rao, a theoretical physicist, and his colleagues, the structure and dynamics of these submicron-sized domains were elucidated ( 79 ). Perturbation of cholesterol levels leads to a loss of homo-FRET between labeled FR-GPI, suggesting that cholesterol plays a key role in their formation and maintenance. Moreover, cells from a wide range of surface density of FR-GPI showed similar levels of homo-FRET, suggesting that they could be organized into submicron domains (<70 nm), whose dimensions are not dictated by the overall concentration of proteins. The fact that a TM form of FR does not exhibit concentration-independent homo-FRET, underlines the importance of GPI-anchor in dictating the nature of clustering. Although these measurements were made at lower resolution scale of whole cells, which precluded direct observation of domains, these results complemented the chemical cross-linking studies on GPI-anchored proteins, which demonstrated that these proteins could form oligomeric clusters on the cell surface ( 80 ).  ( 79,97 ). The data clearly show that nanoclustering of GPI-anchored proteins is concentration independent (A) and the change in the anisotropy with decreased intensity upon photobleaching refl ects the change in homo-FRET due to GPI-anchored proteins nanoclustering (B). Theoretical modeling of the FRET changes upon photobleaching suggested that GPIanchored proteins form nanoscale clusters of two to four molecules (C). D: The top panel shows the presence of GPI-anchored protein cluster-rich regions (blue) from the fl at surfaces of CHO cells expressing FR-GPI labeled with PLF (fl uorescein-conjugated folate analog ligand for FR). The bottom panel represents the thresholded binary map of the region in [D (top)] to clearly show the cluster size 1 ( р 450 nm) and inter cluster distance 2 ( ‫ف‬ 800-1,250 nm). E: Pixel anisotropy distributions from regions mentioned in [D (top)] show a slower exponentially decaying tail (red dots) which appears as a linear decay (black line) in contrast to the simulated Poisson distribution (green line) or the measured distribution of exogenously incorporated NBD-SM at levels that give rise to homo-FRET, which coincides with a quadratic decay profi le (green line). F: Ratio of inter-conversion rates of GPI-anchored protein clustering (represented in an Arrhenius plot) shows a sharp cross-over from temperature dependence to independence at 24°C, implying that cluster remodeling is an active process. Reproduced with permission from ( 83 ). breakthroughs in super-resolution imaging have been closely followed by their application to diverse biological contexts (85)(86)(87). This has opened up the new possibility of directly studying the molecular clustering of membrane molecules at or near scales (10-100 nm) over which membrane domains are most likely to form and operate ( 71,72 ). An elegant near-fi eld scanning optical microscopy (NSOM)-based study from the Garcia-Parajo group directly demonstrated that endogenous GPI-anchored proteins in monocyte apical membrane form cholesterol-sensitive nanoclusters ( 88 ). The inherent single molecule sensitivity of NSOM ( Fig. 5A ) allowed the authors to precisely calculate the presence of three to fi ve molecules per cluster . The clusters were also found to form cluster-rich "hotspots" ( Fig. 5B ). These results agreed well with the homo-FRET measurements described above and provided an independent confi rmation of GPI-anchored protein nanoclustering. Recently, STED imaging by the same group also detected the presence of actin-sensitive FR-GPI clusters on CHO cells ( 89 ). Pair-correlation analysis in conjunction with photo-activation localization microscopy (PC-PALM) imaging also confi rmed that GPI-anchored proteins form nanoscale clusters of three to fi ve molecules in Cos-7 cells ( Fig. 5D ) ( 90 ). Perturbation of cholesterol, sphingolipids, and CA led to a loss of clustering, while antibody-mediated cross-linking of GPI-anchored proteins redistributed the GPI-clusters to co-localize strongly with CA. Together, the super-resolution data on GPI-anchored protein nanoscale organization not only confi rms the presence and confi guration of cholesterol-and actin-sensitive nanoclusters, but also suggests that these clusters can (re)organize to form larger cluster-rich hotspots of ‫ف‬ 100 nm size ( Fig. 5E, F ). Arguably, this also clarifi es our understanding of the mesoscale organization of GPI-anchored proteins and supports the notion of hierarchical organization proposed earlier ( 83 ).

GPI-ANCHORED PROTEIN NANOCLUSTERING REQUIRES TRANSBILAYER ACYL CHAIN COUPLING
TO PHOSPHATIDYLSERINE CA dynamics and cholesterol play a critical role in the clustering and diffusion of GPI-anchored proteins. If CA dynamics are necessary for the creation and maintenance of the nanoclusters, it is imperative that the outer-leafl et GPI-anchored proteins need to sense and interact with the underlying CA. Two possibilities emerge for this connection: either via TM proteins that (in)directly bind CA or by harnessing lipidic interactions laterally (to cholesterol and sphingolipids) and across the bilayer (to inner-leafl et lipids). While TM anchors can certainly corral GPI diffusion, it is unlikely that protein-protein interaction can account for the extremely small and high density GPI clustering, as seen by homo-FRET and other super-resolution microscopies.
The idea of transbilayer coupling has been proposed previously. Experimental studies on symmetric membranes and phospholipid monolayers supported on immobile hydrocarbon fi lms have hinted toward how lipid composition 20-40% of the cell surface FR-GPIs form oligomers or nanoclusters of two to four molecules each ( Fig. 4C ) ( 79 ).
Steady-state anisotropy-based homo-FRET assays also indicated that multiple GPI-anchored proteins, when expressed on the same cell, could intermix and co-cluster at nanoscale ( 79 ). The small cluster confi guration and the low fraction of the GPI-anchored proteins forming nanoscale clusters can explain why earlier hetero-FRET measurements ( 82 ) were not sensitive enough to detect the GPI-anchored protein nanoclusters.
More recently, the implementation of homo-FRET measurement on several microscopy modalities/platforms allowed studying GPI-anchored protein clustering at superior spatiotemporal resolution ( 68,83 ). Homo-FRET imaging at a higher spatial resolution ( ‫ف‬ 300 nm) showed that GPIanchored protein nanoclusters exhibit spatially nonrandom distributions ( Fig. 4E ). Interestingly, cell membrane regions with uniform fl uorescence intensity (and hence concentration of FR-GPI) mapped to intricate spatial anisotropy patterns of variegated membrane patches, which were strongly enriched (or devoid) of GPI clusters. This clearly suggests a hierarchical organization wherein nanoscale FR-GPI clusters ( ‫ف‬ 10 nm) are collected into mesoscale optically resolvable cluster-enriched domains or hotspots ( ‫ف‬ 450 nm) ( Fig. 4D ). The clustering of GPI-anchored proteins was also found to be sensitive to the perturbations of cortical actomyosin, as clustering was completely abrogated on membrane blebs devoid of functional actin cortex ( 83 ).
Homo-FRET-based photobleaching recovery assays provide a window into cluster remodeling and report on dynamics of cluster remodeling. Using assays at different length scale (FRAP-based, as well as single point photolysis), GPI-anchored protein clusters were ascertained to be relatively immobile, but underwent dynamic remodeling by formation/fragmentation more readily at 37°C than at 20°C . This suggested that the cluster remodeling is an activity-dependent process. Indeed, it was also shown that the dynamics of GPI-anchored protein cluster remodeling are also sensitive to acute perturbations of actin polymerization, myosin activity, and cholesterol ( 83 ).
In a more dynamical cellular context, like spontaneously generated membrane blebs which undergo both detachment (during bleb growth) and reattachment (during bleb retraction) to the CA mesh, the anisotropy of EGFP-GPI was monitored, along with the distribution of the Cterminal fragment of ezrin (an F-actin binding protein), to address the interplay between CA activity and local clustering. This assay showed that growing blebs (detached from the underlying cortex) are devoid of EGFP-GPI clusters and clustering recovers gradually on the retracting blebs with reformed functional membrane-cortex links ( 84 ). Together these observations show that the nanometer scale organization exhibits unusual properties and is dependent on the CA mesh.

Super-resolution imaging of GPI-anchored protein clustering
The last decade has witnessed signifi cant advances in the fi eld of super-resolution microscopy techniques. Technical nanoclusters at the plasma membrane ( 5 ) ( Fig. 6A, B ). Though the underlying mechanism of nanocluster formation is not fully understood, recent work has shown the signifi cance of acyl chain interactions between GPIanchored proteins and phosphatidylserine (PS) at the inner leafl et that ultimately leads to the formation of clustered domains at the plasma membrane ( 5 ). Cells incapable of producing a suffi cient amount of PS, due to mutations in enzymes in the biosynthetic pathway of PS, lack the ability to create GPI nanoclusters at the plasma membrane ( Fig.  6D ). This ability could be restored only on the addition of PS with a long saturated fatty acid chain (18:0). The addition of short (12:0) and unsaturated PS (18:1) into mutant cells failed to rescue nanoclustering of endogenous GPIanchored proteins.
Results from atomistic molecular dynamics simulations done on the asymmetric bilayer have indicated that, along in the outer leafl et infl uences the inner-leafl et distribution of lipids ( 91,92 ). Recent studies using free-standing model membrane vesicles ( 93 ) and atomistic molecular dynamic simulations ( 94 ) show that transbilayer asymmetry, lateral heterogeneity, acyl chain length, and saturation are key prerequisites for transbilayer coupling ( Fig. 6E ). This makes the GPI anchor's structure viable for establishing transbilayer coupling between the outer-leafl et protein and the actin cytoskeleton. We fi nd that the chemistry of the fatty acid chain of the GPI anchor is a prerequisite for nanoclustering to occur. Mutation of enzymes affecting lipid remodeling of the GPI anchor, inhibiting the replacement of short and unsaturated acyl chains with long saturated ones, leads to disrupted nanoclustering at the plasma membrane ( Fig. 6C ). The chemistry of the GPI anchor, particularly the nature of the lipid moiety, is extremely important for organizing into cholesterol-sensitive peaks ] of PAGFP-GPI molecules. The correlation due to multiple appearances of a single protein [g(r) stoch ] and the protein correlation [g(r) protein ] were evaluated from the fi t to clustering model . E, F: Cos-7 cells co-expressing PAGFP-GPI and actin-PAmCh in the absence (E) or presence (F) of cross-linking antibodies. This data shows that actin-PAmCh localizes with clusters of antibody-cross-linked PAGFP-GPI (C-F). Reproduced with permission from ( 90 ).
form functional links to the underlying actin cytoskeleton via actin-adaptor proteins capable of binding both the membrane lipids and actin, thus unraveling the basic mechanism that leads to membrane domains.

UNIQUE FEATURES OF GPI-ANCHORED PROTEIN CLUSTERING SUGGESTIVE OF ACTIVE ORGANIZATION
Here we summarize the key results obtained from the homo-FRET and super-resolution imaging of GPI clustering. These quantitative approaches have unraveled unique (and peculiar) features of the clustering behavior of GPI proteins, which suggest some rethinking of the existing paradigms (lipid-raft, critical fluctuations, and picketfence) for the organization of membrane microdomains. 1 ) GPI-anchored proteins are organized as monomers and nanoscale clusters, whose fraction is independent of the total protein concentration ( Fig. 4A ), suggesting an organization maintained away from chemical equilibrium ( 3,79,88,90 ).
2 ) The nanoclusters are enriched in optically resolvable cluster-rich domains at steady state; the statistical distribution of nanoclusters and the statistics and spatial distribution of the optically resolvable domains are affected by acute perturbations of cholesterol levels and CA activity ( 83 ).
3 ) The single-point distribution of the local density of GPI-anchored proteins shows prominent exponential tails at both high and low density. This suggests a nonrandom patterning, where the GPI proteins are held at concentrations well beyond equilibrium mixing and segregation ( 83 ). 4 ) The monomers and nanoclusters are highly dynamic and the rates of aggregation-fragmentation ( ‫ف‬ 0.1 s) are strongly non-Arrhenius (hence they disobey thermal equilibrium). The ratio of the rates is insensitive to temperature in the range 24-37°C ( Fig. 4F ), and shows a sharp fall below this range coinciding with a sharp reduction in CA activity ( 83 ). 5 ) The rates of fragmentation and reformation of nanoclusters show a wide spatial variation, consistent with the spatial heterogeneity of CA activity. The remodeling of the GPI clustering is also sensitive to CA activity and cholesterol levels ( 83 ). 6 ) The nanoclusters are immobile, whereas the monomers exhibit diffusive motion on the cell surface ( 56,83,95 ). 7 ) The nanoclusters fragment into monomers as soon as blebs are spontaneously created on the cell surface, where the cell membrane detaches from the CA ( 96 ). The dynamic nanoclusters reform when the bleb retracts due to repolymerization of actin on the cell membrane, recruitment of myosin, and subsequent actomyosin contractility ( 83 ). 8 ) Transbilayer acyl chain coupling with inner-leafl et PS is critical for the clustering of GPI-anchored proteins. This presumably allows the exoplasmic GPI-anchored proteins to link with the actin cortex ( 5 ). 9 ) Local activation and reorganization of actin machinery by integrin activation triggers enhanced clustering of GPI-anchored proteins ( 88 ). Conversely, cross-linking of GPI-anchored proteins presumably recruits and hence colocalizes strongly with CA ( 90 ).
with the adequate levels of cholesterol and the appropriate lipid chemistry on either leafl et, it is also necessary that either of these components in the outer or inner leafl ets needs to be immobilized. The actin cytoskeleton, along with myosin, gets reoriented and aids in immobilizing the lipids at the inner leafl et. We hypothesized that PS can dynamically remodeling platforms of short actin fi laments ( Fig. 7 ). These dynamic platforms, called asters, are emergent structures resulting from the active hydrodynamics of the short fi laments driven by fi lament treadmilling and motor activity ( Fig. 7 ). The active composite model has provided a consistent explanation for the different peculiar features of GPI-anchored protein nanoclustering mentioned earlier ( 97 ). The active composite model offers a paradigm shift in the way we understand membrane organization, because it gives equal importance to inter-molecular interactions as it does to the interaction of cell membrane molecules with the CA.

ACTIVE COMPOSITE MEMBRANE: EXPECTATIONS AND EXPERIMENTS
The active composite model is a framework to understand membrane organization and has faithfully accounted for the clustering behavior of GPI-anchored proteins. However, the framework also assumes a two-component CA architecture and makes general predictions regarding the organization, cell surface dynamics, and distribution of passive molecules. A systematic experimental verifi cation of these predictions, in turn, can reveal newer qualitative and quantitative properties of the membrane and its resident molecules. Indeed, recent experiments have now verifi ed a number of these predictions.
Considerable evidence now exists for the existence of short dynamic actin fi laments from FCS measurements and SPT measurements ( 97 ). These data are in agreement with earlier biochemical-( 99 ) and electron microscopybased ( 100, 101 ) studies on short actin fi laments. Furthermore, the active composite model predicts that a model passive TM probe with an actin binding domain (TM-ABD) will couple to actin asters and, hence, can be nanoclustered. Homo-FRET ( 97 ) and STED imaging ( 89 ) have These features suggest that GPI-anchored protein clustering is an active "non-equilibrium" process, which requires functional links (via PS) between GPIs to cortical actomyosin activity. Recently, an active composite membrane model ( 97,98 ) of the cell surface as a general framework has been proposed to explain the active organization of cell surface molecules (schematic in Fig. 7 ).

ACTIVE COMPOSITE MEMBRANE MODEL:
A CONCEPTUAL FRAMEWORK The active composite membrane model ( 97,98 ) regards the cell surface as a composite of the multi-component plasma membrane and the membrane juxtaposed CA confi guration that it rests on. The CA has a wide distribution of fi lament length and is expected to be composed of at least two species of actin fi laments. The longer fi laments form a static cross-linked meshwork ( 57 ) along with a pool of short and polar active filaments ( 97 ). Active energyconsuming processes, like fi lament treadmilling and myosin contractility, can drive these fi laments. These dynamic short fi laments can also couple with the membrane transiently via membrane-actin linkers. This coupling, in turn, can dictate the local organization, dynamics, and environment of the membrane molecules, which can interact with this dynamic CA. Hence, in an active composite membrane, cell surface molecules, based on their coupling to this active CA, are classifi ed into three types: inert, passive, and active. Inert molecules (like short chain lipids) do not interact with CA, while passive (GPI-anchored proteins) and active molecules interact with the dynamic CA. While passive molecules cannot regulate local actin architecture, active molecules (like signaling receptors, integrin and T (or B) cell receptors) can do so via triggering nucleation, (de)polymerization, and cross-linking locally. The model also proposes that passive molecules, like GPI-anchored proteins, are driven into nanoclusters by localized and Fig. 7. Clustering in active composite membrane. Active composite membrane model proposes that myosin activity (zoomed inset, right) can organize short dynamic actin fi laments into "aster-like" confi gurations. The "aster" can, in turn, drive nanoclustering of outer membrane proteins and lipids, which can couple (in)directly to the short dynamic actin fi laments. Outer leafl et GPI-anchored proteins (blue) with long acyl chains do so via transbilayer coupling ( 5 ) with inner-leafl et PS connecting to aster via actin adaptors (green). On the other hand, TM-ABD (pink) can couple directly to the underlying actin asters to form distinct nanoclusters. Adapted from ( 96,125 ). sorting ( 113 ). Formation of domains at the plasma membrane or at the endosomal level might help concentrate the cargo, thereby increasing endocytic and cargo sorting effi ciency, respectively. This can even help in determining the fate of the cargo by preferentially sorting them into different domains. A clear example, which links GPIanchored protein clustering at the plasma membrane and endocytosis, is in the context of MA104 cells where clustering of the FR, which is a GPI-anchored protein, is found necessary for ligand uptake ( 114 ). Synthetic GPI analogs, which have minimal structural similarity to endogenous GPI anchor, have been used in studies to dissect out the contribution of the glycan core in protein traffi cking ( 22 ). Similar studies would prove to be useful for understanding the structural and functional relationship of GPI-anchored proteins.

GPI-hotspots and signal transduction
While many cell-surface GPI-anchored proteins are implicated in diverse signaling roles, it is unclear how an outer-leafl et signaling protein can stimulate intracellular downstream effectors. Studying the membrane organization, interactions, and diffusion of these proteins have also contributed to our understanding of GPI-anchored protein signaling. GPI hotspots are regions on the cell surface that have a high local concentration of clusters ( 83,88 ) or show transiently confi ned and compartmentalized diffusion ( 44,56 ). Such regions, in turn, can act as signaling platforms for either GPI-anchored signaling proteins or TM signaling receptors, which associate with GPI microdomains. Early biochemical studies ( 115 ) showed that antibodymediated cross-linking of several GPI-anchored proteins, like DAF, CD59, Thy-1, CD14, and Ly-6, led to the association and activation of Src family kinases. This classical assay has been revisited using high-resolution SPT and imaging tools on live cells to uncover the dynamics of the signaling process. Recent studies have shown that antibodyconjugated gold bead-mediated cross-linking of CD 59 leads to the formation of stabilized clusters, which exhibit stimulation-induced temporary arrest of lateral diffusion (also transient anchorage) and recruit signaling effectors, like G ␣ i2, Lyn kinase, and PLC ␥ , to elicit a transient IP3calcium signal ( 116,117 ). The transient anchorage of Thy-1 clusters requires Src family kinase activity and cholesterol ( 118 ). The same group identified Csk-binding protein (CBP) as a TM anchor which connects the Thy-1 clusters to CA via ezrin ( 119 ).
GPI hotspots also exhibit proximity to nanoclusters of LFA1, an integrin, possibly creating a membrane niche poised for the initiation of cell adhesion ( 88 ). Ligandmediated integrin activation resulted in a dramatic increase in nanoclustering of GPI-anchored proteins, locally driving the formation of nascent adhesion sites ( 88 ), which can eventually stabilize into lipid ordered domains in stable focal adhesions ( 120 ). In the same context, ECM components can also modulate the dimerization and diffusion of uPAR, the GPI-linked receptor for urokinase ( 121 ). Interestingly, uPAR also interacts with integrin and uses integrin as a coreceptor for its signaling ( 122 ), pointing toward confi rmed that a model TM-ABD probe can exhibit actindependent clustering.
The dynamic CA activity can also impart active fl uctuations to drive the diffusion of these molecules. This leads to temperature independence of membrane diffusion of GPI-anchored proteins at length scales similar to those of "actin asters," ‫ف‬ 200 nm. The active fl uctuations can be quelled by perturbations of actin polymerization, myosin activity, and cholesterol, resulting in a reversal of temperature-dependent diffusion akin to inert particles ( 102 ). Similarly, active anomalous density fl uctuations were also detected in the density (or intensity) distribution of GPIanchored proteins on the cell surface ( 97 ).

Functional roles of GPI-anchored proteins in cellular physiology
GPI-anchored proteins are known to critically regulate a wide range of physiological functions, from nutrient uptake to cell migration to immune recognition. Therefore, it is obvious that perturbations resulting in altered GPIanchored protein traffi cking, as well as plasma membrane organization, can lead to several diseases. Recent studies have shown that the genes involved in the biosynthesis of GPI-anchored proteins are linked with various diseases because the fully processed GPI anchor is essential for the proper functioning, as well as targeting, of the protein ( 9,(103)(104)(105)(106). Mutations in GPI remodeling enzymes, like PGAP3, are known to generate intellectual disorders, like hyperphosphatasia ( 9 ), and deletion of this gene results in impaired T cell receptor signaling ( 107 ). Interestingly GPI-anchored proteins are also known to function as cargo transporters, where GPIHBP1, a lipoprotein lipase transporter, carries LPL from subendothelial spaces to the capillary lumen in endothelial cells, thereby initiating lipolysis. The absence of GPIHBP1 leads to diseases such as chylomicronemia, due to elevated triglyceride levels in the plasma ( 108 ).
Ras signaling, which is usually associated with plasma membrane, plays a crucial role in the biosynthesis of GPIanchored proteins in yeast. Experiments done in yeast show that downregulating GPI-anchored protein expression at the plasma membrane leads to hyperactive Ras phenotypes ( 109 ). This is an interesting aspect, which connects cell signaling to expression and organization of GPIanchored proteins at the membrane. Recent studies show that subunits of GPI transamidase can act as oncogenes. For example, overexpression of PIG U leads to an increased cell division rate and a good correlation between the overexpression of these genes and several cancers, like bladder, colon, and ovarian cancers, was also observed ( 110 ). Though the direct functional relevance of GPI anchor is not known, it is essential for localizing the proteins into membrane domains, which might have implications in processes like endocytosis ( 111,112 ) or ER-Golgi cargo yet another functional link between GPI-clusters and integrin signaling.

FUTURE DIRECTIONS AND PERSPECTIVE
GPI-anchored proteins are a diverse set of cell surface molecules with unique structural features that modulate the transport and function of the proteins. Here, we have provided an overview of the contemporary understanding of membrane diffusion and organizational features of the GPI-anchored proteins. We have focused primarily on the biophysical measurements of these properties of GPIanchored proteins. We have also highlighted the bearing of the structure and membrane dynamics of GPI-anchored proteins in a functional context. It is clear that studying GPI-anchored protein organization and dynamics has greatly contributed to the evolving picture of membrane structure. This has now led to the Active Composite Model of the plasma membrane ( 97,98 ), which incorporates the contribution from the association of the dynamic actin fi laments to the membrane, along with the structuring infl uence of the static meshwork on membrane organization (from the picket fence model). Arguably, studying GPIanchored proteins has been fundamental to the evolving landscape of conceptual models describing membrane structure and organization.
However, many unresolved issues remain in GPI-anchored protein cell surface dynamics, organization, and function. GPI-anchored proteins are unique molecules because they require both lipidic interaction and actin activity for their membrane organization and dynamics ( 83 ). This very fact makes them an ideal candidate probe for mechanistic studies on membrane-CA interaction and how the cell might leverage such interactions to create membrane domains or regulate mobilities. To strengthen the role of dynamic actin in organizing GPIs, it will be crucial to study the clustering or diffusion of GPI-anchored proteins in conjunction with the monitoring of the local actin dynamics. This will entail the coupling of proximity or superresolution microscopy of native GPI-anchored proteins or synthetic analogs to the particle tracing approaches for the actin markers. Imaging assays based on dual-color SPT-PALM ( 123 ) or spatiotemporal image correlation spectroscopy ( 124 ) might provide crucial clues.
Moreover, genetic and biochemical approaches would be necessary to unravel the nature of the molecular linker(s) (and their regulation) between the GPI-anchored proteins and actin machinery ( 125 ). Understanding such functional links and their bearing on the membrane dynamics will naturally allow us to probe the properties of the membrane domains GPI-anchored protein's form at the cell surface and how such domains can be endocytosed ( 126 ). Addressing this will necessitate better assays to relate the local composition of the membrane milieu to the clustering of GPI-anchored proteins. Multimodal approaches combining the measure of local lipid order ( 127-129 ) and clustering or imaging-mass spectroscopy tools like NanoSIMS ( 130,131 ) could be key.