Characterizing proteins in their cellular environment: Examples of recent advances in quantitative fluorescence microscopy

Abstract

Quantitative characterization of protein interactions, both intramolecular and intermolecular, is crucial in understanding the mechanisms and regulation of their function. In recent years, it has become possible to obtain such information on protein systems in live cells, from bacteria to mammalian cell lines. This review discusses recent advances in measuring protein folding, absolute concentration, oligomerization, diffusion, transport, and organization at super‐resolution.

Introduction

Following decades of highly quantitative and detailed biophysical investigations of intramolecular and intermolecular protein interactions in vitro, in recent years, it has become increasingly possible to quantitatively interrogate these phenomena in living cells. One particularly useful ensemble of approaches to these problems is quantitative fluorescence microscopy, which has the benefit of providing images of the proteins of interest in live (or sometimes fixed) cells. Although seeing is certainly believing, these advanced microscopy approaches also employ a wide variety of clever imaging modalities in order to move well beyond qualitative pictures to provide quantitative data concerning concentrations, affinities, reaction rates, stoichiometries, diffusion and transport times, and directions, as well as high‐resolution localization information. Indeed, one of the major benefits to carrying out such experiments in live cells stems from the fact that the experiments provide spatiotemporal information that is directly linked to protein function within the cell.

The literature on quantitative fluorescence microscope studies of proteins in live cells is already overwhelming. Hence, this review will not attempt to do justice to the field overall. Rather it will address six functionally important properties of proteins, namely protein production and folding, absolute protein concentrations, protein oligomerization state, protein dynamics, protein transport, and high‐resolution protein localization, and how these properties can be characterized in cells. In each case, one or two examples will be discussed in detail to provide readers with an idea of the growing possibilities and where to begin in the event such studies would be of interest to their research programs. In each case, the specific imaging modality will be explained, such that this review may serve as a quick reference to the various possible experimental strategies. Another important and sometime worrisome aspect of fluorescence imaging approaches is the labeling of the molecules to render them fluorescent. The advantages and disadvantages of the labeling schemes used will be discussed as well in order to allow the reader to consider the current possibilities.

Protein production/folding

One of the first measurements of protein folding in live cells, although not in imaging mode, was made by the Gierasch group on CRABP I protein using a tetra‐cysteine tag and the biarsenical fluorescein dye FlAsH (4',5'‐bis(1,3,2,‐dithioarsolan‐2‐yl)fluorescein).1 Extending this work and using various fluorescent protein fusion strategies, the Clark group has investigated cotranslational folding, the role of the ribosome, and the influence of codon usage, among other aspects of in vivo folding,2, 3 although imaging has not been a central goal in this work. Relatively recent quantitative imaging‐based studies from the Gruebele group have used a variety of imaging modalities to probe folding in live cells (for recent reviews see Refs. 4, 5), including with ReAsH 4',5'‐bis(1,3,2,‐dithioarsolan‐2‐yl)resorufin),6 a resorufin derivative of the FlAsH dye. In most of their folding studies in live cells, a rapid temperature jump is used to unfold the model protein. The technique was named fast relaxation imaging (FReI) [Fig. 1(A,B)],7 in which the live cells (i.e., U2OS, human osteosarcoma, or HeLa, cervical carcinoma) were mounted on a slide on a temperature controlled stage in an inverted fluorescence microscope. A heating infrared (IR) laser (2200 nm) was used to create fast upward temperature jumps (~4°C), and preheating followed by laser extinction was used to create downward jumps in temperature, allowing for study of both unfolding and refolding kinetics. The model protein used was phosphoglycerate kinase (PGK), which the group had long studied in vitro.8 Forster resonance energy transfer (FRET) was used in order to observe the degree of folding of PGK. The cells were transiently transfected with a plasmid coding for a PGK a fusion protein with green fluorescent protein (GFP) (FRET donor) fused to its N‐terminus and mCherry (FRET acceptor) fused to its C‐terminus. Because the perturbation method was based on laser T‐jump using a pulsed IR laser for rapid heating, the GFP and mCherry constructs were chosen to have a melting temperature around 70°C, well above that of the PGK (39°C). Donor fluorescence was excited using epi‐illumination from a high power blue‐light emitting light‐emitting diode (LED) through an aspheric lens. After splitting the emission using dichroic and bandpass filters, green (donor) and red (acceptor) emissions were measured on two separate spots of a charge‐coupled device (CCD) camera. Temperature was monitored using the precalibrated temperature dependence of acceptor fluorescence upon direct excitation with a green laser. The ratio of donor/acceptor (D/A) fluorescence intensity was used to evaluate FRET efficiency and hence the fraction of folded protein.

Figure 1.

FReI studies of folding of PGK in live cells (taken from Ebbinghaus et al.7). (A) FReI rapidly jumps the temperature (purple) or the volume of live cells to perturb the equilibrium constant of biomolecular reactions such as protein folding, protein‐RNA binding, or quinary protein structure (transient but functional structure that forms in cells). A high numerical aperture objective collects multicolor fluorescence (green, red) excited by a laser or LED (blue), and a CCD camera images the dynamics in several channels with millisecond time resolution. (B) FReI can image FRET as a function of time with subcellular resolution to follow how folded a protein is in a specific part of the cell (red = more folded, blue = less folded). Top image shows U2OS cell with models of protein structure on 6 pixels; bottom image shows FReI signal in false color. (C) Equilibrium unfolding. Comparison of results from in vitro and in vivo studies as indicated. Insets show D/A ratio in color code (yellow low ratio/high FRET, green high ratio/low FRET) in a live cell at two temperatures, as indicated; (D) normalized D/A after a temperature jump from 39°C to 43°C, near the melting temperature, resolves unfolding of the less stable construct in vitro and in vivo but not for the more stable mutant.

The first study on PGK thermal denaturation at equilibrium indicated that the protein was more stable live cells than in vitro [Fig. 1(C)]. FReI T‐jump kinetics measurements suggested that this was due to slower unfolding rates [Fig. 1(D)]. In further studies, this group showed that PGK is more stable in the nucleus than in the cytoplasm or endoplasmic reticulum of these cells, thanks to increased folding rates,9 and further that the stability was cell cycle dependent.10 The temperature dependence of PGK stability and folding rates was also investigated using the same approach.11 To monitor coupled folding and diffusion in cells, the Gruebele group used two photo‐bleaching approaches, fluorescence recovery after photo‐bleaching (FRAP) and fluorescence loss in photo‐bleaching (FLIP), which provide diffusion information on larger length scales than FRAP.12 In FLIP, a small spot is photo‐bleached by a laser, whereas an LED is used to illuminate the entire cell. Then the decrease in intensity of the entire cell is monitored as the cytoplasmic pool of fluorescent proteins replenishes the bleached spot. In this manner, the authors found that the diffusion coefficient of unfolded PGK was slowed considerably more than expected due to changes in hydrodynamic radius upon unfolding. This behavior was ascribed to increased stickiness and interaction with cellular proteins such as chaperones when the protein unfolds.

Overall these studies demonstrated that protein folding can be studied in cellulo and that the milieu impacts the observed thermodynamic and dynamic properties of the protein. The approach could in principle be extended to other proteins, although design and construction of the FRET pair constructs for use in this manner is not trivial. For the data acquired in live cells to be interpretable, a detailed characterization of the construct must be carried out first in vitro. Moreover, the mechanisms underlying the modulation of stability of the PGK construct as a function of cellular localization and position of the cell cycle remain to be elucidated.

Protein concentration

One of the most important quantitative pieces of information about proteins in cells that one may wish to obtain is their absolute number and concentration. These quantities determine the ratio of a given protein to its target ligands and the ratio of its concentration to the dissociation constant of their interactions. These ratios in turn, determine the degree to which a given protein interacts with its partners (e.g., small ligands, other proteins, nucleic acids). Such interactions are generally highly regulated and result in key cell state transitions such as metabolic switches, cell cycle progression, differentiation, and transformation.

Fluctuation spectroscopy, or fluorescence correlation spectroscopy (FCS), has long been used to determine protein concentrations and diffusion coefficients in vitro and has been often used to monitor diffusion in cellulo, as well.13 In live cells, FCS has been less popular for determining concentrations, although this has been done in a few instances (e.g., Ref. 14). One of the main drawbacks to the approach is the significant photo‐bleaching arising from parking the laser beam in the sample for several seconds to minutes. Line or circle scanning FCS helped to alleviate this to some extent.15 However, a major breakthrough came with the first implementation by the Gratton group of scanning Number and Brightness analysis (sN&B).16, 17 The sN&B method extends the scanning to full raster scanning of a field of view (FOV), with the pixel dwell‐time much shorter than the diffusion time of the molecules of interest. In sN&B, an FOV is thus scanned many times (e.g., 50), yielding 50 values of intensity at each pixel in the FOV, from which the average fluorescence at that pixel, <F>, and the variance, σ,2 can be computed. As these measurements concern small numbers of molecules whose fluctuations obey Poisson statistics, as demonstrated by Qian and Elson,18 the number of molecules, N, present on average within the excitation or emission volume, V _ex or V _em, can be calculated as

$$N=<F{>}^2/{\sigma }^2,$$ (1)

and the molecular brightness, B, in counts per dwell‐time per molecule is defined as:

$$B={\sigma }^2/<F>.$$ (2)

In sN&B, this moment analysis is carried out at each pixel, and due to the small number of scans, the resulting N and B values at each pixel are then corrected for shot‐noise,17 as follows:

$$n=\left(N\times B\right)/\left(B-1\right);e=B-1,$$ (3)

to yield n, the true number, and e, the true brightness, at each pixel. This procedure provides spatial maps of the absolute number of molecules, n, present in V _ex or V _em at each pixel and their molecular brightness, e. If one calibrates V _ex or V _em with a sample of known concentration and appropriate diffusion coefficient, the n value corresponds to a spatial map of the absolute concentration of molecules. Moreover, if one calibrates the absolute brightness of a single fluorophore, then the e value yields the stoichiometry of the labeled protein of interest. This approach has been used in a significant number of studies to determine protein oligomer stoichiometry, examples of which will be discussed below. Generally, the n and e values for an ensemble of pixels (nuclear or cytoplasmic or membrane, e.g., are averaged to obtain precise and reliable values for true molecular number and brightness. Note that the spatiotemporal information in sN&B data can be extracted using raster scanning image correlation analysis, yielding information concerning the diffusion properties of the molecules.19 Despite the usefulness of the information, sN&B has not been used often in the determination of absolute concentration.

Recently we took advantage of the ability of sN&B to investigate the molecular mechanisms controlling the balance between growth and division in budding yeast.20 We used a calibration method [Fig. 2(A)] in which we determined the absolute molecular brightness of free monomeric GFPmut3, e _GFP, expressed from a plasmid in budding yeast cells and the calibrated 2‐photon excitation volume obtained using fluorescein in glycerol (V _eff), along with Avogadro's number, NA to calculate the absolute concentration of GFP‐fusions (FP) of the proteins of interest (in units of monomer). The brightness of free GFPmut3 was well determined from the hundreds of thousands of pixels over multiple cells in multiple FOV used in the calculation. In our case, the proteins of interest were mostly nuclear; hence, their intensity was averaged over all nuclear pixels using image masking approaches and corrected for the low levels of autofluorescence obtained in 2‐photon excitation at 1 μm. This calibration‐based approach yielded very reliable values for the absolute concentration of the proteins of interest.

Figure 2.

sN&B study of the G1/S transcription factors in budding yeast (taken from Dorsey et al.20). (A) Schematic for the calibration‐based sN&B method used. (B) Schematic of start. (C) sN&B<F> images of the GFPmut3 fusions of Swi4, Mbp1, Swi6, and Whi5 in an asynchronous population of yeast cells grown in glycerol. (D) Plots of absolute concentration of these four proteins versus cell size for hundreds of cells grown in glucose (blue) and glycerol (red).

We were interested in the transcription factors responsible for expression of genes necessary for the transition from G1 phase to S phase [Fig. 2(B)]. This transition is termed Start in budding yeast, and its timing determines cell size. Our goal was to understand how size is set and how it is modulated by nutrients. Yeast grown in poor nutrients divide at a smaller size than when the cells are grown in rich nutrients. The transcriptional activators, SBF and MBF, are made up of a common activator subunit, Swi6, and distinct DNA binding subunits, Swi4 and Mbp1, respectively. SBF is repressed by a Whi5, which when phosphorylated by a cyclin‐dependent kinase, dissociates from SBF and is exported to the cytoplasm. Two of the key genes in the G1/S regulon are cyclins, Cln1 and Cln2, which along with another cyclin Cln3, control their own transcription in a positive feedback loop, which leads to a sharp triggering of Start.

Our sN&B studies of GFPmut3 fusions of Swi4, Mbp1, Swi6, and Whi5 [Fig. 2(C)] expressed from their natural loci in yeast cells revealed that one of the key limitations to triggering Start in small cells is the fact that the copy number of these factors is limiting with respect to the ~200 promoters of the G1/S regulon in small cells, and that the number of molecules of the transcriptional activators increases as cells grow. Indeed, Swi4 concentration doubles as cells grow in G1 (= six‐fold increase in copy number), whereas the concentration of the other proteins remains constant. As the cells triple in size during G1, this corresponds to a three‐fold increase in copy number [Fig. 2(D)]. Comparing cells grown in glucose (rich = blue) and glycerol (poor = red) medium revealed that the three activators exhibited increased concentrations in the poor carbon source. Thus, the activation of the G1/S regulon would occur at a smaller size, as the activator/target site ratio is higher in the poor medium for cells of a given size, whereas the repressor/activator ration was lower. Incidentally, brightness measurements revealed all four proteins to be dimers. The straightforward determination of absolute concentration of these key factors demonstrated that in yeast the mysterious “setting and sensing” of the cell size threshold and its modulation by nutrients results from a simple biochemical titration of promoter target sites by their transcription factors.

While knowledge of the absolute copy number and concentration of key factors in live cells can be extremely useful for understanding and modeling cell state transitions, great care must be taken to obtain the most accurate and precise results possible. In the study discussed here, we used 2‐photon excitation at 1000 nm, which greatly reduces background. In addition, we used a probabilistic background subtraction routine with internal validation prior to calculating concentration from intensity. These steps combined to allow for a lower limit of detection of 1 nM GFP. However, the detection limit is significantly higher using visible excitation. Another key limitation is that the cells, themselves, must remain completely immobile during acquisition. This is achieved using concanavalin A coating procedures for yeast and chitosan coating for bacteria. Mobile appendages in mammalian cells, such as HEK cells, do not allow for reliable N&B measurements for those regions of the cells, which must be masked out prior to calculations. Finally, ideally one desires a system in which the GFP fusions of the proteins of interest are expressed from their natural promoters and hence at their natural levels, as was the case in the study above. With the advent of CRISPR gene editing technologies, this is now possible,21 although not widespread. Despite these precautions, the actual N&B measurements are as simple as a confocal scan, and can rapidly inform on the properties of key proteins and their role in cellular function.

Protein oligomerization

The Gratton group, among others, has used sN&B in a number of cases to determine protein oligomer stoichiometry.22, 23, 24, 25 Following their lead, we investigated the stoichiometry of SNAP‐tagged versions of the glumatate receptor, mGluR2, bound by externally added Alexa488, and expressed at near endogenous levels from a plasmid transfected into primary neurons26 [Fig. 3(A–D)]. Stoichiometry was calculated as the ratio of the observed true brightness divided by the brightness of free Alexa488. We found, that the receptor existed primarily as a dimer in the range of endogenous expression levels, increasing in stoichiometry at higher expression levels [Fig. 3(E)]. At the lowest levels of expression, the stoichiometry decreased below the dimeric level, most likely due to heterodimerzation with endogenous, unlabeled protein. Interestingly, addition of either antagonist (+ positive allosteric modulator [PAM]) or antagonist led to an increase in average observed receptor stoichiometry in an expression level dependent manner [Fig. 3(E–G)]. Fits of the pixel distributions of the true brightness of all cells under a given ligand condition in terms of dimers, tetramers, and higher order oligomers revealed that only tetramers were present in absence of ligand, and subtle differences existed between the two ligand‐bound states. Based on further work with mutants of these proteins, we concluded that the ligand‐dependent conformational dynamics of the receptor resulted in two alternative oligomerization modalities. Note that the timescale of sN&B, in which each pixel is measured only every 2‐3 s, allows for monitoring fluctuations from slowly diffusing molecules such as membrane proteins.

Figure 3.

Ligand dependence of the oligomerization state of mGluR2 in primary neurons. (A) Schematic of the SNAP‐tag approach. (B) Fluorescence intensity 2‐photon image (20 × 20 μm) of a primary neuron transfected with a plasmid expressing the Alexa‐labeled SNAP‐tagged mGluR2. (C) Zoom of the image in (B). Scale bar is 5 μm. (D) Brightness image of the figure in (C). (E) Stoichiometry of A488‐mGluR2 in neurons as a function of expression level (subunits/V _eff). Black points are cells with no ligand present; green points correspond to cells in presence of suturing agonist and a positive allosteric modulator. (F) Stoichiometry of A488‐mGluR2 in absence of ligand (as in E) and in presence of saturating antagonist (red). (G) Whisker plots of A488‐mGluR2 stoichiometry. (H) True brightness pixel distributions of Alexa488‐mGluR2 as a function of ligand. Blue curve represents the fit of the free Alexa488 (=monomer). Fits of the A488‐mGluR2 in presence of antagonist in terms of dimer (D), tetramer (T), and higher order oligomers (X). (J) Bar graph of the fraction of D, T, and X forms of mGluR2 in absence of ligand, in presence of agonist and PAM, and in presence of antagonist, as indicated. Data from Ref. 26.

Protein dynamics

Another very important characteristic of proteins in cells is their translational movement. Reaction rates are often diffusion limited. Moreover, movement from one location within a cell to another, where the reaction can occur, for example, can be limited by static or dynamic barriers to diffusion. Pair correlation methods,27 coupled with scanning approaches published in a series of paper by the Gratton group provide particularly elegant approaches to characterize protein translational dynamics in cell nuclei.28, 29, 30, 31, 32 First,28 the group used the correlations between pairs of pixels along scan lines to characterize GFP diffusion in the nucleus in regions of both low and high DNA density [Fig. 4(A)]. GFP intensity was found to be inversely related to DNA density [Fig. 4(B)]. Multiple scans of the same line in an image were used to construct intensity versus time carpets [Fig. 4(C), left] and autocorrelation function (ACF) carpets [Fig. 4(C), right] of the diffusing GFP, which could then be fit (at each position in the carpet) for the GFP diffusion coefficient and the Go value, where 1/Go is equal to the number of molecules present in the PSF at that position. Regions of high Go correspond to low numbers and vice versa. Note that since these data are acquired in a line scan, the time between points at a given position is equal to the line scan time, 0.472 ms in this case.

Figure 4.

Pair correlation analysis of GFP diffusion in live CHO K1 cell nuclei. Taken from Hinde et al.28 (A) GFP intensity image (left). Scale bar is 5 μm. Hoechst 3342 image overlaid on the GFP image (right). Lower panels correspond to zooms of the scan line in both channels. (B) Intensity profile of GFP along the scan line (top) and Hoechst (bottom). Black trace corresponds to the 1/Go value from panel C, right, and is equivalent to the GFP intensity. (C) Intensity time carpet plot along the scan line (left) and (right) the ACF carpet which is the autocorrelation function at each pixel along the scan line calculated from the intensity versus time values.

In addition to calculating the ACF carpet for each pixel in the line scan, as above, these authors cross correlated the intensity between pixels in pair correlation functions (pCF), where cross correlating a pixel against itself, as in Figure 4(C), right, corresponds to pCF(0). Cross correlation between pixel 1 and pixel 2, for example, would correspond to pCF(1), where the number in parentheses refers to the difference in pixel numbers in the correlation. In this manner, the authors investigated the diffusion of GFP across and within areas of low and high DNA density (Fig. 5). For example, in the pCF(1) plots, the two pixels cross correlated along the line (1 with 2, 2 with 3, etc.) are within the PSF, and the cross correlation value, G _x, is highest at short times [Fig. 5(B, F, I, J), dashed line]. In contrast, when the distance is large between cross‐correlated pixels [e.g., pCF(8) or pCF(14); Fig. 5(C, D, G–I, J), colored full lines], the G _x value is low or even negative at short times and rises at times for which diffusion from one pixel to the other can occur. Low‐to‐low DNA density correlation of GFP moving around/through an intervening high DNA density region occurred within 1–30 ms, whereas high‐to‐high density correlation around a low DNA density region was slower, 10–80 ms, and the fraction of GFP molecules diffusing from low‐to‐low around a high DNA density region was much higher (75%) than those that diffused from high‐to‐high density around/through a low density region (15%). These studies revealed a kind of “molecular flow pattern” in the nucleus which is consistent with the existence of chromosome territories which create channels for and barriers to diffusion.

Figure 5.

pCF carpet analysis of intranuclear diffusion. Taken from Hinde et al.30 (A–D) Intensity profile of the Hoechst 33342 stain across the line measured. The black arrow above each profile indicates to scale the distance that columns were cross correlated in the pCF carpets directly below. The black and red curved arrows indicate, respectively, the DNA environment that is being cross correlated from low or high DNA in the pCF carpets directly below. (E) pCF(0) carpet, which corresponds to calculation of an ACF carpet (90,000 lines analyzed). (F) pCF (1) carpet, which corresponds to cross correlation of adjacent pixels in the same DNA environment within the PSF. (G) pCF (8) carpet which corresponds to cross correlation of pixels in different density DNA environments. (H) pCF (14) carpet, which corresponds to cross correlation of pixels in low‐low DNA around a high DNA density environment or pixels in high‐high DNA around a low DNA density environment. All carpets are depicted on the same color scale: The minimum value was set to 0, and this corresponds to black (no communication). (I) Plot of the amount of correlation for column 4 (low DNA) at each analyzed distance (0, 1, 8, and 14). Curves are normalized to 1 with respect to the ACF. (J) Plot of the amount of correlation for column 13 (high DNA) at each analyzed distance (0, 1, 8, and 14). Curves are normalized to 1 with respect to the ACF.

Protein transport

Another central and functionally important aspect of protein translational dynamics is transport between cellular compartments. An elegant example of detailed, quantitative mapping of nuclear transport was published by the Gratton group.33 In this study, the mechanisms governing transport of cargo between the nucleus and the cytoplasm via the nuclear pore complex (NPC) was investigated using a combination of tracking and correlation approaches. The goal was to distinguish directed motion from diffusion as a means of addressing several models of transport across the NPC33 (and references therein). Orbital particle tracking was used to maintain a reference point in the center of the NPC [Fig. 6(A), top left images]. The NPC comprises about 30 polypeptides, a third of which, including Nup153, are rich in intrinsically disordered Phe‐Gly (F‐G) domains. Cargo are transported by kariopherins such as Kapβ1, and require the Ran GTPase for release. In the work by Caparelli et al. discussed here, the dynamic behavior of C‐terminally GFP‐tagged Nup153, one of the NPC F‐G nucleoporins, was investigated. Nup153 is anchored to the nuclear basket of the NPC [Fig. 6(A), top right schematic]. The ACF of Nup153‐GFP showed a narrow distribution of correlation at positions just off‐center (positions 8–24) of the arc‐like pattern of the ACF with a characteristic time near 10 ms [Fig. 6(A), bottom center], representing cytoplasmic to nuclear (C > N) transport [Fig. 6(A) bottom left schematic]. The pCF(+32) captures movement across the pore, as it correlates intensity between pixels located 32 pixels apart (i.e., pixel 16 with pixel 48). Hence, the pCF for pixels 8–24 correlated with pixels 32 columns away [Fig. 6(A), bottom right], effectively correlates C > N motion between the two red rectangles in the schematic in Figure 6(A), bottom left, whereas pCF(+32) for pixels 40–56 captures motion between these same regions, but in the opposite direction (N > C). The C > N motion of Nup153‐GFP is proposed to involve a collapse‐like conformational transition, whereas motion in the opposite direct involves cargo release and a conformational transition from collapse to elongated structures.

Figure 6.

Orbital tracking/pair correlation analysis of nuclear transport. Taken from Ref. 33. (A) Study of Nup153‐GFP. Top left, two images of the Nup153‐GFP‐labeled NPC complexes. The right‐hand image is a zoom of the boxed area in on the left and shows the tracking orbit which allows to maintain the reference point in the center of the NPC. Scale bars are 5 μm on the left and 1 μm on the right. (Top right) schematic of the NPC, with participating molecules as indicated. (Bottom left) Schematic of the tracking and scanning of an individual NPC. The PSF is scanned along a 64‐points orbit (R = 120 nm) around the pore, above and below the equatorial plane, with the start point indicated by the black dot. Colored regions of pixels (green and red) are indicated in the pCF(+32) in the bottom right. Bottom middle) ACF of Nup153‐GFP and the time profile of the carpet averaged over pixels 8–24 corresponding to the top red rectangle in the schematic to the left. Bottom right) pCF(+32) carpet and time profiles. Pixels implicated in C > N (collapse, blue line in the time profile) and N > C (release, red line in the time profile) transport are indicated. The ACF 8‐24 (dashed black line) is shown for reference. (B) Study of Nup153‐GFP and mCherry‐NLS. (Top left) Nup153‐GFP is cotransfected with NLS‐mCherry (scale bar: 1 μm). (Top right) In the schematic diagram, Nup153‐GFP drives the NLS‐mCherry toward the nucleus (I, II); the intervention of RanGTP detaches NLS‐mCherry and releases Nup153‐GFP (III). (Bottom left) The ACF yields the double‐arc, characteristic of Nup153 activity at the NPC. (Bottom middle) The pCF(+32) carpet in the cytoplasm‐to‐nucleus direction yields the expected distribution of Nup153 collapse times (left carpet, green curve in the time profile); NLS‐mCherry instead shows a bimodal distribution of transit times (middle carpet, red curve in the time profile). Only the faster component (active) pair cross correlates with Nup153 movement in this direction (right carpet, blue curve in the time profile). Bottom right) The pCF(+32) in the opposite direction (N > C) shows the expected distribution of Nup153 release times (left carpet, green curve in the time profile) opposed to a broader distribution of NLS‐mCherry (middle carpet, red curve in the time profile). The pcCF(+32) analysis shows that Nup153‐GFP and NLS‐mCherry are not moving bound together in this direction, as expected (right carpet, blue curve, in the time profile).

The study also addressed how cargo exhibiting a nuclear localization signal (NLS) are transported in and out of the nucleus via the NPC. Cells were transfected with Nup153‐GFP, and mCherry labeled NLS peptide [Fig. 6(B), top row]. The Nup153‐GFP ACF [Fig. 6(B), bottom left] and pCF(+32) carpets were calculated in each channel [Fig. 6(B), bottom center]. The Nup153‐GFP pCF(+32) in the N > C direction was identical to Nup153‐GFP in Figure 6(A), bottom right, whereas the mCherry‐NLS pCF(+32) showed two regions of correlation with characteristic times around 3 and 15 ms. The authors then calculated the pair cross‐correlation function at +32, pcCF(+32) between the GFP and mCherry channels and showed that the faster time implicated correlated motion of the two proteins, whereas the longer time did not. They concluded that C > N transport of the mCherry‐NLS involved transport of the Nup153‐GFP/mCherry‐NLS complex by Nup153, whereas N > C transport implicated unaided mCherry‐NLS diffusion through the pore. These highly detailed and sophisticated dynamics studies of nuclear translocation suggest that directed Nup‐mediated molecular motion may represent an intrinsic feature of the overall selective gating through intact NPCs. We note that given that these experiments require orbital tracking in addition to fluctuation microscopy acquisition, they are not likely accessible to a large number of laboratories. As imaging facilities become more sophisticated, high‐level protein dynamics studies such as these should become more available.

Super‐resolved protein localization

Lastly, it seems appropriate to discuss an obvious property of proteins that determines their functional properties in cells, their exact localization. Since the first demonstrations of super‐resolution approaches such as spatial light modulated microscopy34 or single molecule localization microscopy (SMLM) methods such as photactivatable localization microscopy (PALM) or stochastic optical reconstruction microscopy (STORM),35, 36, 37 a large number of articles have been published concerning the super‐resolution organization and single particle tracking of molecules in cells (e.g., Refs. 38 and 39 for a couple of very nice recent examples). These SMLM methods rely on the stochastic detection of single molecules to localize them with very high precision, beating the diffraction barrier by about 10‐fold. Typical resolution is ~25 nm, while SIM resolution is ~100 nm. In PALM, photo‐switchable proteins are transformed stochastically from green to red emitters by illumination with an ultraviolet (405 nm) switching laser, whereas in direct STORM (dSTORM), organic dyes are switched to a dark state using the same type of illumination and then allowed to stochastically revert to their bright states.

Here, we will focus on two particularly elegant recent studies of a type I bacterial partition complex,40, 41 ParABS, which is responsible for the partition of low copy number plasmids and sister chromosomes in bacteria. This nucleoprotein complex is comprised of a repeat of centromere‐like DNA sequences, parS, the ParB DNA binding protein that specifically recognizes parS sequences, and the ParA Walker‐like ATPase responsible for directing the ParBS complex to opposite poles of the cell using ATP hydrolysis. Two types of models have been proposed to explain segregation by this complex. In the first family of models, filaments of ParA would pull or push the ParBS complex around or through the nucleoid toward the poles, while in the second family of models ParA moves the complexes on the surface of the nucleoid.

The Nollmann and Bouet groups in France first performed PALM imaging of the partition complex of the F plasmid in Escherichia coli following mEos2 fusions of the ParB_F protein. They found that the majority of the 800 ParB_F dimers localized to parS _F sites [Fig. 7(A)], at mid‐cell for cells that had one parS site and at quarter‐cell for those that had two. Splitting the total acquisition time into three bins and color coding the detections within 100 nm of each other at the parS sites for the first (red), second (green), and third (blue) time window [Fig. 7(A‐iii)] demonstrates that the partition complexes display dynamic behavior on the seconds to minutes timescale. These results reveal that ParB binds nearly exclusively and stochastically around the parS sites forming a dynamic lattice confined to this region of the chromosome. Two color SIM experiments monitoring ParB_F‐mVenus and mCherry‐HU in E. coli or ParB_Bsu‐GFP and organic dye‐stained nucleoids in B. subtilis [Fig. 7(B)] that in both cases the partition complexes were located within the nucleoid for both the plasmidic and chromosomal systems, and moreover near the center in the y‐z plane. Similar studies on ParA_F showed the partition complex localized within the nucleoid as well and cells that maintained the plasmid, but that were devoid of ParA_F showed no clustering of ParB_F. A mutant of ParA, ParA^K340A, which is impaired for nonspecific DNA binding showed significantly reduced ParB_F foci with respect to WT, despite the fact that the ParA^K340A variant retains its ability to interact with ParB_F. Single particle tracking sptPALM studies on mEos2 fusions of the WT and mutant ParA proteins revealed two distinct dynamic populations of the WT protein, slowly moving (~0.01 μm²/s) presumably bound proteins localized to the partition complex foci [Fig. 7(C) red] and faster (~1 μm²/s) diffusing species that sample the entire nucleoid [Fig. 7(C), blue]. The K340A mutation that inhibits nonspecific DNA binding leads to loss of the slowly moving species and homogeneous fast diffusion throughout the cells. The studies also showed that the localization of the partition complexes within the nucleoid volume correlates with high‐density chromosomal regions (HDR) and follows the pattern of HDRs throughout the cell cycle. Differences between the behavior of the plasmidic and bacterial (B. subtilis) partition complexes indicated that the replication machinery may play a role. Altogether, the results of these studies allowed for ruling out the fibril‐based models and led the authors to propose a “hitchhiking model” for segregation by partition complexes [Fig. 7(D)]. ParA assembles in small patches at regions of high DNA density (HDR). Stochastic “scanning” allows ParA oligomers to follow changes in the conformation of the nucleoid and to hitchhike on HDRs. Interactions of ParB with HDR‐bound ParA patches may trigger their progressive dissociation from HDRs and subsequently release partition complexes from HDRs. Diffusion would allow partition complexes to bridge two ParA patches without stalling, making partition complex movement a combination of Brownian motion and directional bias.

Figure 7.

Super‐resolution imaging of the ParABS complex (taken from Sanchez et al.41 and Le Gall et al.40). (A) The majority of ParBF assembled into dynamic partition complexes restricted to small confinement zones. (i) Pointillist reconstruction, where each single‐molecule event detected is represented by a single green dot. (ii) single‐molecule density plot. Molecules (N) were grouped together using a maximum exploratory radius of 100 nm (dThresh < 100 nm) and density calculated as N per cell area. The colored scale bar represents the normalized number of detected events per square micrometer. (iii) Temporal localization of single‐molecule events. Total imaging time was divided into three color‐coded fractions (red, green, and blue represent the first, second, and third fraction of total time, respectively). Single molecules were colored according to the fraction of time in which they were detected and combined in a red, green, and blue additive color model (e.g., yellow represents a spatial zone where independent single molecules were detected in the first and second fractions of imaging time, red + green). (B) Typical multicolor 3D‐SIM images of HU‐mCherry‐labeled nucleoids (red) and plasmidic ParB_F‐mVenus (green) in E. coli (DLT3053/pJYB234) left or 4,6‐diamidino‐2‐phenylindole‐stained nucleoids (red) and chromosomal ParB_Bsu‐GFP (green) in B. subtilis (HM671), right. (C) Single particle tracking PALM measurements of mEos2 fusions of ParA variants in live cells. Distributions from of the apparent diffusion coefficient, D (top), and single‐molecule trajectories (bottom) of wild‐type ParA (DLT3284/pJYB286, N = 63 cells) (left) and ParAF^K340A (DLT3286/pJYB288, N = 17 cells) (right). Trajectories were classified as a static species (red) and a fast diffusing dynamic species (blue). (D) Hitch‐hiking model for bacterial chromosome segregation. The hitch‐hiking model for bacterial chromosome segregation. See text for description. Cell outline is shown as a gray mesh, nucleoid as a red cylinder, ParBS partition complexes in green, static DNA‐bound ParA in cyan, free ParA‐ADP in blue, and HDRs as red diffuse circles.

These and many other elegant super‐resolution imaging experiments require numerous control experiments. Copy number determination (or counting) using SMLM methods is not ideal, given the propensity for the dyes or switchable proteins to blink, that is, transition to a dark state and then return to be counted seconds later as a new molecule. While corrections can be made for this blinking by determining the blinking time constant, the numbers obtained should be considered only approximate. In fact, N&B is a better counting method, and should be used in cases in which the absolute number of molecules is of great import to understanding the biological mechanisms. Excitation and switching laser illumination schemes can also strongly modulate the number of molecules observed. We have found that very low and constant levels of the switching laser (405 nm) leads to the least blinking and bleaching and the detection of nearly all molecules in an FOV. Imaging facilities at numerous research institutions have been equipping themselves with commercial versions of these super‐resolution microscopes. It is worthwhile to consult with experts in the field for the design, implementation and analysis of such experiments.

Conclusions

As noted in the Introduction to this review, it was not conceived as a thorough review of all of the literature in the field of quantitative microscopy of protein molecules in cells. Indeed, completing such a task would be nearly impossible. Rather this review was designed to provide the protein scientist with a few detailed examples of what it is possible to measure in live cells, and a feeling for the accuracy and precision of the quantitative results. Examples of protein folding studies, protein concentration determinations, oligomer stoichiometry measurements, diffusion in presence of cellular barriers, directed transport, and super‐resolution localization have been reviewed. In each case, state‐of‐the‐art fluorescence microscopy methodologies were implemented to answer central questions about the functional mechanisms of key cellular events. While highly detailed in vitro experiments will remain an absolute necessity for understanding biological function, it is hoped that this review will convince protein scientists that quantitative studies in cells, which allow for understanding the effects of cellular architecture and dynamic state transitions on these phenomena, are now well within reach.