Every laboratory with a fluorescence microscope should consider counting molecules
Abstract
Protein numbers in cells determine rates of biological processes, influence the architecture of cellular structures, reveal the stoichiometries of protein complexes, guide in vitro biochemical reconstitutions, and provide parameter values for mathematical modeling. The purpose of this essay is to increase awareness of methods for counting protein molecules using fluorescence microscopy and encourage more cell biologists to report these numbers. We address the state of the field in terms of utility and accuracy of the numbers reported and point readers to references for details of specific techniques and applications.
INTRODUCTION
Biology has benefited tremendously from the application of quantitative techniques (reviewed in Mogilner et al., 2012; Morelli et al., 2012). Numbers of molecules, stoichiometries, and concentrations are important for making the most use of quantitative simulations and proposing structural models. In addition, these measurements are critical for in vitro reconstitution and other biochemical assays. The introduction of green fluorescent protein (GFP) and its variants into research laboratories across the globe revolutionized the way we study cells. The linear relationship of signal intensity to the number of GFP molecules reveals more than spatial and temporal information for GFP-tagged proteins. Researchers can count tagged proteins in a living or fixed cell if the fluorescence output of a single GFP molecule is determined.
Counting protein molecules by fluorescence microscopy only requires a fluorescence imaging system and some basic analysis tools such as ImageJ (National Institutes of Health, Bethesda, MD), a free download. Thus research laboratories commonly publishing microscopy data should take advantage of the availability of this technique. Using microscope images to report only qualitative data or even arbitrary units of fluorescence intensity is an underutilization of the data set, especially if the protein is endogenously tagged.
Previously yeast model organisms offered an advantage to anyone counting molecules because of the ability to tag endogenous genes using efficient homologous recombination. The recent rapid development of genome editing techniques using DNA or RNA nucleases targeted to specific sequences makes counting molecules possible in many other cell types (Coffman and Wu, 2012). Current genome editing techniques have been reviewed elsewhere (Ramalingam et al., 2013; Wei et al., 2013; Aida et al., 2014; Chen and Gao, 2014; Mashimo, 2014). Although some of these technologies are still expensive and maturing, they are not absolutely necessary if one can measure the ratio of tagged and untagged protein in the cells and structure of interest (Engel et al., 2009; Johnston et al., 2010).
TECHNIQUES
The two most common techniques for measuring protein molecules by fluorescence microscopy are stepwise photobleaching to count steps and comparing the fluorescence intensity of a protein to a known standard. The specifics of these techniques are reported elsewhere (Wu and Pollard, 2005; Joglekar et al., 2006; Leake et al., 2006; Ulbrich and Isacoff, 2007; Sirotkin et al., 2010; Coffman et al., 2011; Laporte et al., 2011). One advantage of the photobleaching method is that it does not require a standard, but the disadvantage is that it is useful only in a subset of cases, especially when the molecule numbers are relatively low (for discussion see Coffman and Wu, 2012). The ratio method is more broadly applicable, but there are differing opinions about the best way to carry it out. A sum of all z-sections with signal spaced appropriately (Hirschberg et al., 1998; Wu and Pollard, 2005; Sirotkin et al., 2010; Coffman et al., 2011; Coffman and Wu, 2012; Laporte et al., 2011) is necessary to measure all the protein in the cell, and we prefer this method for measurements of specific structures. Others used only the focal plane with maximum intensity for both the standard and the protein of interest (Joglekar et al., 2006), but we showed that this method does not always agree with the sum measurement, depending on the distribution of the signal in the z-direction (Coffman et al., 2011). A change in size or shape of a structure during the cell cycle could contribute some error to the maximum-plane measurement. The ratio method is insensitive to the fluorophore used to some extent, because the same fluorescent protein fused to both the standard and protein of interest will have similar maturation efficiency, brightness, and other features (Coffman and Wu, 2012; Erlemann et al., 2012).
There are two types of measurements that might be useful: global protein content in the cell and local protein concentration in a structure or location of interest. For global protein content, otherwise isogenic cells with no fluorescent protein serve as a background control to subtract out cellular autofluorescence, as well as any offset from the camera and system (Wu and Pollard, 2005). Global protein measurements are best done by comparing fluorescence intensities to immunoblotting or to proteins of known concentration using a linear curve. For local measurements, background subtractions remove the average cytoplasmic concentration from inside the region of interest while simultaneously accounting for autofluorescence and offset. We and others have used a concentric background region of interest or a similar-sized region from nearby (Hoffman et al., 2001; Wu and Pollard, 2005; Joglekar et al., 2006, 2008; Johnston et al., 2010; Sirotkin et al., 2010; Coffman et al., 2011; Laporte et al., 2011). The choice between these two depends on the proximity of other concentrated fluorescence. Local measurements can be taken using the photobleaching method for fewer molecules or a ratio method. Measurements in fission yeast are not obviously affected by quenching from local protein accumulation (Wu and Pollard, 2005; Coffman et al., 2011; Coffman and Wu, 2012).
STANDARDS FOR COUNTING PROTEIN MOLECULES BY FLUORESCENCE MICROSCOPY
Several in vitro and in vivo standards for counting proteins have been reported and widely used. The most basic standard is the fluorescence of a single GFP molecule, which can be measured in several ways: 1) by determining loss of fluorescence intensity during stepwise photobleaching; 2) by directly measuring speckles in a dilute sample of purified GFP; and 3) by making a linear curve using known concentrations of purified GFP-tagged protein adjusted for the ratio of fluorescence of bulk beads to that of single beads (Leake et al., 2006; Graham et al., 2011; Lawrimore et al., 2011). In fission yeast, numbers from fluorescence microscopy that agree with flow cytometry and immunoblotting have been reported up to ∼105 molecules/cell (Wu and Pollard, 2005). Although GFP fluorescence is affected by its environment, measurements in different organisms and comparisons to in vitro GFP seem largely insensitive to differences in environment (Coffman and Wu, 2012). There are many considerations when choosing fluorescent tags and standards to use; these are reviewed elsewhere (Shaner et al., 2005; Chudakov et al., 2010; Coffman and Wu, 2012). The budding yeast homologue of centromere protein A (CENP-A) Cse4 has frequently been used as a counting standard, but problems have arisen (see later discussion). Thus the fluorescence of a single GFP molecule or a calibration curve composed of a range of molecule numbers is a more suitable standard.
ACCURACY OF FLUORESCENCE QUANTIFICATION
Two recent examples yield some useful insight into the accuracy of current methods for counting molecules by fluorescence microscopy. Cytokinesis proteins in the fission yeast Schizosaccharomyces pombe have been counted by live-cell fluorescence microscopy (Wu and Pollard, 2005) and by mass spectroscopy (Marguerat et al., 2012). Comparison of these two data sets shows that 77% of the proteins fall below a fivefold difference (Figure 1A). It is important to note that the growing conditions of the strains were different. Wu and Pollard (2005) used rich medium, whereas Marguerat et al. (2012) used minimal medium. There is a twofold reduction in actin concentration in minimal medium compared with rich medium (Wu and Pollard, 2005), which might explain the differences for most of the proteins. In addition, the proteins with numbers greater than fivefold higher in the fluorescence microscopy data set are all large proteins (>100 kDa), which might affect the accuracy of the mass spectroscopy data (Figure 1A). Indeed, we note that the formin Cdc12 is one such protein for which the fluorescence microscopy value is ∼600 molecules/cell compared with ∼30 in the mass spectroscopy data. Fluorescence microscopy shows that each cell has at least 200 speckles that are believed to be dimers (Coffman et al., 2009), suggesting that the fluorescence microscopy data are more accurate. A recent estimate of total proteins per cell volume (Milo, 2013) indicates the mass spectroscopy data set might underestimate protein numbers by approximately fivefold, which is consistent with the fluorescence data for most of the proteins. Fluorescence microscopy measurements are less susceptible to error arising from protein size or abundance and therefore are likely to be more accurate than mass spectroscopy. Moreover, mass spectroscopy is not useful for counting local concentrations in most cases.

FIGURE 1: Comparisons of protein numbers counted by different quantification methods. (A) The quotient of the protein numbers in fission yeast cells from Table 1 of Wu and Pollard (2005) divided by the data from mass spectrometry for the same proteins (Marguerat et al., 2012) plotted vs. the predicted molecular weights of the proteins (PomBase, www.pombase.org). The majority of quotients (20/26) are <5. Proteins with quotients >5 are labeled. (B) Comparison of fluorescence measurements of CENP-A Cse4 in anaphase clusters in S. cerevisiae using various standards and methods (Coffman et al., 2011; Lawrimore et al., 2011; Erlemann et al., 2012; Galletta et al., 2012; Shivaraju et al., 2012; Aravamudhan et al., 2013). Asterisk indicates that this number was measured indirectly, n is for Spc24 measurement, and the Cse4 number is given by the ratio comparison from Joglekar et al. (2006). (C) Comparison of fluorescence ratio measurement (Coffman et al., 2011) to measurement by PALM (Lando et al., 2012) for S. pombe CENP-A Cnp1 in anaphase clusters. (D) Histogram of the number of articles each year from 1996 to 2012 using fluorescence methods to count proteins. This is by no means an exhaustive tabulation, but it includes >100 cross-references from the key papers on the subject.
The second example that we would like to highlight is the disagreement over measurements of centromere-specific protein CENP-A in budding and fission yeast. The Saccharomyces cerevisiae (budding yeast) CENP-A Cse4 counted by fluorescence microscopy ranges from 32 to 122 per anaphase cluster (Figure 1B) or 2 to ∼8 per centromere, whereas chromatin immunoprecipitation (ChIP) data imply 2 Cse4 molecules per centromere. This is an important distinction, as it might affect structural models of the centromere and kinetochore and the definition of a point centromere. Two of the fluorescence measurements of Cse4 seem to support the number obtained by ChIP (Shivaraju et al., 2012; Aravamudhan et al., 2013), but Lawrimore et al. (2011) showed convincing evidence that ChIP does not yield accurate numbers of proteins bound to centromeric DNA due to its measurement of population averages. There are fewer measurements of the fission yeast CENP-A Cnp1, but ChIP data give a number that lies between the two fluorescence measurements (Figure 1C). Lawrimore et al. (2011) used the ratios reported in Joglekar et al. (2008) to adjust the S. pombe kinetochore numbers, but the tagged Cnp1 in Joglekar et al. (2008) was not the sole copy of Cnp1 (Coffman et al., 2011; Yao et al., 2013). Ndc80 numbers agree closely in three studies (Coffman et al., 2011; Lawrimore et al., 2011; McCormick et al., 2013), suggesting that the photoactivated localization microscopy (PALM) measurement (Lando et al., 2012) might be overcorrected to account for blinking. One possible explanation for the difference between ChIP and fluorescence measurements in both yeasts might be that not all CENP-As in anaphase clusters are associated with centromeric DNA (Haase et al., 2013). In addition, the distribution of Cse4 at budding yeast centromere clusters is not consistent with only 2 molecules per centromere (Haase et al., 2012, 2013). Thus further experiments are needed to determine the amount of CENP-A that contributes to centromere identity in both budding and fission yeasts (Maresca, 2013). However, even the largest and smallest numbers differ by only fourfold (Figure 1B), which might suffice for some applications. Until a consensus is reached, CENP-A proteins are not the best standards to use in fluorescence quantification. Fortunately, the calibration curves for budding (Lawrimore et al., 2011) and fission (Wu and Pollard, 2005; McCormick et al., 2013) yeasts are suitable for measuring protein numbers over several orders of magnitude.
SOURCES OF ERROR
Each method to count molecules has sources of error, and some methods are more technically demanding or require specialized analytical skills. Counting molecules by photobleaching requires a very sensitive imaging system, and the low signal-to-noise ratio introduces errors (Waters, 2009). Detecting the step boundaries in photobleaching data requires user-defined criteria and can be challenging since the data are usually noisy. The modified Chung–Kennedy algorithm was used to aid in defining step boundaries (Leake et al., 2006; Engel et al., 2009; Coffman and Wu, 2012), but the precise boundaries between plateaus might not always be obvious. By assembling a large data set of step sizes, one can attenuate the inaccuracy of defining step boundaries.
Ratio measurements require only a standard fluorescence microscope and a digital camera, but low signal-to-noise ratio is still a concern. Autofluorescence noise can also contribute to errors when GFP intensity is <1.5 times the autofluorescence (Heinrich et al., 2013). The main challenge is the reliability of the standards used to convert fluorescence intensity directly into molecule numbers. A calibration curve is more accurate than a single standard, especially when measured proteins span several orders of magnitude (Wu and Pollard, 2005; McCormick et al., 2013).
Two additional fluorescence microscopy methods not described in detail here have been used to count molecules in live cells. Fluorescence correlation spectroscopy (FCS; Shivaraju et al., 2012) is an established method for determining concentrations of dilute fluorescent proteins in addition to single-molecule dynamics and mobility (Kim et al., 2007). Because this determination is done within a defined volume, the number of molecules is calculable (Meyer and Schindler, 1988). FCS is particularly suited to quantifying molecular dynamics when fluorophores are at nanomolar concentration and are highly mobile. FCS is limited in its application for counting molecules because of its sensitivity to photobleaching and population heterogeneity (Kim et al., 2007).
PALM, a superresolution microscopy technique, has recently emerged as a method to directly count molecules in live cells (Annibale et al., 2011; Lando et al., 2012; Sengupta and Lippincott-Schwartz, 2012). The basic idea of superresolution microscopy is to observe molecules one at a time so that their precise location can be determined. As a result, it should be possible to count molecules directly without the need for separate standards. Unfortunately, the analysis methods for PALM are still fraught with uncertainties, which make it difficult to produce accurate counts. The main difficulty is being able to count each molecule once and only once, partly because photoactivatable fluorescent proteins are able to blink on and off in subsequent images (Annibale et al., 2011; Lando et al., 2012; Sengupta and Lippincott-Schwartz, 2012). As analysis algorithms (Sengupta and Lippincott-Schwartz, 2012) and photoactivatable fluorescent proteins (Zhang et al., 2012) improve, PALM could eventually become the gold standard for counting molecules.
CONCLUSION
Based on the increasing number of articles reporting protein numbers by fluorescence microscopy in the past two decades (Figure 1D), this technique has proven to be useful. Stoichiometries of the budding yeast kinetochore (Joglekar et al., 2006), budding yeast γ-tubulin ring complex (Erlemann et al., 2012), fission yeast cytokinesis node (Laporte et al., 2011), fission and budding yeast endocytic patches (Sirotkin et al., 2010; Galletta et al., 2012), bacterial replisome (Reyes-Lamothe et al., 2010), and many other complexes have been elucidated based on these methods, which are essential for proposing structural models. The next step in many of these cases is to use the stoichiometric data to inform and constrain in vitro reconstitution experiments and mathematical models of the function or assembly mechanisms of these complexes.
FOOTNOTES
Abbreviations used:| CENP-A | centromere protein A |
| FCS | fluorescence correlation spectroscopy |
| GFP | green fluorescent protein |
| PALM | photoactivated localization microscopy |
ACKNOWLEDGMENTS
We thank I-Ju Lee and anonymous reviewers for critical reading of the manuscript and valuable comments. This work is supported by an Elizabeth Clay Howald Presidential Fellowship from The Ohio State University to V.C.C. and National Institutes of Health Grant R01GM086546 to J.‑Q.W.
REFERENCES
- (2014). Translating human genetics into mouse: the impact of ultra-rapid in vivo genome editing. Dev Growth Differ 56, 34-45. Crossref, Medline, Google Scholar
- (2011). Quantitative photo activated localization microscopy: unraveling the effects of photoblinking. PLoS One 6, e22678. Crossref, Medline, Google Scholar
- (2013). The budding yeast point centromere associates with two Cse4 molecules during mitosis. Curr Biol 23, 770-774. Crossref, Medline, Google Scholar
- (2014). Targeted genome modification technologies and their applications in crop improvements. Plant Cell Rep 33, 575-583. Crossref, Medline, Google Scholar
- (2010). Fluorescent proteins and their applications in imaging living cells and tissues. Physiol Rev 90, 1103-1163. Crossref, Medline, Google Scholar
- (2009). Roles of formin nodes and myosin motor activity in Mid1p-dependent contractile-ring assembly during fission yeast cytokinesis. Mol Biol Cell 20, 5195-5210. Link, Google Scholar
- (2012). Counting protein molecules using quantitative fluorescence microscopy. Trends Biochem Sci 37, 499-506. Crossref, Medline, Google Scholar
- (2011). CENP-A exceeds microtubule attachment sites in centromere clusters of both budding and fission yeast. J Cell Biol 195, 563-572. Crossref, Medline, Google Scholar
- (2009). Intraflagellar transport particle size scales inversely with flagellar length: revisiting the balance-point length control model. J Cell Biol 187, 81-89. Crossref, Medline, Google Scholar
- (2012). An extended gamma-tubulin ring functions as a stable platform in microtubule nucleation. J Cell Biol 197, 59-74. Crossref, Medline, Google Scholar
- (2012). Molecular analysis of Arp2/3 complex activation in cells. Biophys J 103, 2145-2156. Crossref, Medline, Google Scholar
- (2011). Counting proteins bound to a single DNA molecule. Biochem Biophys Res Commun 415, 131-134. Crossref, Medline, Google Scholar
- (2013). A 3D map of the yeast kinetochore reveals the presence of core and accessory centromere-specific histone. Curr Biol 23, 1939-1944. Crossref, Medline, Google Scholar
- (2012). Bub1 kinase and Sgo1 modulate pericentric chromatin in response to altered microtubule dynamics. Curr Biol 22, 471-481. Crossref, Medline, Google Scholar
- (2013). Determinants of robustness in spindle assembly checkpoint signalling. Nat Cell Biol 15, 1328-1339. Crossref, Medline, Google Scholar
- (1998). Kinetic analysis of secretory protein traffic and characterization of Golgi to plasma membrane transport intermediates in living cells. J Cell Biol 143, 1485-1503. Crossref, Medline, Google Scholar
- (2001). Microtubule-dependent changes in assembly of microtubule motor proteins and mitotic spindle checkpoint proteins at PtK1 kinetochores. Mol Biol Cell 12, 1995-2009. Link, Google Scholar
- (2008). Molecular architecture of the kinetochore-microtubule attachment site is conserved between point and regional centromeres. J Cell Biol 181, 587-594. Crossref, Medline, Google Scholar
- (2006). Molecular architecture of a kinetochore-microtubule attachment site. Nat Cell Biol 8, 581-585. Crossref, Medline, Google Scholar
- (2010). Vertebrate kinetochore protein architecture: protein copy number. J Cell Biol 189, 937-943. Crossref, Medline, Google Scholar
- (2007). Fluorescence correlation spectroscopy in living cells. Nat Methods 4, 963-973. Crossref, Medline, Google Scholar
- (2012). Quantitative single-molecule microscopy reveals that CENP-ACnp1 deposition occurs during G2 in fission yeast. Open Biol 2, 120078. Crossref, Medline, Google Scholar
- (2011). Assembly and architecture of precursor nodes during fission yeast cytokinesis. J Cell Biol 192, 1005-1021. Crossref, Medline, Google Scholar
- (2011). Point centromeres contain more than a single centromere-specific Cse4 (CENP-A) nucleosome. J Cell Biol 195, 573-582. Crossref, Medline, Google Scholar
- (2006). Stoichiometry and turnover in single, functioning membrane protein complexes. Nature 443, 355-358. Crossref, Medline, Google Scholar
- (2013). Chromosome segregation: not to put too fine a point (centromere) on it. Curr Biol 23, R875-R878. Crossref, Medline, Google Scholar
- (2012). Quantitative analysis of fission yeast transcriptomes and proteomes in proliferating and quiescent cells. Cell 151, 671-683. Crossref, Medline, Google Scholar
- (2014). Gene targeting technologies in rats: zinc finger nucleases, transcription activator-like effector nucleases, and clustered regularly interspaced short palindromic repeats. Dev Growth Differ 56, 46-52. Crossref, Medline, Google Scholar
- (2013). Measuring affinities of fission yeast spindle pole body proteins in live cells across the cell cycle. Biophys J 105, 1324-1335. Crossref, Medline, Google Scholar
- (1988). Particle counting by fluorescence correlation spectroscopy. Simultaneous measurement of aggregation and diffusion of molecules in solutions and in membranes. Biophys J 54, 983-993. Crossref, Medline, Google Scholar
- (2013). What is the total number of protein molecules per cell volume? A call to rethink some published values. Bioessays 35, 1050-1055. Crossref, Medline, Google Scholar
- (2012). Cell polarity: quantitative modeling as a tool in cell biology. Science 336, 175-179. Crossref, Medline, Google Scholar
- (2012). Computational approaches to developmental patterning. Science 336, 187-191. Crossref, Medline, Google Scholar
- (2013). A CRISPR way to engineer the human genome. Genome Biol 14, 107. Crossref, Medline, Google Scholar
- (2010). Stoichiometry and architecture of active DNA replication machinery in Escherichia coli. Science 328, 498-501. Crossref, Medline, Google Scholar
- (2012). Quantitative analysis of photoactivated localization microscopy (PALM) datasets using pair-correlation analysis. Bioessays 34, 396-405. Crossref, Medline, Google Scholar
- (2005). A guide to choosing fluorescent proteins. Nat Methods 2, 905-909. Crossref, Medline, Google Scholar
- (2012). Cell-cycle-coupled structural oscillation of centromeric nucleosomes in yeast. Cell 150, 304-316. Crossref, Medline, Google Scholar
- (2010). Quantitative analysis of the mechanism of endocytic actin patch assembly and disassembly in fission yeast. Mol Biol Cell 21, 2894-2904. Link, Google Scholar
- (2007). Subunit counting in membrane-bound proteins. Nat Methods 4, 319-321. Crossref, Medline, Google Scholar
- (2009). Accuracy and precision in quantitative fluorescence microscopy. J Cell Biol 185, 1135-1148. Crossref, Medline, Google Scholar
- (2013). TALEN or Cas9—rapid, efficient and specific choices for genome modifications. J Genet Genomics 40, 281-289. Crossref, Medline, Google Scholar
- (2005). Counting cytokinesis proteins globally and locally in fission yeast. Science 310, 310-314. Crossref, Medline, Google Scholar
- (2013). Plasticity and epigenetic inheritance of centromere-specific histone H3 (CENP-A)-containing nucleosome positioning in the fission yeast. J Biol Chem 288, 19184-19196. Crossref, Medline, Google Scholar
- (2012). Rational design of true monomeric and bright photoactivatable fluorescent proteins. Nat Methods 9, 727-729. Crossref, Medline, Google Scholar



