Journal of Experimental Botany, Vol. 62, No. 12, pp. 4101–4113, 2011
doi:10.1093/jxb/err176 Advance Access publication 1 June, 2011
REVIEW PAPER
Quantifying immunogold localization on electron microscopic
thin sections: a compendium of new approaches for plant
cell biologists
Terry M. Mayhew*
School of Biomedical Sciences, Queen’s Medical Centre, University of Nottingham, Nottingham NG7 2UH, UK
* To whom correspondence should be addressed. E-mail: [email protected]
Received 4 April 2011; Revised 3 May 2011; Accepted 9 May 2011
Abstract
A review is presented of recently developed methods for quantifying electron microscopical thin sections on which
colloidal gold-labelled markers are used to identify and localize interesting molecules. These efficient methods rely
on sound principles of random sampling, event counting, and statistical evaluation. Distributions of immunogold
particles across cellular compartments can be compared within and between experimental groups. They can also be
used to test for co-localization in multilabelling studies involving two or more sizes of gold particle. To test for
preferential labelling of compartments, observed and expected gold particle distributions are compared by x2
analysis. Efficient estimators of gold labelling intensity [labelling density (LD) and/or relative labelling index (RLI)] are
used to analyse volume-occupying compartments (e.g. Golgi vesicles) and/or surface-occupying compartments (e.g.
cell membranes). Compartment size is estimated by counting chance events after randomly superimposing test
lattices of points and/or line probes. RLI¼1 when there is random labelling and RLI >1 when there is preferential
labelling. Between-group comparisons do not require information about compartment size but, instead, raw gold
particle counts in different groups are compared by combining x2 and contingency table analyses. These tests may
also be used to assess co-distribution of different sized gold particles in compartments. Testing for co-labelling
involves identifying sets of compartmental profiles that are unlabelled and labelled for one or both of two gold
marker sizes. Numbers of profiles in each labelling set are compared by contingency table analysis and x2 analysis
or Fisher’s exact probability test. The various methods are illustrated with worked examples based on empirical and
synthetic data and will be of practical benefit to those applying single or multiple immunogold labelling in their
research.
Key words: Co-distribution, co-labelling, electron microscopy, gold particles, immunolocalization, multiple labelling,
quantification, single labelling.
Introduction
It is widely acknowledged that details of cell fine structure
are below the resolving power of optical (including
confocal) microscopy and are best identified by using
transmission electron microscopy (TEM). When used in
conjunction with immunocytochemical marker probes,
TEM allows interesting molecules (usually, but not always,
protein antigens) to be mapped in the context of cellular
ultrastructure rather than merely visualized at nuclear or
cytoplasmic levels. Furthermore, the application of
particulate colloidal gold conjugated to protein A or
secondary antibodies offers opportunities not only to
immunolocalize identified antigens in different cell or tissue
compartments but also to express these localizations (including co-localizations) quantitatively.
A variety of quantitative methods has evolved to count
gold particles and test for preferential localization in singlelabel experiments or for co-localization in dual- or multilabel
experiments. Methods have ranged in sophistication from
ª The Author [2011]. Published by Oxford University Press [on behalf of the Society for Experimental Biology]. All rights reserved.
For Permissions, please e-mail: [email protected]
4102 | Mayhew
simple gold particle percentage frequency distributions across
subcellular structures, via measures of labelling density (LD),
to point-pattern, clustering and nearest-neighbour analyses
(for reviews, see Griffiths, 1993; Bendayan, 2001; Mayhew,
2007; D’Amico and Skarmoutsou, 2008a, b; Mayhew and
Lucocq, 2008b). In plant cell biology, the same variety of
methods as is seen in other areas of cell biology has been
applied (e.g. Kondo et al., 1998; Anderson et al., 2003;
Follet-Gueye et al., 2003; Seguı́-Simarro et al., 2003; Ju et al.,
2005; Bernal et al., 2007; Christopher et al., 2007; Coronado
et al., 2007; Negi et al., 2008; Wei et al., 2009; Bilska and
Sowiński, 2010; Dafoe et al., 2010; Zechmann and Müller,
2010; Zhou et al., 2010).
In recent years, effort has been accorded to developing
a more rigorous and coherent set of quantitative methods
which utilize sound principles of random sampling, stereological event counting, and statistical evaluation. In fact,
there is now a compendium of new methods for abstracting
quantitative data and testing appropriate null hypotheses
(Mayhew et al., 2002, 2003, 2009; Lucocq et al., 2004;
Mayhew and Desoye, 2004; Mayhew and Lucocq, 2008a, b,
2011). At least some of these technical developments have
been applied to plant cells (Lopes et al., 2006; Ruel et al.,
2009; Chevalier et al., 2010). The purpose here is to review
the available methods and illustrate their practical utility for
quantifying immunogold localization in TEM sections.
Whilst this review is aimed at the plant science community,
applications to other sources of cells, tissues, and organs
have been summarized elsewhere (Mayhew and Lucocq,
2008b; Mayhew et al., 2009).
A given study might begin by defining the subcellular
compartments which are pertinent to the experimental aims
or hypothesis. The biological specimens will need to be
sampled for investigation by TEM and questions asked
about the quantitative methods to be deployed in order to
test the corresponding statistical null hypothesis. In the case
of single-label studies, the aim might be to test for
preferential labelling of certain compartments within a cell
or for shifts in labelling patterns associated with some
experimental factor or treatment. With multilabel studies
(Geuze et al., 1981; Bendayan, 1982; Hagiwara et al., 2010),
the aim is to test for co-localization or co-distribution of
two or more antigens in subcellular structures.
The following sections identify some of the key issues
associated with these aspects of study design.
Defining structures and substructures
Biological structure is divisible into spatial compartments
which may be volume occupying (e.g. cells, organelles),
surface occupying (e.g. cell membranes), or length occupying (e.g. microtubules). On TEM thin sections, volumeoccupying structures (dimensions in lm3) appear as images
(profiles or transections) which have a certain sectional area
(lm2). By the same token, surface-occupying structures
(lm2) appear on section planes as linear features, for
example membrane traces (lm1). Finally, length-occupying
structures (lm1) will appear on section planes as a number
of transections (lm0). In plants, volume-occupying
compartments could be collections of cognate components
(e.g. all chloroplasts within a cell), related components (e.g.
different parts of the Golgi complex), dissimilar components (e.g. a mixture of chloroplasts and cytosol), or spatial
regions (e.g. cuticle, upper epidermis, palisade mesophyll,
spongy mesophyll, or the lower epidermis of a leaf).
Surface- and length-occupying compartments might be
similarly classified. An important initial step is to decide on
the types and numbers of compartments to be included in
the definitive study.
The choice of compartments to include will be determined both by the needs of the investigation and by prior
experience and expectation (based on the findings of a pilot
study or on knowledge of the antigens and antibodies).
Although there may be reluctance to exclude compartments,
in reality a sensible balance must be struck between
conflicting sources of variability, namely precision of
localization and precision of estimation of immunogold
labelling (Mayhew et al., 2002). The former is influenced by
the number of compartments selected and the latter by the
variability of gold particle counts within each of those
compartments. In fact, it is related to 1/OGo where Go is the
observed number of gold particles falling on a given
compartment. Consequently, precision of estimation is
likely to be poorer for rare, small, or poorly labelled
compartments. In general, increasing the number of compartments improves the precision of cellular localization,
but it may also reduce precision of estimation.
Comparing patterns of compartmental gold labelling
within and between different groups of cells involves
calculating the expected numbers of gold particles, Ge. For
statistical testing to be valid using the methods described
below, no expected value should be <1 and no more than
20% of values should be <5 (Daly and Bourke, 2000; Petrie
and Sabin, 2000). If a pilot study shows that these
restrictions are not met, it will be better to reduce the
number of compartments (by omission or combination) or
to count more gold particles. In co-labelling studies, these
considerations may also influence the choice of statistical
test (Mayhew and Lucocq, 2011).
A pilot study provides an effective way of choosing
which compartments to include in the definitive study. It
might be conducted by systematically sampling 1–2
labelled specimen support grids, counting 100–200 gold
particles, and identifying the compartments with which
they are associated (Lucocq et al., 2004). For statistical
reasons, it is prudent to include in analyses not only the
main labelled compartments of interest but also other
compartments not of primary interest (considered to be
unlabelled or ‘background labelled’). For convenience,
compartments not of primary interest may be brought
together as a separate artificial composite compartment
(Mayhew et al., 2002). It is sensible to define at least three
and no more than 12 compartments within a cell (Mayhew
et al., 2002) and, when comparing different groups of cells,
to count approximately equal numbers of gold particles
Quantitative immunoelectron microscopy | 4103
per group (Mayhew and Desoye, 2004; Mayhew et al.,
2004).
is important to use raw particle counts because these satisfy
requirements for statistical comparisons.
Multistage random sampling
Single labelling: comparing compartments
within a cell
In TEM, the final resolution of the image depends in part
on section thickness (usually 50–90 nm). This limits the
amount of material that can be examined and imposes the
need for a multistage or hierarchical sampling scheme in
which the highest stage is the specimen itself (a plant or part
thereof or, alternatively, some cultured cells). During TEM
preparation procedures, specimens give rise to tissue blocks,
some of which are selected for sectioning, and some of the
sections are selected for microscopical examination and
sampled as fields of view (FOVs).
It is critically important that each stage of this selection
process is conducted by some form of random sampling
(Cruz-Orive and Weibel, 1981; Mayhew et al., 2002, 2003;
Lucocq et al., 2004; Mayhew, 2006, 2008; Lucocq, 2008).
The reason is that this form of sampling is fair (unbiased)
and can be efficient. Random sampling at each stage affords
every part of the specimen the same chance of being
selected, and this is important regardless of the nature of
the compartments (volume, surface, or length occupying)
being studied. Random sampling can also allow every
orientation of the specimen an equal chance of being
selected (Baddeley et al., 1986; Mattfeldt et al., 1990;
Nyengaard and Gundersen, 1992). Indeed, whilst randomizing section position satisfies requirements for sampling
volume-occupying structures, the combination of random
position with random orientation is necessary when dealing
with surface- and length-occupying compartments or mixtures of these with volume-occupying compartments. Provided that sampling intervals do not coincide with natural
patterns within the specimen, systematic uniform random
(SUR) sampling tends to be more efficient than simple
random sampling. With SUR sampling, the position and
orientation of the first item can be selected at random and
a pre-determined pattern (the sampling interval) decides the
positions and orientations of subsequent items (Gundersen
and Jensen, 1987; Mayhew, 1991, 2008; Gundersen et al.,
1999; Howard and Reed, 2005).
Once cells have been sampled randomly, and decisions
made about which compartments to include, the magnification of the microscopical FOVs should be kept to the
minimum which allows identification of compartments and
gold particles. This further improves precision of estimation
in the final estimates. It is then a relatively simple matter to
count gold particles associated with each compartment.
This can be achieved by recording FOVs digitally or
photographically, but an efficient alternative is to scan
labelled sections at the TEM (Lucocq et al., 2004; Lucocq,
2008). If undertaken as a pilot study in order to judge where
the bulk of labelled antigen resides, gold particle counts can
be presented as a percentage frequency distribution. In
contrast, for definitive studies using the present methods, it
If the chosen compartments are all volume occupying
(Fig. 1), the first step would be to count gold particles lying
on compartments on all sampled FOVs. The resulting
numerical frequency distribution represents an ‘observed’
distribution (Mayhew et al., 2002). It is intuitively obvious
that, if the labelling of compartments occurred at random,
all compartments would display the same LD value
(expressed, say, as numbers of gold particles per lm2 of
profile area). Therefore, we can test whether or not the
observed labelling of compartments is consistent with
a random pattern by comparing the LD values of different
compartments. For instance, the LD of a given compartment (LDcomp) can be compared with that of the cell as
a whole (LDcell). The resulting ratio, LDcomp/LDcell, provides a measure of the RLI of that compartment (RLIcomp).
When there is random labelling, RLIcomp¼1, but where
there is preferential labelling, RLIcomp > 1. There are two
main ways to obtain RLI values.
Direct estimates of RLI
The ‘expected’ or ‘predicted’ distributions of gold particles
can be obtained by taking advantage of a basic stereological
principle: a set of test points randomly positioned on
sectional images hits compartments with probabilities determined by their relative sectional (or profile) areas. In
practice, randomly positioned lattices of test points
(Fig. 1B) are superimposed on randomly sampled FOVs
(Fig. 1A). The resulting distribution of points represents
that which would be expected if gold particles were
scattered randomly across the cell (Mayhew et al., 2002). If
the total point count is normalized to equal the total
number of observed gold particles falling on all compartments, and the compartmental point counts weighted
accordingly, they will provide the expected distribution,
and the RLI of any given compartment can be estimated as
RLIcomp ¼ Go =Ge
where Go and Ge are the observed and expected number of
gold particles on that compartment.
For a study group with r compartments (arranged in
rows), a convenient inferential test of the statistical significance of apparent differences between observed and
expected distributions (arranged in columns) is the twosample v2 test with r–1 degrees of freedom (df). For a given
compartment, the corresponding partial v2 is calculated as
(GoGe)2/Ge.
If the total v2 value for the given df indicates that
observed and expected distributions are significantly different, preferentially labelled compartments can be identified
on the basis of satisfying two criteria. First, the RLI value
4104 | Mayhew
Fig. 1. Testing for preferential localization. (A) A TEM thin section of a plant cell containing various volume- and surface-occupying
compartments and labelled with a gold-conjugated marker (black squares in red circles). (B) To test whether or not any volumeoccupying compartments are preferentially labelled, expected distributions of random gold particles can be simulated by randomly
superimposing lattices of test point probes. For the lattice shown, points, P, are represented by the bottom left-hand corners of lattice
squares. Gold particles and test points falling on the compartments are counted. Observed and expected distributions of gold particles
are compared by means of v2 analysis. Modifications of the method (see text) can deal with volume- and/or surface-occupying
compartments. chl, chloroplast; cw, cell wall; cyt, cytosol; Gc, Golgi complex; mit, mitochondrion; nuc, nucleus; pm, plasma membrane;
rer, rough endoplasmic reticulum; vac, vacuole.
must be >1 and, secondly, the corresponding partial v2
value must account for a substantial proportion (say, at
least 10%) of total v2.
Comparisons within a given cell can also be undertaken
when the compartments are all surface occupying (i.e.
membranes). In this case, constructing the expected distribution relies on another stereological principle: a set of test
lines randomly positioned and randomly orientated with
respect to sectional images will hit membrane traces with
probabilities determined by their relative trace lengths
(Mayhew et al., 2002).
Indirect estimates of RLI
An alternative way to obtain RLI values is to compare the
LD values of selected compartments with the LD of the
appropriate reference structure (cell, tissue, or region), for
example RLIcomp¼LDcomp/LDcell. Usually, LD values are
expressed as gold particles per lm2 (for profiles of volumeoccupying compartments) or per lm (for traces of surfaceoccupying compartments). However, a simpler and more
efficient approach is to relate gold particle counts to the
observed numbers of test probes (Mayhew et al., 2003) and
express LD values as numbers of gold particles per test
point or per intersection.
It is easier to draw between-compartment comparisons
when all compartments belong to the same category, for
example they are all volume occupying or all surface
occupying. However, some molecules translocate between
membrane and organelle compartments (or vice versa), and
technical refinements have been devised in an attempt to treat
all compartments in a similar manner (Mayhew and Lucocq,
2008a). The refinements recognize that surface-occupying
structures appear on the cut planes of ultrathin sections as
trace lengths, whereas volume-occupying structures appear as
profile areas. Consequently, the traces of surface-occupying
structures do not generate encounters with random test
points (the probability of this happening is zero). One
solution to this disparity is to convert linear membrane traces
into profile areas (Mayhew and Lucocq, 2008a).
To achieve this, an ‘acceptance zone’ is defined on each
side of the membrane trace and its width is determined by
the observed dispersion of gold label or the expected
labelling resolution (say, 20 nm for primary antibody
followed by gold conjugated to protein A). A convenient
guide is to choose a zone width equivalent to twice the
diameter of the immunogold particles being used for
labelling. The profile area of the acceptance zone covering
both sides of the membrane trace is calculated by multiplying its overall width by the profile length estimated by
intersection counting. Alternatively, the number of equivalent test points can be counted after randomly superimposing a lattice with a systematic array of test points (Mayhew
and Lucocq, 2008a, b).
A second issue is the fact that, whilst most membrane
traces are clearly visible on ultrathin sections, some are
indistinct or vague because the parent membrane is tilted
relative to the section plane. To estimate Ge, it is necessary to
correct for this image loss. This can be achieved by confining
counts of gold particles to membrane traces which are clearly
Quantitative immunoelectron microscopy | 4105
Fig. 2. Testing for shifts in localization. Cells from two study groups (e.g. A control and B experimental) contain volume- and surfaceoccupying compartments across which a gold-conjugated marker (black squares in red circles) is distributed. To test for differences in
labelling patterns between the two cell groups, observed numbers of gold particles falling on compartments are compared by
contingency table and v2 analyses. This method can be applied to volume- or surface-occupying compartments or a mixture of the two.
visible because they have been vertically sectioned (Baddeley
et al., 1986). These counts are then corrected for image loss
using correction factors estimated by goniometry or stereology (Mayhew and Lucocq, 2008a, b).
Single labelling: comparing compartments
between cell groups
For volume- or surface-occupying compartments (or a mix
of the two), gold particles on all randomly sampled FOVs
from all randomly sampled grids are counted. Again, the
raw counts represent ‘observed’ distributions and those in
each group of cells (Fig. 2) can be compared statistically by
contingency table and v2 analyses which provide the
corresponding ‘expected’ distributions (Mayhew et al.,
2002). With c groups (arranged in columns) and r compartments (arranged in rows), sets of partial v2 values are
calculated for each compartment and group. Examining the
total v2 value for df¼(c–1)3(r–1) will indicate whether or
not the observed distributions in the study groups differ. If
they do, examining partial v2 values will identify the
compartments which are responsible. Again, a convenient
cut-off is a partial v2 amounting to >10% of the total.
Multiple labelling: testing for compartmental
co-localization
This is performed much as described above for between-group
single labelling (Mayhew and Lucocq, 2011). Analysis begins
with counting Go for each of the gold-labelled antigens being
studied. Usually, the latter will be identified by using different
sizes of colloidal gold particle (Fig. 3). Next, observed
numbers of gold particles are compared by contingency table
Fig. 3. Testing for compartmental co-localization in a dual-labelling
study. A cell has a number of compartments which have been
labelled with large and small gold particles (large squares in green
circles and small squares in red circles). The two sizes of gold particle
seem to co-localize in chloroplasts. To test this possibility, raw
counts of large and small gold particles are compared by contingency table and v2 analyses. Again, the method can be applied to
volume- or surface-occupying compartments or a mix of the two.
4106 | Mayhew
analysis with a antigens (arranged in columns) and c compartments (arranged in rows). This generates the corresponding
expected numbers of gold particles (Ge) and partial v2 values.
The total v2 value, for df¼(a–1)3(c–1), indicates whether or
not the gold labelling distributions of the antigens differ. If
they do, again, partial v2 values will identify the compartments which are responsible.
Multiple (dual) labelling: testing for
co-labelling of structure profiles
The sampling aim now is to select individual profiles (rather
than compartments) in an unbiased and efficient manner
using some form of unbiased counting frame. Several
varieties exist, but a convenient one for dealing with profiles
which are irregular in size and shape is the ‘forbidden line’
counting frame (Gundersen, 1977; Mayhew and Lucocq,
2011). Any profile that interacts with the forbidden lines
and their extensions cannot be counted. Profiles are countable only if they are entirely inside the frame or touch its
‘allowable lines’ (Fig. 4). Since some profiles may extend
beyond the borders of the frame, the frame should be
smaller than the FOV so that it is surrounded by a guard
area. The width of the latter should correspond to roughly
half the maximum caliper diameter of the largest profile. All
profiles sampled in this way must be visible in their entirety,
and this may constrain the size of the compartment which
can be examined and/or the magnification used. An
alternative would be to sample profiles at the TEM using
scanning methods (see Lucocq, 2008; Mayhew and Lucocq,
Fig. 4. Testing for profile co-labelling in a dual-labelling study.
A volume-occupying compartment appears as separate vesicle
profiles on an ultrathin section labelled with large and small gold
particles (large and small squares). A randomly positioned
‘forbidden line’ counting frame has six profiles associated with it.
The ‘allowable lines’ for counting are those at the top and right
side of the frame. One profile at the left touches the forbidden line
and cannot be counted. The five profiles eligible for counting can
be classed as follows: one is double positive, one is double
negative, two (one of which touches the allowable line on the right)
are labelled by small gold particles, and one is labelled by the large
gold particle. The numbers of profiles in these classes are
analysed by contingency table analysis and v2 analysis or Fisher’s
exact test. The method can be modified to deal with surfaceoccupying compartments.
2011), in which case there would be no limit on profile size
as long as each profile can be investigated in its entirety by
examining areas outside the selected FOVs. Once profiles
are identified, they are classified according to the gold
particles of each size which they contain (Fig. 4).
In the case of dual labelling with two sizes of gold particle
(say, 5 nm and 15 nm), the following classes of profile
would be present: (i) those labelled by 5 nm gold particles
but negative for 15 nm particles, Ng5+Ng15–; (ii) those
labelled by both 5 nm and 15 nm particles, Ng5+Ng15+; (iii)
those negative for 5 nm particles but labelled for 15 nm
particles, Ng5–Ng15+; and (iv) those negative for both 5 nm
and 15 nm particles, Ng5–Ng15–.
Unbiased selection of whole profiles will reduce sampling
errors, producing fewer double and single negatives. It may
also improve the chances of detecting double positives.
The counts of profiles in each category are analysed using
a 232 contingency table and, if some of the frequencies are
low and use of the v2 test is inadmissible (e.g. some expected
values are 0 or more than 20% are <5), it may be necessary
to use Fisher’s exact test (Fisher, 1922). This is appropriate
for data classified in two ways, but the calculations are
complicated and best left to a statistical software package.
A convenient site hosted by GraphPad Software Inc. is
available via the internet (http://www.graphpad.com/
quickcalcs/contingency1.cfm) and can be used to undertake
the calculations. With Fisher’s test, a probability level of
P <0.05 indicates that labelling by each of two gold markers
is not independent. In other words, dual labelling of profiles
occurs more (or less) often than would be expected purely
by chance. To resolve these two possible interpretations, an
odds ratio (OR) is calculated. For example, the ratio of
labelled:unlabelled profiles for 15 nm gold particles is
calculated separately for each of the groups that are positive
and negative for 5 nm particles. The OR is then calculated
as the ratio of these two ratios. If the OR is positive, there is
a higher proportion of labelling by 15 nm particles on
profiles that are also labelled by 5 nm particles. If the OR is
negative, there is less labelling concentrated on profiles that
are negative for 5 nm particles (Mayhew and Lucocq, 2011).
Applications of the methods
Single labelling: testing for differences between
compartments within and between groups
To illustrate methods for drawing comparisons within and
between cell groups in single-label experiments, relevant
data have been extracted from a recent study (see Chevalier
et al., 2010) on cells from wild-type and transgenic lines of
tobacco (Nicotiana tabacum). The aim was to localize
glycosyltransferases within different compartments of the
Golgi complex in Bright Yellow 2 (BY-2) suspensioncultured cells. These enzymes are involved in the synthesis
of the hemicellulose polysaccharide xyloglucan found in the
plant cell walls.
Quantitative immunoelectron microscopy | 4107
For immunogold labelling of proteins tagged with green
fluorescent protein (GFP), ultrathin sections of BY-2 cells
expressing AtMUR3–GFP (b-1,2-galactosyltransferase) and
AtXT1–GFP (a-1,6-xylosyltransferase) were exposed to
polyclonal anti-GFP primary antibody followed by rabbit
secondary antibody conjugated with 10 nm gold particles.
Golgi complexes on microscopical FOVs were classified into
cis-, medial- and trans- compartments and the trans-Golgi
network (TGN). In order to compare labelling distributions
in wild-type and transformed cells, gold particles on these
compartments were counted.
For present purposes, data in Tables 1 and 2 of
Chevalier et al. (2010) were modified so as to bring the
totals of observed gold particles and test points to ;200
per cell group (Mayhew et al., 2002; Lucocq et al., 2004;
Mayhew and Lucocq, 2008b). The modified data set was
then used to test two null hypotheses: (i) there is no
difference in localization of gold-labelled anti-GFP in
Golgi compartments of wild-type BY-2 cells expressing
AtMUR3–GFP and (ii) there is no difference in the
distribution of gold-labelled anti-GFP in compartments
of wild-type and transformed BY-2 cells expressing
AtXT1–GFP.
For immunogold labelling, ultrathin sections of
N. tabacum leaf segments were exposed to mouse monoclonal anti-GFP primary antibody followed by 10 nm goldconjugated rabbit secondary antibody. On microscopical
FOVs, the cell wall, plasma membrane, ER, and ERderived vesicles were selected for gold particle counting.
Labelling distributions for GFP–TGBp2 and GFP–TGBp3
were compared. Data in Table II of Ju et al. (2005) were
modified to remove Golgi elements (which did not label)
and, again, to have ;200 observed gold particles in each of
the two groups.
Synthetic data were compiled for a virtual cell type in
which two proteins (designated Ag5 and Bg20) were labelled
using 5 nm and 20 nm gold particles, respectively. Gold
particles of each size were taken to be associated with three
main compartments (cell plasma membrane, Golgi complex,
and ER) and one artificial composite (residual) compartment representing the rest of the cell.
Data sets were used to test the null hypothesis of no
difference in spatial distributions of labelling between
groups (GFP–TGBp2 and GFP–TGBp3 for the real data,
and Ag5 and Bg20 for the synthetic data).
Dual labelling: testing for co-labelling of structural
profiles
Single labelling: testing for differences between groups
To illustrate further between-group comparisons, empirical
data were modified from a study by Zechmann and Müller
(2010). The original aim of this investigation was to
compare the localization of glutathione in the leaves and
roots of different species of dicotyledonous plants. Glutathione is of interest because of its various roles in plant
metabolism and defence.
Ultrathin sections of leaves and root tips were exposed to
anti-glutathione rabbit polyclonal antibody followed by
goat anti-rabbit secondary antibody conjugated to 10 nm
gold particles. Gold particles associated with mitochondria,
chloroplasts, peroxisomes, cytosol, and nuclei were counted
on FOVs. Data in Table 1 of Zechmann and Müller (2010)
were modified to limit comparisons to two species, namely
Arabidopsis thaliana and Cucurbita pepo, and with samples
fixed conventionally in buffered aldehydes. Gold particle
counts were also altered so as to yield totals of ;200 per
group. The resulting data set was used to test the null
hypothesis of no difference in subcellular distributions
between the two plant species.
Dual labelling: testing for compartmental co-localization
Real and synthetic data sets were used to illustrate this
method. The former was modified from findings of an
investigation into the localization of triple gene block
proteins (TGBp2 and TGBp3) in transgenic tobacco leaves
expressing GFP–TGBp2 or GFP–TGBp3 (Ju et al., 2005).
In cells infected by potato virus X, TGBp2 and TGBp3 are
virus-encoded movement proteins associated with the endoplasmic reticulum (ER).
Again, a synthetic data set was compiled for a cell type in
which two proteins (Cg5 and Dg15) were labelled with 5 nm
and 15 nm particles respectively. Data were used to test
whether or not row and column classifications are independent. Profiles of four types (Cg5+Dg15+, Cg5+Dg15–,
Cg5–Dg15+, and Cg5–Dg15–) were identified. The profiles
may be thought of as representing some unspecified subcellular organelle.
Results and worked examples
Single labelling: Golgi localization of AtMUR3–GFP
Quantitative findings for the Chevalier et al. (2010) data
set are summarised in Table 1. The expected gold particles
for each Golgi compartment are calculated from the
observed total number of gold particles (210) and total
test points falling on those compartments (201). For
example, the expected number of gold particles for the
medial-Golgi is estimated to be 773210/210¼80.45. The
RLI for this compartment is obtained by dividing observed by expected gold particles: RLImedial¼135/
80.45¼1.68. It appears that the medial-Golgi is labelled
almost twice as intensely as would be predicted for
a random distribution of gold particles. However, this
must be confirmed by the v2 values. The corresponding
partial v2 for this compartment is calculated from the
observed and expected gold counts as (135–80.45)2/
80.45¼36.99. By the same arguments, the partial v2 values
for the other Golgi compartments are shown in Table 1.
With df¼3 (given by 2–1 columns34–1 compartments), the
total v2 of 75.94 gives a probability level of P <0.001. This
4108 | Mayhew
Table 1. Localization of 10 nm gold-labelled anti-GFP in Golgi compartments of wild-type tobacco BY-2 cultured cells expressing the
glycosyltransferase AtMUR3–GFP
Compartment
Observed gold count, Go
Point count, P
Expected gold count, Ge
RLI, Go/Ge
x2 value
x2 as %
Cis-Golgi
Medial-Golgi
Trans-Golgi
TGN
Column total
39
135
31
5
210
40
77
41
43
201
41.79
80.45
42.84
44.93
210
0.93
1.68
0.72
0.11
0.19
36.99
3.27
35.49
75.94
0.25
48.71
4.31
46.73
100
For v2¼75.94 and df¼3, P <0.001 (v2 analysis). The gold labelling distribution is significantly different from random. Only the medial-Golgi
compartment (RLI¼1.68, v2¼48.71% of total) meets the two criteria for being preferentially labelled (RLI >1 and v2 value >10% of total). Labelling
on this compartment is almost twice as intense as that expected for a random distribution of gold particles. Based on data in Table 2 of Chevalier
et al. (2010).
indicates that the null hypothesis of no difference from
random labelling must be rejected. Only the medial-Golgi
meets the two criteria for preferential labelling, namely
RLI >1 and v2 value >10% of total.
Single labelling: AtXT1–GFP localization in wild-type and
transformed cells
For a given Golgi compartment in a given cell group, the
number of expected gold particles is calculated by multiplying the column sum by the corresponding row sum and then
dividing by the grand row sum. Thus, in Table 2, the
expected number of gold particles associated with the cisGolgi of transformed cells is 197398/401¼48.14. With an
observed gold count of 62, the partial v2 amounts to (62–
48.14)2/48.14¼3.99.
The total v2 value for all Golgi compartments in the two
cell groups is 33.02 and, for df¼3 (2–1 groups34–1
compartments), P <0.001. In other words, there is a difference in labelling distributions between wild-type and transformed cells. Inspection of partial v2 values reveals that the
cis-Golgi, trans-Golgi, and TGN account for the difference.
Wild-type cells have fewer gold particles than expected on
the cis- and trans-Golgi, but more than expected on the
TGN. In contrast, transformed cells display more gold
particles than expected on the cis- and trans-Golgi but fewer
on the TGN.
Single labelling: glutathione localization in Arabidopsis
and Cucurbita
In Table 3, the total v2 value for all subcellular compartments in the leaves of the two plant species is 11.76 and, for
df¼4 (2–1 groups35–1 compartments), P <0.02. The
hypothesis of no difference in labelling distributions between plant species is rejected. Inspection of partial v2
values reveals that mitochondria, cytosol, and nuclei are
responsible. Cucurbita pepo cells have fewer gold particles
than expected on mitochondria but more than expected on
cytosol and nuclei. Arabidopsis thaliana cells have more
particles than expected on cytosol and nuclei but fewer on
mitochondria.
Dual labelling: testing for compartmental co-localization
Quantitative findings for the real data (Ju et al., 2005) are
provided in Table 4. For GFP–TGBp2, 33 gold particles
were on the cell wall, 12 on the plasma membrane, 139 on
the ER, and 53 on ER-derived vesicles. For GFP–TGBp3,
the observed totals were 25, 22, 133, and 0, respectively.
Using contingency table analysis, the expected numbers
of gold particles were calculated by multiplying relevant
column and row totals and dividing by the grand total
(here, 417). Thus, the expected number of GFP–TGBp2
gold particles on the cell wall was calculated as 237358/
417¼32.96. The partial v2 value was (33–32.96)2/
32.96¼0.0000485.
Comparing the two groups, total v2 amounted to 50.33
which, for df¼3 (i.e. 2–1 columns by 4–1 rows), represents
a probability level of P <0.001. The null hypothesis must be
rejected because the two movement proteins are distributed
differently between these compartments. The partial v2
values show that it is the ER-derived vesicles which account
for the difference. This compartment has a greater than
expected number of gold particles in the GFP–TGBp2
group whereas there was no labelling identified in the
GFP–TGBp3 group.
In a synthetic dual-labelling data set, gold particles were
found on four cellular compartments (Table 5): for protein
Ag5, there were 105 particles on the cell plasma membrane,
60 on the Golgi complex, 32 on the ER, and three on the
rest of the cell. Corresponding counts for protein Bg20 were
61, 36, 15, and 1.
By contingency table analysis, total v2 amounted to 0.69
which, for df¼3 (i.e. 2–1 columns by 4–1 rows), represents
a probability level of P¼0.877. The null hypothesis is
accepted; that is, the two proteins are not distributed
differently but co-distribute between these compartments.
Dual labelling: testing for co-labelling of structural
profiles
The pertinent question here is: if a profile is found to label
for one protein of interest, does it also label for a second
protein? To test this, profiles labelled for the first protein
are classified into two groups that are either positive or
negative for the second. Negative profiles for the first
Quantitative immunoelectron microscopy | 4109
Table 2. Localization of gold-labelled anti-GFP in Golgi compartments of wild-type and transformed tobacco BY-2 cultured cells
expressing the glycosyltransferase AtXT1–GFP
Compartment
Wild-type cells
Transformed cells
Row total
x2 value
x2 as %
Cis-Golgi
Medial-Golgi
Trans-Golgi
TGN
Column total
36 (49.86)
66 (79.36)
64 (48.33)
38 (26.45)
204 (204)
62 (48.14)
90 (76.64)
31 (46.67)
14 (25.55)
197 (197)
98 (98)
156 (156)
95 (95)
52 (52)
401 (401)
3.85, 3.99
2.25, 2.33
5.08, 5.26
5.04, 5.22
33.02
11.66, 12.08
6.81, 7.06
15.38, 15.93
15.26, 15.81
100
For v2¼33.02 and df¼3, P <0.001 (contingency table analysis). The labelling distributions are significantly different between wild-type and
transformed cells. Compartmental v2 values indicate that the major contributors to the difference are the cis- and trans-Golgi and TGN compartments
(v2 values >10% of total). Wild-type cells have fewer gold particles than expected on cis- and trans-Golgi stacks but more than expected on the TGN.
Transformed cells have more particles than expected on cis- and trans-Golgi stacks but fewer than expected on the TGN.Values represent observed
(expected) numbers of 10 nm gold particles in each compartment (based on data in Table 1 of Chevalier et al., 2010).
Table 3. Localization of gold-labelled glutathione in subcellular
compartments of leaves of two plant species processed by
conventional fixation
Compartment Arabidopsis Cucurbita Row
total
x2
value
x2
as %
Mitochondria
Chloroplasts
Peroxisomes
Cytosol
Nuclei
Column total
1.78, 1.88
0.37, 0.39
0.00, 0.00
2.39, 2.51
1.19, 1.25
11.76
15.16, 15.95
3.14, 3.31
0.04, 0.04
20.30, 21.37
10.08, 10.62
100
90 (103.59)
13 (15.38)
27 (26.67)
37 (28.72)
53 (45.64)
220 (220)
112 (98.41)
17 (14.62)
25 (25.33)
19 (27.28)
36 (43.36)
209 (209)
202 (202)
30 (30)
52 (52)
56 (56)
89 (89)
429 (429)
For v2¼11.76 and df¼4, P <0.02 (contingency table analysis). The
labelling distributions are significantly different between the two
species. Partial v2 values indicate that the major contributors to the
difference are mitochondria, cytosol, and nuclei (v2 values >10% of
total). Cucurbita cells have fewer gold particles than expected on
mitochondria but more than expected on cytosol and nuclei.
Arabidopsis cells have more particles than expected on cytosol and
nuclei but fewer than expected on mitochondria.Values represent
observed (expected) numbers of 10 nm gold particles in each
compartment (based on data in Table 1 of Zechmann and Müller,
2010).
Table 4. Localization of gold-labelled anti-GFP in compartments
of transgenic Nicotiana tabacum leaf samples expressing GFP–
TGBp2 and GFP–TGBp3
Compartments
GFP–TGBp2 GFP–TGBp3 Row total x2 value
Cell wall
Plasma membrane
Endoplasmic reticulum
ER-derived vesicles
Column totals
33 (32.96)
12 (19.32)
139 (154.59)
53 (30.12)
237 (237)
25 (25.04)
22 (14.68)
133 (117.41)
0 (22.88)
180 (180)
58 (58)
34 (34)
272 (272)
53 (53)
417 (417)
0.00, 0.00
2.77, 3.65
1.57, 2.07
17.38, 22.87
50.33
For v2¼50.33 and df¼3, P <0.001 (contingency table analysis). The
null hypothesis (no difference in distribution patterns) must be rejected:
GFP–TGBp2 and GFP–TGBp3 do not co-localize across these
compartments. Compartmental v2 values indicate that ER-derived
vesicles account for the difference (v2 values >10% of total). More
GFP–TGBp2 than expected is found in ER-derived vesicles but no
GFP–TGBp3 was detected in vesicles.Values represent observed
(expected) numbers of 10 nm gold particles in each compartment
(based on data in Table II of Ju et al., 2005).
Table 5. Synthetic data set of the labelling distributions, across
selected compartments of cultured cells, of two undefined proteins
(Ag5 and Bg20) marked by conjugated gold particles of different
sizes (5 nm and 20 nm)
Compartments
Protein Ag5 Protein Bg20 Row total x2 value
Cell wall membrane
105 (106.07)
Golgi complex
60 (61.34)
Endoplasmic reticulum 32 (30.03)
Residual
3 (2.56)
Column totals
200 (200)
61 (59.93)
36 (34.66)
15 (16.97)
1 (1.44)
113 (113)
166
96
47
4
313
(166)
(96)
(47)
(4)
(313)
0.01,
0.03,
0.13,
0.08,
0.69
0.02
0.05
0.23
0.14
For v2¼0.69 and df¼3, P¼0.877 (contingency table analysis). The null
hypothesis (no difference in distribution patterns) must be accepted:
the two proteins co-localize in these compartments.Values represent
observed (expected) numbers of particles.
protein are also classified as positive or negative for the
second. In the example shown (Table 6), 78 profiles were
selected by the unbiased counting frame and, of these, 19
were double positive (Cg5+Dg15+), four were labelled for the
first protein only (Cg5+Dg15–), 14 were labelled for the
second only (Cg5–Dg15+), and 38 were double negative (Cg5–
Dg15–). Whether analysed by v2 or Fisher’s exact test, the
outcome is a probability level of P <0.001 and the null
hypothesis (protein Cg5 labelling is independent of Dg15
labelling) must be rejected. The calculated OR of 12.89 is
a further indication the two proteins tend to co-label in this
set of profiles.
Discussion
This review has illustrated the potential of a compendium of
methods for comparing single and multiple immunogold
labelling distributions across compartments within and
between experimental groups of cells. Provided that sampling is randomized at each stage of the multistage TEM
selection process, the methods are capable of providing
efficient and unbiased, or minimally biased, estimates of the
numbers of gold particles associated with compartments or
their profiles. Quantifying gold particles distributed between
volume-occupying compartments requires a sampling
scheme which randomizes the positions of encounters
4110 | Mayhew
Table 6. Synthetic data set for profiles of unspecified subcellular organelles labelled for two undefined proteins (Cg5 and Dg15) marked
by conjugated gold particles of different sizes (5 nm and 15 nm)
Protein Cg5+
Protein Cg5–
Column totals
Protein Dg15+
Protein Dg15–
Row totals
Ratios Dg15+/Dg15–
19
14
33
4
38
42
23
52
75
4.75
0.368
OR¼12.89
Given P <0.001 (Fisher’s exact test), the null hypothesis that protein Cg5 labelling is independent of protein Dg15 labelling must be rejected. The
OR of 12.89 is due to a high incidence of double positives coupled with a high incidence of double negatives. Overall, the findings indicate colabelling.Values represent observed numbers of organelle profiles.
between section planes and compartments. For particles
associated with membrane surfaces or tubule/filament
lengths, sample orientations must also be randomized. The
latter also stands when quantifying mixtures of volume-,
surface-, or length-occupying compartments. Efficient SUR
sampling schemes for meeting these conditions have been
presented elsewhere (Baddeley et al., 1986; Mattfeldt et al.,
1990; Nyengaard and Gundersen, 1992; Howard and Reed,
2005; Mayhew, 2008).
The methods for quantifying single labelling are able to
cater for different categories of compartment and not just
those belonging to a single category. This is straightforward
in the case of between-group comparisons, but testing for
preferential labelling of compartments within a given cell
requires extra effort in order to deal with mixtures of
volume- and surface-occupying compartments (Mayhew and
Lucocq, 2008a). With between-group comparisons, it is also
permitted to alter specimen magnification between study
groups provided that this does not compromise the ability to
recognize or resolve gold particles and compartments. Gold
particles <2 nm in size are very difficult to visualize by TEM
without resorting to additional methods such as silver
enhancement. However, the latter is likely to compromise
the ability to resolve particles and, thereby, the ability to
count them unbiasedly and ascribe them to discrete compartments. When multiple labelling is undertaken, it is important
that the different sizes of gold particle can be distinguished.
Usually, this can be achieved when the large particle is at
least 60% larger in diameter than the smaller particle
(Griffiths, 1993). In practice, particles in the range 5–15 nm
in diameter are most often used and this helps minimize the
additional potential problem of steric interaction effects
between larger particles and secondary reagents.
Comparing gold labelling patterns between compartments within a group requires some measure of labelling
intensity (either LD or RLI; Mayhew et al., 2002, 2003)
which is related to the concentration of label within
compartments and may facilitate mechanistic interpretations of biological outcomes. Between-group comparisons
of observed gold counts do not, in general, help to interpret
the mechanisms which underlie shifts in labelling patterns.
For instance, a shift of membrane labelling from one group
of cells to another might be due to changes in membrane
labelling intensities (reflecting differences in the total
number of protein molecules) or amounts of membrane
(due to changes in total membrane surface area).
In the study by Chevalier et al. (2010), GFP-fused
glycosyltransferases within Golgi stacks of tobacco BY-2
suspension-cultured cells were labelled using gold particles
conjugated to anti-GFP antibodies. It was found that all
Golgi cisternae were implicated in xyloglucan assembly. All
fusion enzymes were localized in the Golgi complex, with
AtMUR3–GFP being associated mainly with medial cisternae and AtXT1–GFP with both cis- and medial-Golgi
compartments. The present study has confirmed and
strengthened the former finding by showing that medial
cisternae had a relative labelling index almost twice as great
as that expected for a random distribution of gold particles.
Contingency table analysis has confirmed further that
labelling distributions for AtXT1–GFP were significantly
different between wild-type and transformed cells. The
major contributors to the shift in distribution were the cisand trans-Golgi and TGN compartments. Wild-type cells
had more gold particles than expected on the TGN, whereas
transformed cells had more than expected on cis- and transGolgi stacks.
Originally, Zechmann and Müller (2010) examined the
distributions of glutathione in three species of plants
(A. thaliana, C. pepo, and N. tabacum). They found that
subcellular distributions were similar in these species regardless of whether tissues were subjected to aldehyde or
microwave fixation. In leaves, mitochondria labelled most
intensely, followed by nuclei, cytosol, and peroxisomes.
Similar patterns of labelling were seen in roots. In the
present example, only data from A. thaliana and C. pepo
were included, and significant species differences were
detected. Cells from the former species showed more
particles than expected on cytosol and nuclei but fewer on
mitochondria. Cells from C. pepo had fewer gold particles
than expected on mitochondria but more on the cytosol and
nuclei. This finding suggests that species differences might
exist despite the apparent similarities in labelling density
distributions between the three species noted by the original
authors (Zechmann and Müller, 2010). In fact, contingency
table analysis of all three species (not presented here) also
failed to detect significant differences in the subcellular
distribution patterns of glutathione. Further studies would
be required to resolve this issue.
Apart from the above examples, the RLI has been used to
study the distributions of glutamine synthetase isoforms
(GS1 and GS2) in the glume and flag leaf of winter wheat,
Triticum aestivum (Lopes et al., 2006). The enzymes help
Quantitative immunoelectron microscopy | 4111
catalyse the assimilation of ammonium into glutamine and
glutamate, and have been used as markers of the status and
efficiency of nitrogen usage in different wheat strains. The
authors found that distribution patterns differ between
glume and leaf, partly reflecting differences in the proportions of mesophyll cells. Labelling was associated mainly
with the chloroplasts and cytosol of mesophyll and phloem
cells. In glumes, total labelling was similar in chloroplasts
and cytosol, whereas chloroplast labelling was greater than
that of cytosol in flag leaf.
Ruel et al. (2009) applied the RLI to compare the
patterns of localization of lignin units (condensed or noncondensed) in floral stems of control and knockout
A. thaliana. The mutant plants were knockouts for cinnamoyl-CoA reductase, the primary enzyme specific to the
monolignol pathway. Using polyclonal antibodies against
a range of lignin structures, and gold particles conjugated to
protein A, their findings on immunogold distribution
patterns showed that interfascicular fibres and xylem
vascular bundles were affected differentially by mutation
and suggested that non-condensing lignins are important
for assembling secondary cell walls.
The multiple-labelling methods are the most recently
developed (Mayhew and Lucocq, 2011). Comparing the
distributions of different sizes of gold particles across
compartments is relatively efficient and can be applied to
volume-, surface-, or length-occupying compartments, or
mixtures thereof. It is also suitable for studies in which
more than two molecules are being targeted by gold
conjugates. Furthermore, comparisons can be drawn when
multiple labelling is applied to the same ultrathin sections or
when different antigens are being localized on independent
sets of sections. However, the former scenario is likely to be
more economical in terms of precision per unit cost.
Comparing labelling of compartments by different sizes
of gold particle is also likely to be easier and more flexible
to apply than comparing labelling of individual profiles.
Partly, this reflects the nature of the calculations for
Fisher’s exact test and the fact that, at present, the profile
co-labelling method caters for just two sizes of gold particle.
Future refinements will be required to deal with data
gleaned from triple-labelling studies. In an earlier report
(Mayhew and Lucocq, 2011), it was suggested that sampling
precision and profile recognition could be improved by
biasing towards equatorially sectioned profiles. However, it
should be noted that this could introduce different relative
biases if labelled proteins display different distributions
within the parent organelles. The biases are likely to be less
when the target molecules of profiles are membrane
associated and further reduced when the membranes belong
to large structures (e.g. the plasma membrane of the cell or
rough ER membranes) rather than small structures (e.g.
small vesicles associated with the Golgi complex or ER).
Sampling membranes may also require procedures which
differ from those employed for profile sampling (Mayhew
and Lucocq, 2011).
GFP–TGBp2 and GFP–TGBp3 in subcellular compartments within tobacco leaf samples did not co-localize. In
the original study from which modified data were extracted,
Ju et al. (2005) showed that label was associated with the
ER and that only GFP–TGBp2 was associated with ERderived vesicles. Neither protein was associated with the
Golgi complex. Omitting the Golgi complex, the present
contingency table analysis showed that ER-derived vesicles
contributed most to the difference in compartment labelling. More than expected GFP–TGBp2 was found in ERderived vesicles but no GFP–TGBp3 was detected in
vesicles. The explanation is that the ER-derived vesicles are
induced by TGBp2 and are not present in GFP–TGBp3
transgenic cells (Ju et al., 2005).
Recently, Lucocq and Gawden-Bone (2009) have quantified relative and absolute numbers of gold particles which
lie on compartmental profiles situated at defined positions
with respect to a fixed axis and organelle within the cell.
The appropriate stereological sampling tool is the rotator,
which can be used to estimate total numbers of structures
within cells (Nyengaard and Gundersen, 2006; Lucocq and
Gawden-Bone, 2009). This application of the rotator opens
up the possibility to map compartments in polarized or
dividing cells, or in investigations of protein translocation
within cells.
A potential limitation of the present methods is that they
do not distinguish between gold particles which arise from
specific rather than non-specific labelling. An effective way
of controlling for specificity is to alter the expression or
location of the antigen in some way [e.g. using gene
deletion, small interfering RNA (siRNA), microinjection,
or chemical modification]. Recent developments by Lucocq
and Gawden-Bone (2010) permit quantitative assessment of
the probability that observed gold particles are target
specific. By doing so, it is possible to assign specific labelling
densities to compartments. These authors have shown that
specimen-based controls (analysing wild-type specimens
with normal expression and knockout or knockdown specimens with reduced expression) offer a useful way of
correcting labelling densities for the probability that gold
particles label the target. Corrections for labelling specificity
could be applied to the present methods.
In conclusion, these recently developed methods for
quantitative immunogold TEM (Mayhew and Lucocq,
2008b, 2011; Lucocq and Gawden-Bone, 2009, 2010) offer
high precision, minimal or no bias, and greater rigour of
statistical analysis. Their practical utility for analysing data
from single- and multiple-labelling experiments has been
illustrated here by using real and synthetic data. It is hoped
that these simple but effective methods will contribute to
future studies involving quantitative immunogold labelling
of plant cells and tissues. While some of the methods have
been used recently in immunolocalization studies on plant
cells (Lopes et al., 2006; Ruel et al., 2009; Chevalier et al.,
2010), they have been applied successfully in a wide variety
of other contexts and from subcellular to higher levels of
structural organization (Mayhew and Lucocq, 2008b;
Mayhew et al., 2009). They also allow application to other
types of nanoparticle including quantum dots which have
been used in multiple-labelling studies involving correlative
4112 | Mayhew
optical and electron microscopy (Giepmans et al., 2005;
Mayhew et al., 2009).
Acknowledgements
I am grateful to Dr John Lucocq (Dundee, UK) and
Professor Gareth Griffiths (Oslo, Norway), fellow teachers
on the EMBO- and FEBS-sponsored Practical Courses in
Electron Microscopy and Stereology in Cell Biology.
Without their friendship and collaboration, the set of
immunoEM methods would not have been developed.
References
Anderson JB, Carol AA, Brown VK, Anderson LE. 2003.
A quantitative method for assessing co-localization in immunolabelled
thin section electron micrographs. Journal of Structural Biology
143, 95–106.
Baddeley AJ, Gundersen HJG, Cruz-Orive LM. 1986. Estimation
of surface area from vertical sections. Journal of Microscopy
142, 259–276.
Bendayan M. 1982. Double immunocytochemical labelling applying
the protein A gold technique. Journal of Histochemistry and
Cytochemistry 30, 81–85.
Bendayan M. 2001. Worth it’s weight in gold. Science 291,
1363–1365.
Bernal M, Testillano PS, Alfonso M, del Carmen Risueño M,
Picorel R, Yruela I. 2007. Identification and subcellular localization of
the soybean copper P1B-ATPase GmHMA8 transporter. Journal of
Structural Biology 158, 46–58.
Bilska A, Sowiński P. 2010. Closure of plasmodesmata in maize
(Zea mays) at low temperature: a new mechanism for inhibition of
photosynthesis. Annals of Botany 106, 675–686.
Chevalier L, Bernard S, Ramdani Y, Lamour R, Bardor M,
Larouge P, Follet-Gueye M-L, Driouich A. 2010. Subcompartment
localization of the side chain xyloglucan-synthesizing enzymes within
Golgi stacks of tobacco suspension-cultured cells. The Plant Journal
64, 977–989.
D’Amico F, Skarmoutsou E. 2008a. Quantifying immunogold
labelling in transmission electron microscopy. Journal of Microscopy
230, 9–15.
D’Amico F, Skarmoutsou E. 2008b. Clustering and colocalization in
transmission immunoelectron microscopy: a brief review. Micron
39, 1–6.
Fisher RA. 1922. On the interpretation of v2 from contingency tables
and the calculation of P. Journal of the Royal Statistical Society 85,
87–94.
Follet-Gueye M-L, Pagny S, Faye L, Gomord V, Driouich A. 2003.
An improved chemical fixation method suitable for immunogold
localization of green fluorescent protein in the Golgi apparatus of
tobacco bright yellow (BY-2) cells. Journal of Histochemistry and
Cytochemistry 51, 931–940.
Geuze HJ, Slot JW, van der Ley P, Scheffer RCT. 1981. Use of
colloidal gold particles in double-labelling immunoelectron microscopy
of ultrathin tissue sections. Journal of Cell Biology 89, 653–659.
Giepmans BN, Deerinck TJ, Smarr BL, Jones YZ, Ellisman MH.
2005. Correlated light and electron microscopic imaging of multiple
endogenous proteins using Quantum dots. Nature Methods 2,
743–749.
Griffiths G. 1993. Fine structure immunocytochemistry. Heidelberg:
Springer-Verlag.
Gundersen HJG. 1977. Notes on the estimation of the numerical
density of arbitrary profiles: the edge effect. Journal of Microscopy
111, 219–223.
Gundersen HJG, Jensen EB. 1987. The efficiency of systematic
sampling in stereology and its prediction. Journal of Microscopy
147, 229–263.
Gundersen HJG, Jensen EBV, Kiêu K, Nielsen J. 1999. The
efficiency of systematic sampling in stereology—reconsidered. Journal
of Microscopy 193, 199–211.
Hagiwara H, Aoki T, Suzuki T, Takata K. 2010. Double-label
immunoelectron microscopy for studying the colocalization of proteins
in cultured cells. Methods in Molecular Biology 657, 249–257.
Howard CV, Reed MG. 2005. Unbiased stereology. Threedimensional measurement in microscopy.. Oxford: Garland Science/
BIOS Scientific.
Christopher DA, Borsics T, Yuen CYL, Ullmer W, AndèmeOndzighi C, Andres MA, Kang B-H, Staehelin LA. 2007. The
cyclic nucleotide gated cation channel AtCNGC10 traffics from the ER
via Golgi vesicles to the plasma membrane of Arabidopsis root and
leaf cells. BMC Plant Biology 7, 48.
Ju H-J, Samuels TD, Wang Y-S, Blancaflor E, Payton M,
Mitra R, Krishnamurthy K, Nelson RS, Verchot-Lubicz J. 2005.
The potato virus X TGBp2 movement protein associates with
endoplasmic reticulum-derived vesicles during virus infection. Plant
Physiology 138, 1877–1895.
Coronado M-J, Testillano PS, Wilson C, Vicente O, HeberleBors E, Risueño C- M. 2007. In situ molecular identification of the
Ntf4 MAPK expression sites in maturing and germinating pollen.
Biology of the Cell 99, 209–221.
Kondo A, Nose A, Ueno O. 1998. Leaf inner structure and
immunogold localization of some key enzymes involved in carbon
metabolism in CAM plants. Journal of Experimental Botany
49, 1953–1961.
Cruz-Orive LM, Weibel ER. 1981. Sampling designs for stereology.
Journal of Microscopy 122, 235–257.
Lopes MS, Cortadellas N, Kichey T, Dubois F, Habash DZ,
Araus JL. 2006. Wheat nitrogen metabolism during grain filling:
comparative roles of glumes and the flag leaf. Planta 225, 165–181.
Dafoe NJ, Gowen BE, Constabel CP. 2010. Thaumatin-like
proteins are differentially expressed and localized in phloem tissues of
hybrid poplar. BMC Plant Biology 10, 191.
Daly LE, Bourke GJ. 2000. Interpretation and uses of medical
statistics. Oxford: Blackwell Science Ltd.
Lucocq J. 2008. Quantification of structures and gold labelling in
transmission electron microscopy. Methods in Cell Biology 88, 59–82.
Lucocq JM, Gawden-Bone C. 2009. A stereological approach for
estimation of cellular immunolabeling and its spatial distribution in
Quantitative immunoelectron microscopy | 4113
oriented sections using the rotator. Journal of Histochemistry and
Cytochemistry 57, 709–719.
differential distribution and colocalization in subcellular structures.
Lucocq JM, Gawden-Bone C. 2010. Quantitative assessment of
specificity in immunoelectron microscopy. Journal of Histochemistry
and Cytochemistry 58, 917–927.
Mayhew TM, Lucocq JM, Griffiths G. 2002. Relative labelling index:
Lucocq JM, Habermann A, Watt S, Backer JM, Mayhew TM,
Griffiths G. 2004. A rapid method for assessing the distribution of
gold labeling on thin sections. Journal of Histochemistry and
Cytochemistry 52, 991–1000.
microscopy thin sections. Journal of Microscopy 205, 153–164.
Mattfeldt T, Mall G, Gharehbaghi H, Möller P. 1990. Estimation of
surface area and length with the orientator. Journal of Microscopy
159, 301–317.
Annals of Anatomy 191, 153–170.
Mayhew TM. 2006. Stereology and the placenta: where’s the point?
A review. Placenta 27, (Suppl A), S17–S25.
Mayhew TM. 2007. Quantitative immunoelectron microscopy.
Alternative ways of assessing subcellular patterns of gold labelling.
Methods in Molecular Biology 369, 309–329.
Mayhew TM. 2008. Taking tissue samples from the placenta: an
illustration of principles and strategies. Placenta 29, 1–14.
Mayhew TM, Desoye G. 2004. A simple method for comparing
immunogold distributions in two or more experimental groups
illustrated using GLUT1 labelling of isolated trophoblast cells. Placenta
25, 580–584.
Mayhew T, Griffiths G, Habermann A, Lucocq J, Emre N,
Webster P. 2003. A simpler way of comparing the labelling densities
of cellular compartments illustrated using data from VPARP and
LAMP-1 immunogold labelling experiments. Histochemistry and Cell
Biology 119, 333–341.
Mayhew T, Griffiths G, Lucocq J. 2004. Applications of an efficient
method for comparing immunogold labelling patterns in the same sets
of compartments in different groups of cells. Histochemistry and Cell
Biology 122, 171–177.
Mayhew TM, Lucocq JM. 2008a. Quantifying immunogold labelling
patterns of cellular compartments when they comprise mixtures of
membranes (surface-occupying) and organelles (volume-occupying).
Histochemistry and Cell Biology 129, 367–378.
Histochemistry and Cell Biology 135, 317–326.
a novel stereological approach to test for non-random immunogold
labelling of organelles and membranes on transmission electron
Mayhew TM, Mühlfeld C, Vanhecke D, Ochs M. 2009. A review of
recent methods for efficiently quantifying immunogold and other
nanoparticles using TEM sections through cells, tissues and organs.
Negi SS, Carol AA, Pandya S, Braun W, Anderson LE. 2008. Colocalization of glyceraldehyde-3-phosphate dehydrogenase with
ferredoxin-NADP reductase in pea leaf chloroplasts. Journal of
Structural Biology 161, 18–30.
Nyengaard JR, Gundersen HJG. 1992. The isector: a simple and
direct method for generating isotropic, uniform random sections from
small specimens. Journal of Microscopy 165, 427–431.
Nyengaard JR, Gundersen HJG. 2006. Direct and efficient
stereological estimation of total cell quantities using electron
microscopy. Journal of Microscopy 222, 182–187.
Petrie A, Sabin C. 2000. Medical statistics at a glance. Oxford:
Blackwell Science Ltd.
Ruel K, Berrio-Sierra J, Derikvand MM, Pollet B, Thevenin J,
Lapierre C, Jouanin L, Joseleau J- P. 2009. Impact of CCR1
silencing on the assembly of lignified secondary walls in Arabidopsis
thaliana. New Phytologist 184, 99–113.
Seguı́-Simarro JM, Testillano PS, Risueño MC. 2003. Hsp70 and
Hsp90 change their expression and subcellular localization after
microspore embryogenesis induction in Brassica napus L. Journal of
Structural Biology 142, 379–391.
Wei L, Zhang W, Liu Z, Li Y. 2009. AtKinesin-13A is located on
Golgi-associated vesicle and involved in vesicle formation/budding in
Arabidopsis root-cap peripheral cells. BMC Plant Biology 9, 138.
Zechmann B, Müller M. 2010. Subcellular compartmentation of
Mayhew TM, Lucocq JM. 2008b. Developments in cell biology for
quantitative immunoelectron microscopy based on thin sections—a
review. Histochemistry and Cell Biology 130, 299–313.
glutathione in dicotyledonous plants. Protoplasma 246, 15–24.
Mayhew TM, Lucocq JM. 2011. Multiple-labelling immunoEM using
different sizes of colloidal gold: alternative approaches to test for
associates with plasma membrane and triggers hypersensitive cell
Zhou L, Cheung M-Y, Li M-W, Fu Y, Sun Z, Sun S-M, Lam H- M.
2010. Rice hypersensitive induced reaction protein 1 (OsHIR1)
death. BMC Plant Biology 10, 290.