Detecting fluorescent protein expression and co-localisation on single secretory vesicles with linear spectral unmixing.

Abstract

Many questions in cell biology and biophysics involve the quantitation of co-localisation and the interaction of proteins tagged with different fluorophores. However, the incomplete separation of the different colour channels due to the presence of autofluorescence, along with cross-excitation and emission "bleed-through" of one colour channel into the other, all combine to render the interpretation of multi-band images ambiguous. Here we introduce a new live-cell epifluorescence spectral imaging and linear unmixing technique for classifying resolution-limited point objects containing multiple fluorophores. We demonstrate the performance of our technique by detecting, at the single-vesicle level, the co-expression of the vesicle-associated membrane protein, VAMP-2 (also called synaptobrevin-2), linked to either enhanced green fluorescent protein (EGFP) or citrine [a less pH-sensitive variant of enhanced yellow fluorescent protein (EYFP)], in mouse cortical astrocytes. In contrast, the co-expression of VAMP-2-citrine and the lysosomal transporter sialine fused to EGFP resulted in little overlap. Spectral imaging and linear unmixing permit us to fingerprint the expression of spectrally overlapping fluorescent proteins on single secretory organelles in the presence of a spectrally broad autofluorescence. Our technique provides a robust alternative to error-prone dual- or triple colour co-localisation studies.

Full text

Page 1

BIOPHYSICS LETTER
Fabien Nadrigny Æ Æ Isabelle Rivals Æ Æ Petra G. Hirrlinger
Annette Koulakoff Æ Æ Le ´ on Personnaz
Marine Vernet Æ Æ Myriam Allioux Æ Æ Myriam Chaumeil
Nicole Ropert Æ Æ Christian Giaume Æ Æ Frank Kirchhoff
Martin Oheim
Detecting fluorescent protein expression and co-localisation on single
secretory vesicles with linear spectral unmixing
Received: 18 August 2005/ Accepted: 7 December 2005
? EBSA 2006
Abstract Many questions in cell biology and biophysics
involve the quantitation of co-localisation and the
interaction of proteins tagged with different fluoro-
phores. However, the incomplete separation of the dif-
ferentcolourchannelsdue
autofluorescence, along with cross-excitation and emis-
sion ‘‘bleed-through’’ of one colour channel into the
other, all combine to render the interpretation of multi-
band images ambiguous. Here we introduce a new
tothe presenceof
live-cell epifluorescence spectral imaging and linear
unmixing technique for classifying resolution-limited
point objects containing multiple fluorophores. We
demonstrate the performance of our technique by
detecting, at the single-vesicle level, the co-expression of
the vesicle-associated membrane protein, VAMP-2 (also
called synaptobrevin-2), linked to either enhanced green
fluorescent protein (EGFP) or citrine (a less pH-sensitive
variant of enhanced yellow fluorescent protein (EYFP)),
in mouse cortical astrocytes. In contrast, the co-expres-
sion of VAMP-2-citrine and the lysosomal transporter
sialine fused to EGFP resulted in little overlap. Spectral
imaging and linear unmixing permit us to fingerprint the
expression of spectrally overlapping fluorescent proteins
on single secretory organelles in the presence of a spec-
trally broad autofluorescence. Our technique provides a
robust alternative to error-prone dual- or triple colour
co-localisation studies.
Keywords Spectral imaging Æ Linear unmixing Æ
Fluorescence microscopy Æ Total internal reflection Æ
Exocytosis Æ Protein expression Æ Co-localisation
Abbreviations CMV: Cytomegalovirus Æ
CCCP: Carbonyl cyanide m-chlorophenylhydrazone Æ
EYFP: Enhanced yellow fluorescent protein Æ
FRET: Fluorescence resonance energy transfer Æ
EDTA: Ethylenediaminetetraacetic acid Æ
DMEM: Dulbecco’s modified Eagle’s medium Æ
EF: Evanescent field Æ EGFP: Enhanced green
fluorescent protein Æ FCS: Fetal calf serum Æ
FP: Fluorescent protein Æ FWHM: Full-width half
maximum Æ GFAP: Glial fibrillary acidic protein Æ
P0–1: Postnatal days 0–1 Æ PBS: Phosphate-buffered
saline Æ PS-CFP2: Photoswitchable cyan FP Æ
SD: Standard deviation Æ ci: Confidence interval Æ
SILU: Spectral imaging and linear unmixing Æ
TIRF: Total internal reflection fluorescence Æ
VAMP: Vesicle-associated membrane protein
Electronic Supplementary Material Supplementary material is
available for this article at http://dx.doi.org/10.1007/s00249-005-
0040-8 and is accessible for authorized users.
F. Nadrigny Æ M. Vernet Æ M. Allioux Æ M. Chaumeil
N. Ropert Æ M. Oheim (&)
Molecular and Cellular Biophysics of Synaptic Transmission,
Laboratory of Neurophysiology and New Microscopies,
INSERM U603, CNRS FRE 2500, Universite ´ Rene ´ Descartes
(Paris 5), 45 rue des Saints Pe ` res, 75 006, Paris, France
E-mail: fabien.nadrigny@univ-paris5.fr
E-mail: nicole.ropert@univ-paris5.fr
E-mail: martin.oheim@univ-paris5.fr
Tel.: +33-1-42864221
Fax: +33-1-42864151
I. Rivals Æ L. Personnaz
Applied Statistics Group, Ecole Supe ´ rieure de Physique et Chimie
Industrielles (ESPCI), 10, rue Vauquelin, 75 005, Paris, France
E-mail: isabelle.rivals@espci.fr
E-mail: leon.personnaz@espci.fr
P. G. Hirrlinger Æ F. Kirchhoff
Department of Neurogenetics, Max-Planck Institute
for Experimental Medicine, Hermann-Rein Strasse 3,
37 075, Go ¨ ttingen, Germany
E-mail: petra.hirrlinger@em.mpg.de
E-mail: kirchhoff@em.mpg.de
A. Koulakoff Æ C. Giaume
Laboratoire deNeuropharmacologie INSERM U587,
Colle ` ge de France, 11, place Marcelin Berthelot,
75 005, Paris, France
E-mail: annette.koulakoff@college-de-france.fr
E-mail: christian.giaume@college-de-france.fr
Eur Biophys J (2006)
DOI 10.1007/s00249-005-0040-8

Page 2

Introduction
Classification and feature extraction methods based on
multi- and hyperspectral imaging detectors are routine
in remote sensing and satellite imaging (Chang 2003;
Nielsen 2001). Applied to fluorescence microscopy
(Dickinson et al. 2001; Hiraoka et al. 2002; Neher and
Neher 2004a; Schultz et al. 2001; Shirakawa and Miya-
zaki 2004; Zimmermann et al. 2003, 2005), spectral
imaging and linear unmixing (SILU) improves fluores-
cence resonance energy transfer (FRET) detection (Ec-
ker et al. 2004; Gu et al. 2004; Neher and Neher 2004b)
as well as the discrimination of fluorophores with spec-
tral overlap that otherwise would not be resolved by
dual- or triple band recordings (Lansford et al. 2001;
Shirakawa and Miyazaki 2004; Tsurui et al. 2000).
The purpose of this study was to evaluate the per-
formance of SILU for profiling the expression and co-
localisation of fluorophores on resolution-limited point
objects, which only emit a limited number of photons
before photobleaching. This situation is typical for an
increasing number of multi-colour cell biological,
bioanalytical and biophysical applications that detect
single-molecule FRET (Ishii et al. 1999), concentration
microdomains (Demuro and Parker 2004), semiconduc-
tor nanocrystals (Grecco et al. 2004; Michalet et al.
2001), spectrally karyotype chromosomes (Garini et al.
1999) or image doubly labelled organelles in live cells
(Ellenberg and Lippincott-Schwartz 1999; Stephens et al.
2000). A similar situation is encountered in double and
multi-labelling experiments based on spectral variants of
the green fluorescent protein (GFP) (Ellenberg and
Lippincott-Schwartz 1999). Thus, although a wider col-
our range of fluorescent probes is becoming available, the
choice of spectrally well-separated variants is still very
restricted. For example, a recent study (Hirrlinger et al.
2005) demonstrated that the formation of fluorescent
precipitates limits the use of red-emitting reef coral pro-
teins in transgenic animals. Those FP that workbest have
considerable spectral overlap and cannot be separated
using specific filter sets (Zimmermann 2005). Also, even if
we were able to image two spectrally well-separated FP
proteins, protein expression would still be obscured in
many preparations by the presence of autofluorescence.
Hence, although in a typical experiment the number of
fluorophores is often limited to two or three, these may
be strongly overlapping and therefore require unmixing.
Diffraction-limited imaging of point objects would
benefit from our ability to record the spectral signature.
An additional requirement when imaging, e.g. small
organelles is to make efficient use of the few photons
emitted before irreversible photobleaching. Therefore,
rather than dividing the already low signal into several
discrete non-overlapping colour channels, we optimised
the fluorescence-collected fraction per excitation photon
by ‘‘oversampling’’ (Garini et al. 1999) with few and
wide channels with partially overlapping bands in
sequential acquisitions (Zimmermann 2005).
Spectral unmixing is a pixel-based technique that
quantifies the presence of several fluorophores in mixed
pixels. It assigns a mixed pixel the fluorescence the pixel
would have had if it consisted of a pure spectral com-
ponent only. Due to their simplicity, linear models have
gained significant popularity over mixing models with
higher order moments (Bosdogianni and Petrou 1997)
and are now available with commercial microscope
software. The spectrum of each pixel is considered a
linear superposition of the pure spectra present in that
pixel, weighted by their relative fractional abundance,
which is calculated by a least-squares estimation.
However, adjacent pixels on diffraction-limited ima-
ges contain spatially and spectrally correlated informa-
tion and cannot be considered independent. Further,
due to the detector pixel size diffraction-limited point
objects may only contain a handful of pixels so that the
calculation of statistical descriptors may not be partic-
ularly reliable (Bosdogianni et al. 1997). We therefore
investigated how the microscopic spatial resolution af-
fects our ability to detect protein expression on single
near-membrane astroglial vesicles identified on evanes-
cent-field (EF) images. Robust statistics was ascertained,
in a first step, by searching for and rejecting pixel out-
liers and then binning correlated pixels of Airy disk-
sized regions centred on the single-vesicle images to
extract noise-reduced organelle spectra. A reference
spectral library contained average experiment spectra of
the pure spectral components measured in live cells after
cytoplasmic transfection. We verified that these spectra
were not altered by the lower pH in secretory organelles
(Kneen et al. 1998). The performance of SILU for sin-
gle-vesicle classification was investigated in two realistic
scenarios: (1) the co-localisation on the same organelle
of a green and yellow FP and (2) their mutual exclusion
by targeting citrine and EGFP to secretory vesicles and
lysosomes, respectively. We also derived a statistical
expression for quantifying the precision of the SILU-
based fluorophore abundance estimates and investigated
the relationship between the obtained precision and the
signal-to-background ratio of the spectral images.
At the single-vesicle level, SILU permits us to fin-
gerprint the expression and co-localisation of spectrally
overlapping fluorescent proteins on individual near-
membrane organelles.
Theory and qualitative concept
In this work, a ‘‘mixed pixel’’ is defined as the two-
dimensional (2D) representation on the imaging detector
of the corresponding 3D sample volume (‘‘voxel’’) that
contains one or several fluorophores with non-zero
probability. The emission of a mixed pixel is alternately
viewed through five band-pass filters upon single-wave-
length epifluorescence excitation. In a mixed pixel, some
non-zero fraction of the collected signal arises from
endogenous fluorophores (cellular autofluorescence) and
from the exogenous fluorescent labels, each of which

Page 3

imprint their spectral signature on the recorded emission
spectrum. We consider here the example of mouse cor-
tical astrocytes, cells known to express appreciable levels
of autofluorescence (Anlauf and Derouiche 2005;
Schipper et al. 2002), labelled with a mixture of EGFP
and citrine (Griesbeck et al. 2001). Due to bleed-through
and spectral overlap each channel does not represent a
single species. However, the five-point spectrum is un-
ique for each of the pure species (‘‘endmembers’’ in the
remote sensing terminology), i.e. any mixed pixel is a
linear combination of the pure spectral components. If
these spectral unit vectors are experimentally predeter-
mined in three specially prepared samples in which only
one pure species is present (or dominating), then the
contribution of EGFP, citrine and autofluorescence
present in any unknown pixel can be calculated based on
the intensities they produce in the spectral bands.
In our experimental system, there are three unknown
contributions: autofluorescence, EGFP and citrine.
Therefore, at least three independent measurements are
needed to find a unique solution for all three contribu-
tions. To be less sensitive to image noise and to better
distinguish yellow-green fluorescent probes, we slightly
oversampled the yellow-green part of the spectrum by
detecting three yellow-green spectral bands and one or-
ange and red band, respectively. Although all the nec-
essary information is contained in these coarse spectra,
three important practical questions are raised: (1) How
much variability in the fluorophore spectra can be tol-
erated to reliably distinguish spectrally close fluoro-
phores? (2) How sensitive are the results to spectral
variability (e.g. due to multiple molecular fluorophores
that contribute to the compound autofluorescence
spectrum) or to pH-dependent changes of the fluoro-
phore spectra? and (3) How robust is the SILU tech-
nique for analysing very dim point objects? These
questions are addressed here theoretically and experi-
mentally.
Materials and methods
Cells, labelling and solutions
Primary astrocyte cultures were prepared as described
(Rouach et al. 2003). Briefly, cortices from newborn
(P0–1) OF-1 mice were dissected in PBS/glucose,
meninges peeled off and the tissue mechanically disso-
ciated. Cells were seeded in tissue culture flasks supple-
mented with 10% FCS, 10 U/ml penicillin and 10 lg/ml
streptomycin. After reaching confluence they were har-
vested with trypsin–EDTA, plated as secondary cultures
on 25 mm coverslips (Superior, Marienfeld#1, Lauda-
Ko ¨ nigshofen, Germany) and used 6–11 days after
preparation. We identified astrocytes by their morphol-
ogy, immunocytochemical staining with antibodies
against glial fibrillary acidic protein (>96% were
GFAP-positive) and by the presence of mechanically
evoked spreading [Ca2+] waves (not shown). For
recordings, coverslips were transferred to a holder on the
microscope stage and perfused at 1–2 ml/min with
extracellular solution containing, in mM: 140 NaCl, 5.5
KCl, 1.8 CaCl2, 1 MgCl2, 20 glucose, 10 HEPES
(pH 7.3). All experiments were carried out at 20–22?C.
Plasmids and transfections
To highlight astroglial secretory and lysosomal com-
partments, respectively, we transfected ?105cells per
coverslip with 2 lg/ml synaptobrevin-2-citrine fusion
protein (Griesbeck et al. 2001) under the control of the
cytomegalovirus (CMV-)promoter (Fig. 1a, pCMV-sb2-
citrine-N1, Wojcik et al. 2004, a gift from Dr. Sonja
Wojcik, Go ¨ ttingen, Germany), pCMV-sb2-EGFP-N1
(a gift from Dr. Jean-Pierre Mothet, Gif-sur-Yvette,
France) or pCMV-EGFP-sialine-N1 (Morin et al. 2004)
(Fig. 1b, a gift from Dr. Bruno Gasnier, Paris, France)
using lipofectamin 2000 (Invitrogen, Carlsbad, CA,
USA) and standard protocols. For double transfections,
we maintained the same amount of DNA, which we
equally divided between pCMV-EGFP-sialine-N1 and
pCMV-sb2-citrine-N1. Transfected cells were incubated
overnight and imaged on the following day. Control
transfections with cytoplasmically expressed fluorescent
proteins only (e.g. pEGFP-N1, Clontech, BD Bio-
sciences, Franklin Lakes, NJ, USA) ascertained the
specific targeting of protein constructs (Fig. 1a, b,
insets).
Evaluating the effect of pH changes on experimental
reference spectra
To evaluate the effect of fluorescent protein exposure to
cytoplasmic (7.3) versus vesicular pH (5–5.5), we re-
corded EGFP and citrine reference spectra in control
conditions and in acidified cells using a protocol similar
to Kneen et al. (1998). Cortical astrocytes were trans-
fected with pEGFP-N1 and pCitrine-N1 as above. After
initial incubation in standard extracellular saline, the
medium was replaced by a series of calibration solutions
containing the same salt concentrations and additionally
the protonophore CCCP (20 lM) and K+/H+ex-
changer nigericin(10 lM).
sequentially adjusted to 7.3 (control), 5.4, 5.2, 5.0, 4.8
and 4.3. We monitored the evolution of 488/535 nm
fluorescence every 3 min. Equilibration was reached
after 10 min and the spectrum for each pH value was
recorded thereafter from small cytoplasmic regions. To
ascertain that the variation in fluorescence intensity and
spectrum was indeed due to pH changes, we confirmed
the fluorescence recovery by reversal to pH 7.3 (data not
shown). Although citrine is less pH-sensitive than EYFP
a slight spectral shift was observed (see Fig. 3b) below
pH 5.0, which is not believed to interfere with our
SILU imaging of secretory vesicles (pH 5–5.5), see
Discussion.
ThepHvalueswere

Page 4

Multispectral single secretory organelle imaging
We imaged cells on a custom-inverted microscope
equipped for EF and epifluorescence excitation (see
Schapper et al. 2003 for details). EF excitation was
achieved by focusing the attenuated spectrally and spa-
tially filtered 488-nm beam of a multi-line argon laser
(Reliant 150, Laser Physics, West Jordan, UT, USA)
into the back focal plane (BFP) of a ·60/NA-1.45 oil-
immersionobjective(PlanApochromat,
Hamburg, Germany). The location of the focal spot in
the BFP determined the angle h under which the colli-
mated beam impinged on the surface of the microscope
slide. For h‡hc=arcsin(n2/n1) the 488 nm beam suffers
total internal reflection (TIR). n2=1.35 and n1=1.518
are the refractive indices of the cell and the substrate,
respectively, so that hc=62.8?. We used a ?200 nm
penetration depth of the evanescent field, corresponding
to h=66.2?±0.1?. A Polychrome II (TILL, Gra ¨ felfing,
Germany) provided narrowband (k±8 nm) polychro-
matic epifluorescence excitation.
We used a 500DCLP dichroic mirror (Chroma,
Brattleboro, VT, USA) and discrete emission filters
housed in a motorised and computer-controlled filter
wheel [HQ510/20 (solid line on Fig. 2b, left panel),
HQ535/50 (dashed), HQ560/40 (line-dashed), HQ615/45
(grey, solid), HQ670/40 (grey, dashed)] to acquire epi-
fluorescence spectral image cubes. Gain and exposure
times were identical for all planes. Over dividing the
fluorescence emission into multiple colour channels, e.g.
with a commercial 4-channel emission beam splitter, the
Olympus,
Fig. 1 a
astroglial vesicular compartments. Raw data 514/560 nm evanes-
cent-field (EF) excited fluorescence image of a synaptobrevin (sb2)-
2-citrine transfected astrocyte and cytoplasmic control transfection
with citrine alone (inset). Note the partial targeting of sb2 to the
plasma membrane, seen as a fluorescent haze upon EF excitation. b
Raw data 488/535 nm EF image of a EGFP-sialine-transfected
astrocyte and control (inset). Scale bars are 10 lm for a and b. c
Examples of 488/535 nm in focus xy epifluorescence image and xz-
Evanescent-field imagingofnear-diffraction-limited
and yz-projection images of a 93 nm fluorescent bead (Invitrogen)
and normalised fluorescence intensity line profiles. Pixel size
188 nm, distance between planes 158 nm. Contrast is inverted for
clarity. The experimental point-spread function had a lateral and
axial FWHM of 308±1 nm (red) and 2.19±0.16 lm (black),
respectively. We calculated 342±1 nm (n=27 beads) as the radius
of the Airy disk from the first minimum of a Bessel function with
the same FWHM. Compound magnification was ·120, numerical
aperture (NA) 1.45
60040020010002500 200 1000 40-40200100
c
EGFPcitrine
AF
0.5
0.0
Normalized F
600500
l (nm)
400
b
0.4
0.2
0.0
Normalized F
650600550
l (nm)
0.4
0.2
0.0
Normalized F
650 600 550
l (nm)
EGFP
citrine
AF
a
535 560615 670 nm510
Fig. 2 a Total internal reflection fluorescence (TIRF) image of a
sb-2-EGFP and sb2-citrine co-labelled astroglial vesicle (·120, NA-
1.45) and spectral 510-535-560-615-670 nm epifluorescence image
cube. Scale bar is 1 lm. b Left Filter spectra. Middle 81 mixed
pixels (grey traces) contain the composite spectrum of the
fluorophores present in the corresponding sample voxel. Binning
the nine organelle pixels (black traces) to an Airy disk-sized region
(red on a) reduces the pixel-to-pixel spectral variability and yields a
noise-reduced spectrum (red trace). Pure FP spectra (right) were
measured in the cytoplasm of cells transfected with EGFP or -
citrine alone, and, for autofluorescence (AF), in unlabelled cells,
respectively. c Unmixed reconstituted fluorophore abundance maps
showing the participation to the vesicle shown on a of EGFP,
citrine and AF and their 70% confidence interval of the estimate,
respectively. Error images are almost alike due to equal-energy
normalisation (see Materials and methods). Note the different
intensity scales for the reconstituted images. (A colour version of
this figure is available online)

Page 5

sequential acquisition maximises the number of useful
signal photons per excitation photon, albeit at lower
time resolution. In as much as we did not continuously
acquire spectral image cubes but rather took a spectral
‘‘snapshot’’ to classify the viewed organelles at the
beginning of a single-colour time-lapse acquisition, this
approach is not detrimental to acquiring high-time res-
olution data (see Discussion). To verify that photoble-
aching did not deleteriously affect the sequentially
collected emission spectra (Neher and Neher 2004a), we
collected spectra in both ascending (510–670 nm) and
descending orders. The emission profiles detected for the
various fluorophores were identical (not shown). As a
corollary,thisobservation
robustness of the SILU technique over time.
The resulting images were projected onto a GenIII-
intensified PentaMax charge-coupled device (CCD)
camera (Roper Scientific, Trenton, NJ, USA), gener-
ously provided by Dr. Maı¨te ´ Coppey-Moissan (Institut
Jacques Monod, Paris, France). A custom telescope
matched the camera pixel size to the calculated resolu-
tion. We acquired spectral image cubes with 1 s time
resolution, which was sufficient to track moving organ-
elles visually on consecutive frames, using METAMORPH
software (Universal Imaging, Downington, PA, USA).
equallyillustratesthe
Extracting mixed organelle spectra
We subtracted from the raw spectra background spectra
taken with the same gain and exposure time in cell-free
regions close to the studied organelle. To compensate
the lateral offset resulting from different filter thickness
and organelle movement we excised from the full images
3·3 lm2regions of interest (ROIs) centred on the indi-
vidual fluorescent spot. No systematic criteria were ap-
plied concerning the signal-to-background ratio (SBR;
see, however, Fig. 6), but organelles were rejected when
their lateral full-width half maximal (FWHM) size ex-
ceeded 1.5· the Airy disk so that only near-diffraction-
limited spherical spots were conserved (Fig. 1c). For
spectral analysis, normalized fluorophore spectra to
w?ðiÞ ¼ wðiÞ
shall operate with normalised quantities only the aster-
isk is omitted. We read out nine single-pixel spectra from
the 3·3 central pixels (0.5·0.5 lm2) encompassing the
central maximum of the Airy disk. Due to the diffrac-
tion-limited resolution, these pixels contain spatially and
spectrally related information and should be realisations
of the same spectral vector only varying by additive
noise. We therefore asked whether the spectral vector
equal energy,
.PN
k¼1wkðiÞ
??
;8i: As we
wðx;yÞ ¼ w1ðx;yÞ;w2ðx;yÞ;...;wNðx;yÞ
of pixel x, y belonged to the vector bundle spanned by
the other eight vectors wðx0;y0Þjðx0;y0Þ6¼ðx;yÞby testing if
w(x, y) was an outlier. Here, N denotes the number of
channels. We tolerated two outlier pixels per organelle.
This choice relates to the SBR in the experiment. Testing
against outliers (null hypothesis) in dim images with too
low SBR will falsely discard related but noisy pixel
vectors. If a pixel (x, y) is not an outlier, then (Eq. 2)
,X
is a Fisher variable, F(P?2)N
spectrum of the eight pixels exempt (x, y), rk
1
P?1
Fisher–Snedecor distribution1and ||
vector norm. We accepted a 15% risk of falsely classi-
fying w(x, y), f85%=1.75, which is motivated by the
detectable signal change limited by the shot noise con-
tribution,
S
; to the dim single-vesicle signal S and the
amount of signal per grey value in the image (Oshiro and
ðÞT
ð1Þ
P ? 2
P
wðx;yÞ ? wðx0;y0Þ
??????
N
k¼1
r2
kðx0;y0Þð2Þ
N
. wðx0;y0Þ is the average
2(x¢, y¢)=
Px0;y0ðwkðx0;y0Þ ? wkðx0;y0ÞÞ2and F is the ‘‘F-’’ or
|| denotes the
ffiffiffip
0.4
0.3
0.2
0.0
Normalized intensity
640600560
l (nm)
520
l (nm)
CTR
7.3
5.4
5.2
5
4.8
4.3
a
c
1.0
0.8
0.5
0.3
0.0
F/Fmax
ctr 7.3
5.4
5.2
5
4.84.3
0.4
0.3
0.2
0.0
Normalized intensity
640600560520
CTR
7.3
5.4
5.2
5
4.8
4.3
b
1.0
0.8
0.5
0.3
0.0
F/Fmax
ctr 7.3
5.4
5.2
5
4.84.3
d
Fig. 3 a The spectrum of cytoplasmically expressed EGFP is pH
invariant down to pH 4.8. For citrine (b), no appreciable variation
is detected down to pH 5.2. To ascertain that the pH in the
cytoplasm actually changes despite the absence of a detectable
spectral change, we monitored the fluorescence intensity. For
EGFP (c) as for citrine (d), we observe a pronounced decrease of
the fluorescence intensity between pH 7.3 and 5.4, as expected from
their respective pKa values (5.9 and 5.7 for EGFP and citrine,
respectively). Plots show averages and SD of the fluorescence
recorded from five ROIs in three groups of cells for EGFP and
three ROIs from two groups of cells for citrine. CTR is the value
measured before permeabilisation. Note the larger relative error on
citrine and compare to EGFP traces. (A colour version of this
figure is available online)
1This result supposes that both the numerator and the denominator
follow Pearson’s law. We here assume that the noise is Gaussian.
The anisotropic fluorescent environment aroudn the spot, the
presence of readout and intensifer noise add uncorrelated noise to
each pixel so that they are realisations of events with the same
mathematical expectation value plus noise. As the N components
of the spcetral vector w(x,y) are acquired successively, they are
statistically independent and their quadratic sum in the numerator
of Eq. 2 is described by Pearson’s law. In the same way, the
denominator is Pearson-distributed.

Page 6

Moomaw 2003). We binned correlated pixels to extract a
noise-reduced spectrum,
wðiÞ ¼ w1ðiÞ;w2ðiÞ;...;wNðiÞ
(red trace on Fig. 2b, middle panel), where i is the index
of the spot.
ðÞT
ð3Þ
Linear unmixing
We used a non-constrained supervised ordinary least-
squares linear unmixing (Nielsen 2001) of noise-reduced
mixed organelle spectra. Specifically, we assumed that
(1) endmembers are predefined pure classes with 100%
fluorophore abundance. This implies, that we can mea-
sure pure fluorescent protein spectra by overexpressing
one protein so that cellular autofluorescence becomes
negligible. (2) All pure spectral components present are
known. (3) In contrast to constrained algorithms (see,
e.g. Zimmermann et al. 2003) the non-negativity of the
abundance coefficients akwas not imposed, and (4) the
abundance coefficients were not forced to sum to unity,
P
regions of the parameter space (see, e.g. Fig. 4b). Hence,
if the confidence interval of the estimation was smaller
than the separation distance from the allowed region,
l=1
mal=1. As a consequence, solutions to the unmixing
problem can be located outside physically meaningful
the analysed spot was rejected. (5) Also, we assumed
that the signal of each fluorescent spot consists of a
linear mixture of m pure spectral components, i.e. we
neglected higher orders allowing for fluorophore inter-
action, saturation and quenching effects. Normalised
spectral vectors w(i) were compared with a N · m ref-
erence spectral matrix,
X ¼
x ð ÞðÞkl¼
x11
...
xN1
???
..
???
x1m
...
xNm
.
0
@
B
1
A ¼ x1
C
???
xm
ðÞ;
ð4Þ
where l 2[1, m] varies over pure spectral classes and k
2[1, N] varies over wavelengths k. (6) We did not include
an additional constant column vector, i.e. we considered
background subtraction to completely remove any
transmitted light or amplifier offset.
Pure spectral vectors x1(Fig. 2b, right) were mea-
sured averages, for autofluorescence, from unlabelled
cells, for EGFP and citrine from autofluorescence-free
cytoplasmic regions of astrocytes transfected with
pEGFP-N1 (Clontech) or pCMV-citrine-N1, respec-
tively (Fig. 1, insets). According to our mixing model the
quantity of pure spectra in a mixed spot is a m-vector
^ aðiÞ such that
4
2
0
occurence
1.21.0 0.8
α AF (AF)
α citrine
AF
11
11
11
α
α
EGFP
b
1.0
0.5
0.0
Y
-0.50.00.5
X
EGFPEGFPcitrinecitrine
AF AF
6
4
2
0
occurence
1.041.000.96
α EGFP (EGFP)
10
5
0
occurence
1.041.000.96
α citrine (citrine)
a
c
Fig. 4 a Estimated spectral abundances define vectors in a three-
dimensional (3D) space spanned by the pure component unit
vectors. b We parameterised (see Appendix) the plane shaded in a
so as to represent the endpoints of vectors ^ aðiÞ on a 2D graph.
Green, red and grey spots indicate projected spectral abundance
vectors ofEGFP
^ a ðEGFPÞ; citrine, ^ aðcitrineÞ; and autofluores-
cence, ^ aðAFÞ; respectively. The larger spread observed for AF is
due to the bigger variation of the spectra of autofluorescent
regions. See also Table 1. c Distribution of the pure components
a¢EGFP(EGFP) (green) and a¢citrine(citrine) (red) for 25 unmixed
control regions each containing either pure EGFP or citrine, and
Gaussian fits (red), respectively. Their Gaussian distribution
justifies the used statistical tools to determine the confidence
interval of the fluorophore abundance estimate. (A colour version
of this figure is available online)

Page 7

wðiÞ ¼ X ? ^ aðiÞ þ rðiÞ:
The components of ^ aðiÞ ¼ aEGFPðiÞ;acitrineðiÞ;aAFðiÞ
are the estimated fluorophore abundances, r(i) is the
residual not explained by the linear mixing model. When
the number of detection channels is larger than the
number of spectral components, N ‡ m, ^ aðiÞ is obtained
by minimising the sum of the squared residuals rTÆr. For
ordinary least squares, ^ a is non-biased, i.e. Ef^ ag ¼ a
with dispersion (XTÆRr
the variance of the noise, so that ^ a ¼ XTX
dispersion r2(XTX)?1. By assuming that the noise is
independent among the N channels with equal variance
and is normally distributed with expectation value
E{r}=0, see Fig. 4c, we can interpret the confidence
region for a,by noting that
Þ ?N ? m
ð5Þ
ðÞ
?1ÆX)?1=(XTÆ r2ÆX)?1where r2is
???1X ? w with
^ a ? a
ðÞTXT? X ^ a ? a
ð
m ? rTr! Fm
N?m:
ð6Þ
We chose D^ a ¼ ^ a ? a to be 70% confidence level in
analogy to an m-dimensional SD. The confidence region
of alpha is an m-dimensional ellipsoid.
Detecting protein expression in a resolution-limited
sample voxel
To detect fluorescent protein expression in a fluorescent
spot i (irrespective of the fluorescent protein) we first
tested if its normalised spectral vector w(i) statistically
differed from the autofluorescence bundle {w(AF)},
2,X
Here
ð1=LÞPL
and j is their index. We tolerated a 5% false positive
detection of protein presence (t95%=2.31). Spectral
unmixing permits the estimation of the fractional
abundance of the fluorophores present in the studied
object. To correct for different levels of autofluorescence
within and among cells and to reduce the problem to
two dimensions we re-expressed the 3D spectral abun-
dance vector ^ a by a re-normalised 2D vector,
L ? 1
L þ 1? wðiÞ ? wðAFÞ
wðAFÞ ¼ ð1=LÞPL
autofluorescent pure vectors of the AF vector bundle
???
???
N
k¼1
j¼1wðjÞ
?2: L=25 is the number of
r2
kðAFÞ ! FN
NðL?1Þ:
ð7Þ
and
r2
kðAFÞ ¼
j¼1wkðjÞ ? wðAFÞ
?
^ a0¼ a0
that only contains the exogenous fluorophores. Nor-
malising by 1?aAFinstead of aEGFP+acitrineallows to
take into account the residual. Vectors ^ a0are conve-
niently represented on a parametric plot of a¢EGFPversus
a¢citrine (Fig. 5). Once the problem reduced to two
dimensions, we searched for the presence in voxel i of
EGFP and citrine, respectively, by comparing the mea-
sured ^ a0to the previously characterised pure bundles
f^ a0gEGFPand f^ a0gcitrine: The expression
EGFP;a0
citrine
??¼ aEGFP;acitrine
ðÞ= 1 ? aAF
??;
ð8Þ
a0
lðiÞ ? a0
?
lðEGFPÞ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
! StudentðL ? 1Þ;
if no fluorophore other than EGFP is present, and
similar for citrine. Here, L is the number of ROIs j that
defined the pure spectral vector bundle f^ a0gEGFP; a¢l(i) is
the re-normalised spectral abundance of fluorophore l in
voxel i, a0
L
in Eq. 9 is the SD of the a¢l(j) about their average
a0
presence. Their 95% confidence intervals are shown on
the parametric plots for binary mixtures (m=2) as
broken lines (Fig. 5). Thus, fluorophore presence can
conveniently be read off graphically; an unmixed ROI
within the rectangular region spanned by the broken
lines contains both fluorophores.
Statistical significance between data sets was assessed
by F-test and Student’s t test, when applicable. Param-
eters are given in the text and figures. We defined the
confidence interval of the estimated fluorescence abun-
dance estimates to have the same statistical significance
as an m-dimensional SD.
1
L
PL
j¼1a0
lðjÞ ? a0
lðEGFPÞ
?2
r
?
ffiffiffiffiffiffiffiffiffiffiffi
L ? 1
L þ 1
r
ð9Þ
lðEGFPÞ ¼1
PL
j¼1a0
lðjÞ and the denominator
lðEGFPÞ: Student’s test defines zones of fluorophore
Results
Evanescent-field images of mouse cortical astrocytes
transfected with constructs encoding a sb2-citrine fusion
protein (Fig. 1a, pCMV-sb2-citrine-N1, Wojcik et al.
2004) or the lysosome-specific transporter sialine-EGFP
(Fig. 1b, pCMV-sialine-EGFP-N1, Morin et al. 2004)
revealed a dense pattern of punctate green fluorescence.
To further characterise these diffraction-limited near-
membrane spots, we acquired five spectral images in
rapid succession and excised from the 510-535-560-615-
670 nm epifluorescence image cube 3·3 lm2ROIs)
containing a single near-diffraction-limited spot. We
used epifluorescence excitation instead of TIRF for the
acquisition of spectral images because SILU is an
intensity-based technique and thus sensitive to defocus
and out-of-focus movement artefacts. Wide-field exci-
tation epifluorescence samples a hour-glass-shaped vol-
ume that is fairly insensitive to submicrometre axial
vesicle movement (see PSF in Fig. 1c), whereas the
exponentially decaying evanescent field magnifies axial
movements, resulting in biased fluorophore abundance
estimates (data not shown). The use of epifluorescence
excitation reduces the axial optical sectioning. Negligible
out-of-focus contamination requires to chose fairly iso-
lated spots which is facilitated by the low organelle
density in astrocytes (see Discussion).
Roughly 30% of the spots seen in epifluorescence
appeared on EF excited images and ?50% of the spots
seen on EF images were also visible on the corre-
sponding epifluorescence images. The latter finding is
explained by the higher contrast offered by EF excited

Page 8

αCit
αAF
a
αGFP
b
c
1.0
0.5
0.0
α'GFP
1.00.0
α'Cit
4
2
0
-2
Student's test
EGFP citrine
59 58
3
5756
x10
citrine
26
3
25 24
x10
EGFP
1.0
0.5
0.0
α'GFP
1.00.5 0.0
α'Cit
800 600
EGFP
citrine
300200
20
15
10
5
0
Student's test
EGFP citrine
1.0
0.5
0.0
α'GFP
1.00.50.0
α'Cit
600400
EGFP
citrine
400300
10
8
6
4
2
0
Student's test
EGFP citrine
d
1.0
0.5
0.0
α'GFP
1.00.50.0
α'Cit
1.0
0.5
0.0
α'GFP
1.00.50.0
α'Cit
?
{w(AF)}
w(j)
wCit
wGFP
Fig. 5 Detecting vesicular protein expression and co-localisation. a
Left Fluorescent protein (FP) expression was detected by testing if
spectral vector w(i) was an outlier relative to the autofluorescence
vector bundle {w(AF)}. Middle left To compare across organelles
andcellswithvaryingAF
^ a ! ða0
EGFP and citrine unit vectors (shaded). Middle right Astrocytes
transfected with synaptobrevin-2 (sb2)-EGFP (green) or sb2-citrine
(red) alone showed no co-localisation (see Table 2 for details).
Black crosses indicate spots for which no decision was possible with
the desired 95% confidence. Right Cytoplasmic expression of
EGFP (green) or citrine (red) alone determined the detection
threshold for the presence of both fluorophores (dashed). b
Cytoplasmic co-transfection of EGFP and citrine is detected as
co-expression. Middle panels Comparison on reconstituted images
of FP presence in the central (red square) versus peripheral regions
of interest (ROI) (testing for outliers, Student’s test, 95%
confidence, t95%) discards these false positives (black symbols).
Right Each analysed spot is represented by a barbell. Endpoints
level,were-normalised
EGFP;a0
citrineÞ that lie coplanar in the plane spanned by pure
report t95%values for the presence of EGFP and citrine on the
organelle if t95%>1.76 (dotted line). Red symbols indicate ROIs that
SILU classifies as citrine-positive organelles, green indicates EGFP
expression, black are organelles below t95%. c Same type of analysis
for 25 organelles in sb2-EGFP and sb2-citrine co-transfected
astrocytes. A majority display both EGFP and citrine (yellow). On
the reconstituted fluorophore abundance maps, a central maximum
for both EGFP and citrine fluorescence is detected. See Table 2 for
details. d When expressing citrine and EGFP on secretory vesicles
(sb2-citrine) and lysosomes (by tagging the lysosome-specific
transporter sialine with EGFP), respectively, about two-thirds of
the organelles express EGFP only. This is explained by the
observation that in cultured astrocytes lysosomal compartments
outnumber secretory vesicles. About one quarter of lysosomes also
contained citrine, probably indicating the uptake and degradation
of membranous sb2-citrine. See Table 2 for details. Middle images
show the example of an EGFP-positive and citrine-negative
organelle. (A colour version of this figure is available online)

Page 9

fluorescence that permits the visualisation of puncta that
are drowned in a diffuse haze on epifluorescence images.
Using combined EF detection and spectral epifluores-
cence imaging, we analysed a total of 250 single-spot
images from the footprint regions of 45 mouse cortical
astrocytes making close contact with the coverslip. Of
these ROIs, 128 contained vesicles (or lysosomes), 122
were cytoplasmic control regions. We only analysed and
unmixed spots that were detected on both evanescent-
field and epifluorescence images.
SILU estimates of fluorophore abundance are sus-
ceptible to errors introduced to the raw images by spu-
rious background due to inefficient filtering, the Poisson
noise of the fluorescence signal itself and the detector
readout noise. Also, by collecting a spectrum for each
pixel, the multispectral image picks up the spectral
properties of the corresponding sample voxel (Fig. 1c).
Thus, when imaging fluorescent sub-resolution objects,
adjacent pixels on the diffraction-limited image contain
correlated information. We note that this situation is
different from that encountered in remote sensing and
satellite imaging applications, where mixed pixels arise
from fine variations in soil coverage present in the single-
pixel image. Such mixed pixels are independent and can
(and should indeed) be unmixed individually. Since the
precision of unmixing depends on the accuracy of the
sample and reference spectra (Fig. 2b, right; see Neher
and Neher 2004a for details), binning the nine pixels of
an Airy disk-sized (0.5·0.5 lm2) ROI centred on the
organelle image for extracting the noise-reduced spec-
trum w(i) of organelle i is expected to increase the pre-
cision of fluorophore abundance estimates (red on
Fig. 2). We ascertained the spectral homogeneity within
the Airy disk by testing the pixel spectra for outliers
(F-test, t85%=1.75, see Materials and methods for de-
tails). The participation of a fluorophore to a mixed
0.5·0.5 lm2ROI centred on the single-spot image was
estimated by solving wðiÞ ¼ X^ aðiÞ þ rðiÞ for ^ aðiÞ in the
least-squares error sense. X is the N · m matrix of
measured pure spectra xl(Fig. 2b, right panel), ^ aðiÞ the
vector estimating the fractional abundance of pure
spectra and r(i) the residual not explained by the linear
mixing model. As fully constrained SILU methods are
known to be less effective than partially constrained
techniques (Chang 2003), we used a supervised least-
squareserrorminimisation
constraining alto be non-negative nor imposing a sum-
to-one constraint (see Materials and methods for
details).
A common interferer in biological imaging is auto-
fluorescence (AF) that can obscure the emission of
exogenous fluorophores (see Fig. 2b, right). Therefore,
we first verified whether SILU recognised pure spectral
components in unlabelled (autofluorescent) and mock-
transfected cells that homogeneously expressed in their
cytoplasmpCMV-EGFP-N1
respectively (see Fig. 1a, b, insets). In the absence of
image noise, SILU should reconstitute unitary abun-
dance vectors ^ a ¼ ðaEGFP;acitrine;aAFÞ with only one
unmixing without
orpCMV-citirine-N1,
non-zero component in the dimension corresponding to
the respective fluorophore.
One concern is that SILU relies on the precision of
the reference spectra used. In order not to interfere with
vesicle docking and fusion, sb2-EGFP, sb2-citrine and
EGFP-sialine express the fluorophore on the lumenal
rather than the extravesicular side so that the fluoro-
phore is exposed to the acidic intravesicular environ-
ment. We therefore verified the influence of pH on the
reference spectra and unmixing. In controls using per-
meabilised astrocytes, the successive acidification of the
cytosol did not appreciably alter EGFP and citrine flu-
orescence emission spectra down to pH values of 4.8
(Fig. 3a) and 5.2 (Fig. 3b), respectively. However, while
their shape proved robust against pH changes, their
intensity rapidly declined upon acidification (Fig. 2c, d).
Upon return to pH 7.3, the shape of the spectrum was
recovered, even after having attained pH 4.3. The
incomplete recovery of EGFP intensity beyond pH 5.0
(data not shown) is consistent with Verkman’s earlier
work and, probably due to a decreased solubility, can be
taken as an internal control that the cytoplasm indeed
had reached pH values below 5 (Kneen et al. 1998). We
conclude that the acidic vesicle lumen does not appre-
ciably shift the fluorophore spectral signature compared
to the cytoplasmic controls but, due to the fainter signal
at low pH, increases the impact of noise on the accuracy
of the measured spectra (see Discussion).
Despite their insensitivity to pH effects, the recorded
pure spectra displayed some variability from cell to cell
and even from one organelle to another (see superim-
posed grey traces on Fig. 2b, right panel). The accuracy
of the reference and unmixed spectra and their degree of
variability determine the accuracy of the fluorophore
abundance estimate. This effect is illustrated when
unmixing spectra taken in control situations where only
one fluorophore was present. In this situation, only the
respective fluorophore truly present should be detected
while the proportion of false positives should be close to
zero. We plot on Fig. 4a the pure spectral abundance
vectors ^ aðiÞ
by the pure EGFP, citrine and AF unit vectors. As ex-
pected, the values of ^ aðiÞ
unit vectors for all three fluorophores (see Table 1 for
the mean ± SD). A score between 0 and 1 rates the
estimated fractional abundance of each fluorophore, 1
equalling a perfect match. For greater clarity, we show
the 2D projection of ^ aðiÞ onto the plane 1T? ^ a ¼ 1
(shaded on panel a, Fig. 4b) instead of a pseudo-3D
plot. On the triangular plot (its parameterisation in the
new co-ordinates X and Y is given in Appendix) pure
spectral vectors are expected to populate the corners. As
anticipated from the larger spectral variability of AF, we
observe a larger dispersion of the reconstituted compo-
nent of ^ aðiÞ for AF (1.000±0.016) than for EGFP
(1.001±0.002) and citrine (0.999±0.002). Thus, even for
AF, unit vectors are recovered with less than 2% relative
error and the estimates of EGFP and citrine in pure AF
cells are equal to zero (see Table 1). Our data hence
hil¼ ðaEGFP;acitrine;aAFÞ in a space spanned
hilare close to the orthogonal

Page 10

confirm the previous observation that despite the pres-
ence of image noise and spectral heterogeneity a few
spectral detection bands suffice to correctly classify
spectrally overlapping fluorophores (Neher and Neher
2004a).
In the following, we therefore use the reconstituted
pure EGFP and citrine vector bundles to define zones of
fluorophore presence for testing if a spectral vector be-
longs to f^ a0gEGFPor f^ a0gcitrine; respectively.
The main interest of spectral single-vesicle imaging
and linear unmixing is its potential to detect FP
expression in the presence of a spectrally overlapping
and spatially varying autofluorescence. We consequently
investigated in sb2-EGFP or sb2-citrine-transfected as-
trocytes if SILU recognised fluorescent protein expres-
sion on single secretory vesicles. Spectral vectors w(i) of
FP expressing organelles differed from AF ROIs when
testing for outliers against the pure AF vector bundle
{w(AF)} (Fig. 5a, left, cf. Eq. 7). To compare organelles
across cells with different AF levels, we re-normalised
^ a0¼ aEGFP;acitrine
a¢citrine(middle panel). Testing if ^ a0ðiÞ was an outlier
relative to the previously obtained (and re-normalised)
pure spectral vector bundles f^ a0
fined detection thresholds ^ a0
EGFP and ^ a0
tively, meaning that an
^ a0located in the rectangular
region bounded by dotted lines designates the presence
of both EGFP and citrine. Indeed, in control cells that
were contransfected with EGFP and citrine (Fig. 5b), all
25 analysed spots fell within the bounded region. Their
average was
^ a0
h i ¼ ð0:331 ? 0:008; 0:669 ? 0:008Þ:
We next unmixed 25 fluorescent spots in cells that
were either transfected with sb2-EGFP (green) or sb2-
citrine (red, Fig. 5a, middle panel). On the rightmost
panel we show for comparison the previously measured
pure abundances for the cytoplasmic expression of the
same FPs, in re-normalised co-ordinates f^ a0
f^ a0
pure EGFP or citrine-expressing vesicles displayed a
wider distribution than the corresponding cytoplasmic
controls and have larger confidence intervals than their
cytoplasmic counterparts. For the same gain and inte-
gration time, single fluorescent organelles images are
dimmer (and hence more noisy) and thus display more
spectral variability than resolution-limited homoge-
neously fluorescent sample voxels (see below).
Perhaps SILU-based fluorophore abundance esti-
mates are biased by sampling off-focus fluorescence
above and below the studied (sub-resolution) organelle.
An average vesicle measures 30–300 nm in diameter and
ðÞ=ð1 ? aAFÞ and plotted a¢EGFPversus
EGFPg and f^ a0
EGFP¼ ð0:965 ? 0:034Þ for
citrineg de-
citrine¼ ð0:976 ? 0:023Þ for citrine, respec-
EGFPg and
citrineg: Organelle spectral abundance vectors ^ a0ðiÞ of
is much smaller than the axial extent of the experimental
point-spread function (Fig. 1c). Also, fluorophores may
not exclusively be localised on secretory organelles. For
example, although predominantly targeted to secretory
vesicles, sb2 is only partially recaptured after fusion and
partly diffuses in the plasma membrane (D. Perrais,
personal communication). Corroborating this observa-
tion, sb2-citrine-expressing astrocytes (Fig. 1a) dis-
played a luminous hazy background that was absent in
EGFP-sialine-expressing cells (panel b). To ascertain a
vesicular localisation of the FP chimera and to exclude
false positives that could result from membranous or
cytoplasmically expressed fluorophores above focus, we
introduced a spatial contrast parameter based on
reconstituted fluorophore abundance maps. The partic-
ipation of pure spectra to each pixel (x, y) was calculated
by multiplying ^ a ? ðx;yÞ componentwise with the scaling
factorP
regions in the periphery of spot i (Fig. 5b). On near-
diffraction-limited images, a vesicular or lysosomal flu-
orophore localisation should be detectable as a central
outlier. In contrast, a diffuse cytoplasmic or membra-
nous localisation should produce a flat contrast. We
validated our approach by first studying defined control
transfections. Figure 5b shows a pseudocolour spatial
abundance map for cytoplasmically co-expressed EGFP
and citrine.Asexpected,
co-localisation of cytoplasmically EGFP and citrine
produces a flat contrast, indicating that the estimated
abundance in the central 3·3-pixel ROI was not statis-
tically different from the surrounding 16 ROIs (Stu-
dent’s test, see below). In contrast, EGFP- and citrine
positive were reliably detected as outliers (panels c and
d). We therefore classified an organelle as co-expressing
EGFP and citrine, if (1) both EGFP and citrine were
detected in the central ROI and (2) the spatial contrast
for both EGFP and citrine showed a central outlier. The
results of the statistical test are quantified in the right-
k=1
N
wk(x, y). Spatial contrast was then defined
as the amplitude of the central ^ aðiÞ relative to 16 control
the volumetric(false)
Table 1 Reconstituted pure fluorophore abundances
EGFPCitrine AF
^ aðEGFPÞ
^ aðcitrineÞ
^ aðAFÞ
h
h
h
il
il
1.001±0.002
0.002±0.002
0.000±0.026
0.000±0.004
0.999±0.002
0.000±0.026
?0.001±0.002
0.000 ±0.002
1.000±0.016
il
0.20
0.15
0.10
0.05
ci (70%)
654
SBR
3
Fig. 6 Evolution of the confidence interval (ci) for estimating
aEGFPwith increasing signal-to-background ratio (SBR). Each spot
represents an unmixed organelle in a sb2-EGFP/sb2-citrine co-
transfected astrocyte. Signal was measured as the average intensity
in an Airy disk-sized 3·3-pixel region centred on the spot.
Background intensities were measured in nearby cell-free regions
with the same gain and exposure time. The accuracy of the
estimation of aEGFPincreases with higher signals. Estimates of a
EGFPwith a precision better than 10% (dashed line) require a signal
three times above background

Page 11

most column of Fig. 5 by plotting, for each analysed
ROI, a barbell the endpoints of which indicate the
contrast in the EGFP and citrine-reconstituted images,
respectively. ROIs were dubbed FP labelled when their
contrast exceeded Student’s t95%>1.76. We used red
and green symbols to designate citrine- and EGFP-po-
sitive organelles, respectively. Black symbols indicate
diffuse expression, yellow shows specific co-localisation
on the studied organelle. The same colour code was then
assigned to the spots on 2D abundance maps. A colour
version of Fig. 5 is available online
Both the detection of protein presence on the orga-
nelle and the estimated degree of co-localisation depend
on the desired confidence level (Table 2). We take the
example of control cells that were transfected with the
fusion protein encoding sb2-EGFP only. If EGFP shall
be detected with 99% certainty, 72% of the fluorescent
organelles expressing another fluorophore than AF
alone are recognised EGFP-positive. No organelle is
classified as citrine-positive, but 28% of the fluorescent
organelles are discarded, because no decision can be
reached with 99% confidence. If, however, the confi-
dence for EGFP detection is lowered to 95%, the frac-
tion of non-classified organelles drops to 4%, and 96%
organelles are recognised EGFP-positive. When toler-
ating 10% error for organelle classification, 92%
organelles are classified EGFP-positive. This apparent
drop is explained by the observation that now 8% are
(falsely) dubbed citrine-positive, and none remains
unclassified. Hence, in practice, a compromise must be
reached between detecting a sufficiently large fraction of
fluorescent organelles to constitute a representative
sample and accepting a false classification. In our hands,
95% confidence (t95%=1.76 for Student’s-distribution,
see the dotted line on the rightmost panel of Fig. 5b–d)
proved a good choice for detecting fluorescently labelled
secretory organelles in cell culture.
Finally, we illustrate the performance of our tech-
nique for detecting co-localisation by co-expressing sb2-
EGFP and sb2-citrine together (Fig. 5c). After the suc-
cessful elimination of only-autofluorescent ROIs SILU
finds,with 95% confidence,
EGFP-positive, 8% citrine-positive and 52% expressing
both fluorescent proteins. This observation is in stark
40% organelles
contrast to previous controls in which a single fluores-
cent protein was expressed and virtually no (false) co-
localisation was detected (Table 2). The surprisingly low
fraction (8%) of citrine-(only) positive organelles com-
pared to the fivefold higher fraction of EGFP-expressing
secretory vesicles is most likely due to the faint lumi-
nosity of citrine compared to EGFP (see below and
Fig. 3).
When instead co-expressing the lysosomal trans-
porter sialine tagged with EGFP together with the ves-
icle marker sb2-citrine (Fig. 5d) we found a substantially
lower 28% overlap; 64% of the studied spots expressed
EGFP only, and almost no citrine was detected alone,
probably indicating the degradation of membraneously
overexpressed sb2 in lysosomal compartments. This re-
sult is also plausible in view of the observation that
lysosomes largely outnumber secretory vesicles in cul-
tured astrocytes (Nadrigny and Oheim, unpublished
data).
Discussion
Multi- and hyperspectral imaging are utilised in air- and
spaceborne remote sensing of Earth topography, envi-
ronmental monitoring, mapping and management of
water or agricultural resources to detect the presence in
mixed pixels of several endmembers, the fractional sur-
face coverage of which is often much lower than the
monitored area represented in one image pixel. Micro-
scopic fluorescence imaging differs from these applica-
tions in that it provides diffraction-limited images on
which point objects are spread over several adjacent
image pixels, which are no longer independent. In this
study, we took advantage of their statistic interdepen-
dence to extract noise-reduced organelle spectra and to
unmix resolution-limited single-vesicle images. We also
derived the m-dimensional confidence interval of SILU-
based fluorophore abundance estimates ^ aðiÞ:
Heterogeneous vesicle populations have previously
been studied after immunocytochemical labelling with
specific antibodies by confocal imaging (Parpura et al.
1995), deconvolution-assisted fluorescence microscopy
(Anlauf and Derouiche 2005) or immuno-electron
microscopy (Bezzi et al. 2004) in fixed preparations.
Alternatively, local ultrasound pulses generate semi-in-
tact ‘‘unroofed’’ cells. Due to their flatness, these
‘‘plasma membrane lawns’’ permit the visualisation of
single membrane-attached vesicles (Lang 2003; Wiegand
et al. 2002). Sub-cellular fractionation and subsequent
gradient-density centrifugation (Calegari et al. 1999;
Papini et al. 1995) or vesicle isolation on immunobeads
(Chilcote et al. 1995) are cell-free alternatives for clas-
sifying vesicles.
Clearly, a better understanding of the dynamic
mechanisms and the regulation of the parallel secretory
pathways that coexist both in non (electrically) excitable
cells (Chieregatti and Medolesi 2005) like astroglia
(Coco et al. 2003; Mothet et al. 2005) but also in neu-
Table 2 Profiling protein expression on single vesicles
n=25 eachClassified
EGFP (%)
Classified
citrine (%)
Co-localised
(%)
sb2-EGFP
sb2-citrine
sb2-EGFP+sb2-citrine
EGFP-sialine+
sb2-citrine
72, 96, 92
0, 0, 0
56, 42, 32
64, 64, 56
0, 0, 0
40, 68, 88
8, 8, 4
8, 0, 0
0, 0, 8
0, 4, 4
32, 52, 64
4, 28, 36
Differences to 100% are due to spots for which decision was
impossible. We counted hits for detecting fluorescent proteins with
99, 95, and 90% confidence, respectively. Lower expression detec-
tion thresholds engender a larger probability of detecting false co-
localisation (also)

Page 12

rones and that co-release different transmitters requires
time-lapse imaging of intact cells rather than fractiona-
lised or semi-intact preparations.
The imaging and tracking over time of distinct sub-
cellular single vesicular compartments labelled with
specific fluorescent fusion proteins has become possible
(Steyer et al. 1997; Oheim et al. 1998) with EF excitation
of fluorescence (see, e.g. Schneckenburger 2002; Axelrod
2003 for recent reviews). However, the multiplexed
detection of spectral variants of FP requires spectrally
separated probes and negligible autofluorescence. Al-
though monomeric red fluorescent jellyfish and coral
proteins are increasingly becoming available in addition
to yellow and green fluorescent fusion proteins for
double or triple labelling experiments (Verkhusha and
Lukyanov 2004), we found that the transient expression
of the vesicle-associated membrane protein VAMP-2
(synaptobrevin-2, sb2) tagged with red fluorescent pro-
teins (DsRed, mDsRed, hcRed) failed to produce a
homogenous cytoplasmic expression in controls (data
not shown). Similarly, transgenic mice expressing red
coral proteins produced fluorescent precipitates that
precluded a detailed morphological analysis (Hirrlinger
et al. 2005).
In turn, the available cyan, green, yellow and orange
FPs display a strong cross-excitation and emission
spectral overlap that precludes their separation using
dual or triple emission band-pass filters. Importantly,
even with non-precipitating red fluorescent labels, the
observed spectrally broad autofluorescence (AF) would
still obscure FP expression and preclude an unambigu-
ous identification of labelled organelles. Therefore, dual
band-pass recordings do not provide a suitable way to
reliably quantify co-localisation in the present experi-
mental scenario.
In the present study, we demonstrate that SILU is a
powerfultooltoquantitate
co-localisation of spectrally overlapping proteins in live
cells with diffraction-limited spatial resolution. While
this observation hardly comes as a surprise, our study
provides the theoretical framework for quantitative
SILU of single diffraction-limited point objects and de-
rives the precision that can be attained. Over splitting
the intensity into several multiplexed detection bands,
we opted for the sequential acquisition of spectrally
overlapping bands (Fig. 2). While this choice reduces the
time resolution in proportion to the number of detection
bands it optimises the signal-to-noise ratio and precision
of the technique (see below). In many experiments,
however, the organelle spectra will be constant on the
time scale of the image acquisition, because the under-
lying biological processes (protein expression, sub-unit
composition, etc.) are relatively slow. In these applica-
tions, it is sufficient to take a spectral ‘‘snapshot’’ at the
beginning of the time-lapse acquisition and then to track
the classified organelles with higher temporal resolution
on a single-wavelength movie. The same argument holds
for experiments with only slowly varying spectra: for
example, Duncan et al. (2003) took advantage of the
the expressionand
time-dependent spectral shift of a fluorescent timer
protein to demonstrate that individual chromaffin
granules are segregated functionally and spatially
according to age. Likewise, photoswitchable GFP vari-
ants (Chudakov et al. 2004) display highly overlapping
fluorescence before and after photoactivation. Semi-
conductor nanocrystals (quantum dots) undergo a
gradual blue shift in their emission spectrum due to
surface oxidation in response to prolonged illumination
(Grecco et al. 2004), which will be captured on sequen-
tial spectral snapshots, interlaced with faster time-lapse
imaging. We note that some applications that measure
dynamic changes in overlapping spectra like spectral
FRET detection (Ishii et al. 1999) might benefit from
projecting different images side by side on the same
CCD camera (Mattheyses et al. 2004) or simultaneously
on several detectors so as to increase the time resolution
at the expense of SILU precision.
We deliberately chose a combination of EF and
epifluorescence excitation, to identify individual near-
membranecandidatevesicles
epifluorescence for acquiring spectral images cubes.
Epifluorescence integrates over a micrometre axial
volume (Fig. 1c), thereby rendering SILU estimates
insensitive against small axial organelle movements and
defocus. However, the use of epifluorescence instead of
EF excitation to acquire spectra engenders a loss of
contrast and a decrease in the axial resolution. In as
much as we unmixed only organelles that appeared both
on EF and epifluorescence images and accounted for
out-of-focus blur by statistically testing for organelle
presence (centre vs. surround), the effective resolution is
not affected. However, epifluorescence imaging will
preclude reasonable SILU estimates in situations where
the organelle density is high. To be sure the recorded
spectrum of a spot was not contaminated by nearby out-
of-focus organelles, we selected (and rejected) ROIs in
relation to the experimental PSF (Fig. 1c), which defines
an observation volume in which only one organelle
should be present. The lateral PSF of the microscope is
described by a Bessel function with the second zero lo-
cated at 625 nm (3.3 pixels) from its centre. This dis-
tance corresponds to 99.6% attenuation and defines a
ring around the 3·3 pixels ROI (i.e. an overall diameter
d=9 pixels) in which no other organelle must be present.
In axial direction, the second zero z2is located at 3.4 lm
from the focus plane. Hence, the studied organelle must
be alone in a cigar-shaped volume of V=p d2/
4z2=8.3 lm3. With roughly 0.01 organelles/lm3in cul-
tured astrocytes we expect n=q V?0.08 organelles in
this volume element. So there is a very low probability
that a second organelle would be present in the readout
volume.
One important finding of our experiments is that
ceterum paribus citrine expression is less easily recogni-
sed than EGFP labelling (Table 2). Is this difficulty re-
lated to the dimmer citrine fluorescence, its larger pH
sensitivity (Heikal et al. 2000), or the larger similarity
between citrine and AF spectra than between EGFP and
forunmixing and

Page 13

AF? We can indirectly exclude the last possibility by
noting that a variable amount of autofluorescence does
not influence the precision of our acitrineor aEGFPesti-
mates (Supplementary Fig. 1). Second, SILU does not
detect more AF in pure citrine than in pure EGFP and
also does not falsely recognise more citrine than EGFP
in non-labelled (only AF) cells (Table 1). Furthermore,
our spectral measurements at different pHs demonstrate
that the EGFP spectrum does not appreciably change
down to pH 4.8 so that the reference spectrum recorded
in the cytoplasm is a good template for that expected in
secretory organelles. Indeed, we find 96% sb2-EGFP-
positive organelles (N=25) when unmixing an EGFP-
sb2-transfected cells. In contrary, the citrine spectrum
begins to change at pH 5.2 so that there could be a slight
difference between the cytoplasmic reference spectrum
and the spectrum in organelles. We detect only 68% of
citrine-sb2-positive organelles in citrine-sb2-transfected
cells (N=25). However, the same results were obtained
when we unmixed these organelles with the citrine
spectra acquired at pH 5.2 instead, and bigger errors
and same or lower detection were obtained with the
reference spectra recorded at even lower pH (5.0 or 4.8).
Thus, we conclude that as for EGFP, the vesicular ci-
trine spectrum is not appreciably different from its
cytoplasmic reference and that it can be used for
unmixing citrine expressed in mildly acidic organelles.
Rather, citrine-labelled organelles are dimmer than
EGFP-labelled ones. This is not an indirect pH effect,
because both EGFP and citrine intensities show a sim-
ilar pH dependence (Fig. 3c, d). On average, spots in
sb2-EGFP-transfected cells displayed a centre versus
surround contrast ratio of 0.67±0.30 compared to
0.32±0.14 for citrine (N=25, P<0.0001). Thus, the
slight variation in the spectral shape of citrine at pH
lower than 5.2 at both dimmer ends may be partially due
to its lower brightness and hence larger relative contri-
bution of image noise (cf. the size of the normalised SD
in Fig. 3c, d). Indeed, low-intensity spots generally have
a spectrum slightly different from the brighter ones (data
not shown). Actually, this difference in brightness be-
tween EGFP and citrine was predictable because
458-nm excitation is more efficient for the former.
Longer-wavelength excitation offers little room for
improvement as the transmission light and also the noise
in the first 510 nm emission channel would increase at
the same time.
Thus, the lower probability to detect citrine-labelled
organelles is rather attributable to the lower brightness
of citrine compared to EGFP than to different responses
to acidic pH. Our results illustrate that a sufficiently high
image contrast is required to successfully unmix labelled
organelles. Finally, which signal-to-background ratio
(SBR) is required to attain a given SILU confidence le-
vel? Figure 6 plots, for a representative sample of fluo-
rescent spots from a sb2-EGFP/sb-citrine co-transfected
astrocyte, the 70% confidence interval of the estimated
fractional abundance of EGFP, a¢EGFP, as a function of
the spot’s SBR (n=21). Background intensities were
measured in nearby cell-free regions with the same gain
and exposure time. As expected, a higher SBR favours
the accurate determination of a¢EGFP. To obtain esti-
mates of a¢EGFPwith a precision D a¢EGFP/a¢EGFPbetter
than 10% (dashed line) requires an average vesicle signal
exceeding three times the image background. For
example, the analysed spots in astrocytes that were
transfected with sb2-EGFP or sb2-citrine, respectively,
had an average SBR of 11.5±4.7 and 9.3±5.2 and
hence allowed good precision estimates (Fig. 6).
In conclusion our results demonstrate that SILU of
spatially and spectrally correlated Airy disk-sized regions
provides robust estimates of protein expression and co-
localisation on individual exocytic and non-secretory
compartments (Chieregatti and Medolesi 2005) in live
cells. We expect our technique to be influential in a large
variety of cell and neurobiological applications where
conventional dual or triple colour measurements fail to
produce reliable co-localisation data.
Acknowledgements The authors thank S.L. Shorte for comments on
the manuscript and M. Coppey-Moissan for the loan of equipment.
Supported by the Institut National de la Sante ´ et de la Recherche
Me ´ dicale (INSERM) and a joint grant from the Centre National de
la Recherche Scientifique (CNRS) and the Ministe ` re National de la
Recherche et de la Technologie (AC DRAB no. 03/93-2003) to CG
and MO and a joint Max-Planck/INSERM AMIGO grant
‘‘Molecular Bases of Astroglial Signalling under Physiological and
Pathological Conditions’’ (to CG, MO and FK). FN was the re-
cipient of a Ph.D. Ministe ` re National de la Recherche et de la
Technologie studentship, PGH was financed by a Max-Planck
postgraduate Ph.D. studentship.
Appendix
Parameterisation of the 2D projection of the 3D flu-
orophore abundance vector onto the plane 1T? ^ a ¼ 1:
For the tertiary mixture of AF, EGFP, citrine, the
estimated fluorophore abundance ^ aðiÞ is a vector in 3D
space, spanned by the EGFP, citrine and AF unit vec-
tors, respectively (Table 1). However, its pseudo-3D
representation does not lend itself to a clear graphic
representation of an appreciable number of unmixed
spots. Thus, for greater clarity, we plot on Fig. 4b its 2D
projection onto the plane 1T? ^ a ¼ 1 (shaded on panel a),
parameterised by X and Y,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Y ¼cos arctan 2
X ¼sinua2
EGFPþa2
?
citrine
p
q
;
. ffiffiffi
2
?
?arctan aAF=B
ðÞ
hi
?
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a2
AFþB2
q
;
ð10Þ
where
B¼
ffiffiffi
2
p
2?cosu?
?
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð
ð
a2
EGFPþa2
citrine
q
?????
?????
Þ
u¼
p
4? arctan aEGFP=acitrine
?3p
Þ½?
foracitrine>0
else
4? arctan aEGFP=acitrine
½?
:
ð11Þ

Page 14

Here, / denotes the angle (positive or negative angle)
between the vectors (aEGFP, acitrine, 0) and (1, 1, 0).
References
Anlauf E, Derouiche A (2005) Astrocytic exocytosis vesicles and
glutamate: a high-resolution immunofluorescence study. Glia
49(1):96–106
Bezzi P, Gundersen V, Galbete JL, Seifert G, Steinha ¨ user C, Pilati
E, Volterra A (2004) Astrocytes contain a vesicular compart-
ment that is competent for regulated exocytosis. Nat Neurosci
7:613–620
Bosdogianni P, Petrou M (1997) Mixed pixel classification with ro-
bust statistics. IEEE Trans Geosci Remote Sens 35(3):551–559
Bosdogianni P, Petrou M, Kittler J (1997) Mixture models with
higher order moments. IEEE Trans Geosci Remote Sens
35(2):341–353
Calegari F, Coco S, Tarverna E, Bassetti M, Verderio C, Corradi
N, Matteoli M, Rosa P (1999) A regulated secretory pathway in
cultured hippocampal astrocytes. J Biol Chem 274:22539–22547
Chang CI (2003) Hyperspectral imaging: techniques for spectral
detection and classification. Kluwer Plenum, New York
Chieregatti E, Medolesi J (2005) Regulated exocytosis: new
organelles for non-secretory purposes. Nature reviews. Mol Cell
Biol 6:181–187
Chilcote T, Galli T, Mundigl O, Edelmann L, McPherson P, Takei
K, De Camilli P (1995) Cellubrevin and synaptobrevins: similar
subcellular localization and biochemical properties in PC12
cells. J Cell Biol 129:219–231
Chudakov DM, Verkhusha VV, Staroverov DB, Souslova EA,
Lukyanov S, Lukyanov KA (2004) Photoswitchable cyan
fluorescent protein for protein tracking. Nat Biotechnol
22:1435–1439
Coco S, Calegari F, Pravettoni E, Pozzi D, Taverna E, Rosa P,
Matteoli M, Verderio C (2003) Storage and release of ATP
from astrocytes in culture. J Biol Chem 278:1354–1362
Demuro A, Parker I (2004) Imaging the activity and localization of
single voltage-gated Ca2+channels by total internal reflection
fluorescence microscopy. Biophys J 86:3250–3259
Dickinson ME, Bearman G, Tille S, Lansford R, Fraser SE (2001)
Multi-spectral imaging and linear unmixing add a whole new
dimension to laser scanning fluorescence microscopy. Bio-
Techniques 31:1272–1278
Duncan RR, Greaves J, Wiegand UK, Matskevich I, Bodammer
G, Apps DK, Shipston MJ, Chow RH (2003) Functional and
spatial segregation of secretory vesicle pools according to ves-
icle age. Nature 422:176–180
Ecker RC, de Martin R, Steiner GE, Schmid JA (2004) Application
of spectral imaging microscopy in cytomics and fluorescence
resonance energy transfer (FRET) analysis. Cytometry (Pt A)
59A:172–181
Ellenberg J, Lippincott-Schwartz J (1999) Dual-colour imaging
with GFP variants. Trends Cell Biol 9:52–56
Garini Y, Gil A, Bar-Am I, Cabib D, Katzir N (1999) Signal to
noise analysis of multiple color fluorescence imaging micros-
copy. Cytometry 35:214–226
Grecco HE, Lidke KA, Heintzmann R, Lidke DS, Spagnuolo C,
Martinez OE, Jares-Erijman EA, Jovin TM (2004) Ensemble
and single particle photophysical properties (two-photon exci-
tation, anisotropy, FRET, lifetime, spectral conversion) of
commercial quantum dots in solution and in live cells. Microsc
Res Tech 65:169–179
Griesbeck O, Baird GS, Campbell RE, Zacharias DA, Tsien RY
(2001) Reducing the environmental sensitivity of yellow fluo-
rescent protein. J Biol Chem 276:29188–29194
Gu Y, Di WL, Kelsell DP, Zicha D (2004) Quantitative fluores-
cence resonance energy transfer (FRET) measurement with
acceptor photobleaching and spectral unmixing. J Microsc
215:162–173
Heikal AA, Hess ST, Baird GS, Tsien RY, Webb WW (2000)
Molecular spectroscopy and dynamics of intrinsically fluores-
cent proteins: coral red (dsRed) and yellow (citrine). Proc Natl
Acad Sci USA 24(97):11996–12001
Hiraoka Y, Shimi T, Haraguchi T (2002) Multispectral imaging
fluorescence microscopy for living cells. Cell Struct Funct
27:367–374
Hirrlinger PG, Scheller A, Braun C, Quintela-Schneider M, Fuss B,
Hirrlinger J, Kirchhoff F (2005) Expression of red coral fluo-
rescent proteins in the central nervous system of transgenic
mice. Mol Cell Neurosci 30:291–303
Ishii Y, Yoshida T, Funatsu T, Wazawa T, Yanagida T (1999)
Fluorescence resonance energy transfer between single fluoro-
phores attached to a coiled-coil protein in aqueous solution.
Chem Phys 247:163–173
Kneen M, Farinas J, Li Y, Verkman AS (1998) Green fluorescent
protein as a noninvasive intracellular pH indicator. Biophys J
74:1591–1599
Lang T (2003) Imaging SNAREs at work in ‘unroofed’ cells—ap-
proaches that may be of general interest for functional studies
on membrane proteins. Biochem Soc Trans 31:861–864
Lansford R, Bearman G, Fraser SE (2001) Resolution of multiple
green fluorescent protein color variants and dyes using two-
photon microscopy and imaging spectroscopy. J Biomed Opt
6:311–318
Mattheyses AL, Hoppe AD, Axelrod D (2004) Polarized fluores-
cence resonance energy transfer microscopy. Biophys J
87(4):2787–2797
Michalet X, Lacoste TD, Pinaud F, Chemla DS, Alivisatos AP,
Weiss S (2001) Ultrahigh Resolution multicolor colocalization
of single fluorescent nanocrystals. Nanoparticles and nano-
structured surfaces: novel reporters with biological applica-
tions. Proc SPIE 4258
Morin P, Sagne C, Gasnier B (2004) Functional characterization of
wild-type and mutant human sialin. EMBO J 23:4560–4570
Mothet J-P, Pollegioni L, Ouanounou G, Martineau M, Fossier P,
Baux G (2005) Glutamate receptor activation triggers a cal-
cium-dependent and SNARE protein-dependent release of the
gliotransmitter D-serine. PNAS 102:5606–5611
Neher F, Neher E (2004a) Optimizing imaging parameters for the
separation of multiple labels in a fluorescence image. J Microsc
213:46–62
Neher RA, Neher E (2004b) Applying spectral fingerprinting to the
analysis of FRET images. Microsc Res Tech 64:185–195
Nielsen AA (2001) Spectral mixture analysis: linear and semi-
parametric full and iterated partial unmixing in multi- and
hyperspectral image data. Int J Comp Vis 42:17–37
Oheim M, Loerke D, Stu ¨ hmer W, Chow RH (1998) The last few
milliseconds in the life of a secretory granule. Docking,
dynamics and fusion visualized by total internal reflection flu-
orescence microscopy (TIRFM). Eur Biophys J 27:83–98
Oshiro M, Moomaw B (2003) Cooled vs. intensified vs. electron
bombardmentCCDcameras—applications
advantages. Methods Cell Biol 72:133–156
Papini E, Rossetto O, Cutler DF (1995) Vesicle-associated mem-
brane protein (VAMP)/synaptobrevin-2 is associated with large
dense core secretory granules in PC12 neuroendocrine cells. J
Biol Chem 270:1332–1336
Parpura V, Fang Y, Basarsky T, Jahn R, Haydon PG (1995)
Expression of synaptobrevin II, cellubrevin and syntaxin
butnotSNAP-25incultured
337:489–492
Rouach N, Segal M, Koulakoff A, Giaume C, Avignone E (2003)
Carbenoxolone blockade of neuronal network activity in cul-
ture is not mediated by an action on gap junctions. J Physiol
553:729–745
Schapper F, Gonc ¸ alves JT, Oheim M (2003) Fluorescence imaging
with two-photon evanescent-wave excitation. Eur Biophys J
32:635–645
Schipper HM, Small L, Wang X, Brawer JR (2002) Role of por-
phyrin sequestration in the biogenesis of iron-laden astrocytic
inclusions in primary culture. Dev Neurosci 24:169–176
andrelative
astrocytes.FEBSLett

Page 15

Schneckenburger H (2002) Total internal reflection fluorescence
microscopy: technical innovations and novel applications. Curr
Opin Biotechnol 16:13–18
Schultz RA, Nielsen T, Zavaleta JR, Ruch R, Wyatt R, Garner HR
(2001) Hyperspectral imaging: a novel approach for micro-
scopic analysis. Cytometry 43:239–247
Shirakawa H, Miyazaki S (2004) Blind spectral decomposition of
single-cell fluorescence by parallel factor analysis. Biophys J
86:1739–1752
Stephens DJ, Lin-Marq N, Pagano A, Pepperkok R, Paccaud JP
(2000) COPI-coated ER-to-Golgi transport complexes segre-
gate from COPII in close proximity to ER exit sites. J Cell Sci
113:2177–2185
Steyer JA, Horstmann H, Almers W (1997) Transport, docking and
exocytosis of single secretory granules in live chromaffin cells.
Nature 388:474–478
Tsurui H, Nishimura H, Hattori S, Hirose S, Okumura K, Shirai T
(2000) Seven-color fluorescence imaging of tissue samples based
on Fourier spectroscopy and singular value decomposition. J
Histochem Cytochem 48:653–662
Verkhusha VV, Lukyanov KA (2004) The molecular properties
and applications of Anthozoa fluorescent proteins and chro-
moproteins. Nat Biotechnol 22:289–296
Wiegand UK, Don-Wauchope A, Matskevich I, Duncan RR,
Greaves J, Shipston MJ, Apps DK, Chow RH (2002) Exocy-
tosis studies in a chromaffin cell-free system: imaging of single-
vesicle exocytosis in a chromaffin cell-free system using total
internal reflection fluorescence microscopy. Ann NY Acad Sci
971:257–261
Wojcik SM, Rhee JS, Herzog E, Sigler A, Jahn R, Takamori S,
Brose N, Rosenmund C (2004) An essential role for vesicular
glutamate transporter 1 (VGLUT1) in postnatal development
and control of quantal size. Proc Acad Sci USA 101:7158–7163
Zimmermann T (2005) Spectral imaging and linear unmixing in
light microscopy. Adv Biochem Eng Biotechnol 95:245–265
Zimmermann T, Rietdorf J, Pepperkok R (2003) Spectral imaging
and its application in live cell microscopy. FEBS Lett 546:87–92