Conference PaperPDF Available

Evaluating Individual mRNA Molecules Detection Techniques in Microscope Images



Single molecule fluorescence in situ hybridization followed by microscopic image analysis is one of the prominent methods used to study gene expression on a single cell level. There are various microscopic image analysis methods, leading to differing mRNA spots being detected in images for the same experiment. We present a technique to evaluate different mRNA spots detection algorithms. It is based on image annotation by expert biologists and the receiver operating characteristics. The detection methods can be compared using parameters that withstand imprecise and imbalanced environments. The proposed evaluation procedure highlighted the difference between two microscopic image analysis methods that are frequently used. It can be applied to any image analysis method that seeks to find mRNA spots on a single cell level.
Evaluating Individual mRNA Molecules Detection
Techniques in Microscope Images
Rethabile Khutlang, Loretta Magagula, Musa Mhlanga
Gene Expression and Biophysics Group
Emerging Health Technologies
CSIR Biosciences
Pretoria, South Africa
Abstract—Single molecule fluorescence in situ hybridization
followed by microscopic image analysis is one of the prominent
methods used to study gene expression on a single cell level.
There are various microscopic image analysis methods, leading to
differing mRNA spots being detected in images for the same
experiment. We present a technique to evaluate different mRNA
spots detection algorithms. It is based on image annotation by
expert biologists and the receiver operating characteristics. The
detection methods can be compared using parameters that
withstand imprecise and imbalanced environments. The
proposed evaluation procedure highlighted the difference
between two microscopic image analysis methods that are
frequently used. It can be applied to any image analysis method
that seeks to find mRNA spots on a single cell level.
Keywords—sm-FISH; spot detection; receiver operating
characteristics; F-measure
Gene expression is studied more and more on the single cell
level [1]. One of the methods used to provide mRNA counts in
individual cells is single molecule fluorescence in situ
hybridization (sm-FSIH) followed by microscopic image
analysis [2].
Single molecule FISH is a microscopy-based assay that
allows for the visualization, detection and localization of
specific nucleic acid sequences in their native environment.
Since its origins, over 20 years ago [3], it has become a
powerful molecular tool for the detection of cytogenetic and
molecular genetic alterations. Applications of FISH have even
extended to clinical diagnosis chromosome analysis [4]. In a
molecular setting, FISH has revealed insights in transcriptional
dynamics [5, 6], mechanisms of RNA synthesis [2] and
transport [7] and intracellular distribution [8,9].
The first application of fluorescent in situ detection
involved the use of RNA probes directly labelled on the 3’end
with a fluorophore to bind specific DNA sequences [10]. The
labelling of probe sequences developed to use fluorophore-
coupled amino-allyl modified bases [11] and the use of
enzymatic incorporation of fluorophore-modified bases [12].
These advances in the technology allowed for the simple
chemical production of an array of low-noise probes. Attempts
to improve signal output of this assay came in the form of nick-
translated, biotinylated probes, which were indirectly detected
using fluorescently labelled streptavidin conjugates [13].
Currently, the standard FISH probe is produced by simple
esterification chemistry to couple fluorophore to a 3’amine-
modified base [14]. This method of probe preparation allows
for precise and direct detection with high signal-to-noise ratios,
improving the sensitivity of the assay.
Initially, RNA detection using FISH was constrained to use
of large oligonucleotide probes. This was problematic as large
probes could adhere to samples non-specifically resulting in
false positives as well as lead to high levels in background
noise. The use of reduced probe sizes lead to improved signal-
to-noise-ratio and sensitivity, allowing for the single-copy
detection of RNA entities and even parts of RNA [15, 16]. In
this variation of the assay, 5 oligonucleotides, each about 50
oligonucleotides long, were labelled with fluorophore moieties.
The hybridization of these probes to their mRNA targets
yielded each target to be visualized as a diffraction-limited
fluorescent “spot” [16]. However, the synthesis and
purification of a small number of heavily labelled probes came
with high difficulty and these probes tended to interact with
each other altering hybridization characteristics which lead to
severe quenching [17]. An improvement of the assay was
made by using a tandem array (12-48) of reliably and singly
labelled probes to accurately detect individual mRNA
molecules at high spatial-temporal resolution. This
advancement in the assay has lead to the simultaneous and
accurate detection of multiple targets using spectrally distinct
fluorophores within the same sample [18,19].
Post image acquisition, Femino et al. [16] used a
constrained deconvolution algorithm to quantitatively restore
out-of-focus light to its original points of origin. They could
then calibrate for the fluorescent output per molecule of probe.
In [19], calibration of fluorescent output per molecule of probe
was not performed, however for 48 probes per mRNA they
detected the same number of mRNA spots per image over a
broad range of thresholds, validating the choice of a threshold
parameter. Additionally, they avoided the difficulty in
synthesizing and purifying heavily labelled probes.
Raj et al. [19] used the Laplacian of Gaussian filter to
remove the non-uniform background and enhance particles.
The resulting image conserves spatial resolution of spots, so
does the wavelet transform based filtering as used in [20]. The
procedures are computationally less expensive than constrained
deconvolution algorithms; so is the procedure proposed by
Trcek et al. [21] spatial band-pass filtering and local
background subtraction to remove residual unevenness in the
There are different thresholding techniques that are applied
to a filtered image to eventually find spots [16,19,20,21]. Raj
et al. [19] chose a threshold from a range of thresholds for
which the number of mRNAs detected varied the least. Trcek et
al. [21] used Gaussian mask fitting to find the centre and
intensity of each spot. In any case, the detected spots can be
analysed on a per cell basis if the cell marker is used in an
We present an evaluation of individual mRNA molecules
detection techniques in microscope images. The evaluation
procedure is applied to two detection techniques. It is based on
the use of expert biologists as the gold standard in marking
spots in a microscope image. The evaluation procedure uses the
receiver operating characteristics analysis (ROC) and
performance evaluation metrics used in machine vision and
The organisation of this paper is as follows. The next
section outlines the method of evaluating detection techniques
(methods used to prepare mRNAs are in supplementary data).
Then detection techniques evaluation results are presented.
A. Spots validation
Spots found in a z-stack image by an expert biologist
constituted the gold standard used in evaluating the
performance of a detection algorithm on that stack. Biologists
circled all mRNA spots they could find using a custom made
GUI. Hausdorff distance [22] was used to study intra- and
inter-observer variability in marking spots and compare that to
detection algorithms’ found spots; the modified Williams index
(MWI) [23] was obtained from the Hausdorff distances to
further compare algorithms’ spots boundaries to hand drawn
ones. The index is the ratio between the average computer-
observer agreement and the average observer-observer
agreement. For N observations, MWI is calculated leaving one
observation out at a time, for N-1 observations, resulting in N
B. Detection techniques evaluation
The posterior probability of a detected spot was calculated
by finding the ratio of pixels found by both an algorithm and an
expert to pixels found by an expert; minus fraction of pixels
missed or over-segmented by an algorithm. Background pixels
were regarded as non-target objects. The ROC curves were
plotted using spots as the target class. The area under the ROC
curve (AUC) is used as an evaluation value integrating the
entire ROC. Sensitivity and specificity, typical two-class
detection performance evaluation measures, could be
established from the ROC curve at a chosen operation point.
Since the non-target class far exceeds the target class, the
posfrac-recall ROC [24,25] was used to evaluate detection
algorithms, as this is the imbalanced problem. The prior
probability of the positive class is significantly less than that of
the negative class, their ratio skew, was used to study what
fraction of non-target objects to include in the analysis. Typical
imprecise environment detection evaluation measures can then
be used to compare detection algorithms at one operating point:
posfrac fraction of positive detections (1),
posfrac =TP +FP
precision (2) the fraction of positive detections that are
actually correct and it is usually a meaningful parameter when
detecting rare events because it effectively estimates an overall
posterior probability [25],
precision =TP
recall and F-measure (3) the geometric mean of precision and
recall [25].
denotes the test objects labeled as target and
are truly targets, while
denotes false targets.
- recall,
are calculated by normalizing
by the
total number of positive and negative objects respectively,
is the sum of positive and negative objects.
sensitivity while 1
denotes specificity.
Fmeasure =2TPr
A. Spot validation
Spot validation was studied using a set of 10 z-stack
images. In each stack, the plane that showed spots the most
clearly was chosen. The similarity of spots marked by the two
expert biologists was studied on spots contours extracted using
the custom made GUI. The comparisons in Table I were made
using the Hausdorff distance. T11 and T12 represent the first
expert marking spots the first and second times, more than a
week apart, T2 represents the second expert. AL1 represents
spots detected using the image analysis procedure outlined in
[19], while AL2 represents spots found using wavelets-based
detector [20].
The first expert had the highest intra-observer variability,
4.5518. There was the highest dissimilarity in the ellipses
drawn around spots. The variability is further confirmed by the
standard deviation of the Hausdorff distances between the first
and second times the first expert marked the spots, it is the
highest. The second expert still had high intra-observer
variability, although it was not higher than inter-observer
variability. The standard deviation of inter-observer variability
is the second highest, elucidating the difference in marking
spots between the two experts.
The mean Hausdorff distances between first round of spot
marking by experts and automated detection procedures were
lower than those between and among experts; prompting a
suspicion than maybe experts marked spots differently the
second time, a week later. The Hausdorff distances between
both LoG-based and wavelets-based detections and experts the
second time they marked spots were calculated, Table II.
Instead of experts marking spots differently the second
time, Table II suggests that the first expert has higher
variability in marking spots than the second expert. This is
because variability between the second expert marking spots
the second time and automated detections is stable when
compared to that expert the first time and automated detections.
This observation suggests that the first expert is the source of
variability. The low Hausdorff distances between the first
expert the first time and automated detections imply that
though the first expect had the highest overall variability, the
first expert had high variability the second time they marked
Table II further shows that spot contours found by the
wavelets-based method agree better with experts than those
found using LoG-based method, as this was established in
Table I. Fig. 1 shows typical spots marked by the first expert
side by side with those detected by the two methods. If the first
expert had the highest variability in marking the spots, yet
visually that experts spots marking look consistent then it can
be concluded that the two experts marked spots similarly. Spots
detected by automated detections visually have contours that
differ from those of experts, however are acceptable as
Hausdorff distances for 10 stacks are comparable to those of
The set of expert markings comprised four observations per
object; two experts marked spots twice. The value of the MWI
for the LoG based method was 1.0094; its 95% confidence
interval, assuming the standard normal distribution, was
(1.0070, 1.0118). The value of the MWI for the wavelets based
method was 1.0172; its 95% confidence interval was (1.0148,
1.0196). The upper limit of the confidence interval for both
methods is greater than one, indicating that the methods agree
with the experts at least as well as the experts agree with each
Figure 1. A shows original image, B is spots marked by the first expert the
first time, C highlights those detected by the LoG-based method, D shows
spots marked by the expert the second time and lastly E is spots found using
wavelets-based method.
B. Detection techniques evaluation
Fig. 2 shows ROC plot for both methods using objects on a
z-stack level deemed the most in focus visually. Spots marked
by an expert constituted the gold standard. The AUC for the
LoG-based method was 0.7751, while that of the wavelets-
based method was 0.6070. The LoG-based method had a higher
AUC value; over a range of posterior probabilities cut-offs it
had better performance than the wavelets-based method.
For each method, at the operating point corresponding to
posterior probability threshold set at 0.5, Table III shows the
performance evaluated using parameters deemed suitable for
imprecise environment. Sensitivity versus specificity was
considered not informative enough, as the two classes were
Even though the LoG-based method had the highest AUC
value, it is less precise than the wavelets-based method at the
operating point chosen. Precision, what fraction of detected
spots are actually spots should be an important measure in
evaluating detection algorithms as noise frequently increases
the false positive detections. The gain in precision came at the
loss in sensitivity recall. Sensitivity fell by 10% for an
increase in precision of 20.50%. The wavelets-based method
picks up a lot less non-spots, a quarter of those by LoG-based
method, objects at the expense of missing a few true positive
spots. This leads to the implication that maybe the normal ROC
is not suitable for this problem; the posfrac-recall ROC could
offer better performance evaluation.
Figure 2. Example The ROC curves for the LoG-based and wavelets-based
Figures 3 and 4 show posfrac-recall ROC curves for the
LoG-based and wavelets-based methods respectively, for the
target prior probabilities = 0.5, 0.1 and 0.01. The prior
probability of the non-target class was varied by varying the
fraction of background pixels from the gold standard image
considered as the non-target objects.
The posfrac-recall curves indicate that the two methods
have similar performance with varying skew values. The
choice of skew, fraction of non-target objects to include in
evaluating a method, depends on the percentage of posfrac
deemed acceptable in detecting spots in an application. The
posfrac of both methods lowers with increasing skew for a set
sensitivity. However, precision is fixed as skew varies.
Figure 3. The posfrac-recall ROC curves for the LoG-based method.
Figure 4. The posfrac-recall ROC curves for the wavelets-based method.
According to posfrac-recall curves, the LoG-based method
has better overall sensitivity; but since it is less precise than the
wavelets-based method its posfrac is high due to high false
positives. Table 4 shows the posfrac of the two methods at 80%
sensitivity with varying skew; it also shows their AUC. At
80% sensitivity, the wavelets-based method has lower posfrac
for all skew values.
However, above maximum sensitivity of the wavelets-
based method, its posfrac significantly surpasses that of the
LoG-based method. That is confirmed by the AUC values
LoG-based method values are consistently higher than those of
the wavelets-based method. The choice of the skew value and
sensitivity at which to operate depends on the problem being
investigated. If a method that finds all the spots, even at an
expense of including background noise is desired, the high
posfrac value can be ignored.
When the spot detection algorithms performance evaluation
is treated as an imbalanced case problem, the posfrac-recall
curves can be used to help decide at what skew and sensitivity
different methods can be compared. This is appropriate because
the distribution of spots to be detected is not known a priori.
The methods are evaluated on a per stack basis, but the
evaluations can be conducted on a batch of stacks of images.
Spots can be detected in 3D or maximum projections of stacks,
the evaluation metrics proposed would still hold. The
0 0.2 0.4 0.6 0.8 1
1 ï Specificity
ROC curve
0 0.2 0.4 0.6 0.8 1
Posfrac recall curve
0 0.2 0.4 0.6 0.8 1
Posfrac recall curve
evaluation metrics can be applied to other spot detection
algorithms, not just the two tested here.
We have proposed a procedure to evaluate performance of
spot detection algorithms in microscope images. The procedure
depends on the marking of spots in images by an expert
biologist. The marked spots form a gold standard in
determining accuracy of an algorithm in imprecise and
imbalanced environment. This methodology was demonstrated
on two spot detection algorithms, the LoG-based and wavelets-
based methods. It was able to highlight the differences in
performance between the two methods. It can be applied on
other spot detection algorithms, provided that they seek to find
the entire diffraction-limited spot.
[1] R.D. Larson, H.R. Singer, D. Zenklusen, A single molecule view of
gene expression.” Imaging Cell Biology, vol. 19(11), pp. 630637, 2009.
[2] A. Raj, S.C. Peskin, D Tranchina, D.Y. Vargas, S. Tyagi, Stochastic
mRNA Synthesis in Mammalian Cells.” PLoS Biology, vol. 4(10), pp.
1-13, 2006
[3] E.F. DeLong, G.S. Wickham, N.R. Pace, Phylogenetic stains:
ribosomal RNA-based probes for the identification of single cells.
Science, vol. 243(4896), pp. 1360-1363, 1989
[4] M. Guttenbach, W. Engel, M. Schmid, “Analysis of structural and
numerical chromosome abnormalities in sperm of normal men and
carriers of constitutional chromosome aberrations. A review.” Human
Genetics, vol. 100(10), pp. 1-21, 1997
[5] D.Y. Vargas, K. Shah, M. Batish, M Levandoski, S. Sinha, S.A. Marras,
I.P. Schedl, S. Tyagi, “Single-molecule imaging of transcriptionally
coupled and uncoupled splicing.” Cell, vol. 147(5), pp. 1054-1065, 2011
[6] R.P. Jansen, M Kiebler, Intracellular RNA sorting, transport and
localization.” Nat. Struc. & Mol. Bio., vol. 12, pp. 826-829,2005
[7] Y.D. Vargas, A. Raj, S.A.E. Marras, F.R. Kramer and S.Tyagi,
Mechanism of mRNA transport in the nucleus.” PNAS, vol. 102(47),
pp. 17008-17013, 2005
[8] F.J. Oborra, D.A. Jackson, P.R. Cook, “The path of transcripts from
extra-nucleolar synthetic sites to nuclear pors: transcripts in transit are
concentrated in discrete structures containing SR proteins." Journal of
Cell Science, vol. 115(15), pp. 2269-2282, 1998
[9] C.S. Osborne, L. Chakalova, K.E. Brown, D. Carter, A. Horton, “Active
genes dynamically colocalize to shared sites of ongoing transcription.
Nature Genetics, vol. 36, pp. 1065-1071, 2004
[10] J.G Bauman, J. Wiegent, P. Borst and P. Duijn, A new method for
fluorescence microscopical localization of specific DNA sequences by in
situ hybridization of fluorochrome-labelled RNA.” Expe. Cell Research,
vol. 128(2), pp. 485-490, 1980
[11] P.R. Langer, A.A. Waldrop and D.C. Ward, Enzymatic synthesis of
biotin-labeled polynucleotides: novel nucleic acid affinity probes.
PNAS, vol. 78(11), pp. 6633-6637, 1981
[12] J. Wiegant, T. Ried, P.M. Nederlof, M. van der Ploeg, H.J. Tanke and
A.K. Raap, In situ hybridization with fluoresceinated DNA.” Nucleic
Acids Research, vol. 19(12), pp. 3237-3241, 1991
[13] R.C. Singer and D.C. Ward, “Actin gene expression visualized in
chicken muscle tissue culture by using in situ hybridization with a
biotinated nucleotide analog.” PNAS, vol. 79(23), pp. 7331-7335, 1982
[14] M. Batish, A. Raj and S. Tyagi, Single Molecule Imaging of RNA In
Situ. Jeffrey E. Gerst (ed.), RNA Detection and Visualization: Methods
and Protocols. Methods in Molecular Biology, vol. 714(1), pp. 3-
[15] J.B. Lawrence, R.H. Singer, C.A. Villnaye, J.L. Stein, and G.S. Stein,
Intracellular distribution of histone mRNAs in human fibroblasts
studied by in situ hybridization. PNAS vol. 85(2), pp. 463-467,1988
[16] A.M. Femino, F.S. Fay, K Fogarty and R.H Singer, “Visualization of
single RNA transcripts in situ. Science, vol. 285(5363), pp. 585-590,
[17] A.M. Femino, K.Fogarty, L.M. Lifshitz, W.Carrington, R.H. Singer,
Visualization of single molecule of mRNA in situ. Methods in
Enzymology, vol. 361, pp.245-304, 2003
[18] J.M. Levsky, S.M. Shenoy, R.C. Pezo and R.H. Singer, “Single cell gene
expression profiling. Science, vol. 297, pp. 836-840, 2002
[19] A. Raj, P. van den Bogaard, S.A. Rifkin, A. van Oudenaarden, S. Tyagi,
Imaging individual mRNA molecules using multiple singly labelled
probes. Nature Methods, vol. 5, pp. 877-879, 2008
[20] J.C.Olivo-Marin, Extraction of spots in biological images using
multiscale products. Pattern Recognition, vol. 35, pp. 1989-1996, 2002
[21] T. Trek, A.J. Chao, R.D. Larson, H.Y. Park, D. Zenklusen, M.S. Shenoy,
R.H. Singer, Single-mRNA counting using fluorescent in situ
hybridization in budding yeast. Nature Protocols, vol. 7, pp. 408-419,
[22] D. Huttenlocher, G. Klanderman and W. Rucklidge, Comparing
images using the Hausdorff distances. IEEE Trans. Pattern Anal.
Machine Intell., vol.15(9), pp. 850-863, 1993
[23] V. Chalana and Y. Kim, “A methodology for evaluation of boundary de-
tection algorithms on medical images. IEEE Trans. Med. Imag., vol.
16(5), pp. 642-652, 1997
[24] T. Landrebe, P. Paclik, D.M.J. Tax, S. Verzakov and R.P.W Duin,
Cost-based classifier evaluation for imbalanced problems. Lecture
Notes in Computer Science, vol. 3138, pp. 762-770, 2004
[25] T. Landrebe, P. Paclik, R.P.W. Duin, A.P. Bradley,Precision-recall
operating characteristic (P-ROC) curves in imprecise environments.
18th Inter. Conf. on Pattern Recognition, pp. 123-127, 2006
Methods and Materials
The eGFP gene sequence was found on PubMed and inserted
in 5’-3’ direction into the probe designer algorithm on The parameters set on the
algorithm were as follows:
Number of probes 48
Probe length 20 nucleotides
GC content 45%
No of GFP probes
Lyophilized probes (Biosearch Technologies) were
resuspended in 100 µl of TE (10mM Tris, 1 mM EDTA,
Sigma) buffer (pH 8) to a final concentration of 100 mM each
and stored at -20°C. Equal volumes of thawed probes were
aliquoted (10 mM each) and pooled together for each gene to a
final concentration of 480 mM for genes with 48 probes.
Initially, precipitation was carried out with 10% volume of 3M
Sodium Acetate (pH 5.2, Sigma) and 2.5X volume 100% cold
Ethanol (Minema) according to smFISH protocol by Batish et
al. (2011) Probes were precipitated overnight by incubation at
-20°C. Probes were then spun at 14 500 Xg, 4°C for 20 min.
The pellet was then resuspended in 200 µl 0.1 M Sodium
Bicarbonate (Sigma) or Sodium Tetraborate (Sigma).
Approximately 0.3 mg of ATTO-565 NHS-ester dye (ATTO-
TEC, Germany) was dissolved in 10 µl dimethyl sulphoxide
(DMSO, Sigma). Dissolved dye solution was added to 190 µl
of 0.1 M Sodium Bicarbonate (Sigma). The dye solution was
added to the probe solution and incubated overnight in the
dark at 37°C. Following conjugation reaction, the probes were
reprecipitated at -20°C overnight as previously described.
Probes were then spun at 14 500 Xg, 4°C for 20 min.
Supernatant which consisted of unconjugated dye was
discarded and conjugated probe pellet was rinsed twice with
70% Ethanol at 14 500 Xg, 4°C for 5 min. Supernatant was
discarded and pellet was allowed to air dry. Pellet was
resuspended in 200 µl of Buffer A (0.1 M Triethyl ammonium
(TEA, Sigma)). Conjugated probes were separated and
purified to enrich for dye-conjugated probes by reverse phase
HPLC on a C18 column. Buffer A is the aqueous phase
column which allows sample molecules to adhere to column
and Buffer B (Triethyl ammonium and 70% (v/v) acetonitrile
(Labscan) contains organic solvents in which oligonucleotides
are preferentially soluble. An optimized programme of 2 to
98% Buffer B over 20 min was used to purify probes.
Conjugated probes were detected at two wavelengths, 260 nm
for nucleic acid and corresponding wavelength for dye used
either 565 nm for ATTO-565. The appropriate fractions,
containing conjugated were collected and dried in a Centri-
Vac. Dried probes will were then re-precipitated overnight as
previously described. Probes were then spun down with the
same parameters as previously described. Probes were allowed
to air dry and were re-suspended in a small volume of TE
buffer (pH 8, Sigma). DNA concentrations were then
determined using a Nanodrop. Probes were then diluted to a
final concentration of 50ng and stored at -20°C until
hybridization steps.
Cell Culture
HeLa cells were grown in DMEM (Dulbecco’s Modified
Eagles’s Medium, Gibco) with 10% FBS (Fetal Bovine
Serum, Gibco), 2 mM L-glutamine (Sigma Aldrich) and G418.
Cells were transfected with 1 µg JOMU WT and
LIpofectamine 2000 (Invitrogen) complexes and 1ml Opti-
MEM I Reduced Serum Medium (Gibco). Media was changed
to DMEM after 4 hours and cells were incubated at 37°C and
5% CO2 for 24 hr. Cells were passaged at 1:10 into fresh
growth medium containing kanamycin sulphate (Roche). After
cells had reached 90% confluency, cells were seeded in 12
well plates, each well containing an ethanol cleaned 15mm
coverslip. Approximately 1 X 105 cells were seeded in each
well in 1 ml of media. Cells were grown in a 37°C incubator
with 5% CO2 overnight. Cells were stimulated with 20ng/ml
TNF-α (Tumor Necrosis Factor Alpha, Sigma Aldrich) and
fixed after the following time points: 2hr, 2hr 30min and 3hr.
Cell Fixation
For fixation, culture medium was aspirated off wells and cells
were gently washed 2X with phosphate buffered saline (PBS,
Lonza). 1ml of paraformaldehyde (PFA, Sigma Adrich) was
added to cells and incubated in PFA for at least 10min. PFA
was aspirated off and cells were gently washed 2X with PBS.
Cells were then stored in 70% Ethanol (Minema) at 4°C in
parafilm sealed plates until hybridization experiments.
Probe hydrization and Imaging
Prior to hybridization, cells are gently washed 2X with PBS. A
volume of 50ng of a specific conjugated probe is then added to
hybridization buffer (50% (v/v) deionised formamide
(CalBiochem), 10% (w/v) dextran sulphate (Sigma), 300 mM
NaCl (Sigma), 20 mM NaH2PO4 (Sigma), 2 mM EDTA
(Sigma), 10 µl vanadyl ribonucleoside complex (Sigma), 250
ug/ml E. coli tRNA (Sigma). For each coverslip, 7 µl of
hybridization buffer containing 50ng of probe is used.
Coverslips are then inverted, cell side down, onto 7 µl of
hybridization buffer on parafilm coated glass. Hybridization
was then carried out in 37°C water bath in the dark overnight.
Coverslips were transferred into a 12 and 2X SSC (300 mM
NaCl, 0.3 M tri-sodium citrate, Ambion) at 37°C for 30min.
Wash step was repeated three times in fresh wash buffer. Then
0.125 µg DAPI (Invitrogen) was added 20 min into the final
wash step and incubated under the same conditions for 10 min.
Coverslips were then gently washed 2X in PBS and incubated
with equilibration buffer for 2-5min. Coverslips were then
mounted onto ethanol cleaned coverslips, using glox buffer
containing 3.7 X 10-3 mg/µl glucose oxidase (Sigma) and
164.38U/µl catalase (Sigma) as a mounting buffer. Cells were
imaged on a Nikon widefield TIRF microscope using a 100X
oil immersion objective under lamp illumination. Imaging was
done using mercury lamp illumination through the appropriate
filter sets at low camera gain in each of the fluorescent
channels using an Andor iXion897 camera. The DAPI nuclear
stain was visualized in the 405 channel at 10ms exposure time.
GFP was imaged in the 488 channel with 100ms exposure
time. eGFP mRNA (“spots”) were imaged in the 561nm
channel after 200ms exposure (imaging software, µManager).
JOMU WT Plasmid Map
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
We have used fluorescein-11-dUTP in a nick-translation format to produce fluoresceinated human nucleic acid probes. After In situ hybridization of fluoresceinated DNAs to human metaphase chromosomes, the detection sensitivity was found to be 50 -100 kb. The feasibility and the increase In detection sensitivity of microscopic imaging of In situ hybridized, fluoresceinated DNA with an integrating solid state camera for rapid cosmld mapping is illustrated. Combination of fluoresceinated DNA with biotinated and dlgoxigenlnated DNAs allowed easy performance of triple fluorescence In situ hybridization. The potential of these techniques for DNA mapping, cytogenetics and biological doslmetry is briefly discussed.
Full-text available
Fluorescent in situ hybridization (FISH) allows the quantification of single mRNAs in budding yeast using fluorescently labeled single-stranded DNA probes, a wide-field epifluorescence microscope and a spot-detection algorithm. Fixed yeast cells are attached to coverslips and hybridized with a mixture of FISH probes, each conjugated to several fluorescent dyes. Images of cells are acquired in 3D and maximally projected for single-molecule analysis. Diffraction-limited labeled mRNAs are observed as bright fluorescent spots and can be quantified using a spot-detection algorithm. FISH preserves the spatial distribution of cellular RNA distribution within the cell and the stochastic fluctuations in individual cells that can lead to phenotypic differences within a clonal population. This information, however, is lost if the RNA content is measured on a population of cells by using reverse transcriptase PCR, microarrays or high-throughput sequencing. The FISH procedure and image acquisition described here can be completed in 3 d.
Conference Paper
Full-text available
A common assumption made in the eld of Pattern Recog- nition is that the priors inherent to the class distributions in the training set are representative of the true class distributions. However this as- sumption does not always hold, since the true class-distributions may be dieren t, and in fact may vary signican tly. The implication of this is that the eect on cost for a given classier may be worse than expected. In this paper we address this issue, discussing a theoretical framework and methodology to assess the eect on cost for a classier in imbalanced conditions. The methodology can be applied to many dieren t types of costs. Some articial experiments show how the methodology can be used to assess and compare classiers. It is observed that classiers that model the underlying distributions well are more resilient to changes in the true class distribution than weaker classiers.
Full-text available
This protocol describes a method to image individual mRNA molecules in situ. About 50 oligonucleotides complementary to different regions of a target mRNA species are used simultaneously. Each probe is labeled with a single fluorescent moiety. When these probes bind to their target, each mRNA molecule becomes so intensely fluorescent that it can be seen as a fine fluorescent spot. Several different mRNA species can be detected in multiplex imaging using differently colored probe sets for each species. An automated image-processing program is used to count the number of mRNA molecules of each species that are expressed in each cell, thus yielding single-cell gene expression profiles.
Full-text available
Traditionally, machine learning algorithms have been evaluated in applications where assumptions can be reliably made about class priors and/or misclassification costs. In this paper, we consider the case of imprecise environments, where little may be known about these factors and they may well vary significantly when the system is applied. Specifically, the use of precision-recall analysis is investigated and compared to the more well known performance measures such as error-rate and the receiver operating characteristic (ROC). We argue that while ROC analysis is invariant to variations in class priors, this invariance in fact hides an important factor of the evaluation in imprecise environments. Therefore, we develop a generalised precision-recall analysis methodology in which variation due to prior class probabilities is incorporated into a multi-way analysis of variance (ANOVA). The increased sensitivity and reliability of this approach is demonstrated in a remote sensing application.
Full-text available
We describe a method for imaging individual mRNA molecules in fixed cells by probing each mRNA species with 48 or more short, singly labeled oligonucleotide probes. This makes each mRNA molecule visible as a computationally identifiable fluorescent spot by fluorescence microscopy. We demonstrate simultaneous detection of three mRNA species in single cells and mRNA detection in yeast, nematodes, fruit fly wing discs, and mammalian cell lines and neurons.
We present a new method to detect and count bright spots in fluorescence images coming from biological immunomicroscopy experiments. It is based on the multiscale product of subband images resulting from the à trous wavelet transform decomposition of the original image, after thresholding of non-significant coefficients. The multiscale correlation of the filtered wavelet coefficients, which allows to enhance multiscale peaks due to spots while reducing noise, combines information coming from different levels of resolution and gives a clear and distinctive chacterization of the spots. Results are presented for the analysis of typical immunofluorescence images.
Introns are removed from pre-mRNAs during transcription while the pre-mRNA is still tethered to the gene locus via RNA polymerase. However, during alternative splicing, it is important that splicing be deferred until all of the exons and introns involved in the choice have been synthesized. We have developed an in situ RNA imaging method with single-molecule sensitivity to define the intracellular sites of splicing. Using this approach, we found that the normally tight coupling between transcription and splicing is broken in situations where the intron's polypyrimidine tract is sequestered within strong secondary structures. We also found that in two cases of alternative splicing, in which certain exons are skipped due to the activity of the RNA-binding proteins Sxl and PTB, splicing is uncoupled from transcription. This uncoupling occurs only on the perturbed introns, whereas the preceding and succeeding introns are removed cotranscriptionally. PAPERCLIP:
Analyzing the expression of single genes in single cells appears minimalistic in comparison to gene expression studies based on more global approaches. However, stimulated by advances in imaging technologies, single-cell studies have become an essential tool in understanding the rules that govern gene expression. This quantitative view of single-cell gene expression is based on counting mRNAs in single cells, monitoring transcription in real time, and visualizing single proteins. Parallel advances in mathematical models based on stochastic, discrete descriptions of biochemical processes have provided crucial insights into the underlying cellular mechanisms that control expression. The view that has emerged is rooted in a probabilistic understanding of cellular processes that quantitatively explains both the mean and the variation observed in gene-expression patterns among single cells. Thus, the close coupling between imaging and mathematical theory has established single-cell analysis as an essential branch of systems biology.