F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48 41
quantiﬁcation of the image (densitometry). Parameters that in 5% acetic acid, or with AmphiGreen Fluo (Biostep, Jahnsdorf
inﬂuence the correlation are: blot transfer conditions, blot Germany) according to manufacturer's protocol.
irregularities (Schilling et al., 2005), concentration of anti-
bodies and ECL substrate, exposure time, camera sensitivity, 2.3.3. Antibodies
color format and adjustments, band deﬁnition and back- In order to be able to investigate proteins of different sizes
ground correction. on the same membrane, the membrane was cut into
This paper analyses some critical aspects of the Western horizontal stripes, see Fig. 2. The stripes were probed using
blot procedure with focus on the kinetics of the antibodies the Western blot protocol from Cell Signaling Technology
and ECL reaction. The systematic analysis of the different (http://www.cellsignal.com/support/protocols).
parameters allows the derivation of a mathematical model of The antibodies against phosphorylated STAT3 (Tyr 705),
the kinetics. This is then used to determine the calibration phosphorylated Akt (Tyr 308) and phosphorylated ERK1/2
curve type and to address some sources of inaccuracies in the (Thr 202/Tyr 204) and secondary horseradish peroxidase-
protocol. A program was written that determines the coupled antibodies were from Cell Signaling Technology.
coefﬁcients of the calibration curve.
2. Materials and methods The immunoblots were incubated with enhanced chemi-
luminescence solution from Cell Signaling Technology for
2.1. Materials 1 min and exposed between 30 and 120 s on a Molecular
Imager ChemiDoc XRS System from Bio-Rad which has a
William's medium E (WME), L-glutamine and dexametha- cooled and calibrated CCD sensor with a linear response
sone were obtained from Sigma-Aldrich (Deisenhofen, Ger- curve. Constant substrate concentration was achieved by
many), fetal calf serum (FCS) from Biochrom (Berlin, taking the pictures while the membrane is completely
Germany), penicillin/streptomycin (10,000 U/ml) from submerged in ECL solution.
Gibco (Paisley, Scotland) and Interleukin-6 from Immuno- ECL images were taken in the native format using the
tools GmbH (Friesoythe Germany). system's standard software Quantity One and then exported
to 16-bit TIFF format.
2.2. Sample preparation
2.3.5. ECL image analysis
For the Western blots described in this paper, protein All images were treated in a way that avoids loss of
extracts from mouse hepatocytes were used. The isolation information. This means: No use of lossy image formats such
and culture of the cells is described in the Supplementary as JPEG; no change of resolution; always use the highest
information. The protein extracts were prepared as follows: possible color depth (32 or 16 bit) to preserve precision; no
The medium was completely removed, the cells were washed overexposure when taking pictures to avoid highlight clip-
twice in ice-cold PBS and lysed with 500 μl lysis buffer per ping and no change to contrast and/or brightness that would
well (25 mM Tris, 150 mM NaCl; pH 7.2) supplemented with result in clipped highlights or shadows; also, no “Gamma”
protease and phosphatase inhibitor cocktails (Sigma- correction, as this would destroy the linearity of the images.
Aldrich). Cells were scraped off on the well and lysates were Exceptions were only made for illustration purposes or in the
transferred to microcentrifuge tubes and kept on ice. During form of non-destructive layers within Photoshop®, in order to
incubation (30 min), the samples were sonicated for 10–15 s. increase visibility of some image features.
Lysates were centrifuged at 13,000 rpm at 4 °C for 20 min in a Analysis of ECL images was performed using the public
table centrifuge to remove cell debris. The cleared super- domain ImageJ program (developed at the National Institutes
natants were transferred to fresh tubes and protein concen- of Health and available at http://rsb.info.nih.gov/ij/), using
trations were determined by BCA assay (Pierce, Rockford, the “Gel Analysis” functions. Background correction was done
USA). using a “rolling ball” method with a radius of 4 times the
The reference sample for the calibration curve was width of a band. Result of the analysis is a value for each band
obtained from 25 ⁎ 106 freshly isolated hepatocytes stimu- which is proportional to the Integrated Density Value (IDV) of
lated in suspension with 20 ng/ml IL-6 for 30 min. that band.
2.3. Western blot 2.3.6. Membrane quantiﬁcation
If a quantitative analysis of membrane staining was done,
2.3.1. SDS-PAGE the image of the stained membrane, as well as a reference
Total cellular lysates (20 μg) and a calibration curve of picture of a white sheet of paper (both with the same lighting
different protein amounts of the reference sample were and aperture setting), were linearised, using the DRI
loaded on a 10% SDS-polyacrylamide gel. (Dynamic range increase) technique: Exposure series were
taken instead of single images and converted to a linear HDR
2.3.2. Blot transfer (High Dynamic Range) picture (Debeverec and Malik, 1997).
The separated proteins were transferred to 0.45 μm HDR conversion was done using the free software Picturenaut
nitrocellulose membrane in cold transfer buffer (48 mM (available from http://www.picturenaut.de). Other software,
Tris, 39 mM glycine, 1.39 mM SDS, 20% methanol) at a such as HDRShop and Photomatix, is also freely available and
constant 45 mW/cm2 for 15 min using a semi dry blot provides equivalent functionality. After linearisation, vignet-
chamber. The transfer was assayed either with 0.1% Ponceau S ting and light distribution were corrected by dividing the
42 F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48
pixel values of the membrane image by those of the reference The membrane is then incubated with the primary
image. This was done directly on the HDR images using the antibody. Some authors suggest that the kinetics of immu-
free, open source software Blender (http://www.blender. noassays is hyperbolic (Studnika, 1987; Braitbard et al., 2006).
org). ImageJ provides the same functionality. The resulting Indeed the reaction between antibody and antigen can be
image was then converted to 16 bit grayscale and quantiﬁed described by the law of mass-action (Ekins and Gosling, 1970;
using Adobe® Photoshop®, using a constant background O'Connor and Gosling, 1997; Holland and Holland, 2003) that
value. The IDV was determined using the “average” ﬁlter on shows similarities to the hyperbolic enzyme kinetics pro-
each area of interest, subtracting the background value from posed by Michaelis and Menten (1913). Thus, antibodies can
the result and multiplying the remainder by the size of the be assumed to have a hyperbolic binding behavior. Eq. (1)
area of interest. This is mathematically equivalent to adding shows a general hyperbolic equation:
the single pixel values in the area. a
CAB1 = f Cprot = CAB1;max − ; ð1Þ
Cprot − Cprot;0
2.3.7. Calibration curve
After quantiﬁcation of the ECL bands, a hyperbolic where CAB1 is the local concentration of bound primary
regression was done on the results from the dilution series, antibody and Cprot the concentration of the protein. CAB1,max is
linking the IDVs to the amounts of protein in the dilution the theoretical maximum saturated concentration of bound
series samples. This was done using the Python script that is antibody. a and Cprot,0 have no direct physical signiﬁcance but
part of the Supplementary material to this paper. The are coefﬁcients that describe the curvature of the hyperbola.
resulting curve was inverted and used to compute an The secondary antibody binds to the primary antibody in
equivalent protein amount for each IDV value of the sample the same fashion, i.e. with hyperbolic saturation. It can be
bands. These values are then proportional to the amount of shown (see Supplementary information) that a hyperbolic
speciﬁc protein in the band. A graphical representation of the function of another hyperbolic function is again a hyperbolic
complete data processing is in Fig. 5. curve, with different coefﬁcients, thus the combined transfer
function of two chained antibodies is still hyperbolic. Fig. 1
shows the inﬂuence of a variation in antibody concentrations
and conﬁrms that in case of visible saturation the behavior
can indeed be represented by a hyperbola.
3.1. Western blot signal chain
The ECL reaction is an enzyme–substrate reaction, where
the substrate is available in abundance and the enzyme is
As in any quantitative assay, a standard curve is necessary
bound to the secondary antibody. The local light emission rate
also in Western blot. From a certain protein amount the curve
(or intensity, I) is therefore dependent on the local concen-
becomes non-linear and saturation occurs. A quantitative
tration of enzyme Cenz (which is proportional to CAB2) and the
determination of proteins in samples is usually done only in
enzyme's reaction rate:
the linear range of the standard curve (Blomberg and Klasse,
1988). I = Cenz · VECL : ð2Þ
In order to extend the useful range of the standard curve,
The reaction rate VECL is determined by the Michaelis–
the type of the standard curve regression must be deter-
Menten equation, given that substrate concentration is much
mined. For this, it is necessary to regard the complete process
higher than enzyme concentration:
chain from initial protein concentration in a sample to the
measured band intensity in the ECL image. Every step with Vmax · Csubst
linear behavior preserves the information on relative protein VECL = ; ð3Þ
Km + Csubst
amounts, while non-linear effects such as saturation can
distort these relations. where Vmax and Km are constants and Csubst is the substrate
For each sample, a known quantity of reference protein concentration. It is noteworthy that the reaction velocity is
solution is pipetted into the gel slots. A certain fraction of this not dependent on the enzyme concentration but only on the
is then transferred to the membrane during the blotting substrate in the ECL solution. If Csubst is constant, the reaction
process. Ideally, the amount of protein in the membrane is velocity is also constant over the complete membrane. If the
proportional to the amount in the gel. There are however non- membrane is only netted with ECL solution, the substrate
linear effects in the blot, caused by non-constant distribution concentration can become inhomogeneous because over
of electric resistance depending on local protein concentra- some parts of the membrane the reagent is consumed faster
tion, size, temperature and possibly other factors (Schilling than elsewhere. Therefore, the images are taken with the
et al., 2005). Since these errors are not systematic, the transfer membrane submerged in ECL solution in order to achieve
function is still assumed to be linear in this paper, although constant substrate concentration during the whole image
some of the errors can be corrected later in the process (see acquisition process. This way, there is an exchange between
Section 3.3). spots with high and low reagent consumption.
Fig. 1. Example blot of three times the same dilution series, incubated with different concentrations of antibody. A: Image of membrane stained with Ponceau S. The
staining was quantiﬁed at the vertical coordinate of P-Stat3 ;. B: Quantiﬁcation results show a linear function, thus the amount of protein at the vertical coordinate is in
fact proportional to the amount of whole protein ;. C: ECL image shows P-Stat3 (Tyr 705) bands ;. D: Quantiﬁcation result of ECL image. Depending on the combination
of antibody concentrations, saturation is visible or not. While hyperbolic regression has the same coefﬁcient of determination in all cases, the linear curve becomes
inaccurate if saturation occurs.
F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48 43
44 F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48
Fig. 2. Western blot analysis of several proteins using one calibration curve. A: Membrane staining with Ponceau S. Twelve samples on the left side and 6 samples
for the calibration curve on the right side. Horizontal lines mark the places where the membrane was cut. B: ECL image of the different proteins.
Like most chemiluminescence systems, the one used for this For the hyperbolic regression, a script was written using the
paper has a calibrated sensor and delivers linear output, so scripting language Python (freely available from http://www.
the pixel values in the image are proportional to the local light python.org) and the add-on module Scipy (also freely available
emission rate I as long as they are not overexposed. The image from http://www.scipy.org). The script takes the protein
quantiﬁcation process then integrates the pixel values over amounts and quantiﬁcation results for the dilution series sam-
each band to give an integral value for the brightness of that ples and constructs a correlation using least-squares regression
band which is proportional to the amount of light emitted by with a hyperbolic function, taking an analytical approximation
the whole band. (Studnika, 1987) as a starting point and reﬁning it by using the
Summing up the complete process chain, the result of ECL Scipy routine leastsq to minimize the sum of error squares. The
band quantiﬁcation is proportional to a hyperbolic function of coefﬁcients for the resulting hyperbola are returned and can
the original amount of protein in the samples. be used in any spreadsheet application to linearise the results
This hyperbolic function can be linear over the relevant from the other samples. The script itself, hard- and software
interval, depending on many factors such as protein amount, requirements and a more detailed description of it are
blotting parameters and antibody concentrations, but if a available as part of the Supplementary information.
dilution series is used to derive the parameters of the The number of samples in the dilution series must be at
standard curve, relative quantiﬁcation can be done also in least four because the hyperbolic function used has three
the non-linear range, up to protein concentrations where degrees of freedom, thus will always match three points with
almost full saturation is reached and the calibration curve perfect ﬁt. This is not desirable because in this case, every
becomes horizontal. error in each of the samples is directly transferred to the
regression curve. The more samples are used the larger the
3.2. Calibration curve statistical signiﬁcance of the regression. This is comparable to
using the mean value of several single values. In this paper, 6
Regarding the composition of the calibration curve, samples were used to derive the standard curve.
several aspects are important.
While any protein solution can be used, it is helpful to use 3.3. Membrane staining
one that contains each of the proteins of interest in sufﬁcient
amount. In this study, a protein sample from the same cell type Quantiﬁcation quality of the membrane staining can be
as the one being analyzed was used. As can be seen in Fig. 2, a greatly improved by correcting the membrane images prior to
single dilution series can also be used to quantify several the actual quantiﬁcations. The procedure is based on the
different proteins from one Western blot, if the protein optical and physical properties of the system.
composition is similar to that of the samples. The range of the The ﬁrst step towards a linear staining image is the
dilution series should be large enough to include all of the linearisation of the sensor response. This is done by taking an
protein concentrations that are expected in the other samples. exposure series (multiple images with different exposure
Slight extrapolation outside of the range of the dilution series is lengths) and combining the images into an HDR image
possible, but accuracy is reduced with increasing extrapolation (Debeverec and Malik, 1997) which is by deﬁnition linear.
distance. In order to make sure that the range is sufﬁcient, an When using a light source, illumination is usually not
advance test blot can be made with a very large dilution series constant over the complete membrane area. Also, most
and only few test samples to determine the lowest and highest camera lenses produce vignetting (i.e. the corners of an
dilution that will be needed. image appear darker). Both effects can be corrected by a so-
F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48 45
Fig. 3. Correction of light distribution and lens vignetting (“ﬂat-ﬁelding”). A: Uncorrected image of a Western blot membrane stained with Ponceau S. B: Reference
(“ﬂat-ﬁeld”) image of a white sheet of paper. C: Corrected image. Brightness values have been adjusted to improve visibility of the brightness gradients in the
original and the reference picture.
called ﬂat-ﬁeld correction (Englert and Harlander, 2006). While quantifying the membrane staining is optional, it
Fig. 3 illustrates the procedure, and detailed information can can be used for several purposes:
be found in the Supplementary information. The ﬂat-ﬁeld
correction used in this paper consists of taking a reference • Controlling the quality of the blot transfer. When using
image of a white sheet of paper and dividing the membrane quantiﬁed results, it is easier to spot some systematic or
image by this reference image. Although for the results in this random errors.
paper the software Blender was used, it was found that ImageJ • To some extent: If blotting and/or pipetting errors are
is capable of performing the division (via “Process → Math → detected, the results of the ECL quantiﬁcation may be
Divide”) also for HDR images (provided they are in gray scale, corrected by using the numbers from the membrane
not RGB mode). Because of the simpler interface, the use of staining as the relative amounts of total protein instead of
ImageJ is recommended. a constant value. However, quantiﬁcation errors for the
It has been found that images acquired with ﬂuorescent membrane image appear to be as high as or slightly higher
staining are better suited for quantiﬁcation compared to than those for the ECL image. One reason for this is that the
Ponceau S. This is connected to the fact that a color staining background is usually less regular than in ECL images, and
such as Ponceau S darkens the membrane locally. Since the background correction using available methods becomes
color of the membrane cannot become darker than that of the less accurate. Therefore, staining quantiﬁcation should only
pure staining color, there must be saturation before that color be used to correct ECL results if it is reasonably certain that
value is reached. This leaves only a certain useful range of the error being corrected is larger than the quantiﬁcation
staining concentrations, independent of whether the actual error for the membrane image.
binding of Ponceau S to the protein is saturated. With • In this paper, the quantitative membrane staining analysis
ﬂuorescent staining, the membrane appears brighter with was also used to produce the results shown in Fig. 1, section B.
increasing protein concentration, and there is no theoretical
upper bound for brightness. Thus, saturation with Amphi- 3.4. Data processing
Green Fluo will only occur if the binding itself is saturated.
AmphiGreen Fluo has been found to have a larger dynamic Fig. 5 illustrates the data evaluation. All bands in the ECL
range (relative intensity quotient between weakest and image (Fig. 5.A) are quantiﬁed. This results in a set of numbers
strongest bands in the linear range) in all cases regarded. representing the integrated density values from the ECL
Fig. 4 shows an example comparison of Ponceau S vs. image for each band. Since for the calibration curve the
AmphiGreen Fluo staining. relative amounts of protein are known, the points can be
46 F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48
Fig. 4. Comparison between Ponceau S (left) and AmphiGreen Fluo (right) membrane staining on the same membrane. A: Membrane pictures, contrast enhanced
to improve visibility. The uniform rectangles mark the areas that were used for quantiﬁcation. They are equal for both images. B: Quantiﬁcation results for both
stainings. Ponceau S shows saturation while AmphiGreen produces a straight line for the same membrane.
drawn into a graph that connects the IDV for each band to the proteins and allows simultaneous analysis of different
amount of whole protein that was used for the dilution series. proteins on the same membrane.
For these pairs of values, a hyperbolic regression is done A computer program was written for determination of the
(Fig. 5.B). The resulting formula then can be used to connect parameters of the hyperbolic regression curve and is available
the IDV for each sample to a protein amount (Fig. 5.C). This as part of the Supplementary information.
protein amount is the amount of total protein from the Summarizing the steps according to our protocol:
dilution series sample that contains the same amount of the
• Loading and blotting of samples and calibration curve
speciﬁc protein. In the case presented here, this value has no
• Membrane staining
direct meaning but is proportional to the amount of speciﬁc
• Optionally: staining quantiﬁcation
protein in the sample. It can be normalized by another value,
• Cutting the membrane into stripes for different proteins
for example from a control sample. If the amount of speciﬁc
• Blocking and incubation with antibodies, ECL
protein in the dilution series was known, the method could
• Densitometry of the ECL image with ImageJ or other
also be used to derive absolute protein amounts.
If visible irregularities have been found in the membrane
• Determination of the hyperbolic regression coefﬁcients
staining, it is possible to reduce them by normalizing each
using the regression script
value by the IDV determined during membrane quantiﬁcation
• Computation of the samples' relative protein content
at the respective sample location. This is equivalent to relating
according to the calibration curve
the amount of speciﬁc protein to the amount of whole protein
• Normalization of the calculated amounts to a house-
on the membrane, thus giving a concentration value that is
keeping protein in the same sample.
independent of the quality of the blot transfer. As pointed out
before, this introduces errors from staining quantiﬁcation into Our results show that depending on the relative concen-
the ﬁnal result, thus should only be used if blotting errors are tration of the antibody and antigen, saturation is present in
large and a repetition of the blot is not practical, and was not the ECL image or not. Therefore, use of a dilution series is
used in the Western blots used for this paper. always recommended in order to determine whether the
analyzed samples are in the linear range. Performing a
4. Discussion hyperbolic standard curve regression on the dilution series
requires no big additional effort and provides valid results
We describe a semiquantitative Western blot method with both in the linear and in the non-linear range of the kinetics,
a hyperbolic ﬁtting of the standard curve which includes the while being both more accurate and reliable. Since the
use of a calibration curve also for relative quantiﬁcation of Western blot procedure itself is unchanged, the accuracy is
F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48
Fig. 5. Overview of ECL image quantiﬁcation. A: A stripe of the blot membrane, containing P-Erk bands from samples as well as that contained in the dilution series. Total protein amounts for the dilution series are given. B: Using
the dilution series, a standard curve is derived by hyperbolic regression. The general formula, the regression results, and the resulting function are shown that link the protein amount P to the measured IDV. C: The inverted
standard curve, links IDV to the equivalent total protein P′. P′ is proportional to the amount of protein in each of the samples, also if the measured IDVs are not.
48 F. Heidebrecht et al. / Journal of Immunological Methods 345 (2009) 40–48
on par to that of other quantitative Western blot methods. In Appendix A. Supplementary data
the blots performed for this study, the mean relative error was
generally around 10%, and standard deviation was between 10 Supplementary data associated with this article can be
and 20%. found, in the online version, at doi:10.1016/j.jim.2009.03.018.
The procedure does not compensate for non-systematic
(or partially systematic) errors due for example to inhomo-
geneities of the blot transfer. A method for this has been References
proposed (Schilling et al., 2005). Using the membrane
staining to correct such blot irregularities and also pipetting Blomberg, J., Klasse, P.J., 1988. Quantiﬁcation of immunoglobulin on
electrophoretic immunoblot strips as a tool for human immunodeﬁ-
errors as described in Section 3.3 could potentially serve a ciency virus serodiagnosis. Journal of Clinical Microbiology 26 (1), 111.
similar purpose, but the inaccuracies of the staining quanti- Braitbard, O., Glickstein, H., Bishara-Shieban, J., Pace, U., Stein, W.D., 2006.
ﬁcation were found to be not much smaller than the errors Competition between bound and free peptides in an ELISA-based
procedure that assays peptides derived from protein digests. Proteome
being corrected, so it is at the moment only recommended as Science 4 (12).
an emergency measure for large and obvious blotting errors. Burnette, W.N., 1981. “Western blotting”: electrophoretic transfer of proteins
The situation could be improved by more reliable quantiﬁca- from sodium dodecyl sulfate-polyacrylamide gels to unmodiﬁed nitro-
cellulose and radiographic detection with antibody and radioiodinated
tion methods for membrane staining, including more accu- protein A. Analytical Biochemistry 112, 195.
rate background correction, or by better staining procedures. Chen, E., Kwon, Y.T., Lim, M.S., Dube, I.D., Hough, M.R., 2006. Loss of Ubr1
Using luminescent staining instead of reﬂective does help, but promotes aneuploidy and accelerates B-cell lymphomagenesis in TLX1/
HOX11-transgenic mice. Oncogene 25, 5725.
by itself is not sufﬁcient to make membrane staining a
Chen, J., Guerriero, E., Lathrop, K., SundarRaj, N., 2008. Rho/ROCK signaling in
universal instrument to reliably increase the overall accuracy regulation of corneal epithelial cell cycle progression. IOVS 49 (1), 175.
of the semiquantitative Western blot. Debeverec, P.E., Malik, J., 1997. Recovering high dynamic range radiance maps
The quantiﬁcation method proposed in this paper is based from photographs, SIGGRAPH '97. Proceedings of the 24th Annual
Conference on Computer Graphics and Interactive Techniques, p. 369.
on the physical properties of the Western blot and therefore is Ekins, R., Gosling, J.P., 1970. Theoretical aspects of saturation analysis. In:
able to deal with some of the systematic errors that appear Diczfalusy, E. (Ed.), Karolinska Symposia on Research Methods in
when using a simpler approach, most notably antibody Reproductive Endocrinology: Steroid Assay by Protein Binding, p. 11.
Englert, C.R., Harlander, J.M., 2006. Flatﬁelding in spatial heterodyne
saturation and errors incurred during image acquisition. spectroscopy. Applied Optics 45 (19), 4583.
Being able to include partly saturated bands greatly increases Feather-Henigan, K., 1999. Immunoblot imaging with a cooled CCD camera
the range of protein concentrations which can be determined. and chemiluminescent substrates. American Biotechnology Laboratory
Taking the optical properties of the camera and lighting Holland, T., Holland, H., 2003. A mathematical model of immunohistochem-
system into account, the requirements to the image acquisi- ical preparations, which provides quantitative predictions. Journal of
tion system can be reduced. Still, of course, the error Microscopy 214, 1.
Kim, T.H., Huh, J.H., Lee, S., Kang, H., Kim, G.I., An, H.J., 2008. Down-
correction introduced by using a calibration curve will not regulation of claudin-2 in breast is associated with advanced disease.
yield useful results for fully saturated bands, and ﬂat-ﬁelding Histopathology 53 (1), 48.
images and using DRI techniques will not neutralize noise, nor Kricka, L.J., 1991. Chemiluminescent and bioluminestcent techniques. Clinical
Chemistry 37/9, 1472.
make weak bands easier detectable if the camera system is
Lin, J.C., Choi, E., Pagano, J.S., 1985. Qualitative and quantitative analyses of
not sensitive enough. Epstein–Barr virus early antigen diffuse component by western blotting
Hyperbolic saturation behavior is often found in biological enzyme-linked immunosorbent assay with a monoclonal antibody.
analytical methods, including antibody and enzyme reactions, Journal of Virology 53, 793.
Michaelis, L., Menten, M.L., 1913. Die Kinetik der Invertinwirkung. Biochem-
thus the use of a hyperbolic curve for standards curve ische Zeitschrift 49, 339.
regression might also beneﬁt the quantitative analysis of O'Connor, T., Gosling, J.P., 1997. The dependence of radioimmunoassay
other methods, such as ELISA, Bradford and BCA (see also detection limits on antibody afﬁnity. Journal of Immunological Methods
Studnika, 1987). The script provided in the Supplementary Ornberg, R.L., Harper, T.F., Liu, H., 2005. Western blot analysis with quantum
information executes a general hyperbolic regression and is dot ﬂuorescence technology: a sensitive and quantitative method for
therefore suited for all of these applications. multiplexed proteomics. Nature Methods 2 (1), 79.
Renart, J., Reiser, J., Stark, G.R., 1979. Transfer of proteins from gels to
diazobenzyloxymethyl-paper and detection with antisera: a method for
Acknowledgements studying antibody speciﬁcity and antigen structure. Proceedings of the
National Academy of Sciences 76 (7), 3116.
Roman-Blas, J.A., Stokes, D.G., Jimenez, S.A., 2007. Modulation of TGF-b
We thank Dr. Juan Cabello Pardos for reviewing the signaling by proinﬂammatory cytokinesin articular chondrocytes.
manuscript, Mario Keller for help with laboratory work, Osteoarthritis and Cartilage 15, 1367.
Andrea Robitzki's lab for sharing their Chemiluminescence Schilling, M., et al., 2005. Computational processing and error reduction
strategies for standardized quantitative data in biological networks. FEBS
System with us and Michael Sonntag from Biostep for lending
Journal 272, 6400.
out UV-lights and green ﬁlter for testing the AmphiGreen Fluo Studnika, G.M., 1987. Hyperbolic regression analysis for kinetics, electrophor-
staining solution. esis, ELISA, RIA, Bradford, Lowry, and other applications. CABIOS 3 (1), 9.
This work was partially supported by the German Federal Towbin, H., Staehelin, T., Gordon, J., 1979. Electrophoretic transfer of proteins
from polyacrylamide gels to nitrocellulose sheets: procedure and some
Ministry of Education and Research (BMBF), Projekt System- applications. PNAS 76 (9), 4350.
biologie; (grant number 0313081).