Abstract

Laser-scanning microscopy allows rapid acquisition of multi-channel data, paving the way for high-throughput, high-content analysis of large numbers of images. An inherent problem of using multiple fluorescent dyes is overlapping emission spectra, which results in channel cross-talk and reduces the ability to extract quantitative measurements. Traditional un-mixing methods rely on measuring channel cross-talk and using fixed acquisition parameters, but these requirements are not suited to high-throughput processing. Here we present a simple automatic method to correct for channel cross-talk in multi-channel images using image data only. The method is independent of the acquisition parameters but requires some spatial separation between different dyes in the image. We evaluate the method by comparing the cross-talk levels it estimates to those measured directly from a standard fluorescent slide. The method is then applied to a high-throughput analysis pipeline that measures nuclear volumes and relative expression of gene products from three-dimensional, multi-channel fluorescence images of whole Drosophila embryos. Analysis of images before unmixing revealed an aberrant spatial correlation between measured nuclear volumes and the gene expression pattern in the shorter wavelength channel. Applying the unmixing algorithm before performing these analyses removed this correlation.

© 2007 Optical Society of America

Full Article  |  PDF Article

References

  • View by:
  • |
  • |

  1. K. Carlsson and K. Mossberg, "Reduction of cross-talk between fluorescent labels in scanning laser microscopy," J. Microsc. 167, 23-37 (1992).
    [CrossRef]
  2. J. Vassy, J. P. Rigaut, A. M. Hill, and J. Foucrier, "Analysis by confocal scanning laser microscopy imaging of the spatial distribution of intermediate filaments in foetal and adult rat liver cells," J. Microsc. 157, 91-104 (1990).
    [CrossRef] [PubMed]
  3. K. Mossberg and M. Ericsson, "Detection of doubly stained fluorescent specimens using confocal microscopy," J. Microsc. 158, 215-224 (1990).
    [CrossRef] [PubMed]
  4. M. E. Dickinson, G. Bearman, S. Tille, R. Lansford, and S. E. Fraser, "Multi-spectral imaging and linear unmixing add a whole new dimension to laser scanning fluorescence microscopy," BioTechniques 31, 1272-1278 (2001).
  5. H. Shirakawa and S. Miyazaki, "Blind spectral decomposition of single-cell fluorescence by parallel factor analysis," Biophys. J. 86, 1739-1752 (2004).
    [CrossRef]
  6. J. R. Mansfield, K. W. Gossage, C. C. Hoyt, and R. M. Levenson, "Autofluorescence removal, multiplexing, and automated analysis methods for in-vivo fluorescence imaging," J. Biomed. Opt. 10, 41207 (2005).
    [CrossRef] [PubMed]
  7. H. Ahammer, T. T. J. DeVaney, M. Hartbauer, and H. A. Tritthart, "Cross-talk reduction in confocal images of dual fluorescence labelled cell spheroids," Micron 30, 309-317 (1999).
    [CrossRef] [PubMed]
  8. D. ChorvatJr, J. Kirchnerova, M. Cagalinec, J. Smolka, A. Mateasik, and A. Chorvatova, "Spectral unmixing of flavin autofluorescence components in cardiac myocytes," Biophys. J. 89, L55-57 (2005).
    [CrossRef] [PubMed]
  9. C. L. Luengo Hendriks, S. V. E. Ker¨anen, C. C. Fowlkes, L. Simirenko, G. H.Weber, A. H. DePace, C. Henriquez, D.W. Kaszuba, B. Hamann, M. B. Eisen, J. Malik, D. Sudar, M. D. Biggin, and D. W. Knowles, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution I: data acquisition pipeline," Genome Biology 7, R123 (2006).
    [CrossRef] [PubMed]
  10. S. V. E. Keränen, C. C. Fowlkes, C. L. Luengo Hendriks, D. Sudar, D. W. Knowles, J. Malik, and M. D. Biggin, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution II: dynamics," Genome Biology 7, R124 (2006).
    [CrossRef] [PubMed]
  11. K. R. Castleman, Digital Image Processing (Prentice Hall, Englewood Cliffs, New Jersey, 1996).
  12. S. V. Costes, D. Daelemans, E. H. Cho, Z. Dobbin, G. Pavlakis, and S. Lockett, "Automatic and quantitative measurement of protein-protein colocalization in live cells," Biophys. J. 86, 3993-4003 (2004).
    [CrossRef] [PubMed]
  13. D. Demandolx, and J. Davoust, "Multicolor analysis and local image correlation in confocal microscopy," J. Microsc. 185, 21-36 (1997).
    [CrossRef]
  14. E. M. M. Manders, F. J. Verbeek, and J. A. Aten, "Measurement of co-localization of objects in dual-colour confocal images," J. Microsc. 169, 375-382 (1993).
    [CrossRef]

2006

C. L. Luengo Hendriks, S. V. E. Ker¨anen, C. C. Fowlkes, L. Simirenko, G. H.Weber, A. H. DePace, C. Henriquez, D.W. Kaszuba, B. Hamann, M. B. Eisen, J. Malik, D. Sudar, M. D. Biggin, and D. W. Knowles, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution I: data acquisition pipeline," Genome Biology 7, R123 (2006).
[CrossRef] [PubMed]

S. V. E. Keränen, C. C. Fowlkes, C. L. Luengo Hendriks, D. Sudar, D. W. Knowles, J. Malik, and M. D. Biggin, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution II: dynamics," Genome Biology 7, R124 (2006).
[CrossRef] [PubMed]

2005

J. R. Mansfield, K. W. Gossage, C. C. Hoyt, and R. M. Levenson, "Autofluorescence removal, multiplexing, and automated analysis methods for in-vivo fluorescence imaging," J. Biomed. Opt. 10, 41207 (2005).
[CrossRef] [PubMed]

D. ChorvatJr, J. Kirchnerova, M. Cagalinec, J. Smolka, A. Mateasik, and A. Chorvatova, "Spectral unmixing of flavin autofluorescence components in cardiac myocytes," Biophys. J. 89, L55-57 (2005).
[CrossRef] [PubMed]

2004

S. V. Costes, D. Daelemans, E. H. Cho, Z. Dobbin, G. Pavlakis, and S. Lockett, "Automatic and quantitative measurement of protein-protein colocalization in live cells," Biophys. J. 86, 3993-4003 (2004).
[CrossRef] [PubMed]

H. Shirakawa and S. Miyazaki, "Blind spectral decomposition of single-cell fluorescence by parallel factor analysis," Biophys. J. 86, 1739-1752 (2004).
[CrossRef]

2001

M. E. Dickinson, G. Bearman, S. Tille, R. Lansford, and S. E. Fraser, "Multi-spectral imaging and linear unmixing add a whole new dimension to laser scanning fluorescence microscopy," BioTechniques 31, 1272-1278 (2001).

1999

H. Ahammer, T. T. J. DeVaney, M. Hartbauer, and H. A. Tritthart, "Cross-talk reduction in confocal images of dual fluorescence labelled cell spheroids," Micron 30, 309-317 (1999).
[CrossRef] [PubMed]

1997

D. Demandolx, and J. Davoust, "Multicolor analysis and local image correlation in confocal microscopy," J. Microsc. 185, 21-36 (1997).
[CrossRef]

1993

E. M. M. Manders, F. J. Verbeek, and J. A. Aten, "Measurement of co-localization of objects in dual-colour confocal images," J. Microsc. 169, 375-382 (1993).
[CrossRef]

1992

K. Carlsson and K. Mossberg, "Reduction of cross-talk between fluorescent labels in scanning laser microscopy," J. Microsc. 167, 23-37 (1992).
[CrossRef]

1990

J. Vassy, J. P. Rigaut, A. M. Hill, and J. Foucrier, "Analysis by confocal scanning laser microscopy imaging of the spatial distribution of intermediate filaments in foetal and adult rat liver cells," J. Microsc. 157, 91-104 (1990).
[CrossRef] [PubMed]

K. Mossberg and M. Ericsson, "Detection of doubly stained fluorescent specimens using confocal microscopy," J. Microsc. 158, 215-224 (1990).
[CrossRef] [PubMed]

Biophys. J.

H. Shirakawa and S. Miyazaki, "Blind spectral decomposition of single-cell fluorescence by parallel factor analysis," Biophys. J. 86, 1739-1752 (2004).
[CrossRef]

D. ChorvatJr, J. Kirchnerova, M. Cagalinec, J. Smolka, A. Mateasik, and A. Chorvatova, "Spectral unmixing of flavin autofluorescence components in cardiac myocytes," Biophys. J. 89, L55-57 (2005).
[CrossRef] [PubMed]

S. V. Costes, D. Daelemans, E. H. Cho, Z. Dobbin, G. Pavlakis, and S. Lockett, "Automatic and quantitative measurement of protein-protein colocalization in live cells," Biophys. J. 86, 3993-4003 (2004).
[CrossRef] [PubMed]

BioTechniques

M. E. Dickinson, G. Bearman, S. Tille, R. Lansford, and S. E. Fraser, "Multi-spectral imaging and linear unmixing add a whole new dimension to laser scanning fluorescence microscopy," BioTechniques 31, 1272-1278 (2001).

Genome Biology

C. L. Luengo Hendriks, S. V. E. Ker¨anen, C. C. Fowlkes, L. Simirenko, G. H.Weber, A. H. DePace, C. Henriquez, D.W. Kaszuba, B. Hamann, M. B. Eisen, J. Malik, D. Sudar, M. D. Biggin, and D. W. Knowles, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution I: data acquisition pipeline," Genome Biology 7, R123 (2006).
[CrossRef] [PubMed]

S. V. E. Keränen, C. C. Fowlkes, C. L. Luengo Hendriks, D. Sudar, D. W. Knowles, J. Malik, and M. D. Biggin, "3D morphology and gene expression in the Drosophila blastoderm at cellular resolution II: dynamics," Genome Biology 7, R124 (2006).
[CrossRef] [PubMed]

J. Biomed. Opt.

J. R. Mansfield, K. W. Gossage, C. C. Hoyt, and R. M. Levenson, "Autofluorescence removal, multiplexing, and automated analysis methods for in-vivo fluorescence imaging," J. Biomed. Opt. 10, 41207 (2005).
[CrossRef] [PubMed]

J. Microsc.

K. Carlsson and K. Mossberg, "Reduction of cross-talk between fluorescent labels in scanning laser microscopy," J. Microsc. 167, 23-37 (1992).
[CrossRef]

J. Vassy, J. P. Rigaut, A. M. Hill, and J. Foucrier, "Analysis by confocal scanning laser microscopy imaging of the spatial distribution of intermediate filaments in foetal and adult rat liver cells," J. Microsc. 157, 91-104 (1990).
[CrossRef] [PubMed]

K. Mossberg and M. Ericsson, "Detection of doubly stained fluorescent specimens using confocal microscopy," J. Microsc. 158, 215-224 (1990).
[CrossRef] [PubMed]

D. Demandolx, and J. Davoust, "Multicolor analysis and local image correlation in confocal microscopy," J. Microsc. 185, 21-36 (1997).
[CrossRef]

E. M. M. Manders, F. J. Verbeek, and J. A. Aten, "Measurement of co-localization of objects in dual-colour confocal images," J. Microsc. 169, 375-382 (1993).
[CrossRef]

Micron

H. Ahammer, T. T. J. DeVaney, M. Hartbauer, and H. A. Tritthart, "Cross-talk reduction in confocal images of dual fluorescence labelled cell spheroids," Micron 30, 309-317 (1999).
[CrossRef] [PubMed]

Other

K. R. Castleman, Digital Image Processing (Prentice Hall, Englewood Cliffs, New Jersey, 1996).

Cited By

OSA participates in CrossRef's Cited-By Linking service. Citing articles from OSA journals and other participating publishers are listed here.

Alert me when this article is cited.


Figures (5)

Fig. 1.
Fig. 1.

Overlapping emission spectra of three fluorescent dyes. The shaded areas indicate the wavelength intervals that are acquired in each channel. Note how the channel recording the Sytox Green signal also records the tail of the Coumarin signal, but the Sytox signal is minimal within the Coumarin acquisition window. In the same way, the Sytox signal bleeds through to the Cy3 channel, but not the other way around.

Fig. 2.
Fig. 2.

Comparison of the automated unmixing method with the measured cross-talk using a standard test slide of bovine pulmonary artery endothelial cells. (a) The measured green image (F-actin, BODIPY) and (b) the measured red image (microtubules, Texas Red) were recorded using simultaneous excitation at 488 nm and 543 nm. The gain and offset of each channel were independently set to fill the 12 bit dynamic range of the images. (c) The image of the measured cross-talk from the green to the red channel was then recorded in the red channel using only green dye excitation at 488 nm, and using the same gain and offset as that for the measured red image. (d) The measured pure red image was recorded in the red channel using only red dye excitation at 543 nm, again using the same detector settings. (e) The estimated cross-talk image and (f) the estimated pure red image calculated from the measured green and measured red images shown in (a) and (b). The bar in (a) is 50 µm. The same, small gamma change has been performed on all images to enhance the contrast in the dark areas and thus make the cross-talk better visible.

Fig. 3.
Fig. 3.

Joint histograms are used to estimate cross-talk levels. (a) Red-green joint histogram before and (b) after unmixing with the proposed algorithm. The crosses are the points detected by the algorithm on each horizontal line through the histogram. The dashed line is the fit through these points. The offset along the x-axis (red channel) of this fitted line was a result of the autofluorescence, which had a different strength in the two channels, and was ignored. The continuous line, going through the origin, is of the same slope as this fitted line, and corresponds to the vertical axis in the corrected histogram (b). (c) Joint histogram of the measured green image versus the green-to-red cross-talk. Notice that the red channel axis offset is less than in (a). This is because the 488 nm laser excites less autofluorescence than the 543 nm laser. (d) Red-green joint histogram using the pure red image, which shows the expected shape for the histogram plotted in (b). (e) Joint histogram of the measured red image versus the measured pure red image, using only pixels with green intensity of 2000 or more. The cross-talk shifts the plot off the diagonal towards the right. (f) Histogram as in (e) but using the estimated pure (unmixed) red image rather than the measured red image. The algorithm was able to shift the plot back towards the diagonal. The white dots in (e) and (f) are the center of mass of the pixel data.

Fig. 4.
Fig. 4.

Application of the method to a high-throughput image analysis pipeline. (a) An optical slice through the middle of a 3D confocal image of a fruit fly embryo, stained for DNA with Sytox Green, for ftz mRNA with Coumarin (blue) and for eve mRNA with Cy3 (red). The white rectangle indicates the region of interest used for (b)–(f). (b) Region of interest from the blue channel, (c) the green channel and (d) the red channel as measured. (e) The image from the green and (f) red channel after unmixing using the proposed algorithm. (g) Joint histogram of the blue channel versus the green channel. (h) Joint histogram of the green channel versus the red channel. (g) and (f) show the points detected by the algorithm (crosses) and the linear fit through these points (dashed line). (i) Joint histogram of the blue channel versus the green channel after automatic unmixing. (j) Joint histogram of the green channel and the red channel after automatic unmixing. Comparison of (b) with (e) and (c) with (f) shows how images in the green and red channels are improved after unmixing. The bar in (b) is 20 µm. The same, small gamma change has been performed on (b)–(f) to enhance the contrast in the dark areas and thus make the cross-talk better visible.

Fig. 5.
Fig. 5.

Channel unmixing removes the effect of cross-talk and increases the accuracy of analysis results. (a) A maximum intensity projection along the optical axis from a single two-photon fluorescence image of an embryo stained for eve with Coumarin. DNA is shown in green. (b) An image of an embryo stained for ftz with Coumarin. (c) Plot of averaged eve (yellow line) and ftz (purple line) mRNA expression levels along the anterior/posterior axis on the embryo. (d) Measured nuclear volumes when either eve or ftz is stained for in the Coumarin channel. (e) Measured nuclear volumes from the same images after fully automatic channel unmixing using the algorithm presented here. Plots in (c)–(e) are the averages of the data from a cohort of embryos. The vertical dotted lines in these plots, at the location of maximum eve intensity, indicate the correlation between expression level and measured volumes. (d) and (e) also show the corresponding 95% confidence limits.

Equations (1)

Equations on this page are rendered with MathJax. Learn more.

S 1 =         a 1 f 1 + c 2 , 1 a 2 f 2 + c 3 , 1 a 3 f 3 + g 1 S 3 = c 1 , 2 a 1 f 1 +         a 2 f 2 + c 3 , 2 a 3 f 3 + g 2 S 3 = c 1 , 3 a 1 f 1 + c 2 , 3 a 2 f 2 +         a 3 f 3 + g 2 .

Metrics