COLOCALIZATION ANALYSIS

+ =

Center for Microscopy and Image Analysis

Urs Ziegler (ziegler@zmb.uzh.ch) Caroline Aemisegger (Aemisegger@zmb.uzh.ch)

ColocalizationThe presence of two or more structures on the same location.

Colocalization is always relative!The nearer the objects the higher the chance that they interact.

Colocalization in fluorescence microscopy at subcellular level

Definition: The presence of two or more fluorochromes on the same physical structure in a cell.

Limitation in best case scenario: optical resolution of microscope (Light microscope: 200 x 200 x 400 nm)

Colocalization never measures interaction, itjust states that two dyes are close in a definedvolume.

www.olympusconfocal.com/applications/colocalization.html

Guidelines for preparation and acquisition of samples for colocalization analysis

• Choice of fluorochromes: not too close, not too far (bleedthrough, chromatic aberration)

• Controls: for positive and negative colocalization, autofluorescence, single labelled controls

• Confocal; Widefield/ Deconvolution

• Appropriate objective (high numerical aperture (leads to small detection volume), highly corrected)

• Avoid bleedtrough/ crosstalk

• Minimize noise (Confocal: averaging/low speed; Widefield: camera, optimize exposure time)

• Dynamic range: use whole range, no saturation

• Proper sampling frequency (Nyquist sampling: about 3 pixels over resolution distance)

FITC/Cy3

Analysis Tools

Visualization Quantification

RGB overlay

Imaris, ImageJ,Leica, Photoshop

Intensity based Object based

Correlation of the strength of linear relation between two channels.

Imaris, ImageJ Imaris, ImageJ

Structure identification anddetermination of overlap of objects.

Colocalization visualizationgreen + red = yellow ?

Don`t trust your eyes in colocalization.Never overlay 3D projections.

Bolte et al., Journal of Microscopy 2006,Vol. 224, 213

Analysis Tools

Visualization Quantification

RGB overlay

Imaris, ImageJ, Leica, Photoshop

Intensity based Object based

Correlation of the strength of linear relation between two channels.

Imaris, ImageJ Imaris, ImageJ

Structure identification anddetermination of overlap of objects.

Intensity correlation based analysis

• The scatterplot/ 2D histogramm:

• The coefficients:

Pearson coefficient

Manders coefficients

Costes approach

Software: Imaris, ImageJ and others

LIVE DEMOIntensity based analysis

Data sets : M.Walch, U.Ziegler et al., Uptake of Granulysin via Lipid Rafts Leads to Lysis of IntracellularListeria innocua.The Journal of Immunology, 2005,174:4220-4227.

Walch et al, J. Immunol.,2005,174:4220

The scatterplot

Overlay green/red

Good first visual estimate of colocalization. Information about image quality.

☺

Bolte et al., Journal of Microscopy 2006,Vol. 224, 213

Only qualitative correlation.

Quiz: Which scatterplot belongs to which image series?

1 2

3 4

Result: A3, B1, C4, D2Bolte et al., Journal of Microscopy 2006,Vol. 224, 213

Rr= 1: perfect colocalizationRr= 0: random localizationRr= -1: perfect exclusion

The Pearson Coefficient (PC): The Formula

Understanding the Pearson Coefficient

PC:-0.108 PC:0.169

PC:0.446 PC:0.446 PC:0.446 PC:0.228

The PC is not dependent on a constant background and on image brightness.☺The PC is not easy to interpret and affected by addition of non-colocalizing signals. No perspective of both channels.

Manders coefficients

M1, M2 coefficient:

(Manders et al. , 1993)

Proportion of overlap of each channel with the other.

M = 1: perfect colocalizationM = 0: no colocalization

Understanding the Manders coefficients

PC:-0.108 PC:0.169 PC:0.446

PC:0.446 PC:0.446 PC:0.228

☺

PC:0.425

M1:0.000 M2:0.000 M1:0.250 M2:0.250 M1:0.500 M2:0.500

M1:0.500 M2:0.500 M1:1.000 M2:0.163 M1:0.500 M2:0.286

The Manders coefficients are easier to interpret than the PC. They are not sensitive to the intensity of overlapping pixels.

They are sensitive to background. A threshold has to be set!

The influence of noise on the Pearson and Manders Coefficients

PC:0.446 PC:0.437 PC:0.425

M1:0.500 M2:0.500

All these coefficients are influenced by noise.Minimize noise at acquisition and eventually deconvolveyour dataset prior to analysis.

Deconvolution improves colocalization analysis of multiple fluorochromes in 3D confocal data setsmore than filtering techniques. L. Landmann. Journal of Microscopy 208:2, 134 (2002).

PC:0.422 PC:0.398

M1:0.490 M2:0.260 M1:0.490 M2:0.180

A statistical approach: Costes method*

(A) Estimation of automatic threshold

(B) Statistical significance:

Compares PC for no-randomised with randomized images and calculates significance.

*

Costes method: Randomization

https://info.med.tu-dresden.de/MTZimaging/

17/40 pixels overlap.Significant or random?

If >95% of random images correlate worse than real image, we can trust the correlation coefficient.

Summary Intensity based analysis tools

-sensitive to noise and background-threshold has to be set!

- independent on brightness of image- good indication of the contribution of each channel to the colocalization

M1,M2: proportion of one channel signal coincident with signal in other channel Values from 0 to 1.

Manderscoefficients

-long calculating time (3D)

-statistical approach-minimizes influence of noise

Automatic thresholding,Calculation of significance.

Costesapproach

Correlation of intensity distribution between channels.Values from -1 to 1.

Meaning

-sensitive to noise-only reliable for high correlation- not easy to interpret

-independent on brightness of image- not influenced by constant background

Pearson coefficient

☺

Analysis Tools

Visualization Quantification

RGB overlay

Leica, Imaris,ImageJ, PS

Intensity correlation based Object based

Correlation of the strength of linear relation between two channels.

Imaris, ImageJ Imaris, ImageJ

Structure identification anddetermination of overlap of objects.

Object based analysis

Less dependent on intensities. May be automated.

1. Segmentation: Object / background

2. Connexity analysis: definition of objects

3. Calculation of colocalized volume, area, centroids..

☺Objects need to be segmentable; not for diffuse labelling.

Summary

• Imaging technique:No saturation No bleed throughminimize noise

• Preprocessing of images:Deconvolution, Smoothing (Gaussian or Median Filter)Background subtraction

• Analysis:Combine different methodsAnalyse multiple images

Image J: JACoP Plugin

Links

• Imaris colocalization tutorial:www.bitplane.com/go/web-training/training-archive

• JACoP :http://imagejdocu.tudor.lu/doku.php?id=plugin:analysis:jacop_2.0:just_another_colocalization_plugin:start

• General Colocalization tutorials: http://www.macbiophotonics.ca/PDF/MBF_colocalisation.pdfhttp://www.olympusconfocal.com/applications/colocalization.html

Thank you for your attention!

+ =

www.zmb.uzh.ch