Like any microscopy experiment, **co-localisation** requires experimenters to follow [[https://c4science.ch/w/bioimaging_and_optics_platform_biop/teaching/good_practices/ | Good Practices ]].
= Co-Localization =
We usually distinguish two major schools when it comes to co-localisation: **Qualitative vs. Quantitative**
While the confidence of your results can vary wildly between the two methods, the methodology and approach are the same:
(IMPORTANT) **Biological Question:** You would like to assess if two components are at the same place in the image.
(WARNING) You should then compare two conditions to show a difference between two states (eg. **Together** VS **Independent**).
(NOTE) You must produce controls to assess the staining quality (prove that the signal is not due to [[ TBD | bleedthrough, crosstalk]] or unspecific staning ).
= Qualitative Co-Localisation=
(IMPORTANT) Means **visually** assessing if two components **look like** they are in the same "place" in the image.
From the image set below, some people could try to argue that **Protein X** can be found in **Organelle A**.
Others (like the BIOP) will be more //sceptical// and will ask to compare to a different situation. For example, by adding a drug to increase (or decrease) the co-localisation levels or doing another staining (another organelle).
{F7195478 ,size = full}
**Indeed**, from the image set below, one could easily argue that the **Protein X** is localizing more in Organelle B with respect to Organelle A.
{F7195486 ,size = full}
// It looks like the image we can observe from the //Organelle B// staining is very similar to the image from the //Protein X// staining.
On the contrary, it seems as though almost all the //Organelle A// are devoid of //Protein X//. Even if //Protein X// is (by chance?) found within //Organelle A// at some places, it looks like that //Oganelle B// is much more co-present with //Protein X//.
Quantitative Co-Localisation analysis can help you to decrease the //"by chance?"// factor as well as measure it.
= Quantitative Co-Localisation=
(IMPORTANT) You **measure a parameter**, or some parameters, to **quantify** the degree of co-Localisation of two components.
(WARNING) It is recommended to compare two conditions to show a difference between two states **Together** VS **Independent**.
(NOTE) Controls are required to assess staining quality (proove that signal is not due to [[ TBD | bleedthrough or crosstalk]] ) and are required to do the analysis (depending of the paramater(s) you can use) .
= Pearson's correlation coefficient (PCC) =
(WARNING) It works well for signals with equal (or very similar) Histogram (the distribution of the intensities) and Area coverage within the image.
(IMPORTANT) In EVERY other cases it could lead to a misleading conclusion!
In the figure below, the analysed images (ch.1 VS ch.2) have an equal number of spots.
{F7198970, size = full}
In these cases, one can see that PCC reflects very well the co-localization status of the spots.
For a 100 % co-localisation PCC is equal to 1. The fewer of spots co-localising and the closer to 0.
Note that, if no spots are co-localising PCC is below 0, spots are excluded (they are anti-correlated).
{F7198942, size=full}
**BUT** if the total number of spots are different between the two channels, PCC could lead to mis-interpretation.
One can compare cases AF and AI , which have similar PCC while exhibiting very different scenari.
For me, having 3/4 spots co-localising with 36 other spots , is a very different situtation than having 9/27 still with 36 , but the PCC can not reflect.
{F7198948, size=full}
So, even if we could use PCC to look at Protein-X and Organel-B co-localisation (area coverages are similar)
the Organel-A and Protein-X have area coverages are reallly too different and can't be analysed by PCC.
So one couldn't use PCC to measure Protein-X co-localise more with Organel-A VS Organel-B.
A way to overcome this issue is to use the Mander's coefficients .
= Mander's coefficient =
(IMPORTANT) TBD : link to Mander's article
In the figure below, the analysed images (ch.1 VS ch.2) have an equal number of spots.
In these cases, one can see that PCC reflects very well the co-localization status of the spots so do the Mander's coefficients ( M1 and M2, respectively for ch.1 and ch.2).
{F7198978, size = full}
**BUT** when the total number of spots are different between the two channels, Mander's coefficients appear much more meaningful because they highlight the contribution of each channel into the co-localisation.
Compare AF, AG, AH
{F7198984, size = full}
=Mander's coefficient Drawback=
With "Biological" images (having bakcground, a bit of [[ TBD | bleedthrough or crosstalk]] ), you have to define threshold value for each channels to limit analysis to "True Positive" pixels.
Defining this threshold is not necessary easy but should be doable , and bette than using PCC.
=some refs=
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3074624/