Hi Dorota,

I don't think you did anything wrong.

It seems to me that the discrepancy has to do with two issues: a limited number of samples available for the stability calculation and the very stability (or rather lack thereof) of the selected genes in the tested cell lines.

The stability values of 0.221 and 0.129 suggest acceptable but not perfect stabilities. Also, they suggest that the expression levels vary more than you would like them to. This is where statistics come into play: in general, the more variation you deal with, the larger sample you need to get accurate results.

It's no different in this case. 5 cell lines is already a smallish sample to select a reliable reference from a set of quite variable genes. Throwing one out leaves you with 4, which is just 1 more than the minimal set Normfinder can work with. With a fairly high and random variability, it makes sense that adding or removing 1 cell line from the analysis, when all we have is 5, may dramatically change the end result.

My preference would to be to draw conclusions from more rather than less data, if possible.

I will be curious to hear other views, though.

Warm regards and good luck!

R

How did you normalize gene expression, usually relative expression is either fold change or log change.

You don't quiet have "divergent" results here, only WM266-4 seems to have an opposite trend.

Hi Dorota,

I could not see the images for the results, but I can say it is not easy to find a reference gene stable enough to normalize both tumor and normal cell lines. As mentioned above, you did nothing wrong, but you have a small set of conditions to normalize.

My first recommendation would be to use the geNorm approach (now deprecated as a software, but the method still applies), which consists basically on dividing the gene of interest expression value by the geometric mean of the expression of a set of reference genes. The number of genes you will need to normalize everything will depend on how stable are the genes you chose as references.

Another tip would be to use some alternative reference sequences. Please take a look at Genome Biol. 2010; 11(1): R9. They used Alu repeats to normalize qPCR data. I have tested it and it works really well in melanoma cells. You may have to adapt it to the TaqMan method, though.

Good luck with your experiments and let us know what you decide to do.

Hi Dorota

That is my best algorhytm:

1. Finding of candidate genes for example some classic references (ACTG, GAPDH, HPRT etc.) and some unusual genes (SCARNA5, RNU12 etc.).

2. Analize mRNA expression in expreimental and control tissue. Fof me tumor and morphologicaly unchanged microenvironment. Usually I use about 30-50 tissue samples with accurate quality control of RNA integrity (see MIQE guidelines). It is preferable to use Agilent Bioanalizer but if you can't, you should use simple total RNA electroforesis in agarose (2%) to see 18S and 28S rRNA subunits and estimate the results.

3. Using web-based tool RefFinder or similar you should estimate expression stability of candidate genes, and than choose at least two most stable. It is would be great if you receive Cq standard deviation about 1, but it is quite rare event in my practice.

4. It is crucial to define correlation of choosen reference genes with clinical and morphological parameters of the sample. If you find some significant correlation it is very bad. You should find new reference pair.

Of cource it is very interesting to determine phenotypic heterogeneity of mRNA expression within each tissue sample...

You can find the short description in attachet presentation

Good luck