Blood reference genes help

[mlbinperth]

Hello,
I am a bit hopeless at understanding the various reference gene selection algorithms..

Thought I would ask a general (but wordy) question while I chew through the information available on this forum

I am trying to find an appropriate set of reference genes for analysing blood from melanoma patients. As blood is typically a difficult tissue I decided to test quite a few markers with 10 samples (5 patient, 5 control) and then go from there...the patient samples were chosen with a range of our putative target expression values and the controls were chosen to roughly match the age and gender of the patient samples.

I have made up a total of 20 primer sets for blood markers (ie CD14, HBB etc), common reference genes (ACTB, GAPDH, HPRT1, HMBS etc) as well of some markers that have come out of microarray analyses of normal and diseased blood (not melanoma) (PPIB, FPGS, DECR1 etc) trying to get a nice mix of message abundance as I am primarily intersted in rare messages but would like to have candidate reference genes with a range of abundance levels

I ran the samples in triplicate with the max and min range of all replicates less than a cycle so I think my replicates are reasonable although more blood sample numbers would be desirable I thought I would drive up my sample numbers once I had my reference genes narrowed down to ~3-4 reasonable ones

I am running a demo of GenEx while I am trying to talk my boss into buying a licence, as well as the excel add ons for GenNorm, Normfinder and Bestkeeper

I have Ct values directly from my qPCR machine (BioRad IQ5) as well as from linreg edit after importing my background subtracted (not baseline corrected) optical values, presumably this program calculates the Ct taking into account individual run efficiencies all runs seemed to have behaved nicely (reasonably reaction efficiencies, nice melt curves) so any wobble is likely due to the nature of blood samples as opposed to poorly designed primers, unclean preps and undo inter run variation

so I input either the Ct values directly (Bestkeeper, GenEx GeNorm, GenEx Normfinder) or linearise (using 2E-Ct) the Ct values, then normalise to the highest expressing sample for each marker and plug those values into the Excel and GenEx versions of GeNorm and Normfinder

Although GeNorm and Normfinder generally agree I get slightly different values and best marker suggestions depending on what inputs I have I guess not surprisingly as all markers seem relatively okay ie my individual Genorm M values and NormFinder Sd values are not all that bad but not fantastic and my accumulated SD curve is very shallow ie the minimum does not occur until ~10 markers even though the top three markers vary depending on how you pre-process your data. Bestkeeper chooses quite different markers and seems to favour markers of higher abundance ???? I have no comprehension of the statistics involved between the programs so I am trying to input my data in as many ways as possible and see how it effects the curves I get and try to get a visual consensus of what is going on without really understanding it !!!

I notice that GenEx allows you several options for linearising and normalising Ct data and these transformations give me different results than if I let GenEx use the straight Ct values and different again from when I plug these values into the ExCel add on versions of GeNorm and NormFinder

So is it a matter of picking one approach or are there any suggestions of what approach might be more valid?

Looking at the Normfinder accumulated SD curve it seems to me that the steep front end (best 1-4) then long flat U shape (5-17) suggests that my top 10 to 15 markers are all reasonable and it doesn't matter which I use but I should try to use 3-4 reference genes

Even if slightly different markers are chosen for the top 2 or 3 the accumulated SD curve looks roughly the same

Any thoughts or suggestions if I am not being too vague???

Cheers
Mark in Perth Western Australia

[Mikael Kubista]

Dear Mark,
GenEx Normfinder and geNorm delivers the same results as the Excel spreadsheets provided with the original publications. If you do not get the same results, I suspect you are entering or pretreating the data wrong in any of the applications. There are example files provided with GenEx try analyzing them with GenEx Normfinder/geNorm and the Excel macros to verify you are handling the data correctly

How do you pre-process the data for geNorm/Normfinder analysis in GenEx? Usually, no pre-processing is needed in GenEx; you can analyze the Cq values directly. Pre-processing is only needed if you are running multiplate experiments and must correct for interplate variation and you think the assays have quite different PCR efficiencies and wants to take that into account.

In general the larger the panel of good reference gene candidates the more reliable will be the prediction. However, the reference gene candidates must all be reasonable, which means they should not be differentially expressed between treatment groups. geNorm identifies poor performing candidates and eliminates them, however, it eliminates all the genes successively. In contrast Normfinder calculates intergroup variations (assuming you have treatment groups) based on which the user decides which, if any, genes should be eliminated. Once poor performing reference gene candidates have been eliminated best reference genes are identified with Normfinder by considering all samples equal (i.e., not taking groups into account). The Normfinder prediction performed this way is more reliable than geNorm, since the ranking is based on all reliable data (data from non-regulated reference gene candidates), while geNorm compares different subsets of the data through the elimination process and as consequence the successive M-values are not comparable (by the way, the M value is actually an approximation of the standard deviation)!

The accumulated standard deviation calculated by Normfinder usually drops initially and then slowly levels of reaching a minimum at a fairly high number of reference genes. This is expected behavior, since the improving effect on SD on every reference gene added to the average decreases with number. In fact, if all reference genes in the panel were equal no minimum would be reached; rather the accumulated SD would continue to decrease eventually asymptotically leveling off. The reason we reach a minimum is that we are including the genes in the order of stability.

Hope it helps (and you get your budget  )!

[mlbinperth]

Thanks so much for your reply Mikael,

I put just the Cq values in GenEx and just to compare I then transformed my Cq values:

2E-Cq = transformed value

I then divided all my transformed values by the transformed value of the highest expressed sample for each marker ie lowest Cq hence highest transformed value

so my transformed and normalised values are all less than or equal to 1 but no negative values

When I put these values into genex and tick/untick the boxes as appropriate and into the VBA/excel versions of geNorm and Normfinder I get the same sort of curves but different M and SD values

is there some better way to transform/normalise my values?

Cheers
Mark

[Mikael Kubista]

Mark,
GenEx calculates correctly Norminder and geNorm (but see Note 3). Test as follows:

Load File TATAA_panel.mdf (located among the example files in the folder reference genes)
Select Normfinder and press Run (without changing anything)
The result is that SD of the genes ranges from 0.0511 (PPIA) to 0.1582 (B2M). For convenience you can keep the spreadsheet on the screen for comparison later.

Next edit the file TATAA_panel.mdf in the GenEx editor
Select Pre-processing/Relative quantities/Minimum to change data to linear scale arbitrarily assigning an expression of 1 to the samples with highest expression (done for each gene separately)
Load the file to GenEx Control Panel by pressing the Load button. You will be asked to save the file, which you do with, for example, the name: “Panel RQ.mdf”.
Select Normfinder. Tick the box “Apply log2” and press Run
The result is the same as above, showing that that Cq values can be analyzed directly.

Note 1: the normalization option chosen when calculating relative quantities in the pre-processing has no influence on the result.

Note 2: The same holds true for geNorm: analysis can be made directly on Cq values. If data are normalized to relative quantities, the option chosen does not influence the result.

Note 3. I have seen that if I change back and forth between projects with data in log and linear scale, occasionally analyzing with geNorm and Normfinder it may happen that GenEx looses track on the scaling and calculates wrong result. This is a bug. I have difficulties reproducing it though. If you have a system where you can reproduce the bug, we appreciate if you mail the data and the procedure you use to support@multid.se and they will fix it. Importantly, in normal use when all data are Cq values (or all data are relative quantities) the bug does not appear!

Good luck!