Oncotype DX 21-gene panel: How to do qRT-PCR normalization?

Question

[bill-zt]

Oncotype DX 21-gene panel is a popular panel for breast-cancer recurrent prediction and has been cited ~thousands times (See DOI: 10.1056/NEJMoa041588 ). What confused me a lot is the first step of expression level normalization for 16 cancer-related genes. Almost every literature wrote:

For each sample, normalised expression measurements are calculated as the mean cycle threshold (CT) for the 5 reference genes minus the mean CT of triplicate measurements for each individual gene. Normalized expression measurements are scaled from 0 to 15 units, where 1 unit reflects an ~2-fold change in RNA quantity

Why 0~15? Suppose the mean CT values for 5 reference genes is 28, and the CT for target gene MKI67 is 32. How can I calculate the normalized expression level for gene MKI67 ?

[Kevin Blighe]

I believe the mentioning of the reference-normalised values is more of a finding and that it's not that they have scaled those values to the 0-15 range. That is, after they have subtracted the mean of 3 replicates for each gene of interest from the mean of the 5-panel reference, they made the observation that values ranged between 0-15. They state:

Reference-normalized expression measurements ranged from 0 to 15, where a 1-unit increase reflected an approximate doubling of RNA.

The mentioning of 'doubling' of RNA is purely related to the fact that each cycle difference in PCR reflects a duplication / copying of the original cycle's content.

In situations where a negative reference-normalised value is found, the values can be shifted by a specific factor in order to bring the range to have min=0. The eventual range for your particular data may eventually be 0-8 o 0-23, etc.

------------------------------------------

Note, then, that these reference-normalised values are then used to produce the recurrence score (RS), which is on the range 0-100. As the method is proprietary, they do not go into the stats. However, I imagine that we're talking about a simple regression model here, with the 16-gene panel's reference-normalised values as predictors and some marker of relapse as outcome. Predictions from regression models are on the scale 0-1 (or 0-100%).

Kevin

[qiachen]

Thank you Dr. Blighe, Could you please explain it more clear? What is the value of the specific factor? Is the value different for each sample? Thank you so much!

[Kevin Blighe]

Hey Qian. Thanks for the question. It is just a 'range scaling', whereby they 'compress' or 'expand' the range of the data.

Generate random data

mat <- matrix(rexp(200, rate=.1), ncol=20)

Scale between 0 and 1

mat_range_scaled <- (mat - min(mat)) / (max(mat) - min(mat))

range(mat_range_scaled)
[1] 0 1

Scale between 0 and 15

range(mat_range_scaled * 15)
[1]  0 15

[qiachen]

Thank you Kevin. 'compress' or 'expand' the range of the data? in the original paper: N Engl J Med. 2004 Dec 30;351(27):2817-26.

1. Reference-normalized expression measurements range from 0 to 15, with a 1-unit increase reflecting approximately a doubling of RNA.
2. the recurrence score (RS) is rescaled from the unscaled recurrence score, as follows: RS=0 if RSU<0; RS=20×(RSU–6.7) if 0≤RSU≤100; and RS=100 if RSU>100.

Any comments? Thanks a lot!

[Kevin Blighe]

Sorry, I did not answer about the factor value: if, for example, the range of values is -5 to +10, then we can add a factor of +5 to the values, which will bring the range to 0 to 15.

Reference-normalized expression measurements range from 0 to 15, with a 1-unit increase reflecting approximately a doubling of RNA.

Without the authors showing the actual formula that they used for this 'range', I will not hypothesise further what they did. However, as per my original comment ( above - C: Oncotype DX 21-gene panel: How to do qRT-PCR normalization? ), it could be that the Ct values from their PCR experiment were already measured in the range 0 to 15. On the Ct value scale, each unit increase reflects a doubling of RNA.

the recurrence score (RS) is rescaled from the unscaled recurrence score, as follows: RS=0 if RSU<0; RS=20×(RSU–6.7) if 0≤RSU≤100; and RS=100 if RSU>100.

I do not know much about this - The authors should state from where they derived these values.