FFPE and fresh-frozen samples in different RNA seq batches?

Question

[Axel]

Hello,

We plan to send around 650 tumor samples for RNA-seq. Around 150 of the samples are FFPE samples and the rest are fresh-frozen samples. The samples are old and therefore the RNA quality is far from perfect. We plan to use the Illumina Ribo-Zero plus kit which seems to be suitable for low-quality RNA and FFPE samples. Each batch will contain a maximum of 96 samples. We will extract RNA from all samples and then randomize them into batches to prevent skewness of relevant variables between batches.

I am now wondering if it is preferable to separate the FFPE samples and fresh-frozen samples into different batches due to the large differences in RNA quality, thereby generating more balanced output data from each individual batch. In other words, is it best to randomize the 150 FFPE samples into two batches which will contain only FFPE samples and then randomize the 500 fresh-frozen samples into six batches which will contain only fresh-frozen samples? Or would it be better to combine the samples and then randomize everything, thereby generating a relatively even ratio of FFPE samples to fresh-frozen samples within each batch? We will of course include sample type as a covariate in the contrast matrix when analyzing differentially expressed genes between the groups of interest. To achieve the highest possible power, we would prefer to analyze all samples together.

Thank you in advance!

Axel

Answer 1

[ATpoint]

The preservation method is the known batch variable here, so you should definitely not separate them based on this variable, otherwise samples will be confounded by this and there is no way to later remove that effect via regression. That assumes that you want to analyse all these data as a single dataset. In other words, if you separate FFPE from FF then you later do not know whether differences come from the library prep or from the preservation method. At best you should process samples of both preservation methods in parallel. You have 500 FF vs 150 FFPE so roughly 3:1, and at 96 samples at max per run you could send batches of 72 FF + 24 FFPE in each go. Be sure that for these batches everything is identical, so RNA (if possible) was extracted on the same day and definitely library prep was done in one go. Then you can later use regression to remove the batch effect based on preservation method and analyse samples as one single dataset.

[Axel]

Thank you for a detailed response!