Differential analysis on RNA-seq : Different results with two "equivalent" contrast matrix

Question

[Nathalie]

Hello,

I am performing differential expression analysis on 3 cell lines A, B and C. A and B are responsive (R) to treatment T and C is non responsive (NR).

I build a design matrix to compare cell line C between control and treatment :

cell_treatment <- factor(dge$samples$cell_treatment)
### ### Levels of cell_treatment: A_ctrl B_ctrl C_ctrl A_T B_T C_T
design0 <- model.matrix(~0+cell_treatment, data=dge$samples)
colnames(design0) <- levels(dge$samples$cell_treatment)

c_matrix <- makeContrasts(
A.ctrl.vs.A.T = A_ctrl  - A_T,
B.ctrl.vs.B.T = B_ctrl  - B_T,
  C.ctrl.vs.C.T = C_ctrl  - C_T,
  levels = colnames(design0)
)

If I do an other analysis by comparing ctrl vs T but based on the responsive, non-responsive (R, NR) status of the cell line, ie, :

status_treatment <- factor(dge$samples$status_treatment)
### ### Levels of cell_treatment: R_ctrl NR_ctrl  R_T NR_T
Design1 <- model.matrix(~0+status_treatment, data=dge$samples)
colnames(design1) <- levels(dge$samples$status_treatment)

c_matrix <- makeContrasts(
R.ctrl.vs.R.T = R_ctrl  - R_T,
NR.ctrl.vs.NR.T = NR_ctrl  -NR_T,
  levels = colnames(design1)
)

I would expect to identify the same genes as differentially expressed for comparison C.ctrl.vs.C.T and NR.ctrl.vs.NR.T as they concern the same non responsive cell line C. However, I don't. I understand that it is because of the design matrix are different and therefore the dispersion estimation and the fit steps lead to different results but how can I choose between these 2 options ? How am I supposed to interpret the results ?

Best regards