If you can ensure that

• the amount of cells is equal
• the efficiency of RNA isolation is equal
• the efficiency of the RT is equal
• the amount of cDNA subjected to the rtPCRs is equal

then you can omit any normalization.

However, it's usually rather difficult to ensure all these things. Its easier and safer to to rely on all that and simply normalize to a loading control.

Note that not every housekeeping gene is a valid reference gene (loading control). A reference gene must be expressed in the target cells and its expression must not be altered due to the treatments (that's not neccesarily true for housekeeping genes! Example: Gapdh is regulated by hypoxia and thus not a valid reference gene in experiments involving hypoxia; Actb is regulated upon cell migration, growth and division and is thus not a valid raference gene when such processes are involved).

Ideally use a whole panel on (putative) reference genes. You may so possibly identify genes that are regulated (which you may not use to normalize then) and/or individual (slight) changes in their expressions will likely average out, giving you more reliable results.

Thank you. I'll select the same gene as others did then. You are right, normalization sounds easier, although the crll quantities will be the same amount simply because they will be split into two equal parts.

Since the experimental protocol (that imposes a constant mass of total RNA) makes the data compositionnal, quantified cDNA quantities are related to mass fractions of the RNAs in the original sample, and not to their absolute amounts. Because of that, changes in amount of a single RNA (even not quantified) will change mass fractions of all RNAs of the sample. Hence, in principle, changes in individual quantified cDNA amounts cannot be interpreted.

Normalization to so-called housekeeping genes is the most common way to make results interpretable, and simultaneously controls for the several problems mentionned in the above posts.

You may check the following paper for alternative methods, that strongly lowers the need for control genes, and even more of the assumption that their expression does not change.

https://academic.oup.com/bioinformatics/advance-article/doi/10.1093/bioinformatics/bty629/5053322