How To Compute Local Rna Duplex Stability At Variable Temperature (Other Than 37 Degr.)?

Question

[Michael]

siRNA's are 19-21 long RNA duplexes with 2 base 3' overhangs present in the RNA-interference pathway. I would like to calculate the local internal stability of several RNA duplexes in siRNA, as described in Khvorova et al. (2003) based on (delta G) calculations according to Freier et al (1986). I also want to calculate local stability at other temperatures than 37 degrees (not every bug lives at 37 degr. C) to achieve plots similar to Figure 2 in Khvorova et al. How well does the stability calculation transfer?

In Kvorova et al., they used the proprietary OLIGO software for this purpose. Is there an open-source alternative to this procedure (Khvorova et al.):

The Oligo 5.0 Primer Analysis Software (National Biosciences, Inc., Plymouth, MN) was used for calculations of internal stability profiles using the free energy values in Table 1 (−kcal/mol, adjusted to 25°C).

... The internal stability values reflect the stability of pentamer subsequences within the sequence under investigation and were calculated according to the nearest neighbor method (Freier et al., 1986). For the calculation of the average internal stability values of the terminal four bases on the 3′ end of the molecule, the 19 bp targeting sites were extended based on the target mRNA or miRNA precursor sequences.

I guess the calculation should be rather basic, therefore a formula for the nearest neighbor method of Freier should be sufficient.

Sorry if this is a very basic and partially biochemistry/biophysics question. Edit: Seemingly, it is not at all as easy as I thought, even though any primer designer software and RNA-folding software should internally compute these stabilities. I tried for example Pairfold and mfold, but that outputs only minimum free energy (which is 0 for perfect pairing). Another hint: the claculation should be similar as for "3' Max stability" in eprimer3 but with thermodynamic parameters adjusted for RNA.

Answer 1

[Eugen Buehler]

If your goal is to design siRNAs that have efficacy at temperatures other than 37C, it still may not be advisable to include melting temperature as a selection feature. Tm and thermodynamics in general are not great predictive tools for siRNA design (although they are crucial for PCR primer design). You would (IMHO) be better off using existing siRNA design algorithms that work primarily through selecting sequences with good GC-asymetry and other base position preferences, and then testing for efficacy of knockdown.

[Michael]

Hi Eugen, thank you for the recommendation. In fact, I am not planning to use Tm, but the difference in 5-prime stability for strand selection bias as described in the paper I cited (Khvorova et al.). This is also named asymmetry rule. As a first step, I want to be able to generate such plots as given in that paper to display the local energy for each siRNA and miRNA. I have gone through some validation studies and found that the 'asymmetry rule' (be it in thermodynamic form, or specific nucleotide preference, e.g. AT vs. GC) is one of the few features (if not the only one) that is equivocally confirmed throughout the studies and used by most siRNA design tools. Therefore, without any further evidence nor the potential for an unbiased evaluation on this particular organism, it seems to have the largest potential to be transferable work for a very different (non-mammalian).

Answer 2

[Brad Chapman]

Have you tried RNAfold in Vienna RNA?

• main page: http://www.tbi.univie.ac.at/~ivo/RNA/
• latest version: http://www.tbi.univie.ac.at/~ronny/programs/vrna2g.html

-T allows temperature scaling.

[Michael]

Thank you Brad, I have just tried it, but it seems to have problems with short duplex sequences, like the other folding programs as well, or I do it wrong (likely ;) ) I think, RNAduplex, is the application that does what I describe. Can you maybe explain this result (a typical calculation according to the paper I cited):

$ RNAduplex     
Input two sequences (one line each); @ to quit
....,....1....,....2....,....3....,....4....,....5....,....6....,....7....,....8
AAAAA
TTTTT
lengths = 5,5
((((.&)))).   1,5   :   1,5   ( 1.10)

I mean, I would expect

(((((&)))))

as a perfect pairing structure. Also, all these programs insist on computing a structure, while the structure is totally flat. Am I doing something wrong?

[Brad Chapman]

Most of the work in this area is in folding or primer design so there are useful models but you may need to massage the usage. You might want RNAeval, where you can specify the "structure" explicitly since you've gotten it out of an alignment already (( and ) indicate matches in structure and . are mismatches):

TTTTTAAAAA
(((((&)))))

length1 = 5
length2 = 5
UUUUU&AAAAA
(((((&)))))
energy =   1.50


GGTGGCCCCC
((.((&)).))

length1 = 5
length2 = 5
GGUGG&CCCCC
((.((&)).))
energy =  -1.90

Does that give what you're looking for?

[Michael]

Yes, that looks like right way to do it (i also mixed up the dot-bracket notation before). Thank you for the pointer. Also, this software uses a more recent parameter set. Do you by chance have an idea how good the approximation might be if one goes from T=37 to e.g. T=5?

[Brad Chapman]

Sorry I'm not experienced enough with lower temperature comparisons to give you firm numbers on accuracy. My experience is at higher temperatures: identifying errors during PCRs. I found the models useful and relative deltaG differences meaningful. As you get further from the experimental inputs with longer sequences and very different temperatures, the absolute melting predictions from these aren't as useful. A lot of things that we found experimentally weren't expected to form based on melting temperature predictions alone. Hope this helps.

Answer 3

[Alex Paciorkowski]

The Mathews lab at my institution has developed RNA structure prediction algorithms. You may want to check to see whether any of their software meets your use case.