Melting temperature
Which melting-temperature model each tool computes with, why two panels can print different numbers for the same primer, and which number to take to the bench.
This page exists because more than a dozen tools report a Tm, and they do not all use the same model. Each tool page names its own; this page puts them side by side. The model-by-model detail, with the eight choices compared on one primer, is on Melting Temperature.
What a melting temperature is, and why there is more than one
The melting temperature is the temperature at which half of a duplex is annealed. It is not a property of a sequence alone: it depends on the salt and magnesium in the buffer, on the strand concentrations, on additives such as DMSO, and on the chemistry of the two strands (DNA/DNA, RNA/RNA or an RNA/DNA hybrid). Three families of model estimate it:
| Family | How it works | Where it holds |
|---|---|---|
| Nearest-neighbour | Sums a measured enthalpy and entropy for each adjacent base-pair stack, adds initiation and end terms, corrects for salt, and solves for the half-annealed temperature at the stated strand concentration | Oligos of any practical length. The only family that responds to concentration and to the order of the bases, not just their count |
| GC formula | A linear function of GC fraction and length, with a salt term | Long duplexes, and protocols that were calibrated against one (QuickChange is the standing example) |
| Wallace rule | 2 °C per A or T, 4 °C per G or C (Wallace et al., Nucleic Acids Res 6, 1979) | Short oligos only, about 14 bases and under. Above 20 bases it is wrong by tens of degrees |
The nearest-neighbour tables themselves differ, because they were measured by different groups under different conditions: SantaLucia & Hicks 2004 (the unified table), Allawi & SantaLucia 1997, Sugimoto 1996 and Breslauer 1986. On one 20-mer the 2004 table gives 61.5 °C and the 1986 table 71.2 °C. Neither is a mistake; they are different measurements.
Which tool uses which model
| Tool | Model it computes with | Can you change it? |
|---|---|---|
| Melting Temperature | SantaLucia & Hicks 2004 with the Taq-mix buffer by default; eight DNA models and four RNA or hybrid models | Yes, every model and all eight salt corrections |
| PCR | The polymerase decides the convention: SantaLucia & Hicks 2004 for Taq, Allawi & SantaLucia 1997 with the dsDNA-binding-domain convention for Phusion and Q5. Both annealing temperatures are shown side by side | Yes, the polymerase and, independently, the table |
| Primer Designer | Nearest-neighbour for the search. Hairpins and dimers are always primer3's thermodynamic alignment | The displayed model, yes. A GC or Wallace choice changes the display only, and the panel says the search still ran under nearest-neighbour |
| Site-directed mutagenesis, Saturation mutagenesis | SantaLucia nearest-neighbour | Yes, including the GC formula a published method was calibrated against |
| Gibson | Two numbers per junction: a SantaLucia nearest-neighbour Tm, and a 2 °C/4 °C sum that the assembly's minimum and maximum Tm filters are applied to | No. The filters read the second number only |
| In-Situ Probes | Unified SantaLucia, salt-corrected, at the probe and target concentrations you set | Salt and concentrations |
| In-Situ Probes, padlock arms | Sugimoto 1996 at 0.1 µM probe, 75 mM monovalent and 10 mM divalent salt, corrected for 20% formamide | No: these are the published padlock conditions |
| Primer Library | Wallace rule, and stated as out of range above 20 nt | No. It is a fallback readout for a list, not a prediction |
| Multiplex PCR, qPCR, LAMP | primer3's thermodynamic alignment for structures and dimers, judged against primer3's own published limits | The limits, by preset |
Why two panels disagree, and which number to read
The same primer can show different temperatures in two panels, and both are right for the model and buffer each states. The three causes, in order of how often they explain a difference:
- A different buffer. The salt correction alone moves a 20-mer by about 8 °C between two of the standard buffers. A Tm quoted without its buffer is half an answer.
- A different polymerase convention. Phusion and Q5 carry a DNA-binding domain, and their annealing rule uses a different table and takes the product length into account. On the PCR page's worked example the same pair anneals at 60.5 °C under the Taq rule and 57.1 °C under the Phusion/Q5 rule; the panel shows both so the difference is in front of you.
- A different scale entirely. Gibson's filter value and the primer library's readout are 2 °C/4 °C sums. On the 40 bp overlap in Gibson's worked example that is 106 °C where the nearest-neighbour answer is 67.9 °C: a ranking number, not a temperature.
Which to take to the bench: the nearest-neighbour value under the preset for your polymerase's vendor, and then stay with it. Comparing a number from one model against a threshold calibrated on another is the usual way a good primer pair gets rejected.
Corrections
- Salt. Eight published methods, applied inside the nearest-neighbour calculation, named rather than numbered (Biopython's methods 0 to 7).
- DMSO and formamide. Applied after the model, at the rates Biopython's code uses. Its documentation states a different DMSO factor than its code; the code is what was reproduced, because it is what other tools agree with.
- Betaine and glycerol are not offered. Betaine's effect is non-linear and depends on GC content, and there is no single agreed conversion from percentage to a temperature shift, so no number is printed rather than an approximate one.
How the numbers are checked
Every model is reimplemented from the published library that defines it (Biopython's
MeltingTemp, pydna, QUEEN and primer3) and compared result for result against that library on
the same inputs. The tables of citations, with the four nearest-neighbour sets and the RNA sets, are
on Melting Temperature.
Related
Melting Temperature · PCR · Primer Designer · Gibson · In-Situ Probes · Primer Library