Correcting for heat capacity and 5'-TA type terminal nearest neighbors improves prediction of DNA melting temperatures using nearest-neighbor thermodynamic models.

Nearest-neighbor thermodynamic (NNT) models currently provide some of the most accurate predictions of melting thermodynamics, including melting temperature (T(m)) values, for short DNA duplexes. Inherent to all existing NNT models is the assumption that ΔH° and ΔS° for the helix-to-coil transition are temperature invariant. Here we investigate the impact that this zero-ΔC(p) assumption has on the accuracy of T(m) predictions for 128 DNA duplexes. Previous and new melting thermodynamic data are analyzed to establish an estimate of ΔC(p)(bp), the heat capacity change per base pair, of 42 ± 16 cal mol(-1) K(-1) bp(-1), as well as an optimal thermodynamic reference temperature (T(ref)) of 53 ± 5 °C. These results were used to modify the unified NNT model to properly account for the temperature dependence of ΔH° and ΔS° and thereby extend the range over which T(m) is accurately predicted. This new approach is shown to be especially useful for duplexes that melt at a T(m) greater than 70 °C. Thermodynamic data collected by differential scanning calorimetry (DSC) for 16 duplexes designed to melt over a broad temperature range were used to verify the values of ΔC(p)(bp) and T(ref) and to show that ΔC(p)(bp) is essentially constant above 37 °C. Additional DSC analysis of 12 duplex sequences containing all 10 nearest neighbors allowed for errors associated with different terminal nearest neighbors to be examined and showed that duplexes containing one or more terminal 5'-TA groups are significantly more stable than predicted by the unified NNT model. A correction to improve prediction of the hybridization thermodynamics of duplexes with terminal 5'-TA groups is provided.