CHARMM 36m Additive Force Field

• CHARMM 36m force field is a refined fixed-charge potential that enhances biomolecular MD simulations by optimizing charge scaling and nonbonded interactions.
• It implements targeted NBFIX corrections to reduce overbinding in charged interactions and achieve better agreement with experimental data.
• The force field delivers improved structural fidelity, realistic protein dynamics, and balanced nucleic acid conformations across diverse biomolecular systems.

The CHARMM 36m additive force field is a refined all-atom, fixed-charge empirical potential developed to improve the accuracy and transferability of molecular dynamics (MD) simulations of biomolecules, including proteins, nucleic acids, and complex assemblies with lipids and ionic species. As an evolution of the CHARMM36 family, it addresses key deficiencies identified in earlier force fields, particularly the tendency toward overbinding in charged group interactions, excessive stabilization of certain biomolecular conformations, and an incomplete treatment of electronic polarization effects. CHARMM 36m achieves these improvements by introducing targeted modifications to atomic partial charges, nonbonded interaction parameters—including extensive use of NBFIX corrections—and by systematically benchmarking against experimental structural, thermodynamic, and dynamic observables.

1. Rationale for Parameter Refinement and Charge Scaling

A central limitation in conventional non-polarizable force fields is the neglect of electronic polarization, which results in overestimated ion–ion and charge–charge interactions and poorly balanced salt bridge strengths. CHARMM 36m incorporates insights from the Molecular Dynamics in Electronic Continuum (MDEC) model (Leontyev et al., 2015), which recommends scaling the charges of ionized groups to account for the screening effect of the medium’s electronic polarizability. In this framework, the effective charge $q_{\mathrm{eff}}$ is determined by

$$q_{\mathrm{eff}} = \frac{q}{\sqrt{\epsilon_{\text{el}}}},$$

where $\epsilon_{\text{el}}$ is the high-frequency dielectric constant (typically $\epsilon_{\text{el}}\approx1.7-2.0$ for organic materials). This yields a scaling factor of $q_{\mathrm{eff}}\approx0.7q$, which is applied across ionized groups such as carboxylates and amines. The adoption of scaled charges reduces the electrostatic interaction strength, mitigates excessive pairing and aggregation artifacts, and results in more physically realistic protein dynamic behavior—such as increased salt bridge fluctuations and altered solvent accessibility in catalytic sites.

2. NBFIX Corrections and Pair-Specific van der Waals Parameterization

CHARMM 36m introduces systematic pair-specific adjustments to Lennard-Jones parameters (NBFIX corrections) to correct the exaggerated direct binding affinity between charged groups, particularly amine–carboxylate and amine–phosphate pairs (Yoo et al., 2015). The standard Lennard-Jones 12-6 potential is

$$V_{\mathrm{LJ}}(r) = 4\varepsilon \left[ \left(\frac{\sigma}{r}\right)^{12} - \left(\frac{\sigma}{r}\right)^6 \right],$$

where $\sigma$ and $\varepsilon$ are atom-type-specific parameters. In CHARMM 36m, the cross-term minimum distance parameter $r_\text{min}$ is increased by $\Delta r_\text{min}$ for key pairs:

• Amine–carboxylate N–O=C: $\Delta r_\text{min}=0.08$ Å, yielding $r_\text{min}=3.63$ Å.
• Amine–phosphate N–O=P: $\Delta r_\text{min}=0.16$ Å, yielding $r_\text{min}=3.71$ Å.

These adjustments are narrowly targeted and do not globally alter solvation free energies or the packing of non-interacting groups. The NBFIX strategy effectively reduces spurious aggregation of proteins, improves osmotic pressure predictions in charged solute solutions, and brings peptide-mediated nucleic acid and membrane assembly simulations into closer alignment with experimental structural data.

3. Treatment of Electrostatics and Electronic Solvation Effects

The fixed-charge paradigm in CHARMM 36m, while efficient, omits explicit electronic polarization. To compensate, the methodology incorporates both charge scaling and corrections to the calculated solvation free energies. The total solvation free energy is treated as

$$\Delta G = \Delta G_\mathrm{nuc} + \Delta G_\mathrm{el},$$

with $\Delta G_\mathrm{el}$ accounting for the purely electronic (fast) polarization component. Non-polarizable simulations using unscaled (“bare”) charges overestimate both pair interactions and nuclear solvation energies, but omitting $\Delta G_\mathrm{el}$ can lead to systematic errors, particularly in systems where ion–ion interactions are significant or in low-dielectric environments. Incorporating explicit scaling in CHARMM 36m thus improves the transferability and physical fidelity of the force field across aqueous and biomolecular environments (Leontyev et al., 2015).

A further implication of this approach is the reconciliation between empirical water model dipoles and ab initio values. The effective dipole moment in non-polarizable water models such as TIP3P is reduced compared to the real liquid value:

$$\mu_{\text{eff}} = \frac{\mu}{\sqrt{\epsilon_{\text{el}}}},$$

with $\mu \approx 3.0$ D and $\mu_{\text{eff}} \approx 2.35$ D for $\epsilon_{\text{el}} \approx 1.78$. This adjustment ensures consistency in MD simulations of solute–solvent electrostatics.

4. DNA and Protein Parameterization: Conformational Balance and Structural Flexibility

CHARMM 36m extends the CHARMM36 strategy for nucleic acids and proteins by carefully rebalancing torsional and nonbonded parameters. For DNA, the force field displays an “A-philic” propensity in base stacking and a “C-philic” bias in backbone conformational energetics, yielding a fragile balance that can produce B-form duplexes as a dynamic equilibrium (Strelnikov et al., 2022). The corresponding potential energy is expressed as

$$E = \sum_{\text{bonds}} k_b (r-r_0)^2 + \sum_{\text{angles}} k_\theta (\theta-\theta_0)^2 + \sum_{\text{dihedrals}} V_n [1 + \cos(n\phi - \delta)] + \sum_{i<j} [V_{\mathrm{LJ}}(r_{ij}) + V_{\text{Coulomb}}(r_{ij}) ],$$

where $$V_{\text{Coulomb}}(r_{ij}) = \frac{q_i q_j}{4\pi\epsilon_0 r_{ij}}$$.

Simulations reveal that in CHARMM36 (and, by extension, CHARMM 36m), sequence-dependent transitions between A-, B-, and C-forms of DNA are captured via the interplay between stacking terms and backbone flexibility. However, the B-form is maintained only due to opposing biases, resulting in pronounced sequence and salt sensitivity, and a weaker dependence on salt compared to experimental observations. This behavior contrasts with AMBER bsc1, which over-stabilizes the B-form via excessively strong base stacking energies and yields inflexible DNA that does not access non-canonical forms (Strelnikov et al., 2022).

For proteins, the effect of charge scaling, NBFIX parameters, and optimized torsional energetics in CHARMM 36m results in biomolecular assemblies with more realistic flexibility, salt bridge dynamics, and better agreement with experimentally derived structural ensembles.

5. Comparative Performance Versus Polarizable Force Fields

CHARMM 36m is an additive force field: atomic partial charges are fixed and polarization is not explicitly modeled. Recent benchmarks against polarizable models, such as Drude-2019, demonstrate that CHARMM 36m yields lower and more structured root-mean-square fluctuations (RMSF) of residues, more extensive networks of correlated motions, and improved preservation of native hydrophobic contacts (Milinski et al., 6 Oct 2025). Specifically, residue correlation maps derived from CHARMM 36m simulations show stronger and more coherent islands of positive and negative dynamical correlations, mirroring experimental allosteric networks. In community network analysis, key functional modules (e.g., in the PPARγ ligand-binding domain) form well-coupled communities, supporting robust allosteric transmission, whereas polarizable force fields yield more fragmented, less interconnected communities. Table 1 summarizes core differences:

Property	CHARMM 36m Additive	Drude-2019 Polarizable
Correlated Motions	Strong & coupled	Weaker & fragmented
RMS Residue Fluctuations	Moderate, closer to experiment	Higher
Hydrophobic Cluster Retention	High	Lower

These findings indicate that, despite a simplified treatment of polarization, CHARMM 36m often provides a superior description of long-range dynamical couplings and native-state stabilization critical for allosteric signaling and correlated motions in biomolecules.

6. Applications and Integration with Existing CHARMM Frameworks

The full compatibility of CHARMM 36m parameters with the CHARMM36 family enables seamless use in mixed biomolecular simulations: proteins with nucleic acids, proteins with lipid membranes, or systems incorporating amino acid–based ionic liquids (Fileti et al., 2015). For ionic liquid studies, the CHARMM 36m strategy of charge reassignment (based on ion-pair electrostatic potential fitting) and targeted vdW cross-term modification enables accurate reproduction of hydrogen bonding geometries, densities, and thermodynamics while retaining transferability and computational efficiency. In systems requiring integration with polymer, carbohydrate, or synthetic molecule parameters, CHARMM 36m’s additive, modular design, and the preservation of parameter philosophy streamline the modeling workflow.

7. Limitations and Prospects for Further Refinement

Despite substantial advances, residual limitations persist. The force field’s weak prediction of salt-dependent DNA transitions, the delicate and sequence-sensitive balance between base stacking and backbone flexibility, and the empirical nature of charge scaling represent active areas for further development (Strelnikov et al., 2022). Electronic polarization, while approximated via scaled charges, is not dynamically propagated, and some limitations in the simulation of ion–protein and ion–nucleic acid interactions remain. Ongoing efforts in the field involve not only further NBFIX refinements and backbone torsion tuning, but also hybrid approaches that blend explicit polarization with additive force field efficiency.

Conclusion

CHARMM 36m represents a significant advance in fixed-charge empirical force field design, balancing the need for physical realism in electrostatics, base stacking, and solvation energetics with the practical constraints of computational scalability. Through an overview of charge scaling, pair-specific NBFIX corrections, and additive parameter optimization, it delivers improved structural and dynamical fidelity across a broad range of biomolecular assemblies. Continuous benchmarking against experimental observables and polarizable force fields informs ongoing parameterization efforts aimed at further increasing accuracy and transferability for complex systems spanning proteins, nucleic acids, membranes, and ionic environments.