AMBER Archive (2006)

Subject: Re: AMBER: water diffusion

From: Thomas Cheatham (
Date: Tue Jul 18 2006 - 01:36:36 CDT

It's late, but...

> I've been trying to run MD of DNA in a periodic box using AMBER 99 implemented
> in Hyperchem 7.52 and found that I could not mimic DNA denaturing. Then I ran

Do you have an idea of what the time scale for DNA denaturation is?
Also, for your particular sequence, do you know at what temperature the
structure should melt/denature?

It is true, if you run with an improper treatment of the electrostatics,
such as using a truncated group-based cutoff, you can rapidly (on a
100-1000 ps time scale) distort a DNA duplex. Full strand separation
likely takes longer (and in my experience is rarely seen, although we did
see it recently in GB simulation after a 500 ns run).

> just water in a 20A x 20A x 20A box for 200 ps and found that at 300K it

[this is a small box for a DNA duplex; how large is the structure and do
you have an idea of what the (periodic) DNA concentration is at that
volume and if it is reasonable?]

> behaved like a crystal (no diffusion, only libration). The water is apparently
> implemented as TIP3P. Hypercube people told me that it is likely the intrinsic
> property of this potential. Is it true?

What is the time scale for DNA rotation and translation? For a duplex of
5-20 base pairs, the rotational correlation time is likely on the
nanosecond time range, so the observation of little motion on a 200 ps
time scale seems fully consistent to me. You could measure the mean
squared displacement of the solvent (water) to see if it matches
expectation. TIP3P diffuses at twice the experimental value.

> I also tried anther force field, MM+, and saw a 'normal' water behavior their.

What is "normal" behavior?

If you search on Google for "DNA melting temperature", a variety of
servers appear where you can predict (based on detailed experimental
thermodynamic measurements by Turner, Santa Lucia and others and
heuristics) duplex melting temperatures. In many cases (for systems of
more than 6-8 base pairs), the melting temperatures are above room
temperature (~300 K). If the melting temperature is above room
temperature, should the DNA denature or melt in simulation of DNA at 300K?
If so, what do you expect the time scale to be? 200 ps is not very long
in terms of the "motion" of a DNA duplex (or single strand). Fast bond
vibrations occur on the 10 femptosecond time scale (which is why we need
to run MD with timesteps in the 1-2 fs time range) such that a 200 ps
simulation only samples ~200 X 100 or 20000 (fast) bond vibrations and
even fewer angle and dihedral transitions. Likely the collective motions
necessary to denature the DNA require more time (particularly if there
is any appreciable activation barrier)... Moreover, if you start
the system in a "folded" (duplex or pre-organized) state (at a given
concentration/temperature) you may not be at equilibrium (i.e. you have to
likely run the simulation for multiple "relaxation" times to let it get to
the equilibrium state).

If you delve into the literature, some conformational transitions (in DNA
fibers which are not the same as small duplexes) require minutes to days
(for A->B transitions and even longer for others). I would estimate that
"melting" or "denaturation" of small DNA duplexes in solution require at
least many nanoseconds up to minutes (which is admittedly a large range in
time!). Experimentally, nucleic acid annealing is performed over
minutes to hours. Note further that the melting temperature usually
refers to the inflection point in a melting curve where half of the
molecules are denatured so at the melting temperature, given a single
molecule in a periodic box you only have a 50% chance of it being

In my experience, even very small duplexes at 300K, such as d(GCGC)2 in
explicit solvent with a proper treatment of the electrostatic
interactions, do not "denature" on the ps - 500 ns time range (even though
the melting temperature is likely < 10 degrees!). We do see d(GC)2 or
d(GCG)2 denature on this timescale. For single strands, helix like
structures appear metastable on the nanosecond time range. It
is not clear if 200 ps is sufficient.

Accurately modeling the melting temperature (or relative stability of
sequence specific DNA interactions) is really a grand challenge simulation
project. I would like to think we can do this, but I am not yet
convinced that we can do it...

So short answer: The "potentials" we apply are not perfect, likely contain
artifacts, and may not properly represent what we are trying to simulate.
On the other hand, they have proven surprisingly powerful in the
understanding of A<->B equilibria, structure and dynamics of various DNA
molecules ranging from G-DNA to triplexes to i-DNA and odd structures.
There is a vast literature available (albeit only since the mid 80's when
the first nucleic acid simulations were performed but since then there are
literally hundreds of publications). Check out the literature and try to
better understand what is measured experimentally; ultimately, make sure
you strive to validate the simulation results in comparison with what we
know (from experiment).

Good luck in your simulations,

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to