team of Lawrence Livermore researchers is developing a new approach
to biological research by linking advanced simulations with laboratory
experiments to explain biological phenomena at an unprecedented
level of detail. In addition to promoting close collaborations with
other biological scientists at the Laboratory, the research involves
newly developed simulation methods that often run on Livermore's
teraops (trillion operations per second) supercomputers.
"The emerging explanation
of biological functions in terms of their underlying chemical processes
is creating an important role for predictive chemical simulations
in biological research," says Mike Colvin, head of the Computational
Biology Group in the Biology and Biological Research Program (BBRP).
The group is currently involved in a wide range of projects that
includes studies of the action of anticancer drugs, the DNA-binding
properties of mutagens in food, the mechanisms of DNA repair enzymes,
and the biophysics of DNA base pairing.
All of the projects strongly
tie modelers to experimental researchers in BBRP because, says Colvin,
"There is a growing consensus that the integration of computation
and experiment will accelerate progress in biology."
The work is funded in part
by Lawrence Livermore's Laboratory Directed Research and Development
program as a strategic initiative combining experts from the BBRP,
Physics and Advanced Technology, and Computation directorates. The
research is also attracting new funding from agencies such as the
National Institutes of Health.
The group's simulation methods
range from molecular dynamics that use classical laws of physics
to first-principles methods that use quantum mechanics to exactly
describe the electronic structure of every atom and thus their chemical
properties. (See the box below.) The group also uses some of the
world's most powerful computers, including massively parallel supercomputers
that are part of the Department of Energy's Accelerated Strategic
Computing Initiative (ASCI). Several simulations completed on ASCI
computers have established new standards for size and accuracy in
the chemical modeling of biological processes.
a Virtual Microscope
Lawrence Livermore computational biologist Mike Colvin,
the most difficult aspect of designing and running a biological
simulation is fully understanding the biological problem
at hand and reducing it to one or more key chemical reactions.
An example is determining why one molecule is vastly more
toxic to a cell than another chemically similar molecule.
"The first step
in our job is to dig through the literature and talk with
the experimental collaborators to identify the essential
reactions that we need to simulate in order to discover
the differences in toxicity," says Colvin. Such simulations
reveal the exact physical mechanisms and energetics of
Barsky calls advanced biochemical simulations a "virtual
microscope." This microscope, he explains, combines powerful
computers and software programs that link the laws of
physics and chemistry to structures of biological molecules
that have been determined by x-ray diffraction and other
He cautions, however,
that one danger with the microscope analogy is that scientists
new to such techniques may be tempted to simulate everything
without carefully identifying the biological question
being addressed. "We have found that the simulations must
be very closely tied to experiments. Without having specific
things to test and look for, your simulations may produce
a mountain of data that do not reveal much."
simulations principally take two forms: molecular dynamics
and first-principles quantum chemistry.
dynamics is used for DNA and protein studies involving
tens of thousands of atoms at a time. These simulations
depict atoms like balls interconnected with springs to
approximate their motions and their interactions with
on first-principles quantum chemistry accurately predict
the chemical properties of atoms and molecules. The technique
uses quantum mechanics to determine the distribution of
electrons around each atom. From this electron distribution,
any chemical property can be determined, including the
structures and energies of molecules.
have made it possible to do so-called first-principles
molecular dynamics in which the motion of molecules is
simulated using accurate quantum chemical interactions.
These methods have been developed by Lawrence Livermore
physicists Francois Gygi at the Center for Applied Scientific
Computing and Eric Schwegler and Giulia Galli in the Physics
and Advanced Technologies Directorate. The new techniques
are being applied to selected biochemical problems, such
as how DNA is cleaved by repair enzymes and how an anticancer
drug is activated by the body.
simulation methods are used together to solve a single
problem. For example, a researcher can do molecular dynamics
on DNA structure, then pull out a small area of interest
and do first-principles simulations for understanding
the exact electronic forces at play.
of two modeling methods applied to the interaction of
hydrogen fluoride and formaldehyde. (a) Molecular dynamics
models depict atoms like balls interconnected with springs
to simulate their motions. (b) First-principles simulations
use quantum mechanics to predict the chemical properties
of atoms and molecules.
Lesions Need Repair
of the computational biology work focuses on the dynamics and structure
of DNA and repair mechanisms for specific forms of DNA damage. Human
genes are made up of long, double-stranded DNA molecules that contain
the instructions for building the proteins that make up the machinery
of every cell. Understanding the role of DNA repair and other processes
that protect cells from radiation and chemical insults has been
a long-term interest at Lawrence Livermore.
One simulation project has
shown how human repair enzymes recognize a common form of DNA damage
called an abasic lesion. These lesions occur when a stretch of DNA
loses one of its constituent bases, leaving a gap in the chain.
Such lesions arise spontaneously more than 10,000 times per day
in every cell. They can also be caused by exposure to pesticides,
food mutagens, and ionizing radiation from the sun. If unrepaired,
the damage can lead to disease, notably cancer, through a mutation
in the genetic code.
Cells have a set of repair
enzymes (proteins) that scan DNA looking for damage such as abasic
lesions. In humans, the major enzyme responsible for repairing abasic
DNA is the endonuclease Ape1. Livermore scientists have been addressing
the question of how repair proteins such as Ape1 recognize and bind
specifically to damaged DNA.
"We want to know what changes
to the DNA shape caused by the lesions are recognized by the repair
protein," says Computational Biology Group member Daniel Barsky.
Until his simulations were completed last year, scientists had suggested
a variety of possibilities.
Using a structurally related
protein-DNA complex as a template, Barsky first built a model of
DNA bound to Ape1 to guide BBRP biologist David Wilson III and colleagues
in determining which amino acids of Ape1 might be most important
for its activity. Wilson also synthesized different forms of DNA
to see which, if any, would be recognized by Ape1. Surprisingly,
all of the altered forms attracted Ape1 to a significant degree.
Spurred by Wilson's findings, Barsky used a cluster of advanced
scientific workstations in the Livermore Open Computing Facility
to complete the first full molecular dynamics simulations of the
The simulations focused on
a stretch of DNA with an abasic lesion surrounded by thousands of
water molecules. Although the computations simulated a time span
of only 2 billionths of a second, they required the equivalent of
several months of processing time on a fast single-processor computer.
model of Ape1, a DNA-repair protein, binding to a strand of
DNA. DNA strands are in yellow and green, the protein is in
blue, and ultraviolet-absorbing amino acids are in red.
Changes Flag Enzyme
The results showed that the
abasic site does not form a permanent hole or gap in the DNA, as
some researchers had postulated. Instead, the missing base causes
changes to the internal motions of the DNA-changes that are thought
to be important for damage recognition by the repair enzyme.
For example, at various intervals
in Barsky's simulations of damaged DNA, a thymine base unpaired
with adenine and paired instead with the cytosine base that was
opposite the abasic site, where a guanine would normally be. This
transient thymine-cytosine base-pair mismatch had not been previously
observed. Another clearly visible change was that the sugar molecule,
formerly attached to the missing guanine, flipped out of the chain.
The results indicated that
the abasic DNA chain has a great deal more flexibility and bending
than normal. This unnatural flexibility apparently "flags" the repair
protein, says Barsky. Further proof for this concept was supplied
last year when x-ray crystallography studies of abasic DNA bound
to Ape1 showed a kink in the DNA at the abasic site.
The results from the simulation
are being used to determine how specific differences in the Ape1
protein that have been found in a portion of the human population
affect the DNA repair capacity of those individuals. Such knowledge
will help researchers predict which people are at greater risk of
developing disease from environmental exposures that induce the
formation of abasic lesions.
Barsky has also been studying
the base pairings in parallel-stranded DNA, a novel form of DNA
in which both strands are oriented in the same direction instead
of aligning in opposite directions. His quantum chemical calculations
contradict previous theories of how the guanine-cytosine base pairing
might occur in parallel DNA. The results predict that the greatest
stability occurs through an orientational "wobble" in which guanine
and cytosine form only two hydrogen bonds instead of the three bonds
they form in normal DNA.
In on Key Ions
Barsky's DNA simulations
are magnified a millionfold in simulations done by colleagues Eric
Schwegler and Felice Lightstone. Their focus is on the chemical
reactions involved in phosphate hydrolysis, an essential part of
DNA repair. To study this reaction, they first slice a strand of
DNA in two to see how other enzymes repair the damage.
molecular dynamics simulations use 65 water molecules, one dimethyl
phosphate (the simplest repeating structure comprising the DNA backbone),
and one magnesium ion in a cubic "box." The model includes a magnesium
ion because many DNA-cutting enzymes such as Ape1 require the ion
to catalyze the DNA cleavage. In fact, a high concentration of magnesium
ions alone can catalyze the cleavage.
The simulations track the
movement of every atom and its cloud of electrons. Of particular
interest is what happens to the magnesium ion. It attracts six water
molecules and momentarily adheres to the dimethyl phosphate, thus
making the cleavage reaction possible.
The simulations are run
on the ASCI Blue supercomputer using Lawrence Livermore software
adapted for multiprocessor machines. Despite the enormous computational
power of the computer, Lightstone can only simulate one trillionth
of a second per month. "As a result," says Lightstone, "we have
to be selective in what we simulate."
The payoff is new understanding
of the pathways of reactions involved in DNA repair. "Experimentalists
can't tell you all the steps involved in a biological process,"
Lightstone says. "Seeing all the steps gives us ideas about how
we might modify the reaction or even control the enzyme's actions."
model at left shows a DNA molecule missing one of its bases.
The simulation at right reveals its kinked, unnatural shape
that "flags" the repair protein Ape1.
Docking to Enzymes
Lightstone is also working
on so-called computational docking to help identify small molecules
called ligands. Ligands can uniquely bind to selected sites on proteins,
including DNA-repair enzymes and deadly neurotoxins such as tetanus
and its relative, botulinum. The identified ligands would be used
in sensors to indicate the presence of neurotoxins. The effort is
being pursued in conjunction with BBRP researcher Rod Balhorn, Livermore's
Chemical and Biological Nonproliferation Program, and computational
experts at Sandia National Laboratories. (See S&TR, April
Biology Looks at theTies that Bind.)
docking calculations were first used to screen the Available Chemicals
Directory—a listing of some 250,000 purchasable compounds for which
the three-dimensional structure is known—for chemicals that would
bind to the tetanus protein. A cluster of scientific workstations
using molecular mechanics techniques required only 10 seconds to
analyze the shape of each candidate molecule and determine the extent,
if any, to which it could bind to a small depression in the tetanus
After three days of calculations,
the simulation had ranked all of the compounds in the chemical library,
out of which Lightstone compiled a list of 11 candidates for laboratory
tests. BBRP experimentalists found that 5 of the 11 molecules successfully
bound to the tetanus protein.
Lightstone points to the
efficiency of using methods such as computational docking as a screening
tool. "Experimentally testing 250,000 compounds would take years
of work," she says. She notes that the docking simulations are done
at a coarse level of accuracy. "If you did them at first-principles
level, we would get bogged down in long computer times. Our purpose
is to screen lots of compounds very quickly."
Lightstone moved on to finding
candidate ligands for binding to botulinum toxin, which is viewed
as a more dangerous threat. This time, she applied a second computational
step, flexible docking, to the top 2,000 compounds identified in
the first step. Flexible docking, which requires two additional
weeks of computer time, rotates molecular bonds to find the optimum
shape of a molecule for binding to a selected protein. The top-scoring
ligands are being tested for their affinity to the botulinum protein.
To increase the sensitivity
and selectivity of a portable sensor, Lightstone is searching for
an additional ligand for a second, nearby site on the botulinum
protein. Together, the two ligands could be used to detect the toxin
at low concentrations.
simulation focuses on magnesium, which is required by many DNA
repair enzymes. The simulation includes many water molecules
and one dimethyl phosphate molecule (the simplest repeating
component of DNA). Oxygen is red, hydrogen is white, carbon
is black, magnesium is purple, and phosphate is orange.
longest running collaborative project in the Computational Biology
Group is studying the function of mutagenic chemicals called heterocyclic
amines. These compounds are formed in the cooking of several foods
and may be a risk factor associated with cancer in the human digestive
tract. As with most substances that damage DNA and cause cancer,
the food mutagens must be activated by metabolic reactions the body
uses to break down chemicals. Once activated, the mutagens bind
to DNA and can interfere with the accurate duplication of the genetic
code, thereby leading to mutations and eventually cancer.
the past four years, the group has been applying simulations to
help experimentalists identify a subset of heterocyclic amines called
2-aminoimidazole-azarene (AIA) compounds. The group is looking at
several properties of these mutagens, including the metabolic steps
that actuate the compounds, the initial attachment site of these
mutagenic compounds on DNA, how binding to the mutagens depends
on the DNA sequence of bases, and the effect of different cooking
processes on mutagen formation.
is that all two dozen molecules in the AIA family are chemically
similar, and yet there is a 10-millionfold range in mutagenicity,"
says Colvin. Identifying the factors that vary the potency will
help scientists better predict the human health risks associated
with exposure to food mutagens.
most recent simulations involve the action of cytochrome P450, the
enzyme responsible for first activating the mutagens. Because the
structure of the human form of P450 is unknown, the group built
a computer model using related enzymes whose structure is known.
The simulations are used to dock different AIA compounds into the
enzyme's active site to determine if the more potent species make
a better fit and are thus more likely to become activated.
group is also simulating the interaction between P450 and members
of the bioflavanoid family, which inhibit food mutagens. About the
same size and shape as AIA compounds, bioflavanoids are found in
fruits and vegetables. One hypothesis is that the bioflavanoids
lower the incidence of food mutagens by competing for the same activation
site on P450 as the AIA compounds.
simulation reveals the binding mechanism between a food mutagen
and cytochrome P450, the enzyme that catalyzes the initial activation
step for this mutagen.
fully instrumented hamburger patty is fried to determine its
temperature as a function of depth as well as the corresponding
concentrations of food mutagens. The data are used to develop
computer simulations of the cooking process and to predict the
formation of mutagens.
A new effort in the Computational
Biology Group is to develop an accurate computer model of cooking
hamburger patties and other meats. The goal, says student researcher
Ngoc Tran, is to successfully simulate what BBRP investigators do
on a hot stove to study how cooking methods affect food mutagen
production. Thus, virtual cooking would help reduce the number of
experiments done in the kitchen by identifying the key experimental
measurements that must be made.
Developing a simulation of
the cooking process and the formation of mutagens is not straightforward
because it must reflect such factors as the fat and moisture content
of the meat, the frying temperature, the heat conductivity of the
pan and the meat, and the manner of cooking. For example, BBRP researchers
discovered last year that flipping a hamburger frequently during
frying reduces the number of mutagens. (See S&TR,
January/February 2001, The
Laboratory in the News.)
Tran spent last summer working
with Mark Knize and Cyndy Salmon accumulating raw data for the simulations
by cooking a couple dozen hamburgers. She measured the temperature
at different depths of the patties as well as the corresponding
concentrations of AIA compounds. She found that the first millimeter
contained 50 percent of the food mutagens, the second millimeter
contained 25 percent, and the third contained 10 to 15 percent.
The simulations accurately reproduced the temperature profiles measured
while cooking beef patties and correctly predicted how the concentration
of mutagens varied at different meat depths.
The current goal is to refine
the model so that it accurately reflects every aspect of cooking.
Simulations will then be used to determine the sensitivity of mutagen
formation to such parameters as fat content and pan temperature.
The ultimate goal is to design new cooking procedures that minimize
the formation of mutagens.
anticancer drug cyclophosphamide (depicted in green) is seen
in this simulation forming an interstructural cross-link to
DNA. The formation of such cross-links is believed to be key
to the anticancer activity of these drugs.
dynamics simulations indicate how antifreeze proteins in the
winter flounder may prevent ice formation by forming several
stable structures, like this one, that act as an insulating
are also helping scientists to understand the functioning of one of
the oldest families of anticancer drugs, the phosphoramidic mustards.
The group includes the widely used drugs cyclophosphamide and ifosfamide.
These drugs are closely related to the poisonous mustard gas used
in World War I. Doctors noticed at the time that the gas killed rapidly
dividing body cells and reasoned that a derivative might work on cancer
cells because they continually divide.
being used for more than 40 years, several important questions about
the drugs' biological activity remain unanswered. The key to the drugs'
therapeutic activity—and toxic side effects—are the activation steps
they undergo in the body before they bind to DNA. Colvin has been
collaborating with scientists at Duke UniversityŐs Comprehensive Cancer
Center to understand how the activation reactions affect the drugs'
potency. Fortunately, the drugs are small enough to be modeled using
first-principles methods. Simulations of most of the steps in the
activation pathway have helped to explain several unexpected properties
of these drugs and are suggesting improved versions of the standard
drugs are known to kill cancer cells by forming cross-links between
the two strands of a cell's DNA. These cross-links are particularly
difficult for a cell to eliminate; just a few cross-links kill a cancer
cell. Colvin and Dat Nguyen, a graduate student researcher from the
University of California at Davis, are simulating how cyclophosphamide
forms cross-links. The goal is to understand how the drug's molecular
structure could be changed to make it more effective in forming cross-links.
at Duke University and the University of Chicago will synthesize the
best molecular candidates identified in the simulations and test their
cross-linking capacity on DNA. The most promising molecules will then
be tested for their effectiveness in killing cancer cells.
Computational Biology Group is also collaborating with Nguyen and
several UC Davis professors to examine a phenomenon that has puzzled
biologists for more than four decades: the ability of Antarctic
fish to survive sea temperatures well below freezing. The fish contain
a variety of so-called antifreeze proteins in their bloodstream
that inhibit the growth of ice crystals in their bodies. Similar
proteins have also been identified in some insects and plants that
can withstand freezing temperatures.
exact mechanisms of how they function have been a mystery, but molecular
dynamics simulations by Nguyen provide new clues. The simulations
tested the hypothesis that these proteins depress the freezing temperature
by binding to ice crystals and acting like an insulating blanket.
The results showed that pairs of antifreeze proteins can form several
stable structures. The proteins may be able to absorb and store
heat by undergoing transitions between these structures.
says that Lawrence Livermore's new capabilities in computational
biology are being recognized by the greater scientific research
community in the form of invited talks, requests for review articles
and textbook chapters, and new collaborations by colleagues at universities
computational modeling is successfully applied to more biological
problems," Colvin says, "it is clear that simulation will have a
growing role in the training and research of biological scientists."
There is also little doubt that advanced simulations will continue
to change the nature of biological research.
abasic lesions, Accelerated Strategic Computing Initiative (ASCI),
Ape1, bioflavanoids, botulinum, Center for Applied Scientific Computing,
computational biochemistry, cyclophosphamide, DNA, first-principles
molecular dynamics, food mutagens, molecular dynamics, quantum mechanics,
information contact Mike Colvin (925) 423-9177 (firstname.lastname@example.org).
is leader of the Computational Biology Group in Livermore's
Biology and Biotechnology Research Program (BBRP). He received
B.S. degrees in chemistry and humanities from the Massachusetts
Institute of Technology and a Ph.D. in chemistry from the University
of California at Berkeley. He joined the Laboratory in 1986
as a postdoctoral fellow in the Institute of Scientific Computing
Research and then became a staff physicist in O Division, where
he concentrated on designing biologically inspired algorithms
for faint object detection. In 1990, Colvin transferred to the
Center for Computational Engineering at Sandia National Laboratories,
California, to develop quantum chemical methods for massively
parallel computers. In 1997, he returned to Livermore and joined
BBRP. His research there has focused on using advanced computer
simulations to study biological phenomena.