Vol 460|13 August 2009

NEWS & VIEWS

MATHEMATICAL PHYSICS

A tight squeeze
Henry Cohn
How can identical particles be crammed together as densely as possible? A combination of theory and
computer simulations shows how the answer to this intricate problem depends on the shape of the particles.
We all know from experience with luggage just
how difficult it is to pack objects efficiently
into a limited space. These difficulties are even
greater when huge numbers of objects, such
as grains of wheat, are involved. From Luke
the evangelist’s reference to “a good measure,
pressed down, shaken together and running
over” all the way to modern disclaimers that
contents may have settled during shipping,
nobody has been able to analyse how to achieve
the densest possible packings. On page 876 of
this issue, Torquato and Jiao describe1 remark-
able computer-simulation results that show
how subtle this problem can be, while offering
new hope for understanding important cases.
Identical, perfect spheres are among the
simplest objects to pack. It is not difficult to
guess how — just look at the way cannonballs
are stacked at war memorials. But theoretical
analysis of the problem presents profound dif-
ficulties that were overcome only recently by
Thomas Hales2, nearly four centuries after the
answer was conjectured by Johannes Kepler3.
Of course, most granular materials do not con-
sist of spherical grains, and this complicates
matters tremendously. For most grain shapes
we cannot guess or even closely approximate
the answer, let alone prove it, and it is difficult
to develop even a qualitative understanding of
the effects of grain shape on packing density.
Instead of perfect spheres, Torquato and
Jiao study1 packings of the five Platonic solids:
the tetrahedron, cube, octahedron, dodecahe-
dron and icosahedron (see Fig. 1 on page 877).
These are the simplest and most symmetrical
polyhedra. Needless to say, nobody expects the
grains in physical materials to have these pre-
cise shapes, but they are a beautiful test case for
understanding the effects of corners and edges
and the role of symmetry.
The cube-packing problem is easy — cubes
can fill space completely. But the densest
packings of the other Platonic solids are much
less obvious. They do not tile space — fill
space with no gaps or overlaps — despite Aris-
totle’s incorrect assertion4 that tetrahedra do.
In their simulations, Torquato and Jiao find
a striking difference between two cases: the
octahedron, dodecahedron and icosahedron
have central symmetry (that is, each remains
© 2009 Macmillan Publishers Limited. All rights reserved
a b
Figure 1 | Two-dimensional view of packings. a, In a two-dimensional Bravais lattice packing, there is one
particle per lattice cell, and the particles are all aligned with each other. In this illustration, the particle
is represented by a triangle and is located in the upper left corner of the cell. b, In a general, non-Bravais
lattice packing, there are multiple particles per cell, and they can be arbitrarily rotated. In their computer
simulations, Torquato and Jiao1 show that the best packings of centrally symmetric Platonic solids turn
out to be Bravais lattice packings, which is remarkable given how restricted such packings are.
unchanged by a 180° rotation about a point at use a powerful simulation technique. Starting
its centre). But the tetrahedron does not have with an initial guess at a dense packing, they
central symmetry, and this turns out to be the gradually modify it in an attempt to increase
crucial distinction. its density. In addition to trying to rotate or
In the centrally symmetric cases, Torquato move individual particles, they also perform
and Jiao show that the highest-density pack- random collective particle motions by means
ings belong to the simplest kind, called Bravais of deformation and compression or expansion
lattice packings (Fig. 1a), although this con- of the lattice’s fundamental cell. With time, the
straint is never directly imposed on their simulation becomes increasingly biased
simulations. In such arrangements, all the par- towards compression rather than expansion.
ticles are perfectly aligned with each other, and Allowing the possibility of expansion means
the packing is made up of lattice cells that each that the particles are initially given consider-
contain only one particle. The densest Bravais able freedom to explore different possible
lattice packings had been determined previ- arrangements, but are eventually squeezed
ously 5,6, but it had seemed implausible that they together into a dense packing.
were truly the densest packings, as Torquato The new tetrahedron packing is a variant of
and Jiao’s simulations and theoretical analysis an ingenious construction found by Chen last
now suggest. By contrast, for the tetrahedron it year7. Using physical models, she observed that
has long been known that Bravais lattice pack- tetrahedra pack well when arranged in a form
ings are far from optimal, and in this case the Torquato and Jiao call ‘wagon wheels’: wheels of
authors achieve a record density: they find a five tetrahedra sharing an edge, with the wheels
non-Bravais lattice configuration (Fig. 1b) joined in pairs at right angles (see Fig. 3a on
that fills up 78.20% of the space available page 878). How best to arrange these pairs of
(an improvement over the previous record7 of wagon wheels is not clear, but Chen used a
77.86%, or 36.73% for Bravais lattices8). computer algebra system (a software program
To find their packings, Torquato and Jiao that manipulates mathematical formulae)
801
NEWS & VIEWS
to optimize their placement and achieved a
density of 77.86%, which is a vast improve-
ment over the previous record9 of 71.75%.
The authors’ simulations1 suggest that Chen’s
solution was nearly, but not quite, optimal.
Although Torquato and Jiao’s improvement
on Chen’s packing is noteworthy, perhaps the
most intriguing implication of their work is the
apparent optimality of Bravais lattice packings
for the centrally symmetric Platonic solids (as
well as generalizations such as Archimedean
polyhedra). This part of the work may seem
less exciting, because the densest packings
turned out to be known already. However, in a
field with few clear organizing principles, this
latest insight into the part played by symmetry
might take on the role of a twenty-first-century
Kepler conjecture. It will surely inspire
many future research papers, and with any
STEM CELLS
Escaping fates with open states
Robert J. Sims III and Danny Reinberg
The ability of embryonic stem cells to give rise to any cell type relies on
a remodelling protein that maintains open chromatin. But the chromatin
landscape of these cells may be more complex than previously thought.
Embryonic stem (ES) cells are ideal models for
studying the molecular principles that dictate
cell fate. ES cells can self-renew and form every
type of cell in an organism — a property called
pluripotency. It is well established that DNA-
binding proteins set the stage for determining
cell-type specificity by orchestrating complex
patterns of gene expression1. However, DNA
is tightly assembled with accessory proteins
into a structure called chromatin, which
limits its accessibility to regulatory proteins.
Chromatin can be loosely packed, or ‘open’,
allowing DNA-binding proteins ready access
to the genome. Alternatively, it can be densely
compacted, or ‘closed’, making the DNA rela-
tively inaccessible2. On page 863 of this issue,
Gaspar-Maia et al.3 provide evidence that
chromatin structure is intimately connected
with cell fate by showing that highly accessible
chromatin is essential for the unique properties
of stem cells.
ES cells have to achieve a delicate balance
of gene regulation: they must suppress genes
that result in premature differentiation, while
maintaining expression of genes that allow
self-renewal. Previous work4,5 has shown that
ES cells maintain an open chromatin structure
and express a large proportion of their genes.
Differentiation of ES cells into mature cell
types, such as neural cells, is accompanied by
closing of chromatin4 and by widespread gene
silencing5. As chromatin structure strongly
dictates whether or not genes can be expressed,
Gaspar-Maia and colleagues3 examined the
802
luck we won’t have to wait 400 years for a full
understanding of it. ■ The interplay between Chd1, H3K4me3
Henry Cohn is at Microsoft Research New
England, One Memorial Drive, Cambridge,
Massachusetts 02142, USA.
e-mail: cohn@microsoft.com
1. Torquato, S. & Jiao, Y. Nature 460, 876–879 (2009).
2. Hales, T. C. Ann. Math. 162, 1065–1185 (2005).
3. Kepler, J. Strena Seu de Nive Sexangula [A New Year’s Gift of
Hexagonal Snow] (Godfrey Tampach, 1611).
4. Aristotle On the Heavens Book III, Pt 8 (transl. Guthrie,
W. K. C.) (Harvard Univ. Press, 1939).
5. Minkowski, H. Nachr. Akad. Wiss. Göttingen Math. Phys.
Kl. II, 311–355 (1904).
6. Betke, U. & Henk, M. Comput. Geom. 16, 157–186
(2000).
7. Chen, E. R. Discrete Comput. Geom. 40, 214–240
(2008).
8. Hoylman, D. J. Bull. Am. Math. Soc. 76, 135–137 (1970).
9. Conway, J. H. & Torquato, S. Proc. Natl Acad. Sci. USA 103,
10612–10617 (2006).
role of enzymes that can modify this structure
in keeping ES cells pluripotent.
The authors found that decreasing the
amounts of a chromatin-remodelling protein
called Chd1 impaired ES-cell proliferation
and reduced expression of the DNA-binding
transcription factor Oct4, which is a master
regulator of ES-cell function. Chd1 has previ-
ously been shown to facilitate gene expression6,
consistent with its role in Oct4 regulation.
However, in this case3 its depletion did not
result in a global decrease in gene activity, as
would be expected for a factor widely associ-
ated with active gene transcription. Instead, ES
cells lacking Chd1 upregulated the expression
of genes involved in nervous-system develop-
ment, and they tended to differentiate into cells
of the neuronal lineage3.
Whether chromatin structure is accessible
or compact is determined in part by chemical
modification of its core protein components,
histones. For example, trimethylation of lysine 4
on histone H3 (H3K4me3) is a marker of
open chromatin. Chd1 contains a motif that
recognizes H3K4me3 (refs 7, 8) and indeed,
when Gaspar-Maia and colleagues performed
genome-wide studies of Chd1 localization
on chromatin in normal ES cells, they found
that Chd1 co-localizes with H3K4me3. When
the authors depleted ES cells of Chd1, they
observed an increase in the proportion of con-
densed chromatin in these cells. On the basis
of their results, they surmise that Chd1 main-
tains an open chromatin structure in ES cells,
© 2009 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 460|13 August 2009
which is required for their pluripotency.
and open chromatin (Fig. 1) highlighted by
Gaspar-Maia and colleagues3 is reinforced
by findings from other studies9–13. H3K4me3
is incorporated into chromatin during active
transcription9; therefore, promotion of gene
expression by Chd1 might ensure that chro-
matin is rich in H3K4me3 and maintained
in an open conformation. Chd1 has also
been shown10 to regulate DNA-replication-
independent deposition of chromatin enriched
in H3K4me3. Moreover, binding of Chd1 to
chromatin may protect H3K4me3 from turn-
over through histone demethylation. This
histone mark also prevents the binding of
factors that mediate gene silencing, such as the
NuRD histone-modifying complex11,12 and the
DNA methyltransferase subunit DNMT3L13.
Loss of H3K4me3 would therefore allow
increased chromatin compaction induced
by NuRD and DNA methylation, highlight-
ing the need to maintain a correct balance of
H3K4me3. In addition, the histone methyl-
transferase Suv39H1 facilitates the formation
of silent chromatin domains by methylating
lysine 9 on histone H3. Suv39H1 is known
to act on histones that lack methylated H3K4
(Fig. 1); thus, open chromatin would serve to
counteract this repressive activity.
What Gaspar-Maia and colleagues’ work3
does not explain is how, despite depletion
of Chd1, global ES-cell gene activation is
largely unaffected while genes that drive dif-
ferentiation towards the neuronal cell fate
are upregulated. It is possible, though, to rec-
oncile these findings by taking into account
Chd1-mediated regulation of the distribution
of H3K4me3. For instance, the p400/TIP60
complex, which turns genes on or off by chang-
ing the composition of chromatin, also seems
to bind H3K4me3 (ref. 14). When the amount
of H3K4me3 is reduced in ES cells, bind-
ing of p400/TIP60 to its chromatin targets
is impaired14. In this way, the perturbation
of H3K4me3 in the absence of Chd1 could
impair gene silencing through its effects on
factors such as p400/TIP60. Although unusual
for somatic cells (non-gametes), in ES cells,
H3K4me3 may be associated with silencing
of genes at focal locations throughout
chromatin15. Alternatively, widespread gene
reactivation could be a consequence of indirect
effects of the downregulation of ES-cell master
regulatory factors such as Oct4.
Studying differentiated cells undergoing
reprogramming to a pluripotent state16 allows
insight into the core properties of stem-cell
chromatin. Cells that fail reprogramming reac-
tivate subsets of stem-cell-related genes, but
simultaneously fail to repress lineage-specific
transcription factors16. Thus, the correct mix of
gene expression to achieve stem-cell properties
may depend more on local alterations in chro-
matin structure than on global chromatin reor-
ganization. Gaspar-Maia et al.3 show that Chd1
loss prevents the reprogramming of somatic