J. theor. Biol.

(1997) 187, 297306

A Symmetrical Theory of DNA Sequences and Its Applications

C-T Z

Department of Physics, Tianjin University, Tianjin 300072, China

(Received on 1 August 1996, Accepted in revised form on 3 January 1997)

A unified symmetrical theory of DNA sequences has been established based on the basic symmetry of
the DNA bases. It is shown that the symmetry of DNA sequences is inherently related to that of a cube
and its inscribed regular tetrahedron. A DNA group is defined as a particular alternating group of order
4, in which the permuted objects are four bases. The symmetry of DNA sequences is described by the
DNA group which is isomorphic to the tetrahedral group. The matrix representation for the DNA
group has been obtained, and used to establish the relationships between the transforms of bases and
the rotations of the tetrahedron. It is found that any DNA sequence can be uniquely described by three
independent distributions, i.e., the distributions of the bases of purine/pyrimidine, of amino group/keto
group and of strong/weak hydrogen bonds along the sequence. The three distributions are invariant
in some sense under the transforms of the DNA group, indicating that the three distributions are
inherent for the sequence. The mathematical format of the theory lays a foundation for further
development. The applications of the theory to analyse some DNA sequences are presented.
7 1997 Academic Press Limited

1. Introduction bonds are formed in the G-C pair, while two

We are living in the age of information explosion,
especially the explosion of biological information.
The DNA sequences constitute the primary biologi-
cal information. As pointed out by Gilbert (1991),
for 15 years, the DNA database has grown by 60%
a year, a factor of ten every 5 years. The human
genome project has accelerated this rate of increase.
In response to a large amount of DNA information
in the database, a new paradigm, now emerging, is
that the starting point of the study will be theoretical
cases where most gene sequences are known (Gilbert,
1991). This paper is devoted to a theoretical study on
the symmetry of DNA sequences. The symmetry of
DNA sequences is an important and basic theoreti-
cal topic, although to our knowledge, it seems that
only a few studies on this topic have been published
(Magarshak & Benham, 1992). The DNA sequence
consists of four kinds of bases, A, C, G and T. The
sequence exhibits some symmetry which is based on
that of the DNA bases. For example, three hydrogen
hydrogen bonds are formed in the A-T pair. This
symmetry of WatsonCrick pair leads to a basic
symmetry of DNA double helix: the double helix
keeps invariant under the transform A t T and
G t C. The symmetry of the WatsonCrick pair
also leads to the fact that the G + C (or A + T)
content in the sequence, rather than those of
individual G or C, possesses a more clear biological
meaning. More examples regarding the symmetry of
the DNA sequences are related to the multiple
recognition sequences by some enzymes. For in-
stance, the restriction enzyme AccI recognizes the
sequences 5'-GT (A or C) (G or T) AC-3'
(Cornish-Bowden, 1985). There exists a common
amino group in both bases A and C, and a common
keto group in both bases G and T. Therefore, the
above multiple recognition sequences exhibit a
symmetry of amino and keto groups. It is the aim
of this paper to establish and present a theory to
deal with all of the symmetry of DNA sequences in
a unified form.

00225193/97/150297 + 10 $25.00/0/jt970401 7 1997 Academic Press Limited

298 .-.

A C
2. Symmetrical Theory of DNA Sequences NH 2 NH 2

N 5
7
6
1
N 5 4 3N
8
9 2 6 1 2
The starting point of our study is based on the N 4
3
N N O
observation of symmetry in the four DNA bases. The
chemical structures of adenine (A), cytosine (C),
guanine (G) and thymine (T) are shown in Fig. 1. The
nomenclatures of bases and the sets of bases are G T
adopted according to Recommendation 1984 O O
proposed by the NC-IUB (Cornish-Bowden, 1985). H CH 3 H
N 5 6 4
The bases can be divided into two classes, i.e., 7 1
N 5 3
N
8
9 2 6 1 2
purine, R = A, G, 3

Bases
~ N 4
N NH 2 N O
_ pyrimidine, Y = C, T.
The bases can be also divided into another two F. 1. The chemical structures of the four DNA bases.
classes,

Bases
~ amino group, M = A, C,
_ keto group, K = G, T.
The above two classifications are based on the pure
chemical structures, as shown in Fig. 1. However,
there are stronger hydrogen (H) bonds in the G-C
pair and weaker hydrogen bonds in the A-T pair in
the double helix. Therefore, the bases can be further
divided into another two classes according to the
strength of the hydrogen bonds, i.e.,
~ strong H-bonds, S = G, C,
Bases
_ weak H-bonds, W = A, T.
The symmetry of DNA bases observed above may
be represented by some symmetrical geometrical
entities in somewhat analogous style to that used by
elementary particle theorists in assigning individual
particles to some symmetrical geometrical entities.
The first geometrical entity that we used to represent
the above symmetry is a cube. Each cube has six
identical faces which can be divided into three pairs
according to the three directions. They are the
right/left faces (x direction), the front/back faces
(y-direction) and the up/down faces (z direction).
Without generality, we assign the purine/pyrimidine
bases to the right/left faces, respectively. Similarly, we
assign the bases of amino/keto group to the
front/back faces, respectively. Finally, we assign the
bases of weak/strong H-bonds to the up/down faces,
respectively. Drawing the diagonals at each face in the
manner as shown in Fig. 2, we obtain a regular
tetrahedron. The four vertices are referred to as A, C,
G and T, respectively, as shown in Fig. 2. In addition
to the vertices, there are four triangular faces in the
tetrahedron, i.e., DCGT, DAGT, DACT and DACG.
They correspond to the bases of non-A, non-C,
non-G and non-T, and denoted by B, D, H and V,
respectively, (Cornish-Bowden, 1985). There are six
edges in the tetrahedron, i.e., those of AG, CT, AC,
GT, AT and GC. They correspond to the bases of
purine, pyrimidine, amino group, keto group, weak
H-bonds and strong H-bonds, respectively. The edges
AG and CT are situated at the faces of purine and
pyrimidine, respectively, and AG_CT, as shown in
Fig. 2. The edges AC and GT are situated at the faces
of amino and keto group, respectively, and AC_GT.
Finally, the edges AT and GC are at the faces of weak
and strong H-bonds, respectively, and AT_GC. The
above symmetrical goemetrical entities, i.e., the cube
Z
A
Y
T
O
X
C
G
F. 2. A cube and its inscribed regular tetrahedron, which are
used to represent the symmetry of the DNA bases. As usually, A,
C, G and T are referred to as the four DNA bases, respectively.
Note that the coordinate system is set up by using the three middle
lines of the regular tetrahedron.

and the regular tetrahedron, are the foundation of the
symmetrical theory presented in this paper.
It should be pointed out that Trainor and
co-workers made use of a similar tetrahedral
representation for the codon sequences (Trainor
et al., 1984).
T-group. On the other hand, the set of all possible
Consider a DNA sequence with N bases. Inspect
the sequence by stepping one base at a time. Let the
steps be denoted by n, n = 1,2, . . ., N. Calculate An ,
Cn , Gn and Tn for the nth step, where An (Cn , Gn , Tn )
is the cumulative number of occurrence of base A (C,
G, T) in the sub-sequence situated at the region from
the first to the nth base in the DNA sequence
inspected. Obviously, An , Cn , Gn and Tn are all positive
integers. The four integers (An , Cn , Gn , Tn ) can be
mapped onto a point Pn in the three-dimensional
space, based on the symmetry of the regular
tetrahedron (Zhang & Zhang, 1994). In the case of
coordinate system set up in Fig. 2, the coordinates xn ,
yn and zn of point Pn can be expressed in terms of An ,
Cn , Gn and Tn (Zhang & Zhang, 1991a, b), i.e.,
F xn = 2(An + Gn ) n,
g yn = 2(An + Cn ) n,
f zn = 2(An + Tn ) n,
xn ,yn ,zn $[n,n], n = 1.2, . . .,N. (1)
When n runs from 1 to N, we have points P1, P2,
. . ., PN . The appropriate curve connecting points P1,
P2, . . ., PN one by one is called the Z curve of the
DNA sequence (Zhang & Zhang, 1994). The points
P1, P2, . . ., PN are referred to as the nodes of the Z
curve. A DNA sequence is uniquely determined by
the coordinates of the nodes of the Z curve.
Therefore, a DNA sequence may be represented by a
3 N matrix of the following form
2 3
x1 x2 . . . x N
(DNA sequence) = y1 y2 . . . yN . (2)
z1 z2 . . . zN
There are apparent biological meanings in eqns (1),
where xn , yn and zn display, respectively, the
distributions of bases of purine/pyrimidine, amino/
keto group, and strong/weak H-bonds along the
sequence. When the occurrence number of bases of
purine (amino group, weak H-bonds) in the
sub-sequence situated from the first base through the
nth base is greater than that of pyrimidine (keto
group, strong H-bonds), then xn q 0 (yn q 0, zn q 0,
otherwise, xn Q 0 (yn Q 0, zn Q 0). The three distri-
butions can be uniquely determined by the given
299
DNA sequence and the DNA sequence can be
uniquely described by the given three distributions.
A regular tetrahedron (RT) is a body of high
symmetry. The set of all possible motions which keep
the RT fixed in the space forms a terahedral group or
permutations on the set of four symbols forms a
symmetrical group S4. The alternating group A4 is an
invariant subgroup of S4. The T-group and the A4
group are isomorphic to each other. A DNA group
is defined as a particular A4 group in which the
permuted objects are the four bases of a DNA
molecule. In fact, the DNA group and the T-group
are the same group from a point of view of the
abstract group. They have the same group structure
and the same matrix representation.
To find the matrix representation of the DNA
group, we study the rotational motions on the RT.
Let R be a 3 3 rotational matrix representing a
rotation of the coordinate system around some axis.
Under the rotation R, the representation matrix
[eqn (2)] for a DNA sequence is transformed into
2 3 2 3
x'1 x'2 . . . x'N x1 x2 . . . x N
y'1 y'2 . . . y'N = R y1 y2 . . . yN , (3)
z'1 z'2 . . . z'N z1 xz . . . zN
where x'1 , y'1 , z'1 etc. are the new coordinates after the
rotation. The order of the DNA group is 12.
Referring to Fig. 2, we find that the elements of the
DNA group consist of I, i.e., the identity operation;
Rx , Ry , and Rz , i.e., the 180 rotation around the x,
y and z axis, respectively; RA , RC , RG , and RT , i.e., the
120 rotation around the OA, OC, OG and OT axes,
respectively; and RA2 , RC2 , RG2 , and RT2 , i.e., the 240
rotation around the OA, OC, OG and OT axes,
respectively. The rotational matrix for a rotation of
angle u around the x axis is
2 3
1 0 0
Rx (u) = 0 cosu sinu . (4)
0 sinu cosu
when u = 180, we find
2 3
1 0 0
Rx (180) = Rx = 0 1 0 . (5)
0 0 1
The matrix representations for the other elements of
the DNA group can be found similarly. The total
representation of the DNA group is summarized as
300 .-.

T 1
Multiplication table of the DNA group
I Rx Ry Rz RA RC RG RT RA2 RC2 RG2 RT2
I I Rx Ry Rz RA RC RG RT RA2 RC2 RG2 RT2
Rx Rx I Rz Ry RC RA RT RG RT2 RG2 RC2 RA2
Ry Ry Rz I Rx RT RG RC RA RG2 RT2 RA2 RC2
Rz Rz Ry Rx I RG RT RA RC RC2 RA2 RT2 RG2
RA RA RT RG RC RA2 RG2 RT2 RC2 I Rx Rz Ry
RC RC RG RT RA RT2 RC2 RA2 RG2 Rx I Ry Rz
RG RG RC RA RT RC2 RT2 RG2 RA2 Rz Ry I Rx
RT RT RA RC RG RG2 RA2 RC2 RT2 Ry Rz Rx I
RA2 RA2 RC2 RT2 RG2 I Rz Ry Rx RA RT RC RG
RC2 RC2 RA2 RG2 RT2 Rz I Rx Ry RG RC RT RA
RG2 RG2 RT2 RC2 RA2 Ry Rx I Rz RT RA RG RC
RT2 RT2 RG2 RA2 RC2 Rx Ry Rz I RC RG RA RT

23 2 3
follows 1 1
(A) = 1 , (C) = 1 ,

2 3 2 3
1 1
1 0 0 1 0 0
I= 0 1 0 , Rx = 0 1 0 ;

23 23
0 0 1 0 0 1 1 1
(G) = 1 , (T) = 1 . (7)

2 3 2 3
1 1
1 0 0 1 0 0
Ry = 0 1 0 , Rz = 0 1 0 ;
0 0 1 0 0 1 Considering the rotation Rx first, we find

2 3 23 2 3
1 0 0 1 1

2 3 2 3
0 0 1 0 0 1 Rx (A) = 0 1 0 1 = 1 = (G), (8)
RA = 1 0 0 , RC = 1 0 0 ; 0 0 1 1 1
0 1 0 0 1 0
and similarly,

2 3 2 3
0 0 1 0 0 1 Rx (C) = (T), Rx (G) = (A), Rx (T) = (C). (9)
RG = 1 0 0 , RT = 1 0 0 ;
So, the transforms of the DNA bases under Rx , Ry Rz
0 1 0 0 1 0
may be expressed as follows
Rx : A \ G, C \ T,

2 3 2 3
(10)
0 1 0 0 1 0
2
R = 0
A 0 1 , RC2 = 0 0 1 ; Ry : A \ C, G \ T, (11)
1 0 0 1 0 0
Rz : A \ T, G \ C. (12)

2 3 2 3
0 1 0 0 1
RG2 = 0 0 1 , RT2 = 0
1 0 0 1 0
The multiplication table for the DNA group can be
simply obtained by the products of the matrices in
eqns (6). The result is listed in Table 1.
We turn to study the transform of the DNA bases
under each of the 12 operational elements of the DNA
group. Consider a DNA sequence with only one base,
i.e., N = 1. In this case we have four possible DNA
sequences, i.e., A, C, G and T. By using eqns (1),
we obtain their matrix representation in the form of
eqn (2), i.e.,
Usually, the transform Rx is called transition, while Ry
0
and Rz are called transversion. In this paper, Rz is also
0 1 (6)
called a complementary transform specially. The four
0
elements I, Rx , Ry and Rz form an invariant subgroup
of the DNA group, as can be seen in Table 1. It is
isomorphic to the Klein 4-group (K4 group). For the
other operations we find
RA : A : A, C : T, T : G, G : C, etc. (13)
The 12 operational elements of the DNA group are
listed in Table 2, which can also be obtained by
rotating the RT in Fig. 2. For example, a 180
rotation of the RT around the z-axis yields the
transform of A \ T and G \ C, as can be seen
clearly in Fig. 2. The K4 group and its two cosets
 301

T 2
Transforms of the bases under the Any DNA sequence has an inverted sequence
operations of DNA groups corresponding to it. For example, a sequence is
A C G T ACGT, and the inverted sequence is TGCA. The
I A C G T
Rx G T A C
operation to inverse a sequence is denoted by Rr . The
Ry C A T G operation Rr can be described as follows. Suppose
Rz T G C A that the Z curve representing a DNA sequence with
RA A T C G
RC G C T A
N bases is described by xn , yn and zn . Then the Z curve
RG T A G C representing the inverted sequence can be described
RT C G A T by xn , yn , and zn , where
RA2 A G T C
RC2 T C A G x n = xN xN n ,
RG2 C T G A
RT2 G A C T yn = yN yN n , (14)
zn = zN zN n ,
n = 0, 1, 2, . . ., N.
Therefore, if a Z curve representing the DNA
exhaust the DNA group. So, the elements of the DNA
sequence read from the direction of 5' to 3', then the
group are divided into four classes: i.e., (I), (Rx , Ry ,
Z curve representing the complementary strand in the
Rz ); (RA , RC , RG , RT ); and (RA2 , RC2 , RG2 , RT2 ).
same direction (5' to 3') can be obtained by performing
The three components of the Z curve i.e., xn , yn and
a joint operations RrRz to the original Z curve. In this
zn represent the three independent distributions which
case, in addition to the Rz operation, an operation Rr
completely describe the DNA sequence. The trans-
is still needed. This is due to the fact that the DNA
forms of xn , yn and zn under the 12 elements of the
double helix consists of two anti-parallel strands.
DNA group are shown in Table 3. It is seen that the
three distributions are invariant except their signs are
changed or their positions are exchanged probably
under the 12 operations of the DNA group. Note that What do we mean by the symmetry of DNA
all of xn , yn , and zn are composed of the sums of the sequences? Generally speaking, symmetry means that
numbers of bases. This implies that the sums of the some properties of things keep invariant under some
numbers of bases, rather than those of the individual transform. For example, an object and its image are
bases, have more biological meaning. For a long time, said to be of symmetry because the object keeps
it is well known that the G + C content, rather than invariant under the reflection transform. Similarly, in
the content of individual G or C, is a meaningful the case of DNA sequence, symmetry means that
biological quantity widely used by biologists to some properties of the sequence keep invariant when
describe the DNA sequence. In our theory, the sums a total or a part of its bases undergoes the transforms
A + G, C + T, A + C, G + T, A + T and G + C described by the DNA group. There are two kinds of
appear naturally. symmetry of the DNA sequences, i.e., the global and
the local symmetry. If the property of a sequence
keeps invariant when a total of bases in the sequence
undergoes some transform, the sequence is said to be
of global symmetry with respect to that transform.
T 3
Otherwise, if the property of a sequence keeps
Transforms of x, y and z under the
invariant only when one base or a part of bases in the
operations of DNA group
sequence undergoes some transform, the sequence is
x y z
I x y z
said to be of local symmetry with respect to that
Rx x y z transform.
Ry x y z One of the examples of the global symmetry of
Rz x y z
RA z x y
DNA sequences that we have found is the symmetry
RC z x y of Z curve. The Z curve keeps invariant under the 12
RG z x y operations of the DNA group. To illustrate this,
RT z x y
RA2 y z x
suppose that there are two DNA sequences both
RC2 y z x having N bases. They are represented by Z1 curve (P1,
RG2 y z x P2, . . . PN ) and Z2 curve (Q1, Q2, . . . QN ), respectively,
RT2 y z x
where Pi , and Qi are the nodes of the Z1 and Z2
302
curves. The two Z curves are said to be of the same
structure, if for any n = 1, 2, . . ., N, we always
have =OPn = = =OQn =, where =OPn =(=OQn =) is the
distance between the origin and the point Pn (Qn ). It
can be proved that any Z curve representing a DNA
sequence is invariant under the operations of the
DNA group, in the sense that the Z curve and its
transformed Z curves have the same structure. This
can be proved as follows. Using eqns (1), we find
rn 0 =OPn = = (xn2 + yn2 + zn2 )1/2
= 2(An2 + Cn2 + Gn2 + Tn2 n 2/4)1/2, (15)
n = 1, 2, . . ., N.
Therefore, rn is invariant under the operations of the
DNA group. Generally speaking, there exist 12 DNA
sequences having the same length, which are
represented by the same Z curve in the sense that the
structures of their Z curves are the same. For
example, the DNA sequence ACGT and its
transformed sequences under the operations of the
DNA group are GTAC, CATG, TGCA, ATCG,
GCTA, TAGC, CGAT, AGTC, TCAG, CTGA, and
GACT, respectively. All the 12 sequences have the
same Z curve in the sense that the corresponding 12
Z curves have the same structure. In other words,
starting from the Z curve of the sequence ACGT, the
Z curves of the remaining 11 sequences can be
obtained by the rotational operations described by
the DNA group.
Another invariant quantity of the DNA group is
the information entropy H of the DNA sequence,
which is defined by
4
H = s pi log2 pi , (16)
i=1
where pi (i = 1, 2, 3, 4) are the occurrence frequencies
of bases A, C, G and T in the sequence, respectively.
The concept of isentropic surface has been introduced
based on this symmetry and used to study the
molecular evolution (Zhang & Zhan, 1994; Wang &
Zhang, 1996).
The Fourier transform of xn , yn , zn will be an
important tool for studying the three distributions.
Suppose that X(k), Y(k) and Z(k) are the discrete
Fourier transforms of xn , yn , and zn , respectively,
where k = 1, 2, . . ., N, then the power spectrum is
defined as
P(k) = =X(k)= 2 + =Y(k)= 2 + =Z(k)= 2,
where =X(k)= means the module of X(k), etc. It is easy
to see from Table 3 that for any given value of k, P(k)
is invariant under the operations of the DNA group.
.-.
In summary, the Z curve, the information
entropy H and the Fourier power spectrum P(k)
defined by eqn (17) have the global symmetry with
respect to the 12 operations of the DNA group.
Among the 12 operations of the DNA group, the
complementary transform Rz has a direct biological
meaning. It means an exchange of bases at each site
between the strand and its complementary strand.
Therefore, the Z curve, the information entropy H
and the Fourier power spectrum P(k) are invariant
for both strands.
The use of a single symbol to designate a variety
of possible bases at a single position indicates that
there is a local symmetry of the DNA sequence
according to our nomenclature. For example,
isoleucine is coded by the codons 5'-ATH-3', where
H means the bases of not-G. This means that the
coding property (coding for isoleucine) is invariant
when the bases at the third position undergo the Rg
or Rg2 operations. In fact, all the coding sequences
exhibit the local symmetry. Another example of the
local symmetry is about the DNA sequences
recognized by some enzymes. For example, consider
the recognition sequences by the restriction en-
zymes. Most of the restriction enzymes of type II
recognize a simple unique DNA sequence. How-
ever, some of the restriction enzymes recognize
series of derivative sequences, where two or more
bases may present at a particular postion in the
recognition sequence. For example, SduI recognizes
the sequence 5'-G (A or G or T) GC (A or C or
T) C-3' (Cornish-Bowden, 1985). The local sym-
metry exists at the second and fifth positions in the
sequence. It would be difficult to find common
biochemical features in the set of bases (A, G, T)
or (A, C, T). In the case of set of (A, G, T), they
are neither all purines nor pyrimidines; they are
neither all the bases of amino group nor keto
group; they are neither all the bases of strong
hydrogen bonds nor weak hydrogen bonds. The
existing phenomenon takes place naturally in our
theory. The recognition feature is invariant under
the operations Rc and Rc2 at the second position and
Rg , Rg2 at the fifth position. The local symmetry
described by the DNA group widely exist in the
DNA sequences. Besides the recognition sequences
by the restriction enzymes, the local symmetry also
exists in the sequences recognized by some other
enzymes or proteins. Generally speaking, there must
exist a local symmetry in any DNA sequence which
(17)
includes at least one of the following 11 symbols
within the sequence: R, Y, M, K, S, W, H, B, V,
D, and N, where N means any base. We will
discuss this problem in more detail later.
 303

T 4
Transforms of 10 codes under the operations of DNA group
R Y K M S W B D H V
I R Y K M S W B D H V
Rx R Y M K W S H V B D
Ry Y R K M W S D B V H
Rz Y R M K S W V H D B
RA M K S W Y R B V D H
RC K M W S Y R H D V B
RG K M S W R Y V B H D
RT M K W S R Y D H B V
RA2 W S Y R K M B H V D
RC2 W S R Y M K V D B H
RG2 S W R Y K M D V H B
RT2 S W Y R M K H B D V

3. Applications
In Table 1 of Recommendation 1984 by the
NC-IUB (Cornish-Bowden, 1985), 14 single-letter
codes representing incompletely specified bases in
nucleic acid sequences were recommended. They are
A (Adenine), C (Cytosine), G (Guanine), T
(Thymine), R (puRine), Y (pYrimidine), M (aMino),
K (Keto), S (Strong interation), W (Weak inter-
action), H (not- G, H follows G), B (not-A, B follows
A), V (not-T or not -U, V follows U), and D (not-C,
D follows C). As discussed in Section 2, the 14 codes
can be represented by four vertices, six edges and four
triangular faces of a regular tetrahedron. There is an
obvious equality: 4 + 6 + 4 = 14. The tetrahedral
representation of the 14 codes is useful for
memorizing these symbols and for comprehending the
relations of them.
The transforms of these 14 codes under the 12
operations of the DNA group are listed in Table 4.
Of the 12 operations the complementary element Rz
has apparent biological meaning. Under Rz , the
symbols are transformed into those in the comp-
lementary strand, respectively. Since the z-axis in
Fig. 2 is perpendicular to the edges of weak (W) and
strong (S) hydrogen bonds, the symbols W and S
are invariant under the rotation around the z-axis.
The transform of other symbols under Rz can be
obtained easily by rotating the tetrahedron around
the z-axis. The details of the transform for the 14
codes under Rz are specially shown in Fig. 3. Our
work gives both Table 1 and Table 2 of Recommen-
dation 1984 by the NC-IUB (Cornish-Bowden,
1985) an intuitive explanation and a simple way for
memory.
Since the discovery of the first type II restriction
endonuclease, the volume of type II restriction
endonuclease available in a purified form has been
increasing at a remarkable speed. The type II enzymes
represent those that both recognize and cleave the
DNA molecule at a specific sequence. As is well
known that most restriction endonucleases recognize
simple unique DNA sequences. However, a growing
class of endonucleases includes those that recognize
series of derivative sequences, where two or more
bases may be present at a particular position in the
recognition sequence. As mentioned previously, the
sequence displays a local symmetry. Since the
symmetry of the DNA sequences, according to our
theory, should be described by the DNA group, there
should be some rules to control the multiple
recognizing DNA sequences. For example, the
enzyme SduI recognizes the sequence 5'-GDGCHC-3'
(Perbal, 1990). Performing the operation Rz to this
sequence and considering Table 4 or Fig. 3, we obtain
the transformed sequence 5'-CHCGDG-3'. Perform-
ing a reflection operation Rr to the recognition
sequence, we find that the reversed sequence is
5'-CHCGDG-3'. The two sequences are identical, or
simply, Rz = Rr , indicating that the two operations
are equivalent in this case. A multiple DNA sequence
is defined as a DNA sequence which consists of at
least one of the following 10 symbols: R, Y, M, K, S,
W, H, B, V and D. A multiple DNA sequence with
R Y M K S W D H B V
R Y M K S W D H B V
F. 3. Display intuitively the transforms of 10 codes under the
Rz (complementary) operation.
304 .-.
the property that Rz = Rr is called a self-complemen-
tary sequence. Looking at the list of restriction
endonucleases (Perbal, 1990), we find that all the
multiple recognizing sequences are self-complemen-
tary. So, the first empirical rule that describes the
multiple recognizing sequence may be summarized as
follows:
Rule 1: all the multiple recognizing sequences of the
type II restriction endonucleases should be self-comp-
lementary.
Note from Fig. 3 that the two symbols in any one
of the following pairs (R, Y), (M, K), (D, H) and (B,
V) are complemenatary with each other, while S and
W are complementary with itself, respectively.
Therefore, as a complement to Rule 1, we still have
Rule 2: the codes (R, Y), (M, K), (D, H) and (B,
V) occurring in the multiple recognition sequences
must appear in pair, while the codes W and S in the
recognition sequences may either be in pair or single.
So far as we know that there is no exception of
these two rules for the multiple recognizing sequences
now available (Perbal, 1990). Furthermore, we can
predict the possible recognizing sequences not
available at present. For example, the sequences
SGGCCS, GBGCVC, CCSSGG, RGGSCCY etc. are
possible substrates of the restriction endonucleases,
based on the two rules proposed above. However,
whether such enzymes exist that recognize the above
multiple sequences is still an unsolved problem. The
two rules are only the necessary but not the sufficient
conditions for the multiple sequences which could be
recognized by the restriction endonucleases.
28 bases. The Z curve for this sequence is shown by
the gray curve in Fig. 4. Performing the Rr operation
The technique of computer graphics of the Z curves
provides a possibility to detect some symmetrical
patterns of the DNA sequences visually. Let us take
the inverted repeat sequence as an example, which
exhibits an important symmetry of the DNA
sequences. The inverted repeat sequences usually play
a key role in the recognition process between the
DNA molecules and the proteins or enzymes. The
format of the Z curve provides a method for finding
such a symmetry visually and quickly. One kind of
inverted repeat sequence is the so-called self-comp-
lementary sequence mentioned above, having the
form: 5'-XYZZ'Y'X'-3', where X, Y and Z are bases
or sets of bases and X', Y' and Z' are complementary
to X, Y and Z, respectively. Such sequences may form
the palindrome structures. Performing the Rz
operation to the above sequence, we obtain the
8
7
6
5
4
z axis
3
2
1
0
1 8
1.0 6
0.5 0 2 4
0
0.5 4 2 y axis
x axis 1.0 8 6
F. 4. The Z curve (the gray one) for the DNA sequence
5'-TGTGTGAATTGTACGTACAATTCACACA-3', which has
28 bases. Performing the Rr operation to the gray curve, as
described by eqn (14), we obtain the Z curve corresponding to the
inverted sequence, which is represented by the dark black one. Both
curves are smoothed by the method of spline function (Zhang &
Zhang, 1994). Note that the two curves are symmetrical with
respect to the z-axis, indicating that the sequence studied is an
inverted repeating one. For more details, see the text.
sequence 3'-X'Y'Z'ZYX-5'. Performing the Rr oper-
ation to the obtained sequence again, we have the
sequence 5'-XYZZ'Y'X-3', which is identical to the
inverted repeat sequence. Therefore, the condition for
a sequence to be an inverted repeat sequence may be
expressed by a simple formula: RrRz = RzRr = I,
where I represents the identity operation. Using the
multiplication table listed in Table 1, the above
formula may be rewritten as Rr = Rz . For example,
5'-CCGCGCGCGG-3' is a typical inverted repeat
sequence. A more complicated example is 5'-TGTGT-
GAATTGTACGTACAATTCACACA-3', which has
to the gray curve, as described by eqn (14), we obtain
the Z curve corresponding to the inverted sequence,
which is represented by the dark black curve in Fig. 4.
Both curves are smoothed by the method of spline
function (Zhang & Zhang, 1994). Rotating the gray
curve 180 around the z-axis, we find that the rotated
Z curve coincides with the dark black curve. In other
words, the two curves are symmetrical with respect to
the z-axis, indicating that the sequence studied is an
inverted repeat sequence. Thus, we find the symmetry
of the inverted repeat of the DNA sequence by a pure
geometric approach. This method provides a simple
and quick way to find the inverted repeat sequences
visually. Some other symmetrical patterns of DNA
sequences can be detected visually by any trained
researcher. It would be of considerable advantage to
analyse a great amount of DNA sequences quicky

and efficiently by this approach because the human
brain and eyes are generally more sensitive to
an intuitive graph than to an abstract symbolic
sequence.
4. Discussion and Summary
As a language (Trifonov & Brendel, 1986), the
DNA sequence may have a number of distinct forms
of expression, although all of them are equivalent.
The traditional form of the DNA sequence is a letter
string, consisting of four kinds of letters A, C, G and
T. This form is widely used in biochemistry and
molecular biology. It is necessary for the storage of
information and some statistical analysis. Therefore,
as a main form, it will be used continuously.
However, the form of letter sequence is extremely
difficult to recognize and find some special patterns
intuitively in the sequence. For a long sequence, the
form is too abstract to impress the observers. To
overcome the drawback of letter sequence, the
geometrical representation form of the DNA se-
quence has been proposed by many authors (Hamori
& Ruskin, 1983; Lathe & Findlay, 1984; Gate, 1985;
Pickover, 1992; Zhang & Zhang, 1994). Among the
geometrical representations of the DNA sequences,
the Z curve proposed by Zhang & Zhang (1994) is
the most general one. All of the remaining
geometrical forms mentioned above are the special
cases of the Z curve (Zhang & Zhang, 1994).
As we have seen in this paper, the format of the
Z curve is based on the symmetry of the DNA
sequence. The symmetry is described by a cube and
its inscribed regular tetrahedron. We believe that
there must be some inherent connection between the
symmetry of the DNA sequence and that of the
regular tetrahedron. In the present work, the
connection between the symmetry of the DNA
sequence and the rotational symmetry of the regular
tetrahedron has been established. The latter is
described by the tetrahedral group. Therefore, the
symmetry of the DNA sequence may be described by
the DNA group, which is isomorphic to the
tetrahedral group. The matrix representation of the
DNA group has been worked out. Each rotational
operation of the tetrahedron is represented by a
unique 3 3 matrix. At the same time, any DNA
sequence can be represented by a 3 N matrix,
based on the format of the Z curve, where N is the
length of the sequence. The transform of the bases
under an operation of the DNA group corresponds
to simply a matrix multiplication. Based on this
mathematical skill, the transforms listed in Tables 2
4 are obtained. Such a mathematical format is not
305
only very nice, but also very useful for the further
analysis of DNA sequences.
The concept of the three distributions in a DNA
sequence presented in this paper is an important and
a novel one. So far as we know no similar concept
has been reported in the literature. We can see from
Table 3 that the three distributions are invariant in
some sense under the 12 operations of the DNA
group. The three distributions are inherent for the
DNA sequence. Any DNA sequence is uniquely
described by the three distributions. Therefore, the
study of the DNA sequence may be carried out by
the study on the three distributions. This new
approach for studying the DNA sequence seems to
be very promising.
In conclusion, the study presented here and those
published previously (Zhang & Zhang, 1994) have
opened a new research area in which the geometrical
methods and concepts are widely introduced into the
study of DNA sequences. The symmetrical theory of
DNA sequences is based on this geometrical
approach. Our theory really reflects the symmetry of
some realistic DNA sequences in a unified form. This
theory provides a necessary complement to a great
amount of literature on the study of DNA sequences
available now.
The present study was supported in part by the grant
39570187 from the China Natural Science Foundation.
REFERENCES
C-B, A. (1985). Nomenclature for incompletely
specified bases in nucleic acid sequences: Recommendation 1984.
Nucl. Acids. Res. 13, 30213030.
G, M. A. (1985). Simpler DNA sequence representations.
Nature 316, 219.
G, W. (1991). Towards a Paradigm shift in biology. Nature
349, 99.
H, E. & R, J. (1983). H curves, A novel method of
representation of nucleotide series especially suitable for long
DNA sequences. J. Biol. Chem. 258, 13181327.
L, R. & F, R. (1985). Novel DNA sequence
representation. Nature 314, 585586.
M, Y. & B, C. J. (1992). An Algebraic Represen-
tation of RNA Secondary Structures. J. Biomol. Str. & Dyn. 10,
465488.
P, B. (1990). A Practical Guide to Molecular Cloning
(2nd Edn). Chapter 5. New York: John Wiley & Sons.
P, C. A. (1992). DNA and Protein tetragrams: Biological
sequences as tetrahedral movements. J. Mol. Graphics 10, 26.
T, L. E. H., R, G. W. & S, V. L. (1984). A
Tetrahedral Representation of Poly-Codon Sequences and a
Possible Origin of Codon Degeneracy. J. theor. Biol. 108,
459468.
T, E. N. & B, V. (1986). GNOMIC, A Dictionary of
Genetic Codes. Rehovot-Philadelphia: Balaban Publishers.
W, J.-H. & Z, C.-T. (1996). Study on the Isentropic
Equations of Nucleotide Sequences and Their Application. J.
theor. Biol. 181, 197202.
306 .-.

Z, C.-T. & Z, Y. (1994). Analysis on the Distribution of Z, C.T. & Z, R. (1991b). Diagrammatic representation
Bases in 1487 Human Protein Coding Sequences. J. theor. Biol. of the distribution of DNA bases and its applications. Int. J.
167, 161165. Biol. Macromol. 13, 4549.
Z, C.-T. & Z, R. (1991a). Analysis of distribution of Z, R. & Z, C.-T. (1994). Z curves, An Intuitive Tool for
bases in the coding sequences by a diagrammatic technique. Visualizing and Analyzing the DNA sequences. J. Biomol. Stru.
Nucl. Acids Res. 19, 63136317. Dyn. 11, 767782.