J Mol Evol (1995) 41:98-103
jou.. o, MOLECULAR
Distributions of the Use Frequencies of Amino Acids in the Hypervariable
Regions of Immunoglobulins
F. Lara-Ochoa, 1 E. Vargas-Madrazo, 2 J.C. Almagro I
1Instituto de Quiraica, UNAM, Apdo. Postal 70-213, Coyoacan04510, Mexico D.F.
2 Instituto de InvestigacionesBiologicas, UniversidadVeracruzana, Xalapa, Ver. Mexico
Received: 6 February 1994 / Accepted: 10 August 1994
Abstract. Two types of distributions for the frequen-
cies of occurrence of amino acids in each position of
hypervariable regions CDR-1 and CDR-2 were obtained
for 2,000 immunoglobulins. The results show that some
positions fit an inverse power-law distribution, while
others fit an exponential-type distribution. As a result of
comparison with structural data in the literature it is pro-
posed that sites in which the frequency distribution fits
the inverse power law are critical to maintaining canon-
ical shapes of the recognition regions or are involved in
modulating these canonical conformations, while those
sites where the distribution fits the exponential law are
those which should be exclusively involved in the rec-
ognition mechanism.
Key words: Immunoglobulins - - Canonical confor-
mations - - CDR - - Zipf law - - Inverse power law
Introduction
Immunoglobulins (Igs) are key proteins that mediate the
immune response by binding foreign epitopes, hindering
their ability to either bind to receptors on target cells, or
by marking invading microorganisms for destruction. An
important challenge in immunology deals with the unre-
solved question of how the immune system recognizes
foreign epitopes.
The antibody binding site is formed by six hypervari-
able or complementarity-determining regions (CDRs):
three in the variable domain of the light chain, V L, and
Correspondence to: F. L a r a - O c h o a
© Springer-VerlagNew York Inc. 1995
three in the variable domain of the heavy chain, V H
(Padlan 1994; Wu and Kabat 1970; Kabat and Wu 1971).
It has been proposed that the specificity of the Igs is
determined by the sequence and size of the CDRs (Kabat
1978).
CDR- 1 and CDR-2 of both V L and V H are codified by
a continuous gene fragment, while the CDR-3 is codified
by recombination events (Tonegawa 1983; Alt et al.
1987). It has been found that except for the third region
of V H, binding site loops have one of a small number of
main-chain conformations termed canonical structures
(Chothia and Lesk 1987). These specific backbone con-
formations are believed to reflect the existence of a few
key conserved residues in the loop and framework of the
antibody molecule (Chothia and Lesk 1987). This sug-
gests that some of the amino acids in the CDRs may have
a structural role, while the rest are either exclusively
involved in the recognition function or are irrelevant.
Likewise, Kabat et al. (1977) postulated the existence of
evolutionarily conserved residues within CDRs, and that
the antigen specificity is determined by a few hypervari-
able positions in the immediate neighborhood of such
conserved positions. Along similar lines, Ohno et al.
(1985) proposed that if some sites on the CDR-1 and
CDR-2 are conserved for the shaping of the primordial
antigen binding cavity, then the remaining sites must be
properly involved in the recognition process.
in a previous study (Vargas-Madrazo et al. 1994), it
was found that the frequency of use of amino acids in the
CDRs was skewed. Lara-Ochoa et al. (1994) performed
a statistical analysis of the first two regions of the anti-
body-combining sites in order to determine whether or
not the frequency of use of individual amino acids is
constrained by the role played by each position. The
 99

Table 1. Relative frequencies of amino acids in percent with the highest constancy on the CDR-1 and CDR-2 of the heavy and light domainsa

Predominant Relative Secondary Total Structural Conserved

Site amino acid frequency amino acids frequency (Chothia) (Kabat)

Heavy chain
26 a (98%)
27 F (48%) Y (42%) 90%
28 T (62%) S (22%) 84%
29 F (76%) L (11%) 87%
30 T (50%) S (41%) 91%
31 S (48%) T (5%) 53%
32 Y (65%) F (11%) 76%
34 M (61%) I (14), V (8%) 83% + +
51 I (88%) +
52a P (56%) S (10%) 66% +
52b K (87%)
54 N (27%) S (28%), G (27%) 82% + +
55 G (61%) S (14%) 75% +
57 T (76%)
59 Y (94%) F (2%) 96%
60 N (48%) A (21%), S (12%) 81%*
63 F (46%) V (28%), L (19%) 93%
64 K (77%)
65 G (61%) S (23%) 83%
Light chain
24 R (48%) K (16%) 64%
25 A (55%) S (30%), G (10%) 95%
26 S (87%) T (2%) 89%
27 Q (51%) N (1%) 52%
27a S (81%) G (11%) 92%
27b L (47%) V (19%), I (12%), A (11%) 89%
27c V (42%) L (32%) 74%
32 Y (72%) F (5%) 77%
33 L (58%) M (17%), V (10%) 85%
52 S (79%) T (7%) 86%
54 L (48%) R (48%) 96%*
56 S (72%) W(11%) 83%

The frequencies of amino acids for all positions listed fit an inverse power distribution
b In these special cases the secondary amino acids do not share the same properties of hydrophobicity and volume

distribution o f the f r e q u e n c y o f use o f t h e s e sites w a s fit
to t w o d i f f e r e n t t y p e s o f d i s t r i b u t i o n s ( L a r a - O c h o a et al.
1994). T h e thrust o f the p r e s e n t article is c o n c e r n e d w i t h
t w o aspects: first, to p e r f o r m statistical analysis o f the
f r e q u e n c y o f use o f a m i n o acids for e a c h p o s i t i o n o f the
C D R - 1 a n d C D R - 2 o f the V H and V L c h a i n s o f 2,000 Igs;
a n d s e c o n d , the c o m p a r i s o n o f this analysis w i t h the
o b s e r v e d use f r e q u e n c i e s o f a m i n o acids in the r e p o r t e d
s e q u e n c e s for the I g ' s g e r m - l i n e g e n e s ( G e n e - b a n k , re-
l e a s e 79).
Methodology and Results
The amino acid distribution table reported by Kabat et al. (1991) is used
to establish the percent frequencies of the most-used amino acids for
each site. From these data, we observe in Table 1 that amino acid uses
of some sites have frequencies that are of the same order or higher than
those of sites buried in the core of a protein family (Lira and Sauer
1989). As an illustration, and following the site numeration suggested
by Kabat et al. (1991), we observe in Table 1 that Gly-26 and Ile-5t on
VH have a probability of use of 0.98 and 0.88, respectively, which
should skew their distribution frequencies significantly. For some po-
sitions, the use frequencies may be as low as 0.48 (for instance, Phe-27
on Vu); in these cases, alternative amino acids which are second or
third in abundance, sharing similar properties of hydrophobicity and
volume, can be found in the same site (Table 1). Hence, all these cases
should correspond to a nonrandom frequency distribution. It is there-
fore of importance to determine for each site the type of distribution for
the amino acid use frequency and to attempt to infer from these dis-
tributions the role of each site relative to the CDR-1 and CDR-2.
The inverse power law gives a model of a skewed distribution that
seems appropriate for modeling the use frequency data. This distribu-
tion has been used in different contexts, including biological systems
(West 1985; Nicolis 1987), linguistics (Nicolis, 1986), and social sci-
ences (Montroll and Badger 1974) and to model diverse physical phe-
nomena (Montroll and Shlesinger 1983; Meijer et al. 1981). A version
of this distribution was proposed by Zipf (1949) for the first time. Here
we will use a modified version of Zipf's relation that was proposed by
Mandelbrot (1977), namely
P(r) = K (r + p)-13 (1)
where K, p, and [3 are parameters ([3 > 0) intrinsic to each system, while
P(r) denotes the probability of use of each amino acid with rank order
r. In linguistics, the physical meaning of the constants K, p, and 13has
been widely studied (Schroeder 1991, and references therein cited), K
 100
1 . . . . . . . . . . . . . . . . . . 0,5

0,~ . __ ...................................................................
0 .........................................................................

0 ............................................
-0.51 . + +

o,

_2_.~
-2 ~~_ -"I .~. . . . . .~. . . - ~I .......
-25

-,3,5 [--- ~ I----q--~ , ----T-----T-----I

-.02 0 0,2 0.4 0,6 0.8 I 1,2 1.¢
log (r + Ro) 4 8 12 '~6
2 6 10 14. 18

1 - -

0,5
0.5
0
-0,5
-I . . . . L4

-I .5

-2
-2.5 -2

-.3 L - 2.5 " ~

--5,5 -- I I ......... I --1--- I T-"-'-I-I
-3
0 0.2 0,4 0,6 0.8 1 1.2 14 /

Io0 (r + Ro) -5.5 - - i ~. . . . . . . . r--............

5 10 15 2Q
Fig. 1. Plots of the logarithm of the rank vs the logarithm of their
relative frequencies for those sites identified as structural. The corre- I

lation coefficients for all the positions were higher than 0.95. The Fig. 2. Plots of the rank vs the logarithm of their relative frequencies
illustrated positions are for VH: site 29: +, and site 55: *, and for Vrf for those sites identified as exclusively involved in the recognition
site 25: +, and site 52: * mechanism. The correlation coefficients for all the positions were
higher than 0.95. The illustrated positions are for VH: site 53: +, and
site 58: *, and for VL: site 29: +, and site 34: *
is interpreted to be a measure of the extension of the text--that is, the
total number of words in the text; 13gives a measure of the diversity or
richness of words employed in the text; and p gives the probability of
first-rank words (small or large). The meaning of these parameters in Discussion
our application needs to be modified to take into account different types
of constraints. For instance, for some sites on the CDRs the use fre-
quency of amino acids is nonrandom (Vargas-Madrazo et al. 1994),
A set of sites proposed to be conserved on the CDR-1
which reduces the available number of amino acids for the site, limiting and CDR-2 have been reported in the literature (Kabat et
the total number of elernents (and richness ?) for the site. This and other al. 1977). These sites are listed in the seventh column of
constraints should be considered before attempting to extrapolate di- Table 1. Chothia and Lesk (1987) reported a subset of
rectly the meanings used in linguistics. sites on the CDRs which are key for the maintenance of
The linearized form of relation (1) was plotted for each of 27 sites
on the heavy chain and for each of 21 sites on the light chain. The
the canonical conformations (structural set). These sites
results of plotting the logarithm of the probability of use of each amino are listed in the sixth column of Table 1. In this table we
acid vs In (r + p) are illustrated in Fig. 1 for some of the analyzed observe that 14 of the 17 cases reported as structural sites
positions. The results were statistically tested and for 19 sites of V H and by Kabat et al. (1977), and all the sites considered by
12 sites of V L (see Table 1) the fit with a straight line was reasonably Chothia and Lesk (1987) as critical for maintaining the
good, with correlation coefficients higher than 0.95.
The rest of the positions, that is, 8 on VH and 8 on Vr_, could not be
canonical conformations, within the analyzed regions,
fit to the inverse power law. Instead, they fitted well an exponential correspond to those sites whose frequency distributions
distribution of the form fit the inverse power law. This correspondence with sites
previously proposed as conserved, and the fact that the
P(r) = kexp(-Xr), X a constant parameter (2) more abundant amino acids in each site share similar
properties of hydrophobicity and volume, physicochem-
with correlation coefficients greater than 0.95. Plots of data fitting the ical properties typical of structural amino acids (Miyata
exponential distribution are illustrated in Fig. 2. et al. 1979; Lim and Sauer 1989; Sneath 1966; Grantham
 101

1974), strongly suggests that all sites whose frequency Table 2. Relative frequencies of amino acids in percent with the
distributions fit the inverse power law have a structural highest constancy as obtained from the germ-line gene data base
role in the CDRs.
Predominant Relative Secondary Total
What is interesting then is to have found in this work Site amino a c i d frequency aminoacids frequency
that the amino acids responsible for keeping the canon-
ical conformations fit well a skewed distribution such as Heavy
chain
the inverse power law. Distributions of this type natu-
26 G (81%)
rally arise in probabilistic processes whose primary 27 F (34%) Y (37%) 70%
achievements require the unfailing completion of numer- 28 T (52%) S (17%),
ous independent secondary subtasks (Montroll and N(ll%) 70%
Shlesinger 1982). In fact, the failure of any of these tasks 29 F (60%) I (18%),
V (3%),
is sufficient to cause the failure of the main task. To
L (2%) 83%
illustrate this point let Pi denote the probability of achiev- 30 T (51%) S (18%),
ing the ith step, in a sequence of events oriented toward L (11%) 80%
achieving a general goal. If P denotes the probability of 31 S (41%) D (31%) 72%
achieving this general goal at a given time (Montroll and 32 Y (61%) W(9%)
F (3%) 73%
Shlesinger 1982), then we have
34 M (55%) I (13) 68%
51 I (70%)
P = P a P 2 . . . P~ (3) 54 N (37%) s (16%),
D (13%) 66%
where n represents the total number of successful steps 55 G (55%) S (8%) 64%
57 T (65%)
required to complete the general task.
59 Y (71%)
The relation of the above process to some of the Dar- 6o N (35%) S (14%),
winian ideas of evolution is suggestive. In our context D (10%) 59%
these ideas would suggest that the actual shapes of the 63 F (40%) V (17%),
canonical conformations of the CDRs are achieved as the L (10%) 67%
64 K (56%) Q (10%) 66%
result of an evolutionary process, in which the adopted 65 G (47%) s (19%) 66%
conformations are privileged (optimal?) shapes that in- Light
teract more efficiently with the antigens. The adoption of chain
optimal shapes, that is, shapes that interact more effi- 24 R (69%) K (12%) 81%
ciently with antigens, may be understood considering 25 A (76%) S (7%) 83%
26 S (74%) G (7%) 81%
that protein interactions are increased in surfaces that
27 Q (47%) K (20%),
facilitate the formation of clefts or holes (Lewis 1991). E (9%) 76%
Chothia et al. (1992) indicate that indeed canonical con- 32 Y (46%) N (11%),
formations generate a corresponding number of surfaces F (4%) 61%
involved in the antigen-antibody interactions. In fact, it 33 L (59%) M (8%),
Y (6%),
has been reported that specific antibodies for small hap-
V (3%) 76%
tens often have a concave combining site, frequently 52 s (40%) T (21%) 61%
found as a deep pocket or groove on the surface. On the 54 L (35%) R (30%) 65%
other hand, antibodies that bind large molecules, such as 56 s (40%) T (21%) 61%
proteins, tend to have flat combining sites so that the
surface of antibody-antigen is larger (Bolger and Sher-
man 1991). More general prototypes of surfaces have surface that the canonical structures present to the anti-
been reported by Lehn (1990), who has also categorized gens (Chothia et al. 1992). If this is the case, then it does
receptor-substrate interactions as being either endo or increase substantially the number of available surfaces
exo in character. Endo surfaces correspond to cyclic or (prototypes) to contend with a larger number of epitopes.
severely bent molecules that contain holes, clefts, or cav- Thus, canonical conformations are determined by a few
ities having interior interactive sites that converge upon conserved residues which, as supported by our results,
a central locus while exo surfaces contain exterior inter- may be the result of an evolutionary process.
active sites (Gutsche 1992). That the aforementioned sites mainly play a structural
In Table 1 we may observe that the number of con- role is striking, since it has usually been assumed that
served residues is greater than the known critical sites sites 26-30 and 53-55 of V H and 26-32 and 50-56 of
responsible for maintaining the canonical conformations V L, by being located in the loop region, were recognition
(Chothia and Lesk 1987). Nonetheless, these other con- positions. Thus, our results suggest that an enhanced
served sites, particularly those placed in the center of the concept of a recognition site must be based not only on
loop, must be important in modulating the shapes of the whether or not the site is on a loop but also on the type
 102

30-
25-
20-
15-
10-

0-

a
c
d
e

g
h

k
I

arnin0 acids rl
P
B q
r s-3
l " / S-1
V
% 25- tOlal
W
20- Y

15-

5-
0-

c
d

f
g

k l'-<..
r
amino acids ra
P
q

v (etal
w
Y

Relative use frequencies of each amino acid in the total sample (total), and in the two subsamples with one (s-l), and five (s-3) randomly
F i g . 3.
chosen sequences from each one of the specificities present in the Kabat et al. (1991) compilation. A Position 52 of V H. B P o s i t i o n 50 o f V L.

of distribution of use followed by the amino acids on
each site.
To eliminate the possibility that the described behav-
ior may be due to some artifact generated as a conse-
quence of somatic hypermutations, or may be due to a
bias in the analyzed sample (2,000 Igs) that is a conse-
quence of a predominance of certain specificities, e.g., a
greater number of Ig antihaptens, two control tests were
carried out.
The reported sequences for the Ig's germ-line genes
were aligned (Gene-bank, release 79), and the amino
acid use frequencies for each site were analyzed to de-
termine the type of distribution. These use frequencies
showed a behavior similar to that of the protein sample.
The results are shown in Table 2. The fact that two types
of sites persist at the gene level gives additional support
to the suggestion that the observed behavior of the pro-
tein sample should indeed be a consequence of an evo-
lutionary process, and not be generated during the rec-
ognition process.
To eliminate the possibility of an existing bias in the
sample, the total sample was spliced into two subsam-
ples, each one with a different specificity, and their
amino acid use frequencies were compared with that of

the total sample. The two subsamples w e r e built as fol-
lows: (1) one s e q u e n c e was r a n d o m l y c h o s e n f r o m each
specificity in the sample; (2) a r a n d o m sample o f five
sequences was selected f r o m specificities that h a v e five
or m o r e sequences in the sample. The results of this
analysis for one position in VL and one in VrI are s h o w n
in Fig, 3, w h i c h shows that the pattern o f a m i n o acid
usage found in the two subsamples is the same as that in
the total sample. T h e s e m e a s u r e s rule out the possibility
o f any bias in the a n a l y z e d s a m p l e
If the positions discussed are critical in maintaining
the structures o f the family, then our reported findings
i m p l y that the rest of the positions o f the a n a l y z e d C D R s
must be properly i n v o l v e d in the recognition p r o c e s s - -
that is, those positions w h o s e f r e q u e n c y distributions are
e x p o n e n t i a l are those w h i c h m u s t be e x c l u s i v e l y in-
v o l v e d in the recognition process. This e x p e r i m e n t a l fre-
q u e n c y distribution is w e l l known, and is typical o f pro-
cesses that contain no memory (Gilman 1991),
c o n f i r m i n g the traditional v i e w (Ohno et al. 1985) o f a
r a n d o m substitution for these sites. (See h o w e v e r , M i a n
et al. 1991, and V a r g a s - M a d r a z o et al. 1994.)
In summary, our results suggest that the existence o f
two different types o f sites is the key to success in m o -
lecular recognition, in w h i c h the principle i n v o l v e d is
that an e f f e c t i v e identification requires surfaces o f c o m -
plementarity size, shape and functionality.
Acknowledgments. We gratefully acknowledge the valuable advice
received from M. Jimenez-Montano. Furthermore, we wish to thank
Prof. E. Zuckerkandl, E. Lara, C. Castillo-Chavez, and the referee for
their critical review of the manuscript. We also are indebted to Prof.
Zuckerkandl for his generous help with our previous article published
in J. Mol. Evol. J.C.A. was supported by DGAPA grant IN-206093;
E.V.M. was supported by CONACyT.
References
Alt FW, Blackwell TK, Yancopoulos GD (1987) Development of the
primary antibody repertoire. Science 238:1079-1087
Bolger, MB, Sherman MA (1991) Computer modeling of combining
site structure of anti-hapten monoclonal antibodies. Methods En-
zymol 203:21-45
Chothia C, Lesk AM (1987) Canonical structures for the hypervariable
regions of immunoglobulins. J Mol Biol 186:651-663
Chothia C, Lesk AM, Gherardi E, Tomlinson IM, Walter G, Marks JD,
Llewelyn MB, Winter G (1992) Structural repertoire of the human
VH segments. J Mol Biol 227:799-817
Gilman JJ (1991) Research management today. Phys Today March:
42-49
Gutsche CD (1992) Supramolecular chemistry. In: Parker SP (eds)
Encyclopedia of chemistry, 2nd ed. McGraw Hill, Inc., NY
Grantham R (1974) Amino acid difference formula to help explain
protein evolution. Science 185:862-864
Kabat EA (1978) The structural basis of antibody complementarity.
Adv Protein Chem 32:1-75
Kabat EA, Wu TT (1971) Attempts to locate complementarity deter-
103
mining residues in the variable portions of light and heavy chains.
Ann NY Acad Sci 190:382-393
Kabat EA, Wu TT, Bilofsky H (1977) Unusual distributions of amino
acids in complementarity-determining (hypervariable) segments of
heavy and light chains of immunoglobulins and their possible roles
in specificity of antibody-combining sites. J Biol Chem 252:6609-
6616
Kabat EA, Wu T, Perry HM, Gottesman KS, Foeller C (1991) In:
Sequences of proteins of immunological interest, 5th ed. National
Institutes of Health, Bethesda, MD
Lara-Ochoa F, Vargas E, Jimenez-Montano MA, Almagro JC (1994)
Patterns in the complementarity determining regions of immuno-
globulins (CDRs). Biosystems 32:1-9
Lehn JM (t990) Perspectives in supramolecular chemistry, from mo-
lecular recognition towards molecular information processing and
self-organization. Angew Chem Int Ed Engl 29:1304-1319
Lewis AR (1991 ) Clefts and binding sites in protein receptors. Methods
Enzymol 202:126-156
Lim WA, Saner RT (1989) Alternative packing arrangement in the
hydrophobic core of proteins. Nature 339:31-36
Mandelbrot BB (1977) Fractals: form, chance, and dimension. Free-
man, San Francisco
Meijer PHE, Mountain RD, Souler Jr RJ, eds (1981) Sixth International
Conference on Noise in Physical Systems. National Bureau Stan-
dards, Washington, DC, Special Publication No. 614
Mian IS, Bradwell AR, Olson AJ (1991) Structure, function and prop-
erties of antibody binding sites. J Mol Biol 217:133-151
Miyata T, Miyasawa S, Yasunaga T (1979) Two types of amino acid
substitution in protein evolution. J Mol Evol 12:219-223
Montroll E, Badger WW (1974) Introduction to quantitative aspects of
social phenomena. Gordon and Breach Science Pub
Montroll EW, Schlesinger MF (1982) On 1/f noise and other distribu-
tions with long tails. Proc Natl Acad Sci USA 79:3380-3383
Montroll E, Shlesinger MF (1983) Maximum entropy formalism, frac-
tals, scaling phenomena, and 1/f noise: a tale of tails. J Statist Phys
32:209-230
Nicolis JS (1986) Chaotic dynamics as applied to information process-
ing. Rep Prog Phys 49:1109-1187
Nicolis JS (1987) Chaotic dynamics of logical paradoxes. In: Bothe,
Ebeling, Kurzhanski, Peschel, (eds) Dynamical systems and envi-
ronmental models. Academic-Vertag, pp 105-113
Ohno S, Mori N, Matsunaga T (1985) Antigen-binding specificities of
antibodies are primarily determined by seven residues of Vh. Proc
Nail Acad Sci USA 82:2945-2949
Padlan EA (1994) Anatomy of the antibody molecule. Mol Immunol
(in press)
Schroeder M (1991) Fractals, chaos, power laws. WH Freeman and
Company, New York, p 35
Sneath PHA (1966) Relation between chemical structure and biological
activity in peptides. J Theor Biol 12:157-195
Tonegawa S (1983) Somatic generation of antibody diversity. Nature
302:575-581
Vargas-Madrazo E, Lara-Ochoa F, Jimenez-Montano M (1994) A
skewed distribution of amino acids at recognition sites of the hy-
pervariable region of immunoglobulins. J Mol Evol 38:100-104
West BM (1985) An essay on the importance of being non-linear.
Lectures notes in biomathematics, 62, Springer-Verlag, Berlin
Wu TT, Kabat EA (1970) An analysis of the sequences of the variable
regions of Bence Jones proteins and myeloma light chains and their
implications for antibody complementarity. J Exp Med 132:211-
250
Zipf GK (1949) Human behavior and the principle of least effort.
Addison-Wesley, Cambridge