A molecular clock based on the expansion of gene families
A molecular clock based on the expansion of gene families
Yuri A.
Trusov*
and
Paul H.
Dear
1
Institute of Cytology and Genetics, Siberian Branch of Russian Academy of
Sciences, Acad. Lavrentiev Avenue 10,
Novosibirsk
630090,
Russia
and
1
MRC Centre, Hills Road,
Cambridge
CB2 2QH,
UK
Received January 25, 1996;
Accepted February 8, 1996
ABSTRACT
There is evidence to suggest that eukaryotic genomes are subject to frequent insertions and deletions of non-coding DNA. This may lead to a gradual increase or decrease in genome
size, or to a dynamic equilibrium in which the overall size remains constant. We argue, however, that there is a
bias favouring an accumulation of non-coding DNA in the proximity of genes. Such bias causes a progressive
change in genome structure regardless of whether the overall genome size increases, decreases or remains constant. We show that this change may serve as a `molecular clock', supplementing that provided
by nucleotide substitution rates.
INTRODUCTION
Eukaryotic genomes vary tremendously in size, not only between genera but between closely-related species or even between individuals of a single species whose
coding requirements must be closely similar (
1
-
3
). This so-called `C-value paradox' implies that most eukaryotes carry a considerable
burden of non-coding DNA with no direct phenotypic effects (
4
,
5
).
Such DNA runs the risk of deletion (
6
,
7
). Occasionally, such deletion occurs on a large scale, leading to a reduction
in genome size (
8
,
9
). In most cases, however, the amount of non-coding DNA increases (
4
,
8
,
10
,
11
), leading in extreme cases to immense genomes such as those of Latimeria and
many chordate and teleost species (
10
,
12
). This suggests that genomes are in a state of dynamic equilibrium, undergoing
frequent insertions and deletions of non-coding DNA (
13
).
We argue here, however, that there is a bias favouring the accumulation of non-coding DNA in close proximity to genes. This arises because deletions in
such regions are liable to disrupt genes and are hence selected against.
Conversely, deletions of non-coding DNA in large intergenic regions are less likely to have adverse
phenotypic effects.
In consequence, closely-spaced genes will tend to drift apart over time as intervening insertions
outweigh deletions. This will lead towards a state where genes are uniformly
spaced throughout the genome, and hence where all regions are equally
susceptible to deletions. Such `normalisation' of intergenic distances will
tend to occur regardless of whether the overall genome size increases,
decreases or remains constant. We present evidence based on clustered gene-families to show that this is the case, and that the process is
sufficiently regular to serve as a molecular clock.
THEORETICAL BASIS FOR THE MODEL
Vertebrate genomes contain a large proportion of apparently functionless DNA,
consisting of both unique and repeated elements interspersed amongst genes (
6
,
10
). A variety of mechanisms have been proposed to account for the insertion,
duplication and deletion of such DNA (
10
,
14
-
20
), and the balance between these processes determines whether the genome as a
whole expands, contracts or remains of constant size.
Special selective pressures exist on non-coding DNA in the vicinity of genes (we take a `gene' to include
regulatory sequences lying immediately upstream and downstream of the coding
region). In a region densely populated with genes, any deletion of non-functional DNA is liable to take with it part of a gene, and so disrupt gene function. Hence deletions will be selected against, the more
so as their size (and hence the likelihood of their including a part of a gene)
increases. In contrast, insertions will be selectively neutral unless the
point
of insertion lies within a gene, regardless of the size of the insertion. Even
tandem duplications within a gene stand a good chance of being tolerated, as
they may duplicate part of the gene but leave a copy of the original sequence
intact.
The difference between the selective pressure on insertions and on deletions
becomes most acute near the ends of genes. Here, insertions which arise by
tandem duplication will frequently leave an intact copy of the gene (for
example, abcDEFGHIJ... -> abcDEcDEFGHIJ..., where uppercase letters indicate the gene), whereas deletions are particularly liable to destroy control sequences
required for the correct initiation and termination of transcription. (An
exception to the selective bias against deletions occurs in the case of
transposon-like elements which can excise themselves precisely. Deletion of such an
element will simply reverse a prior insertion and is not likely to disrupt gene
function.)
Qualitative support for this selective bias against deletions comes from mutations in the human [beta]-globin cluster (
21
). Of several insertions found here, none had apparent effects on health.
Deletions, in contrast, were associated with a variety of thalassaemias. Hence, our model states that dense clusters of genes will tend to expand
relative to the genome as a whole, due to a `ratchet effect' whereby insertions are more likely to be tolerated than deletions. The
`ratchet' has its greatest influence at the boundaries between genes and
adjacent non-functional sequences. If continued indefinitely and in the absence of
other effects, this would lead to genes becoming uniformly distributed
throughout the genome.
Quantitative interpretation of our model is less straightforward. To predict the
rate of expansion of a gene cluster accurately, we need to know the frequency
with which insertions and deletions of various sizes occur in the absence of
selection; these frequencies have not been directly determined, and indirect
estimates have been made only in a few special cases (
15
,
16
,
22
-
24
). However, we can at least make an estimate of the maximum rate at which
expansion might be expected to occur. We assume that the initial duplication of
a gene gives an intergenic region of ~1 kb (this is supported by examples given below), and that all insertions
in this region are selectively neutral whilst all deletions are harmful. The
majority of insertions and deletions (other than those involving mobile
elements) occur through slippage/mispairing (
15
,
23
-
25
). Based on the data of Di Rienzo
et al.
(
22
) and Strand
et al.
(
25
), this process would introduce DNA at an initial rate of ~0.1 kb/Myr. In practice, some insertions will of course be harmful and some
deletions selectively neutral, so this rate represents an approximate upper
limit on the rate of expansion.
EVIDENCE IN SUPPORT OF THE MODEL
We have sought to test our model by examining the evolution of mammalian gene
families. Such families normally arise by tandem duplication of an ancestral gene (often including flanking sequence;
26
) and family members usually lie close together when they first arise.
Nucleotide substitutions then accumulate in both genes, often followed by
further duplications and continued sequence divergence.
According to our model, the accumulation of non-coding DNA between family members should occur progressively, regardless
of fluctuations in overall genome size. Hence, physical distances would be
greatest between the most distantly-related family members and least between those which, having recently
arisen by duplication, show greatest sequence homology. In other words, the
evolutionary tree of the gene family (based on sequence divergence) should
correspond approximately with its physical map.
A number of factors might conspire to mask this effect. For example, if an
ancestral gene `A' suffered two duplications in rapid succession to yield the
gene-family A-B-C, then the distance A-C would obviously be greater than A-B or B-C, despite all three genes having
arisen almost simultaneously. Some gene families (for example the Hox clusters)
appear to have evolved as functional units in which the spatial relationships of individual genes have probably been
selectively preserved (
27
); such clusters will not expand as predicted by our model. Duplications or
inversions involving more than one gene would also confuse the situation; the
human immunoglobulin heavy-chain locus has clearly suffered many such events (
28
), as well as gene-conversions which disguise the apparent ages of duplications. Finally,
occasional large insertions or deletions may overwhelm the gradual expansion-due to smaller, more frequent events-which we predict. However, we would expect to see evidence for our
model in relatively small gene families which do not appear to have undergone
such disruptions. Therefore, we examined those such families for which maps,
sequence and, where possible, phylogenetic histories are available.
The human [beta]-globin gene cluster contains five functional genes. Intergenic
distances are known precisely and the times of the duplications which gave rise
to family members can be inferred phylogenetically (
15
). These data are summarised as a combined evolutionary tree and physical map in
Figure
1
; the physical distances between genes correspond well to the estimated times of
their divergence. Data for this and other gene families are summarised in Table
1
. In the goat and rabbit [beta]-like globin clusters, we find again a correlation between physical distance (
29
,
30
) and estimated age since duplication. It should be noted that all of these
clusters contain pseudogenes but, for the purposes of our model, we make no
distinction between these and any other non-coding sequence. The mouse [beta]-globin-like cluster (comprising genes in the order y-bh0-bh1-b1-b2) fits our model less
perfectly. The distances bh0-bh1 and bh1-b1 are as we predict (Table
1
), but y-bh0 (2.2 kb) is far smaller than we would expect whilst b1-b2 (14 kb) is larger. The degree of sequence divergence between y
and bh0, combined with their orthology to the rabbit b4-b3 and human [epsilon]-[gamma] pairs, would lead us to expect an intergenic
distance of ~9 kb. We can only assume that a sizeable deletion (~7 kb) has occurred between these genes. Such a deletion must have
occurred between the divergence of mouse from rabbit (60 million years ago) and
that of mouse from deer-mouse (30 million years ago), as the deer-mouse has a [beta]-like cluster similar to that of mouse (
31
). In the case of mouse b1-b2, the low degree of sequence divergence leads us to expect an
intergenic distance of ~7 kb, as compared with the observed 14 kb. A LINE element insertion
accounts for 4.7 kb of the excess. Moreover, the corresponding regions in rat and deer-mouse both contain three genes, of which the outermost members are
homologous to the mouse b1-b2 pair (
31
,
32
). This strongly indicates that mouse has suffered a recent deletion of a gene
between b1 and b2, leaving behind most of the two regions of intergenic
sequence.
The human kallikrein gene-cluster comprises three genes (
33
; Table
1
). Although independent estimates of the dates of the duplication events in this
cluster are not available from phylogenetic studies, sequence divergences
indicate that the pair APS/KLK2 represent the most recent duplication, their
common ancestor having diverged from KLK1 much earlier. Again, physical
distances between these three genes correspond with the inferred ages of
duplications.
The chimpanzee haptoglobin gene cluster consists of three genes in the order Hp-Hpr-Hpp, with intergenic distances of 2.5 and 16 kb respectively
(Table
1
). The age of the divergence between Hp and Hpr is estimated at 30 million years
ago (
34
), whilst dates for other divergences in the cluster can be inferred only from
sequence divergence. Although this cluster fits our model qualitatively (the
greatest intergenic distances correspond to the greatest sequence divergence
and hence the greatest inferred age since duplication), the distance between
Hpr and Hpp (16 kb) is far greater then we would expect. McEvoy and Maeda (
35
) found a retroviral insertion of ~8 kb in this region; the remaining 8 kb is approximately in line with our
model, though it is slightly larger than expected and may indicate that some
additional DNA accompanied the retroviral insertion.
In those cases where the dates of duplication events cannot be determined
phylogenetically, we have had to use dates inferred from sequence divergence to
test our model. It is therefore important that we should be able to identify
any gene-conversion events which may have occurred, as these would greatly reduce
such inferred ages. Fortunately, we can recognise gene-conversions in the following way. If gene conversion has occurred between
two members of a gene family, then the sequence divergence between them will be
less than the divergence between homologues which have existed in related
species for comparable lengths of time. That is, gene conversion will lead to
apparent intraspecific divergence rates which are far less than interspecific
rates. Table
2
lists interspecific divergences between pairs of globin gene homologues in
several species, together with the time since the divergence of the species.
For most gene pairs, the rate of intraspecific sequence divergence (~0.1%/Myr) is comparable with that of interspecific sequence divergence. The
most notable exception is the human [gamma]A/[gamma]G pair, which show a divergence of only 0.02%/Myr. This supports
the suggestion of Slightom
et al.
that this pair has undergone a recent gene conversion (
36
).
If we can thus recognise gene-conversion events and exclude them from our analysis, we should expect to
find a strong correlation between sequence divergence and physical distances between genes, since both are assumed (the latter by our model) to arise from
regular, gradual processes. To test this, we considered the pairs of adjacent
genes indicated by asterisks in Table
1
. Because of the suspected gene conversion between the human [gamma]-genes, we have omitted this pair from our analysis and used in
their place the corresponding genes from orangutan, which has a [beta]-like globin cluster of similar structure to human (
37
). We have also excluded the mouse b1-b2 pair (due to the probable deletion of an intervening gene as described
above) and the chimpanzee Hpr-Hpp pair which has suffered a retrovirus-like insertion. We have, however, included the mouse y-bh0 pair, as there is no independent evidence for the large
deletion we have postulated. The comparison between sequence divergence and
physical distance for these gene-pairs is shown in Figure
2
A.
The coefficient of correlation between sequence divergence and physical distance
is 0.797 (z = 3.61; 0.001 <
P
< 0.005). We stress that this strong correlation does not reflect an
interdependence of nucleotide substitution and physical separation. Rather, it
reflects the fact that these two independent processes are both time-dependent.
Sequence divergence, physical distance and approximate age since divergence of
some pairs of mammalian genes
Divergence (%)
Distance (kb)
Age (million years)
Human
[epsilon]/[gamma] globin*
11.1 " 1.9
12
100
[gamma]
G
/[gamma]
A
globin
0.35 " 0.1
3.5
20
[gamma]/[delta] globin*
19.0 " 2.6
14
200
[delta]/[beta] globin*
3.3 " 1.0
5.4
40
KLK2/APS*
14.1 " 1.7
12
150
APS/KLK1*
30.0 " 2.7
31
310
Orangutan
[gamma]
1
/[gamma]
2
globin*
0.9 " 0.1
3.5
20
Chimpanzee
Hp/Hpr*
4.1 " 0.7
2.5
30
Hpr/Hpp
4.8 " 0.8
16
35
Rabbit
[beta]
1
/[beta]
3
globin*
20.2 " 2.7
14
200
[beta]
3
/[beta]
4
globin*
15 " 2.4
8
100
Goat
[beta]
c
/[epsilon]
II
*
22.2 " 2.9
13.3
200
[epsilon]
I
/[epsilon]
II
*
14.5 " 2.4
7
65
Mouse
y/bh
1
globin*
16.5 " 2.0
2.2
100
bh
0
/bh
1
globin*
3.5 " 1.1
6.7
45
bh
1
/b
1
globin*
25 " 2.7
15.9
200
b
1
/b
2
globin
3.3 " 1.0
14
45
Sequence divergence was calculated according to ref. 38. Asterisks indicate
pairs of adjacent genes for which data are plotted in Figure 2. The ages of
human globin genes are from ref. 15; those of APS/KLK1/KLK2 were estimated from
sequence divergence using the calibrated scale of human globin genes (15). Goat
[epsilon] and rabbit [beta]-4 genes are homologous to human [epsilon]; rabbit [beta]-3 to human [gamma]; goat [beta]-c and rabbit [beta]-1 to human [beta]: the
histories of these homologues were therefore assumed to be similar to their
human counterparts. Mouse y is homologous to human [epsilon]; bh0 and bh1 to human [gamma]; b1 and b2 to human [beta]: the ages of y/bh0 and bh1/b1 duplications are therefore
equal to those of their human counterparts. The bh0/bh1 and b1/b2 duplications
both exist in mouse and in deer-mouse but not in rabbit. Hence they must both have occurred between 60
(divergence of Murinae and Lagomorpha) and 30 million years ago (divergence
Muridae and Cricetidae; 39); we have assumed a value mid-way between these limits. The goat [epsilon]-I/[epsilon]-II genes correspond to their bovine
counterparts, implying that this duplication occurred between 100 (time of mammalian
radiation) and 30 million years ago (time of radiation of Bovidae; 39); we have
assumed a value mid-way between these limits. The [gamma]-globin gene pairs are orthologous between human, gorilla and
orangutan, whilst the lemur and owl monkey have only one [gamma] gene (37); therefore the duplication of the ancestral [gamma] gene occurred between the appearance of the anthropoids (22 million years ago) and the parting of
Hominidae from Pongidae (17 million years ago; 39). The age of the Hp/Hpr
duplication is from ref. 34; that for Hpr/Hpp is based on data from refs 34 and
35.
A direct estimate of the rate of physical separation of genes proposed by our
model is complicated by the fact that exact ages for the relevant duplications
are often not known. Where possible, we have used ages based on phylogenetic
evidence; where this has not been possible, ages have been based on sequence
divergence. Physical distance is compared with age since duplication in Figure
2
B. As we would predict, there is a strong correlation (r = 0.913; z = 5.12,
P
< 0.001) between distance and age.
DISCUSSION
Figure
2
B suggests that gene separation can be used as a molecular clock of comparable
reliability with that based on nucleotide substitution rates, at least for
mammalian gene-clusters over periods of a few tens of millions of years. The intercept of
the line with the vertical axis (~0.6 kb if we assume a linear relationship) represents the amount of
flanking DNA which is typically duplicated along with genes, and the slope of
the line (~0.08 kb/Myr) gives the net rate of accumulation of non-coding DNA between genes. Unlike the clock based on nucleotide
substitution, that described here is immune to the effects of gene conversion
and may be less susceptible to selective pressures on coding sequences.
However, it is vulnerable to disruption by duplications or inversions involving
more than one gene, to translocations, or to occasional large insertions or
deletions which may overwhelm the gradual, average process.
.
Sequence divergence and age of species divergence between homologous pairs of
globin genes
Divergence (%)
Age (million years)
[epsilon] human/[epsilon] lemur
5.8 " 1.3
40
[delta] human/[delta] tarsier
7.3 " 1.5
40
[epsilon] human/[epsilon] goat
5.3 " 1.3
100
[beta] human/[beta]
maj
mouse
14.0 " 2.2
100
[beta] human/[beta] bovine
9.8 " 1.8
100
[epsilon] human/[epsilon] chicken
20.6 " 2.7
200
[epsilon] goat/[epsilon] chicken
21.7 " 2.8
200
Although both clocks are error-prone, the fact that they are disrupted by different factors means that
comparison between them can draw attention to interesting events. For example,
if two genes are widely separated in space but show little sequence divergence, it may be inferred that either a gene conversion has recently occurred between
two long-diverged genes, or that some major upheaval (such as a single large
intervening insertion) has occurred. Conversely, an anomalously small
intergenic distance between genes of highly-diverged sequence would imply that a large deletion has brought together
two genes which were formerly widely separated. Amongst the genes which we have
considered we have found some exceptions to our model. All but one of these,
however, can be accounted for by insertions, deletions or gene conversions for
which there is independent supporting evidence. Only one exception (mouse y-bh0 globin) is unaccounted for by independent evidence, and we would
suggest that there has been a substantial deletion which may be revealed as
more comparative sequence data become available.
Another molecular clock, based on the overall rate of genome expansion in salamanders, has recently been proposed (
11
). This may be regarded as a special case of our clock: in salamanders the
entire genome tends to expand whereas in mammals (and, we would expect, most
species) gene families tend to drift apart despite the overall genome size
remaining constant.
In conclusion, we feel that the phenomenon of gene family expansion is worth
further investigation. The recent increase in sequencing and mapping efforts
should make it possible to test our model in a wider variety of circumstances.
ACKNOWLEDGEMENT
We would like to thank Dr V. A. Berdnikov for helpful criticism and discussion
of our model.
REFERENCES
1 Kumar, A. and Rai, K.S. (1990) Theor. Appl. Genet. 79, 748-752.
2 Kuriyan, P.N. and Narayan, R.K.J. (1988) J. Mol. Evol. 27, 303-310.
3 Laurie, D.A. and Bennet, M.D. (1985) Heredity, 55, 307-313.
4 Doolittle, W.F. and Sapienza, C. (1980) Nature 284, 601-603.MEDLINE Abstract
6 Dover, G., Brown, S., Coen, E., Dallas, J., Strachan, T. and Trick, M. (1982) in Dover G. and Flavell, R.B. (eds) Genome Evolution,Academic Press, London.
7 Saitou, N. and Ueda, S. (1994) Mol. Biol. Evol. 11, 509-512.
8 Sessions, S.K. and Larson, A. (1987) Evolution 41, 1239-1251.
9 Bullock, D.G. and Rayburn, A.L. (1991) Maydica 36, 247-250.
22 Di Rienzo,A., Peterson, A.C., Garza, J.C., Valdes, A.M., Slatkin, M. and Freimer, N.B. (1994) Proc. Natl Acad. Sci. USA 91, 3166-3170.MEDLINE Abstract
23 Tautz, D. and Schlotterer, C. (1994) Curr. Opin. Genet. Dev. 4, 832-837.MEDLINE Abstract
25 Strand, M., Prolla, T.A., Liskay, R.M. and Petes, T.D. (1993) Nature, 365, 274-276.MEDLINE Abstract
26 Bostok, K.J. and Tyler-Smith, K. (1982) in Dover, G. and Flavell, R.B. (eds) Genome Evolution,Academic Press, London.
27 Acampora, D., D'Esposito, M., Faiella, A., Pannese, M., Midlinccio, F.M., Stornaiuolo, A., Nigro, V., Simione, A. and Boncinelli, E. (1989) Nucleic Acids Res. 17, 10385-10402.MEDLINE Abstract
28 Cook, G. P., Tomlinson, I.M., Walter, G., Riethman, H., Carter, N.P., Buluwela, L., Winter, G. and Rabbitts, T.H. (1994) Nature Genet.7, 162-168.MEDLINE Abstract
29 Lacy, E., Hardison, R.S., Quon, D. and Maniatis, T. (1979) Cell 18, 1273-1283.MEDLINE Abstract
30 Townes, T.M., Fitzgerald, M.C. and Lingrel, J.B. (1984) Proc. Natl Acad. Sci. USA 81, 6589-6593.MEDLINE Abstract
31 Padgett, R.W., Loeb, D.D., Snyder, L.R.G., Edgell, M. H. and Hutchison, C.A. (1987) Mol. Biol. Evol. 4, 30-45.MEDLINE Abstract
32 Paunesku, T., Stevanovic, M., Radosavljevic, D., Drmanac, R. and Crkvenjakov, R. (1990) Mol. Biol. Evol. 7, 407-422.MEDLINE Abstract
33 Riegman, P.H.J., Vlietstra, R.J., Suurmeuer, L., Cleutjens, C.B. and Trapman, J. (1992) Genomics 14, 6-11.MEDLINE Abstract
