Content uploaded by Michael Behe
Author content
All content in this area was uploaded by Michael Behe on Jun 13, 2016
Content may be subject to copyright.
450
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Getting There First: An Evolutionary Rate Advantage
for Adaptive Loss-of-Function Mutations
Michael J. Behe1
1Department of Biological Sciences, Lehigh University, Bethlehem, PA 18015, USA.
mjb1@lehigh.edu
Abstract
Over the course of evolution organisms have adapted to their environments by mutating to gain new
functions or to lose pre-existing ones. Because adaptation can occur by either of these modes, it is
of basic interest to assess under what, if any, evolutionary circumstances one of them may predomi-
nate. Since mutation occurs at the molecular level, one must look there to discern if an adaptation
involves gain- or loss-of-function. Here I present a simple, deterministic model for the occurrence
and spread of adaptive gain-of-function versus loss-of-function mutations, and compare the results
to laboratory evolution experiments and studies of evolution in nature. The results demonstrate that
loss-of-function mutations generally have an intrinsic evolutionary rate advantage over gain-of-
function mutations, but that the advantage depends radically on population size, ratio of selection
coefficients of competing adaptive mutations, and ratio of the mutation rates to the adaptive states.
Key words: gain-of-function mutation, loss-of-function mutation, rate of fixation
1. Introduction
In On the Origin of Species Charles Darwin emphasized that natural selection is
relentless, continuously monitoring each organism for its fitness, selecting those
with an advantage and weeding out the disadvantaged [1]. However, as Darwin
also knew, an organism’s advantage in a particular set of circumstances did not
have to involve the gain of a new ability, such as the power to fly or swim. Indeed,
it could involve the loss of those abilities. Flightless birds had adapted to their
habitats partially by abandoning such a faculty. Some organisms went even fur-
ther. Darwin described some barnacles in which the male was reduced to a trans-
parent sac, with little but a reproductive system remaining [2]. By specializing in
this way, the barnacles and their descendants presumably gained an adaptive
advantage over competitors.
In the nineteenth century Darwin and his contemporaries could identify muta-
tions only through their phenotypic effects. However, with the progress of biology
especially in the last half-century, contemporary science can now characterize
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 451
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
mutations also by their molecular effects to the genetic material of a species. In
order to understand the roles of loss-of-function (LOF) versus gain-of-function
(GOF) mutations, one must keep phenotypic versus molecular changes separate.
An altered, visually observable phenotype may be due to any of a number of dis-
parate underlying molecular mutations. For example, a mutant mouse that is 50%
larger than its litter mates might have had the gene for a repressor protein that
switches off production of growth hormone deleted. At the molecular level, that
would be an LOF mutation, since a functional molecular feature was deleted, even
though the increased size of the mouse may strike the casual observer as a
gain-of-function. On the other hand, a large mutant mouse might be due to the
formation of a new promoter site for a transcription factor near a gene involved in
growth, which would be a GOF mutation, since a new functional molecular feature
(the promoter site) was produced. In this paper I will consider LOF and GOF
mutations as affecting functional molecular features such as genes and regulatory
elements, no matter what their possible phenotypic effects may be.
2. The model
Consider a population of organisms that comes under a new selective pressure. To
respond to the pressure two different adaptive mutations are postulated to be
potentially available: one which results in the gain of a molecular function, and
another which results in the loss of one. What factors might affect the probabilities
of either kind of mutation becoming fixed in the population in competition with
the other? One factor of immediate importance is the rate of appearance of the
adaptive mutations. It is very often possible to eliminate a molecular function by
a variety of mutations. GOF mutations, on the other hand, are generally much
more specific, sometimes being produced in only one way.
As an illustration, consider several mutations to human genes that give a meas-
ure of resistance to malaria. The best known such mutation is the sickle cell gene
in which, by means of a single A→T transversion, the codon for a glutamic acid
residue in the sixth position of the β-chain globin gene is converted to a codon for
valine [3]. This can be considered a GOF mutation, because the hemoglobin gains
a self-association site on its surface, allowing the individual proteins upon deoxy-
genation to aggregate into microtubular-like structures [4]. By an as-yet-unknown
mechanism, the polymerization negatively affects the growth of the malarial
parasite (which spends part of its life cycle in the red blood cell) [5, 6]. Another
mutation which confers a measure of resistance to malaria is deficiency of
glucose-6-phosphate dehydrogenase (G6PD), in which a mutant gene produces
little or no functional enzyme [7]. For reasons that are unclear, this interferes with
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
452 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
parasite viability. Population genetic studies have shown that hundreds of separate
mutations have led to deficiency of wild-type G6PD in populations at risk for
malaria. On the other hand, the mutation producing the sickle gene is thought to
have arisen de novo only a few times in the last 10,000 years, or perhaps only
once [8].
The reason for the disparity in the number of de novo mutations is straightfor-
ward. To secure a sickle mutation a particular nucleotide of the β-globin gene must
be substituted. Since the nucleotide mutation rate of humans is on the order of 10−8
substitutions/ generation, that is also the de novo rate of appearance of the sickle
gene [9]. On the other hand, there are many ways to produce a nonfunctional pro-
tein such as malaria-resistant G6PD. For example, during replication the insertion
of a nucleotide anywhere within the coding sequence results in a frame-shift and
likely an inactive polypeptide. Deletion of a nucleotide in the coding region will
have the same affect, as will alteration of a codon from sense to nonsense. Longer
insertions and deletions will frequently have the same effect. Missense mutations,
although likely not completely inactivating the protein, will often make the protein
less stable or less functional. Thus, considered as a class, the mutation rate from a
functional to a nonfunctional gene may be several orders of magnitude greater
than the basic nucleotide mutation rate. (Indeed, the adaptation rate of E. coli,
whose generational nucleotide mutation rate is 50-fold lower than that of humans,
has recently been measured as 10−5)[10]. For the two classes of mutations, in this
paper I explore the effect of this factor on the evolutionary rate of spread of
adaptive mutations as a function of population size, mutation rate, selection coef-
ficient, ratio of selection coefficients of the competing adaptive mutations, and
ratio of mutation rates to the adaptive state.
Calculations were performed using Mathematica [34].
3. Results
3.1 Rela vely small popula on sizes
In this section I consider small population sizes (Ne << 1/v), where Ne is the
effective population size and v is the mutation rate per generation. Unless other-
wise stated, organisms are assumed to be haploid (because most laboratory evolu-
tion experiments have been done with haploids), and the model is developed
accordingly. The resulting equations can be applied to diploid organisms by
replacing Ne by 2Ne.
In order for an adaptive mutation to become fixed in a population of relatively
small size two separate processes must occur, each with its own time scale: (1) if
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 453
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
the mutation does not yet exist in the population when the selective pressure
begins, then the expected waiting time to the appearance of the selected mutation
is tw1 = 1/(2Nevs), where s is the selection coefficient; (2) once the selected muta-
tion appears, the time for it to fix in the population is tfx1 = (2 ln Ne)/s [11].
If one is comparing two distinct mutations in the same population that are
responsive to the same selective pressure, however, both the rates of mutation to
the adaptive state and the selection coefficients may differ. For the second muta-
tion, the expected waiting time to the appearance of the selected mutation may be
written as tw2 = 1/(2Nevsrvrs), where rv is the ratio of the mutation rates to the
adaptive state and rs is the ratio of the selection coefficients for the two cases. The
expected time for the second mutation to spread to fixation in the population can
be written tfx2 = (2 ln Ne)/rss. Considering the case of a GOF versus LOF mutation,
if we take v to be the nucleotide mutation rate, then in general rv will range from
1 to ∼1000 for an LOF mutation. rs can take any positive value (both selection
coefficients are positive because both the GOF and the LOF mutations are postu-
lated to be adaptive).
A useful metric for comparing the prospects of fixation for the GOF versus
LOF mutations is rD/fx, which is defined as the expected time to appearance of an
adaptive GOF mutation minus that for an adaptive LOF mutation, divided by the
time for the LOF mutant to spread to fixation in the population. If the difference
in the expected waiting times between the selected GOF versus LOF mutations is
greater than the time required for the LOF mutation to spread, then the LOF muta-
tion will have already fixed in the population before the expected appearance of
the selected GOF mutation. The expected difference in waiting time to appearance
of the selected mutations is
1111
1
22 2
DwGwL
eevsevs
tt t Nvs Nvsrr Nvs rr
ʈ
=-= - = -
Á˜
˯
(1)
The ratio of the time difference tD to the time for the LOF mutation to spread to
fixation in the population, tfxL, is
/
11
1
211
24ln
l n
evs
D
Dfx s
fxL e e v
e
s
Nvs rr
t
rr
tNvNr
N
rs
ʈ
-
Á˜ ʈ
˯
== = -
Á˜
˯
(2)
Thus whenever rD/fx > 1, the LOF mutation is expected to fix in the population
before the selected GOF mutation appears. Figure 1 illustrates this situation. Two
curves are plotted for the appearance and subsequent spread of an LOF and a GOF
mutation in a population of 106 organisms. The selection coefficient for the GOF
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
454 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
is 0.1 and for the LOF is 0.01; thus rs is 0.1. The basic nucleotide mutation rate is
taken to be 10−9, and rv, the ratio of the mutation rate to the adaptive state for the
LOF vs GOF mutation, is set at 100. The expected waiting time to the appearance
of the selected LOF mutation under these circumstances is 500 generations, while
for the GOF mutation the time is 5,000 generations. On average the GOF mutation
would take 276 generations to fix in the population; the LOF mutation would
require 2763 generations. Figure 1 shows that under such circumstances the
selected LOF mutation would be expected to fix in the population before the
selected GOF mutation appeared. Equation 2 determines the ratio rD/fx for this situ-
ation to be 1.62.
If rs = 1/rv, then equation 2 evaluates to zero, which means there is no expected
difference tD in the waiting time to the appearance of the selected LOF versus GOF
mutations — the rate advantage of the LOF mutation is exactly offset by the rela-
tive weakness of its selection coefficient. If rs < 1/rv, then rD/fx will be negative,
which means that there is less time to the appearance of the selected GOF muta-
tion than to the LOF mutation — the rate disadvantage of the GOF mutation is
more than offset by the relative strength of its selection coefficient.
Fig. 1. Time in generations to occurrence and spread of an adaptive LOF mutation versus GOF
mutation. The LOF mutant (———) has a selection coefficient 0.1-times that of the GOF mutant
(— — —), but a mutation rate to the adaptive state 100-times that of the GOF mutant. The effective
population size Ne is set at 106. The GOF mutation rate v is 10−9 per generation and the GOF selection
coefficient s = 0.1.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 455
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Figure 2 plots the value of rD/fx versus the effective population size Ne for several
values of rv, with rs held constant at one. As can be seen, the value of rD/fx is largely
insensitive to changes in rv, the ratio of the mutation rates to the adaptive state.
Decreasing rv 100-fold from 1000 to 10 leaves the value of rD/fx little changed. In
all of these circumstances (except where rv = 2 at effective population sizes very
near 107) the ratio of the time for the LOF mutation to spread to the difference in
the expected waiting time to the selected GOF versus LOF mutations, rD/fx, is well
above one.
Figure 3 examines the relationship between the value of rD/fx versus the effective
population size Ne for several values of rs, with rv held constant at 1000, its likely
maximum for a typical gene. In this case rD/fx depends linearly on the ratio of the
selection coefficients: at any population size in the range, a decrease of a factor of 10
in rs decreases rD/fx by approximately the same factor. (The magnitude of s, the selec-
tion coefficient itself, which is absent from equation 2, does not affect the results.)
Thus, when rs is 0.01 (that is, when the selection coefficient for the LOF mutation is
only 1% of that of the GOF mutation), rD/fx decreases to a value of one at a population
size of about 1.5 × 105, versus a population size of 1.5 × 107 when rs is one.
Fig. 2. The ratio rD/fx versus effective population size Ne. rD/fx is the time to appearance of an adaptive
GOF mutation minus that for an adaptive LOF mutation, divided by the time for the LOF mutant to
spread to fixation in the population. In this figure the LOF and GOF selection coefficients are equal.
The mutation rate v is 10−9 per generation. (———) rv = 1000; (············) rv = 10; (— — —) rv = 2.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
456 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Figure 4 shows the dependence of rD/fx on rv and rs at a fixed value of Ne = 106.
As can be seen rD/fx is essentially independent of rv over a wide range, but is line-
arly dependent on rs. The pronounced curvature for both plots at lower values on
the x-axis reflects the approach of the factor (rs - 1/rv) to zero.
3.2 Rela vely large popula on sizes
In this section I consider relatively large population sizes (Ne ≥ 1/v). As population
size increases, the expected waiting time to the appearance of either or both
selected mutations can shrink to much less than the expected time for the muta-
tions to spread in the population. In fact, one or both mutations may be present
continuously in the population at a low percentage as a neutral or detrimental
allele before the new selective pressure makes it adaptive. Thus in this population
size range a different metric is required to follow the relative advantage of LOF
versus GOF mutations.
Fig. 3. The ratio rD/fx versus effective population size Ne. rD/fx is the time to appearance of an adap-
tive GOF mutation minus that for an adaptive LOF mutation, divided by the time for the LOF mutant
to spread to fixation in the population. In this figure the rate of mutation to the adaptive state of the
LOF mutant is 1000-times that of the GOF mutant. The mutation rate v is 10–9 per generation.
(———) rs = 1; (············) rs = 0.1; (— — —) rs = 0.01.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 457
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
A useful measure is the ratio of the fractions of LOF to GOF mutations in the
population when the sum of those fractions first increases to 1.0. The time t in
generations required to increase the frequency of a selected mutation from a value
of q0 to qt can be calculated from [11]:
0
0
1
1
1
t
st
qqe
q
-
=ʈ
-
+Á˜
˯
Thus (ignoring double mutants) the number of generations required for the
fractions of an LOF and GOF mutation to sum to one can be calculated from:
00
00
11
1
11
11
s
rst
st
GL
GL
qq
ee
qq
-
-
+=
ʈ ʈ
--
++
Á˜ Á˜
˯ ˯
(3)
Fig. 4. The ratio rD/fx versus rs and rv. rD/fx is the time to appearance of an adaptive GOF mutation
minus that for an adaptive LOF mutation, divided by the time for the LOF mutant to spread to fixa-
tion in the population. rs is the ratio of the LOF to GOF selection coefficients. rv is the ratio of the
rate of LOF to GOF mutation to the adaptive state. The effective population size Ne is set at 106 and
the GOF mutation rate v is 10−9 per generation. (———) rs is set at 1 and rv ranges from 1 to 1000;
(— — —) rv is set at 1000 and rs ranges from 0.001 to 1.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
458 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
The initial fraction q0 when a selected mutation begins to increase in a haploid
population is at a minimum 1/Ne. However, for population sizes greater than the
inverse of the mutation rate, numerous mutants are expected to be present in the
initial population. For example, if the mutation rate v is 10−9 and the population
size is 1012, then there will be 103 mutants produced in the first generation. So the
initial fraction q0G is at least
11
,
e
eee
Nv v
NNN
+=+
and q0L is at least
1
v
e
rv
N+
.
The time t in equation (3) is the time required for the selected mutation to
spread. Thus if we are counting generations from the first application of the selec-
tive pressure, then the expected waiting time for the selected mutation must be
accounted for. As mentioned previously, for a haploid GOF mutation this is twG =
1/(2Nevs) and for an LOF mutation twL = 1/(2Nevsrvrs). Equation (3) can then be
re-written as:
() ()
00
00
11
1
11
11
fx wG s fx wL
st t rst t
GL
GL
qq
ee
qq
-- - -
+=
ʈ ʈ
--
++
Á˜ Á˜
˯ ˯
(4)
where (tfx - tw) is the time for the mutations to spread to a sum fraction of 1.0 after
the waiting time for at least one kind of selected mutation to first appear in the
population. Given Ne, v, s, rv, and rs, equation 4 can be solved for tfx and the value
used to determine rfx, which is the fraction of adaptive LOF mutations divided by
the fraction of adaptive GOF mutations in the population when the two fractions
first sum to one:
()
0
0
()
0
0
1
1
1
1
fx wG
sfxwL
st t
G
G
fx rs t t
L
L
qe
q
rqe
q
--
--
ʈ
-
+Á˜
˯
=ʈ
-
+Á˜
˯
(5)
Figure 5 plots rfx from equation 5 at rs = 1 and rv = 1000 for population sizes Ne
ranging from 107 to 1010. It is seen that at lower values of Ne, rfx increases very
rapidly. Indeed, at population sizes of 107 or less, rfx is greater than Ne, reflecting
the fact that less than one GOF mutant is expected to be present in the population
when the LOF mutant has fixed. As Ne increases, rfx approaches a constant value
of approximately 31.6. Thus when the population initially consists entirely of LOF
and GOF mutants and Ne ≥ 109, under the circumstances described in Figure 5
LOF mutants will represent about 97% of the population.
Figure 6 plots rfx as a function of rv for population sizes from 106.5 to 1012, with
rs = 1. At the smallest population sizes the fixation ratio is extremely sensitive to
the ratio of mutation rates. As Ne increases, however, and it becomes more likely
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 459
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Fig. 5. The ratio rfx versus the effective population size Ne. rfx is the fraction of adaptive LOF muta-
tions divided by the fraction of adaptive GOF mutations in the population when the two fractions
first sum to one. rs = 1; rv = 1000; s = 0.1; v = 10−9 per generation.
Fig. 6. The ratio rfx versus the ratio rv. rfx is the fraction of adaptive LOF mutations divided by the
fraction of adaptive GOF mutations in the population when the two fractions first sum to one. rv is
the ratio of the rate of LOF to GOF mutation to the adaptive state. rs is set at 1; v is 10−9. (———)
Ne = 106.5; (············) Ne = 107; (- - - - -) Ne = 107.5; (– ·· – ·· – ·· –) Ne = 108; (— — —) Ne = 109;
(– · – · – · –) Ne = 1012.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
460 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Fig. 7. The ratio rfx versus the effective population size Ne. rfx is the fraction of adaptive LOF muta-
tions divided by the fraction of adaptive GOF mutations in the population when the two fractions
first sum to one. For all curves rv is set to 1000. (———) s = 0.1, rs = 1, v = 10−9; (············) s = 0.0001,
rs = 1, v = 10−9; (— — —) s = 0.1, rs = 0.5, v = 10−9; (- - - - -) s = 0.1, rs = 1, v = 10−10.
that the mutants are present in the population from the first generation, the
sensitivity decreases. As seen in the figure, the plots of rfx versus rv for values of
Ne ≥ 1/v are essentially superimposable, and closely approximate the relationship
.
fx v
rr=
Figure 7 plots rfx versus Ne for several variables. The solid curve reproduces the
values from Figure 5 of s = 0.1 and rs = 1. Coinciding with the solid curve is a
dotted curve for which s = 0.0001, demonstrating the insensitivity of the curve to
changes in the selection coefficient itself. The long-dashed curve uses the same
parameters as the solid curve except that the value of rs has been decreased to 0.5.
As can be seen, this decreases the value of rfx by several orders of magnitude, so
that at large population sizes the value is below one, and the GOF mutation pre-
dominates at fixation, despite the initial 1,000-fold advantage of the LOF mutation
rate. The short-dashed curve uses the same parameters as the solid curve except
that the value of v has been decreased from 10−9 to 10−10. As can be seen, this has
the effect of simply moving the curve an order of magnitude to the right on the
population axis.
Figure 8 plots rfx versus rv at three values of rs with Ne »1/v. As seen, modestly
varying the ratio of the selection coefficients displaces the curve considerably
along the rfx axis and slightly alters its slope. Figure 9 compares rv and rs versus
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 461
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Fig. 8. The ratio rfx versus the ratio rv. rfx is the fraction of adaptive LOF mutations divided by the
fraction of adaptive GOF mutations in the population when the two fractions first sum to one. rv is
the ratio of the rate of LOF to GOF mutation to the adaptive state. For all curves, Ne is set at 1012 and
v is 10−9. (———) rs = 1; (············) rs = 0.8; (- - - - -) rs = 1.25.
rfx, showing the relative sensitivity of the fixation ratio to those parameters at large
Ne. Figure 9 plots values for rs including from one to 100; that is, for situations in
which the selection coefficient of the LOF mutation is greater than or equal to that
of the GOF mutation. rfx is greater than one and increases rapidly in this region. In
general, whenever rs ≥ 1 and rv > 1, rfx will be greater than one at any population
size. That is, the LOF mutation will always be the majority of the population when
the entire population is initially comprised of LOF and GOF mutations.
4. Discussion
4.1 LOF versus GOF adap ve muta ons
Organisms can adapt to their environment either by acquiring new abilities or by
abandoning old ones. This can be observed in such examples as legless snakes and
sightless cave fish. Science has learned especially in the last fifty years that altered,
observable phenotypes are the manifestation of changes to the genetic endowment
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
462 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
of an organism. It has also learned that there is not a necessary correlation between
loss or gain of an ability at the phenotypic level and loss or gain of a functional
genetic element at the molecular level. In other words, what strikes an observer as
a phenotypic gain of function may be caused by either a molecular loss or gain of
function. The same holds for a phenotypic loss of function: it may be the result of
a genetic gain or loss of function. Because organisms can adapt by either molecu-
lar GOF mutations or LOF mutations it is of basic interest to determine which, if
either, kind of mutation will dominate under various circumstances.
Research over the past fifty years has shown that many genetic elements consist
of multiple nucleotides. Protein coding regions can be thousands of nucleotides in
length; RNA genes can be hundreds of nucleotides; regulatory elements and pro-
cessing signals can be several nucleotides to dozens of nucleotides long. A sub-
stantial portion of possible mutations in these elements will result in the diminution
or loss of their function. Thus, as a class, LOF mutations for a particular genetic
element will occur at a rate from several times to several-orders-of-magnitude
times the basic nucleotide substitution rate.
Fig. 9. The ratio rfx versus rs and versus rv. rfx is the fraction of adaptive LOF mutations divided by
the fraction of adaptive GOF mutations in the population when the two fractions first sum to one. rs
is the ratio of the LOF to GOF selection coefficients. rv is the ratio of the rate of LOF to GOF
mutation to the adaptive state. For both curves, Ne is set to 1012 and v is 10−9. (———) rfx versus rs,
rv is set at 1000; (— — —) rfx versus rv, rs is set at 1.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 463
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
That is not the case for GOF mutations. Consider two examples: First, a tran-
scription factor binding site that is 10 nucleotides in length, and a second DNA
sequence which has 9 of 10 nucleotides that are necessary to form a second regula-
tory site. Suppose that in response to a new selective pressure an adaptive effect
could be secured either by mutating the first site so that it lost its function or by
mutating the single mismatching residue of the second site so that it gained
function. The LOF mutation would on average appear at 10-times the nucleotide
substitution rate, simply because there are multiple ways to break the functioning
element. The GOF mutation, however, would appear at even less than the basic
rate of nucleotide substitution (because for a currently-nonfunctional, potential
genetic element there it is possible that one of the “correct” nucleotides in the
sequence will mutate before the “incorrect” one [12]). Second, consider a recently
duplicated gene which could provide an adaptive effect in response to a new
selective pressure if a certain nucleotide in the gene were altered (allowing the
duplicate gene product to, say, diverge productively in activity from the parent
gene product). Suppose, however, that an adaptive effect could also be had by
reducing or eliminating the activity of another, separate gene. Because of the many
ways in which a gene can be altered to lose function, the LOF mutation would
have a rate several orders of magnitude greater than that of the GOF mutation for
the duplicated gene.
There can be cases in which a GOF mutation may appear at several times the
nucleotide substitution rate. I discussed earlier the sickle mutation, in which a
single particular nucleotide in the β-globin gene must be changed. Yet in other
cases of GOF, there can be several possible nucleotides to change, each of which
will suffice. For example, Couñago et al. [13] replaced the essential gene for ade-
nylate kinase in Geobacillus stearothermophilus — a moderate thermophile —
with that of Bacillus subtilis, a mesophile, which they then grew in a turbidostat
at increasing temperatures. Over the course of 1500 generations they isolated six
thermostable mutants of the enzyme — one single point mutant and five double
point mutants derived from the single mutant. Thus in this circumstance the
enzyme could gain the function of being active in a hostile environment by alter-
ing any of six positions. Nonetheless, the number of ways to break a functional
element will almost always be much greater than the number of ways to construct
one, so that in almost all cases rv would be expected to be greater than one.
4.2 Eff ect of disparity in adap ve rate
In this chapter I investigate the effect of the disparity in rate of mutation to an
adaptive state for LOF and GOF mutations as a function of several parameters.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
464 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
The model presented here is a simple, deterministic one, which does not consider
the probabilistic nature of changes in allele frequencies [11]. Because of its sim-
plicity, the general behavior of the investigated model is visible with considerable
clarity and the issue of the evolutionary rate advantage of adaptive LOF mutations
is highlighted.
The behavior at relatively small population sizes is governed by equation 2,
which accounts for the two separate phases of fixation of a new mutation: the
expected waiting time to the appearance of the selected mutation, and the time
taken for the mutation to spread within the population. An interesting aspect of the
equation is that it does not contain the selection coefficient s; that is, the ratio of
the selection coefficients rs influences the competition between the two mutations
rather than the absolute value of either or both selection coefficients. (This also is
the case at relatively large population sizes, as shown by Figure 7.) Whenever
equation 2 evaluates to rD/fx >1, then the LOF mutation is expected to fix in the
population before a selected GOF mutation appears. Thus, as illustrated in Figure
1, an LOF mutation whose selection coefficient is ten-fold weaker than an adap-
tive GOF mutation can outrace it to fixation, due to its greater rate of mutation to
an adaptive state.
Figures 2 and 3 show that this effect exerts substantial influence at relatively
low population sizes. For a population size of < 107, if rs ≥ 1 and rv > 1, then an
LOF mutant is always expected to fix in the population before a selected GOF
mutant appears. Because of an increasing disparity in waiting times, at population
sizes <<107 an LOF mutant may be fixed in the population first even if its selective
advantage is considerably less than that of a GOF mutant. For example, for a
population size of 105, an LOF adaptive mutation will become fixed first at rv ≥ 1
even if its selection coefficient is only one-hundredth that of a GOF adaptive muta-
tion; that is, if rs ≥ 0.01. At smaller population sizes, the advantage for the LOF
mutation increases linearly with 1/Ne.
If an LOF mutation with a smaller selection coefficient is first fixed in a
population, what scenario is most likely to occur after the GOF mutation
eventually appears? The answer to that question is likely to depend sharply on
the specific genetic elements involved. One possible scenario is that the GOF
mutation also spreads to fixation, and the LOF mutation remains fixed. A second
possibility is that, depending on the physical nature of the mutation, the LOF
mutation may be repaired by subsequent mutation after the GOF mutation
spreads in the population. If it cannot be repaired, it may be replaced by
horizontal gene transfer or by having its function taken over by another genetic
element, or the organism may adapt in other ways to its loss. Penman et al. [14]
recently demonstrated that the outcome in competition between the sickle
mutation (which is highly protective against malaria) and various thalassemic
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 465
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
disorders (which are less protective) is quite difficult to predict because of
epistatic effects unrelated to their anti-malarial activities. Thus the future course
of the evolution of a system after initial fixation of an LOF mutation might be
considerably more complex than a linear succession of mutations with increas-
ing selective value.
For v = 10−9, at population sizes Ne >108 an LOF mutation is no longer
expected to fix in the population before a selectable GOF mutation appears,
even if rs = 1, because the larger population sizes produce both types of
mutations within the time it would take for the LOF mutation to spread in the
population. Nonetheless, even though the metric rD/fx decreases below one in this
range, in many cases the LOF mutation will become the dominant mutation in
the population. In order to assess the advantages of LOF versus GOF mutations
in this population range, a new metric, rfx, was introduced in equation 5. rfx is
the ratio of LOF to GOF mutants when their fraction of the population first sums
to one.
Figure 6 shows that LOF mutations always possess a rate advantage over GOF
mutations if the respective selection coefficients are equal; that is, if rs = 1. Under
these circumstances at large population sizes (Ne ≥ 1/v),
,
fx v
rrª
and the ratio of
LOF to GOF mutations when their fraction first sums to one will range from 1.41
to 31.6 for values of rv ranging from 2 to 1000. Thus the LOF mutant will comprise
from 59% of the population to 97% of the population. If at this point the mutants
then drift neutrally in the population (because it is postulated that neither has a
selective advantage over the other), the LOF mutant is expected to become fixed
with a probability equal to its population fraction [15].
Under what circumstances would two selection coefficients be expected to be
equal? If two mutations both met the new selective pressure without causing
deleterious pleiotropic effects, then their selection coefficients would be expected
to be the same. Thus whenever such a situation presents itself, the LOF mutation
would have an advantage.
If the selection coefficients are not equal, how likely is it that a GOF mutation
will have a value of s greater than that of an LOF mutation, or vice-versa? The
answer to that question is not known, but both LOF and GOF mutations can have
significant selection coefficients. The selection coefficient for LOF mutations of
the rpoS gene of E. coli has been measured at 0.217, a substantial value [16]. The
selection coefficient for the GOF sickle mutation has been estimated as 0.05 to
0.18, again a large value [17]. If in general there is no overall correlation between
adaptive GOF versus LOF mutations and the magnitude of the selection coeffi-
cient, then the intrinsic rate advantage enjoyed by LOF mutations will bias long-
term evolution in that direction.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
466 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
4.3 Comparison to laboratory evolu on experiments
Over the past forty years many laboratories have conducted evolution experi-
ments, observing adaptation of micro-organisms to varying environmental condi-
tions, and in many cases identifying the molecular changes that comprised the
adaptive mutation [4]. How do the results obtained in this chapter bear on the
interpretation of those experiments?
Comparison to experiments where Ne < 1/v: The most extensive laboratory evo-
lution experiment to date has been performed under the direction of Richard
Lenski at Michigan State University [18]. Starting in the early 1990s, Lenski and
colleagues began growing 10 ml cultures of E. coli, which undergo six to seven
doublings per day. Each day they transferred 1% of the culture to fresh medium.
Over the years the cultures have undergone more than 50,000 generations. All
adaptive mutations identified to date appear to be LOF ones [4]. The single most
beneficial mutation was the destruction of the rbs operon by insertion sequences.
The value of the selective coefficient for this was approximately 0.02 [19]. Other
identified LOF mutations include ones in the pykF, nadR, pbpA-rodA, hokB/sokB,
malT, and topA genes. A number of other adaptive genes have been identified to
date, but the natures of the mutations, whether LOF or GOF, have not yet been
reported [20].
The rate of nucleotide mutations per generation of E. coli is ∼5 × 10−10 [21]. The
effective population size Ne of Lenski’s [18] cultures of E. coli is ∼2 × 107, which
is the harmonic mean between the initial population of the day’s culture (5×106)
and the final population of the day (5×108) if the population is assumed to double
in discrete generations [11]. Substituting these numbers into equation 2 shows that
rD/fx would be 1.47 — greater than one — if rs were one and rv were 100. An LOF
mutation would thus be expected to be fixed in the population before a GOF muta-
tion appeared if their selection coefficients were equal. How great of a selective
advantage must a GOF mutation have to outcompete an LOF mutation under these
circumstances? Using equation 2 it is seen that if rs is 0.68, then rD/fx falls slightly
below one. In other words, a GOF mutation would have to have a selection coef-
ficient about 50% greater than an LOF mutation in these circumstances in order to
at least appear in the population before the LOF mutation were fixed.
To find out how much greater the selection coefficient must be to actually out-
compete the LOF mutation, we must use equations 4 and 5 to calculate rfx.
Assuming rs were 0.68, there would be approximately one GOF allele in the popu-
lation per ∼2×107 LOF alleles. In order to overcome the LOF rate advantage,
however, rs would have to fall to ∼0.25. In other words, if the selection coefficient
of the GOF mutation were approximately four times that of the LOF mutations,
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 467
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
then the GOF mutation would be slightly more than half the population. In order
to dominate the population by ∼90% rs would have to be ∼0.2; that is, the selection
coefficient of the GOF mutation would have to be about five-fold that of the LOF
mutation. Since no GOF mutations have yet been seen, we can tentatively con-
clude that there are no GOF mutations available whose selective value is five-fold
greater than the least-adaptive LOF mutations seen in this series of experiments.
(Lenski’s group recently reported a very adaptive Cit+ phenotype, which appar-
ently required both LOF and gene duplication mutations [22]). If an LOF mutation
appeared within the first 25,000 generations, it would require a minimum selection
coefficient of 0.00076 to spread to fixation in the next 25,000 generations. To
outcompete it, a GOF mutation would require a minimum selection coefficient of
five-times this value, i.e. ∼0.0038. Thus it can be concluded that there are no GOF
mutations available under the circumstances of the experiment whose selection
coefficients exceed that number.
The question might be asked, what if a potential GOF mutation with a suffi-
ciently strong selection coefficient existed, but simply failed to arise during the
term of the experiment? That is always a possibility, but an unlikely one. Given
the scale of the Lenski experiment [20], with an effective population size of
2×107 over 50,000 generations and a nucleotide mutation rate of ∼5×10−10, each
nucleotide is expected to be substituted 500-fold over the course of the experi-
ment. Deletions, additions, and other kinds of mutations would similarly be
expected to occur multiple times. There were many redundant opportunities for
all simple mutations to arise (the Cit+ phenotype apparently needed several muta-
tions to arise). Thus we can be confident that if a particular mutation, or kind of
mutation, was not observed, then it is very unlikely to have the necessary selec-
tion coefficient.
Comparison to experiments where Ne > 1/v: As seen in Figures 7–9, at Ne ≥1/v,
rfx is much more sensitive to rs than at smaller population sizes. Just a slight advan-
tage in the selection coefficient for a GOF mutation is sufficient to offset a 1,000-
fold advantage in the rate of LOF mutation. This great sensitivity can be used to
infer whether such a GOF mutation is available under particular environmental
circumstances. That is, if a certain selective pressure is applied, one or more LOF
mutations are observed, and Ne ≥1/v, then the failure to observe a GOF mutation
would imply that no GOF mutation is available within a single mutational step
that had a somewhat greater selection coefficient than the LOF mutations(s).
Conversely, if a GOF mutation were observed but no LOF mutation, we could
deduce that no LOF mutation was available that had a selection coefficient greater
than or equal to the GOF mutation.
As mentioned earlier, Couñago et al. [13] replaced the essential gene for adenylate
kinase in Geobacillus stearothermophilus — a moderate thermophile — with that of
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
468 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
Bacillus subtilis, a mesophile, which they then grew in a turbidostat at increasing
temperatures. Over the course of 1500 generations they isolated six thermostable
mutants of the enzyme — one single point mutant and five double point mutants
derived from the single mutant, which can all be classified as GOF mutations. The
mutation rate of G. stearothermophilus can be estimated by using a value for the
mutation rate of approximately 0.003 per genome per generation for DNA-based
microbes, which yields a value of about 5×10−10 mutations per generation [21]. Since
the authors maintained a continuous culture, the population was not subject to the
large changes in size seen in Lenski’s experiments, so the effective population of
microbes per generation in the turbidostat was ∼5×1010. In other words, in the
Couñago et al. [13] experiment, Ne >1/v. Inserting these values into equations 4 and
5 shows that if a potentially adaptive LOF mutation were available with the same
selection coefficient as a GOF mutation, then it would dominate the population with
an rfx of 9.9; in other words, it would comprise ~91% of the population. Thus it can
be concluded that, despite the frequency of adaptive LOF mutations in Lenski’s
work, no LOF mutation with an rs ≥ 1 compared to the observed GOF mutations was
available in the experiment conducted by Couñago et al. [13]. The likely reason for
the disparate results is the differing experimental regimens. Lenski did not put strong
constraints on the direction for E. coli to evolve, but Couñago et al. [13] replaced an
essential gene with a substitute optimized for a different growth temperature before
applying selective pressure, which they termed a “weak link” method. Furthermore,
Couñago et al. [13] used an Ne that was more than three orders of magnitude greater
than Lenski’s group. The activity of the thermophilic adenylate kinase activity had
to be replaced or compensated for. Apparently, the fastest way available to do so at
high Ne was by GOF point mutations to the mesophilic substitute gene.
4.3.1 Comparison to experiments where two selective routes were
potentially available
An interesting conceptual blend of the Lenski [18] and Couñago et al. [13]
approaches was recently published by Gauger et al. [23]. This group mutated two
amino acid residues of a plasmid-borne trpA gene of E. coli, transfected a Trp−
bacterial strain with the plasmid, and grew it in a tryptophan-limiting medium.
One of the mutations (E49V) alone completely inactivates the gene product; the
other mutation (D60N), when present alone, allows weak Trp+ activity and sup-
ports growth in Trp− media when the plasmid-borne gene is overexpressed. The
authors expected cells containing the double mutant plasmid to take a short,
selected route to full Trp+ activity when grown in tryptophan-limiting medium by
first reverting the inactivating mutation at position 49 (allowing the resumption of
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 469
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
weak Trp+ activity) and then reverting the second mutation at position 60 to regain
full activity. However, almost all mutants recovered after sustained growth had not
taken even the first step on that expected pathway. Rather, the expression of the
trpA gene was decreased either by deletion, insertion of an IS element, or by
various point mutations, apparently saving the cell the energy of overproducing
the protein.
The E. coli point mutation rate is 5×10−10. Gauger et al. [23] grew liquid
cultures to an effective population size Ne of ∼0.6×107 cells per generation.
Substituting these numbers into equation 2 shows that rD/fx would be 5.3 —
greater than one — if rs were one and rv were 100. That is, if the selective advan-
tage the cell received from shutting down overexpression of the plasmid-borne
gene were equal to the selective advantage it would receive from taking the first
GOF mutational step to partial Trp+ activity, the LOF mutation would be
expected to easily be fixed in the population well before a GOF mutation
appeared. For one partial-revertant to be expected to appear before the LOF
mutant fixed under the conditions of the experiment rs would have to be about
0.2. That is, the selection coefficient for the GOF mutation would have to be
approximately five-fold greater than that of the LOF mutation. Equations 4 and
5 can be used to show that for the GOF mutant to be expected to dominate the
population to >90%, the GOF selection coefficient would have to be about 12.5-
times that for the LOF mutation. Apparently, regaining merely limited Trp+
activity did not have 12.5-times the selective value of the decrease in expression
of the plasmid gene caused by the LOF mutations. Thus, under the conditions of
the experiment, the selective pathway back to full Trp+ activity is blocked at the
first step. Interestingly, if cells transfected with either singly-mutated plasmid
(E49V or D60N) were grown in liquid culture, Trp+ revertants quickly took over
the culture, indicating the selection coefficient for full-reversion was greater than
12.5-times the selection coefficient for saving the cell the energy of overproduc-
ing the protein [23].
4.4 Comparison to short-term evolu on in the wild
A possible objection to results from laboratory evolution experiments is that they
are artificial. The organisms are housed in special environments and not exposed
to the rigor and variety of challenges they would encounter in nature. Thus the
many advantageous LOF mutations observed in experimental work may not
reflect what happens in nature, since presumably the great majority of an organ-
ism’s genes are required in the wild, and therefore few if any adaptive LOF
mutations are available in nature.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
470 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
While that may turn out to be the case, and more data will be required to come
to a definitive conclusion, an increasing number of results from nature appear to
ratify the importance of adaptive LOF mutations in the wild. One class of such
LOF mutations which I have mentioned previously includes genes that help adapt
humans to the presence of malaria [4]. Other important human adaptive mutations
are also LOF mutations: immunity to HIV due to a deletion variant of CCR-5 [24];
and resistance to tuberculosis by a deletion variant of SLC11A1 [25]. Development
of lactose tolerance in adult humans [26] also seems a good candidate for an
adaptive LOF mutation, perhaps by loss of a repressor binding site, although that
has not yet been confirmed. In a recent survey of multiple human genomes it has
been determined that for humans, “On average, each person is found to carry
approximately 250 to 300 loss-of-function variants in annotated genes...,” over 1%
of the total number of human genes [27].
A second example of LOF mutation in nature is seen in the evolution of the
plague bacterium Yersinia pestis. A plausible evolutionary scenario to explain its
great virulence is that it serially acquired several plasmids which conferred on it
the ability to be transferred between mammalian hosts by flea bite [28, 29]. After
the acquisition of these plasmids (which are GOF events), the Y. pestis genome
lost several hundred genes, apparently because they were no longer necessary for
its new life cycle [29, 30]. Thus, after several GOF events, the plague bacterium
adjusted to its new environment by much more numerous and rapid LOF adaptive
mutations.
Nadeau and Jiggins [31] have recently reviewed genomic studies of adaptation
in natural populations and note that “Many of the well-studied examples of
adaptive evolution have involved trait loss, such as the loss of bony structures in
freshwater stickleback populations and the reduction of pigmentation and eyes in
cavefish.” Although, as mentioned earlier in this chapter, there is not a necessary
correlation between phenotypic trait loss and adaptive LOF mutations, in the cases
mentioned by Nadeau and Jiggins [31] they coincide. Loss of pelvic spines in
freshwater sticklebacks has been traced to deletion of a Pitx1 enhancer [32]. Eye
reduction in cavefish apparently involves multiple genes [33]. Of those that have
been identified three involve decreased expression of the gene (γ-M crystallin,
rhodopsin, and αA crystallin). One gene, hsp90α, has increased expression, and it
appears to be involved in promoting apoptosis.
5. Conclusion
Organisms have adapted over evolutionary history both by gaining and losing
functions. Therefore it is of basic interest to determine if one or the other
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 471
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
dominates during particular circumstances. Until the past few decades, however,
the molecular events underlying these processes were obscure. In recent decades
science has in some cases gained the ability to determine whether the events
behind a phenotypic adaptation involve an adaptive GOF mutation or an adaptive
LOF mutation [4].
Both experimental laboratory work over the past few decades and recent
genomic studies of adaptation in natural populations attest to the importance, even
dominance, of LOF mutations in short term evolutionary episodes. The work pre-
sented in this paper helps show why this should be the case. Functional genetic
elements such as genes and regulatory regions are built of multiple nucleotides,
and a substantial fraction of mutations to these elements will cause them to lose
their function. Thus the LOF mutation rate can be orders of magnitude greater than
the nucleotide substitution rate. On the other hand, GOF mutations tend to be quite
specific. So the rate for adaptive GOF mutations tends to be equal or very similar
to the nucleotide mutation rate. As shown here, for some population size regions
and for some values for the ratio of selection coefficients, the greater rate of muta-
tion to the adaptive state for LOF versus GOF gives adaptive LOF mutations an
intrinsic edge over adaptive GOF mutations.
In retrospect, the result is straightforward. Yet it also seems somewhat surpris-
ing because, as Nadeau and Jiggins [31] write, “there clearly are complex
structures that are gained during evolution ... and we currently know little about
how this process takes place.” It may be hoped that understanding how organisms
survive in the short term by adaptive LOF mutations will be a step toward
understanding how complex structures are built over the long term.
References
1. Darwin C (1859) The Origin of Species. Bantam Books, New York.
2. Stott R (2003) Darwin and the Barnacle. W. W. Norton, New York.
3. Chang JC, Kan YW (1981) Antenatal diagnosis of sickle cell anaemia by direct
analysis of the sickle mutation. Lancet 2:1127–1129.
4. Behe MJ (2010) Experimental evolution, loss-of-function mutations, and “The First
Rule of Adaptive Evolution”. Quart Rev Biol 85:419–445.
5. Friedman MJ (1978) Erythrocytic mechanism of sickle cell resistance to malaria. Proc
Natl Acad Sci USA 75:1994–1997.
6. Verra F, Simpore J, Warimwe GM, Tetteh KK, Howard T, Osier FH, Bancone G,
Avellino P, Blot I, Fegan G, Bull PC, Williams TN, Conway DJ, Marsh K, Modiano
D (2007) Haemoglobin C and S role in acquired immunity against Plasmodium
falciparum malaria. PloS ONE 2:e978.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
472 M. J. Behe
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
7. Ruwende C, Khoo SC, Snow RW, Yates SN, Kwiatkowski D, Gupta S, Warn P,
Allsopp CE, Gilbert SC, Peschu N, Newbold CI, Greenwood BM, Marsh K, Hill AVS
(1995) Natural selection of hemi- and heterozygotes for G6PD deficiency in Africa by
resistance to severe malaria. Nature 376:246–249.
8. Cavalli-Sforza LL, Menozzi P, Piazza A (1994) The History and Geography of Human
Genes. Princeton University Press, Princeton, NJ.
9. Lynch M (2010) Rate, molecular spectrum, and consequences of human mutation.
Proc Natl Acad Sci USA 107:961–968.
10. Perfeito L, Fernandes L, Mota C, Gordo I (2007) Adaptive mutations in bacteria: high
rate and small effects. Science 317:813–815.
11. Li WH (1997) Molecular Evolution. Sinauer Associates, Sunderland, Mass.
12. Lynch M, Conery JS (2000) The evolutionary fate and consequences of duplicate
genes. Science 290:1151–1155.
13. Couñago R, Chen S, Shamoo Y (2006) In vivo molecular evolution reveals biophysi-
cal origins of organismal fitness. Mol Cell 22:441–449.
14. Penman BS, Pybus OG, Weatherall DJ, Gupta S (2009) Epistatic interactions between
genetic disorders of hemoglobin can explain why the sickle-cell gene is uncommon in
the Mediterranean. Proc Natl Acad Sci USA 106:21242–21246.
15. Kimura M (1983) The Neutral Theory of Molecular Evolution. Cambridge University
Press, Cambridge.
16. Ferenci T (2008) The spread of a beneficial mutation in experimental bacterial popu-
lations: the influence of the environment and genotype on the fixation of rpoS
mutations. Heredity 100: 446–452.
17. Li WH (1975) The first arrival time and mean age of a deleterious mutant gene in a
finite population. Am J Hum Gen 27:274–286.
18. Lenski RE (2004) Phenotypic and genomic evolution during a 20,000-generation
experiment with the bacterium Escherichia coli. Plant Breeding Rev 24:225–265.
19. Cooper VS, Schneider D, Blot M, Lenski RE (2001) Mechanisms causing rapid and
parallel losses of ribose catabolism in evolving populations of Escherichia coli B. J
Bacteriology 183: 2834–2841.
20. Barrick JE, Yu DS, Yoon SH, Jeong H, Oh TK, Schneider D, Lenski RE, Kim JF
(2009) Genome evolution and adaptation in a long-term experiment with Escherichia
coli. Nature 461: 1219–1221.
21. Drake JW, Charlesworth B, Charlesworth D, Crow JF (1998) Rates of spontaneous
mutation. Genetics 148:1667–1686.
22. Blount ZD, Barrick JE, Davidson CJ, Lenski RE (2012) Genomic analysis of a key
innovation in an experimental Escherichia coli population. Nature 489: 7899-7906.
23. Gauger AK, Ebnet S, Fahey PF, Seelke R (2010) Reductive evolution can prevent
populations from taking simple adaptive paths to high fitness. BIO-Complexity
2:1–9.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.
Getting There First … Adaptive Loss-of-Function Mutations 473
b1567 Biological Information — New Perspectives b1567_Sec3.3 8 May 2013 2:34 PM
24. Samson M, Libert F, Doranz BJ, Rucker J, Liesnard C, Farber CM, Saragosti S,
Lapoumeroulie C, Cognaux J, Forceille C, Muyldermans G, Verhofstede C, Burtonboy
G, Georges M, Imai T, Rana S, Yi Y, Smyth RJ, Collman RG, Doms RW, Vassart G,
Parmentier M (1996) Resistance to HIV-1 infection in caucasian individuals bearing
mutant alleles of the CCR-5 chemokine receptor gene. Nature 382:722–725.
25. Barnes I, Duda A, Pybus OG, Thomas MG (2010) Ancient urbanization predicts
genetic resistance to tuberculosis. Evolution doi:10.1111/j.1558–5646.2010. 01132.x
26. Tishkoff SA, Reed FA, Ranciaro A, Voight BF, Babbitt CC, Silverman JS, Powell K,
Mortensen HM, Hirbo JB, Osman M, Ibrahim M, Omar SA, Lema G, Nyambo TB,
Ghori J, Bumpstead S, Pritchard JK, Wray GA, Deloukas P (2007) Convergent
adaptation of human lactase persistence in Africa and Europe. Nat Genetics 39:31–40.
27. The 1000 Genomes Project Consortium (2010) A map of human genome variation
from population-scale sequencing. Nature 467:1061–1073.
28. Wren BW (2003) The yersiniae–a model genus to study the rapid evolution of bacte-
rial pathogens. Nat Rev Microbiol 1:55–64.
29. Carniel E (2003) Evolution of pathogenic Yersinia: Some lights in the dark. In: Skurnik
M, Bengoechea JA, Granfors K (eds) The Genus Yersinia: Entering the Functional
Genomic Era, Kluwer Academic/Plenum, New York.
30. Chain PS, Hu P, Malfatti SA, Radnedge L, Larimer F, Vergez LM, Worsham P, Chu
MC, Andersen GL (2006) Complete genome sequence of Yersinia pestis strains Antiqua
and Nepal516: evidence of gene reduction in an emerging pathogen. J Bacteriol
188:4453–4463.
31. Nadeau NJ, Jiggins CD (2010) A golden age for evolutionary genetics? Genomic stud-
ies of adaptation in natural populations. Trends Gen 26:484–492.
32. Chan YF, Marks ME, Jones FC, Villarreal G Jr., Shapiro MD, Brady SD, Southwick
AM, Absher DM , Grimwood J, Schmutz J, Myers RM, Petrov D, Jónsson B, Schluter
D, Bell MA, Kingsley DM (2010) Adaptive evolution of pelvic reduction in sticklebacks
by recurrent deletion of a Pitx1 enhancer. Science 327:302–305.
33. Jeffery WR (2009) Regressive evolution in Astyanax cavefish. Ann Rev Genetics
43:25–47.
34. Wolfram S (2003) The Mathematica Book. Wolfram Media, Champaign, IL.
Biological Information Downloaded from www.worldscientific.com
by 173.49.178.220 on 06/08/13. For personal use only.