A germ-line-selective advantage rather than an
increased mutation rate can explain some
unexpectedly common human disease mutations
Soo-Kyung Choi, Song-Ro Yoon, Peter Calabrese*†, and Norman Arnheim*†
Molecular and Computational Biology Program, University of Southern California, 1050 Childs Way, Los Angeles, CA 90089-2910
Edited by James F. Crow, University of Wisconsin, Madison, WI, and approved April 16, 2008 (received for review February 7, 2008)
Two nucleotide substitutions in the human FGFR2 gene (C755G or
C758G) are responsible for virtually all sporadic cases of Apert
syndrome. This condition is 100–1,000 times more common than
genomic mutation frequency data predict. Here, we report on the
C758G de novo Apert syndrome mutation. Using data on older
donors, we show that spontaneous mutations are not uniformly
distributed throughout normal testes. Instead, we find foci where
C758G mutation frequencies are 3–4 orders of magnitude greater
than the remaining tissue. We conclude this nucleotide site is not
a mutation hot spot even after accounting for possible Luria–
Delbruck ‘‘mutation jackpots.’’ An alternative explanation for such
foci involving positive selection acting on adult self-renewing Ap
spermatogonia experiencing the rare mutation could not be re-
jected. Further, the two youngest individuals studied (19 and 23
years old) had lower mutation frequencies and smaller foci at both
mutation sites compared with the older individuals. This implies
that the mutation frequency of foci increases as adults age, and
thus selection could explain the paternal age effect for Apert
syndrome and other genetic conditions. Our results, now including
the analysis of two mutations in the same set of testes, suggest
that positive selection can increase the relative frequency of
premeiotic germ cells carrying such mutations, although individu-
als who inherit them have reduced fitness. In addition, we com-
pared the anatomical distribution of C758G mutation foci with
both new and old data on the C755G mutation in the same testis
and found their positions were not correlated with one another.
Apert syndrome ? paternal age ? positive selection ? spermatogonia ? testis
of sporadic cases of human autosomal dominant or sex-linked
diseases (1, 2) or by DNA analysis of sperm (3–5). In a recent
mutations by dissecting the whole human testis and measuring
the mutation frequency at a specific nucleotide site in individual
of mutations in this organ and to compare different models of
the human mutation process.
Individuals born with Apert syndrome (Online Mendelian
Inheritance in Man no. 101200) exhibit prematurely fused
cranial sutures and fused fingers and toes. Most Apert syndrome
2 gene (FGFR2) mutation in a normal father’s germ line that is
transmitted to his offspring. The birth frequency for sporadic
cases is between 10?5and 10?6(7, 8). Surprisingly, ?98% of
sites (C755G and C758G), and a single copy of either mutation
is sufficient to cause the disease. The birth frequency of indi-
viduals with new mutations at either of these two nucleotide sites
suggests that the mutation frequency at either site is 100- to
1,000-fold greater than expected based on what is known about
transversion mutations since humans and chimpanzees last had
a common ancestor (9) and mutation data at many human
disease loci (10). Our previous study (6) allowed us to reject the
he frequency at which human germ-line nucleotide substi-
tutions arise in each generation can be measured by analysis
mutation hot-spot model (the nucleotide has a higher-than-
sites (C755G). Positive selection for spermatogonial cells car-
rying a newly arisen mutation is an alternative explanation (4–6,
11, 12) for an apparently high nucleotide substitution frequency.
In our earlier study (6), we were unable to reject the selection
hypothesis as an explanation for the C755G data.
In this work, we studied the other common de novo Apert
nucleotide substitution (C758G). We examined four individual
testes from three older individuals and two testes from younger
donors. By comparing the data on both the C755G and C758G
mutations, we were able to ask whether the two mutations were
dependent or independent with respect to their anatomical
distribution throughout each testis. Overall, our results support
the selection hypothesis as an explanation for the high frequency
of both Apert syndrome mutations for being responsible for the
increased chance of older fathers having an affected child
(paternal age effect: refs. 1, 11, 13–15).
C758G Mutation Distribution in Individual Testes. We studied the
spatial distribution of de novo C758G mutations within six testes
testes were from older donors. Two of these testes were from one
donor (374, age 62) and one testis from each of two additional
donors (854, 54 years; 59089, 45 years). We also examined two
testes from younger individuals: one testis each from donors
60832, 23 years, and 59056, 19 years.
Each testis was dissected into 192 pieces (six slices each with
32 pieces), the DNA extracted and a highly specific single-
molecule PCR assay was used to estimate the number of mutant
molecules in each DNA aliquot (see Methods). A sperm sample
taken from the epididymis of each testis was also analyzed for
C758G mutations. The summary data are shown in Table 1 and
Fig. 1 [the total DNA content and mutation frequency of each
piece for all testes are presented in supporting information (SI)
Table S1]. The false-positive frequency of the C758G assay was
3.8 ? 10?7(based on experiments using 31.6 million control
Among the three older individuals (374, 854, and 59089), the
average mutation frequency per testis ranged from 7.3 ? 10?4to
1.0 ? 10?4(Table 1). Each of the four testes is characterized by
Author contributions: P.C. and N.A. designed research; S.-K.C., S.-R.Y., and P.C. performed
research; S.-K.C., S.-R.Y., P.C., and N.A. analyzed data; and P.C. and N.A. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
*P.C. and N.A. contributed equally to this work.
†To whom correspondence may be addressed. E-mail: email@example.com or petercal@
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2008 by The National Academy of Sciences of the USA
July 22, 2008 ?
vol. 105 ?
no. 29 ?
a very small number of ‘‘hot’’ pieces with mutation frequencies
3–4 orders of magnitude greater than most of the remaining
pieces (Fig. 1). In every case, 95% of the mutant molecules were
found in no more than 12 (average 6.5) of the 192 pieces
examined, whereas, according to a random spatial distribution,
many more pieces, which together contain 95% of the genomes,
would have been expected. In many cases, several pieces with
high mutation frequencies appear to form foci adjacent to one
another in the same slice or even between slices (Fig. 1).
For the two younger donors, the average C758G mutation
frequencies were 2–3 orders of magnitude lower than for the
with such low frequencies, there is not much variation in the
mutation frequencies among the different testis pieces (Fig. 1).
C755G Mutation Distribution in Individual Testes. We next analyzed
the frequency of de novo C755G Apert syndrome germ-line
mutations in testes 59089-1, 59056-1, and 60832-1 with a similar
site-specific assay used to examine the C755G mutation in the
testes of donors 374 and 854 (6). The new results are shown in
Table 1 and Fig. 1 along with the previously obtained data. The
average C755G mutation frequency for testis 59089-1 from the
45-year-old donor (1.6 ? 10?4) is very similar to the data
observed for the other older donors [374 and 854 (6)]. Likewise,
this testis features a small number of pieces with very high
mutation frequencies among a general background of lower
mutation frequencies (Fig. 1).
The average testis mutation frequency at the C755G nucleo-
tide site in the 23-year-old donor (60832-1) is 9.4 ? 10?7, 2–3
orders of magnitude lower than for the older donors. For the
19-year-old donor (59056-1), the frequency is 1.6 ? 10?5. This
frequency is lower than for any of the older donors; however, it
is not orders of magnitude lower. Indeed, it is only a factor of 4
lower than for testis 374-2 (62 years). Further, the testis piece
with the highest frequency in the 19 year old (0.004) has a lower
frequency than the comparable piece in any of the older
individuals; however, it is only a factor of 1.75 times lower than
in testis 374-2. For the two younger donors and the two mutation
sites, testis 59056-1 and mutation C755G is the only testis–
mutation combination to have a piece with a mutation frequency
great enough to be colored differently than gray in Fig. 1.
Comparison of Testis to Epididymal Sperm Mutation Frequencies. For
7 of the 12 testis–disease mutation pairs, the ratio of the
mutation frequency in the testis to the mutation frequency in the
sperm (found in that testis’s epididymis) is near one, ranging
from 0.29 to 1.76 (see Table 1). For two more pairs, this ratio is
higher at 5.33 and 6.18. In ref. 6, we reported that the relatively
high ratio of 6.18 for the C755G mutation in 854-2 could result
from a highly viscous material found only in the one epididymis,
possibly contaminating that sperm sample with nongerm-line
DNA; this explanation is now unlikely, because for the same
testis and epididymal sperm sample, the ratio for mutation
C758G is less than one (0.44). For testis 374-2 and muta-
tion C758G, the ratio is much greater at 32.3. This testis–
found in a single testis piece (Table 1, Fig. 1). It seems possible
that the sperm in the epididymis results from a less-than-perfect
sampling of the testis, perhaps because of pathological effects
associated with aging (ref. 16 and refs. therein), leading to local
tubule blockage. An underrepresentation of this one testis piece
could cause the observed high ratio. Although we could detect
sperm from the epididymal washes of testis 60832-1 microscop-
ically, the amount of sperm DNA obtained was insufficient to
make any mutation frequency estimates. We cannot account for
this observation other than it being a reflection of some aspect
of this donor’s epididymal or testicular function or sperm-
Ratio of C755G to C758G Mutations. Data on the relative frequency
of the two common types of Apert syndrome mutations found
among affected individuals indicate that about twice as many
carry the C755G as the C758G mutation (262 C755G mutations,
128 C758G mutations; reviewed in ref. 17). In our studies,
pooling data from the four testes from the older individuals,
there were 5,871,596 mutations at the C755G site, whereas
5,843,766 were found at the C758G site. The exact binomial test
strongly rejects (P value ?2.2 ? 10?16) that these data conform
to the 2:1 ratio reported for affected individuals. Alterations
from the expected ratio could simply result from the small
data supporting the 2:1 ratio come from children whose normal
fathers, on average, are considerably younger than the older
testes donors we studied. Although far fewer mutations were
counted (see Table S1), the exact binomial test also rejects the
2:1 ratio (P value ?2.2 ? 10?16) for the two younger donors.
Positional Independence of C758G and C755G Mutation Occurrences.
Having data on both the C758G and the C755G mutation
frequencies for each individual testis allows us to examine
Table 1. Summary data on six testes
No. pieces containing
95% of mutants*
Percentage of all
in those pieces
5.6 ? 10?4
3.1 ? 10?6
2.7 ? 10?4
8.2 ? 10?5
2.0 ? 10?6
7.3 ? 10?4
1.0 ? 10?4
1.2 ? 10?4
1.1 ? 10?4
6.5 ? 10?7
5.8 ? 10?7
?4 ? 10?6–0.043
?4 ? 10?6–0.039
?4 ? 10?6–0.011
?4 ? 10?6–0.005
?4 ? 10?6–9.0 ? 10?6
?4 ? 10?6–3.1 ? 10?5
4.5 ? 10?4
3.9 ? 10?5
1.1 ? 10?4
9.1 ? 10?5
3.0 ? 10?6
3.8 ? 10?4
6.7 ? 10?5
6.8 ? 10?4
1.6 ? 10?4
9.4 ? 10?7
1.6 ? 10?5
?2 ? 10?6–0.047
?4 ? 10?6–0.008
?4 ? 10?6–9 ? 10?6
?4 ? 10?6–0.004
*Based on a resolution of 192 pieces/testis.
†Previously published data (6).
‡Not enough sperm could be collected.
www.pnas.org?cgi?doi?10.1073?pnas.0801267105Choi et al.
whether a testis piece, which has a high mutation frequency for
the one mutation, will also have a high mutation frequency for
frequencies ?0.001 (the colors representing the four highest-
six testes, there are no pieces with high frequencies for both
mutations. For testis 374-1, there is one piece with a frequency
above the threshold for both mutations. Testis 374-1 has more
pieces with high frequencies than any of the other testes, so to
test whether this one overlapping piece is significant, we per-
formed a permutation test (SI Text). This test determined that
the one overlapping piece is not significant, because, for a
random model, 43% would have one or more overlapping pieces.
We conclude there is no correlation between the high-frequency
pieces for the two mutations and that, not unexpectedly, the two
different mutation events arise independently of one another.
Testing the Mutation Hot-Spot Model. One possible explanation for
the incidence of Apert syndrome being so much higher than
expected is that the C755G and C758G sites might have a
higher-than-average chance of undergoing a base substitution.
To test this possibility in our previous article on the C755G
mutation (6), we proposed a model for mutation based on what
is known about human germ-line development and maturation
(1, 15, 18–25) (also see ref. 6 for more references). A key
parameter in this model is the mutation rate per cell division.
The mutation hot-spot model is simply that these two disease
sites have much higher-than-average nucleotide substitution
rates per cell division. A complete description of the model is in
the previous article (6); below, we outline the main points.
The model has two phases. The first phase, called the growth
phase, the male germ-line cells divide symmetrically, and the
number of such cells increases exponentially. Because of this
exponential increase, an early mutation will be shared by many
later germ-line cells. This phenomenon is similar to the Luria
and Delbruck ‘‘mutation jackpot’’ in bacteria (24). The primor-
dial germ cells migrate to the site of gonad formation and form
the seminiferous cords early in embryogenesis (see refs. 24 and
26). Thereafter, the germ cells are expected to remain physically
close to their ancestors, so that any early mutation will result in
a region of the testis with a high mutation frequency. In the
previous article (6), we considered a range of growth-phase cell
the same. In this article, we have fixed the growth phase
generations at 30.
The germ-line cells of the growth phase eventually form the
adult self-renewing Ap spermatogonia (SrAp). The second
phase, called the adult phase, models the testis from puberty to
death. In this phase, the SrAp divide asymmetrically to produce
a daughter SrAp (self-renewal) and another daughter cell whose
During the adult phase, the number of SrAp are assumed to
remain constant (see Discussion). Any new mutation in this
phase produces only one mutant SrAp lineage (and the descen-
dent mutant meiotic and postmeiotic cells, including sperm),
unlike the clusters produced in the growth phase. Thus, muta-
tions arising in the adult phase will be distributed uniformly
throughout the testis. An experimental estimate (20) indicates
that, in the adult human male, SrAp cells divide every 16 days.
In the model, this division rate persists from age 13 until death
(see Discussion). Based on a testis donor’s age at death, we
estimate the number of adult-phase generations.
To test the mutation hot-spot hypothesis, we wrote a computer
program to simulate the model. Both the testing procedure and
with one subtle change (see SI Text). We performed a goodness-
of-fit test; the statistic we considered was the fraction of the
genomes present in the minimum number of testis pieces, which
contain 95% of the mutant cells in the testis. For the older
donors, this statistic was near 5% (Table 1). In contrast, for most
of the simulations, this statistic was near 95%. Based on these
simulations (see Methods), we can strongly reject (P value
?10?6) the hot-spot model (Table 2). For the one younger
donor–mutation combination (testis 59056-1, mutation C755G)
for which there was a relatively high mutation frequency, the
hot-spot model was also rejected by the goodness-of-fit test (P
value ?10?6). For the three other younger donor–mutation
combinations, which had much lower mutation frequencies, the
hot-spot model was not rejected after a Bonferroni correction
for multiple tests (see Table 2). Even for these data, the statistic
testes are depicted twice, once in the C758G column and once in the C755G
In every case, the orientation of each testis relative to the head and tail
portions of the epididymis is the same: the head portion is on the left (slice 1),
the tail on the right (slice 6); the long epididymal axis runs along the upper
surface. The color code shows the number of mutant molecules per million
genomes. In each column, the testes are arranged by age from oldest (Upper)
to youngest (Lower). The data on C755G from donors 374-1, 374-2, and 854-2
have been published (6).
Distribution of mutants in dissected human testes. Each of the six
Choi et al.
July 22, 2008 ?
vol. 105 ?
no. 29 ?
was substantially ?5%, but this was not due to ‘‘hot spots,’’
most testis pieces had zero observed mutants (Table S1).
Germ-Line Selection. In our previous article (6), we modified the
model to incorporate selection. It is known from model organ-
isms that stem cells can switch from an asymmetric to a
symmetric division pattern and back again, and that such be-
havior can depend on factors that are intrinsic and extrinsic to
the stem cells (27, 28). The form of selection that we introduced
is that in the adult phase, mutated SrAp cells occasionally divide
symmetrically (in the hot-spot model, all adult phase divisions
are asymmetric; in the selection model, nonmutated SrAp still
always divide asymmetrically). Because these new SrAp cells are
expected to remain near their progenitors, these rare symmetric
divisions enable mutation clusters to form and grow locally with
time increase the overall mutation frequency in the testis. Our
selection model is based on the model described above, but we
add a selection parameter p: at each adult phase generation, a
mutated SrAp divides symmetrically with probability p and
divides asymmetrically with probability 1?p (after a symmetric
division, each daughter SrAp reverts to asymmetric divisions
until the next rare symmetric division). A similar model was
independently proposed by Crow (11).
Unlike the hot-spot model, the selection model qualitatively
matches the distribution of mutation frequencies in the data. In
simulations, foci of high mutation frequency emerge, and these
foci often intersect several adjacent testis pieces. To quantita-
tively test the selection model, we used the ?2test to compare
simulations to the actual testis data (see Methods). After apply-
ing the Bonferroni correction for multiple tests, we could not
reject the selection model for any of the testis–mutation com-
binations. Regarding the selection parameter, for the older
donors, the inferred p is ?0.01 (on average, 1 of every 100
divisions is symmetric; Table 2). For the 23-year-old donor
(60832-1), this parameter is zero, implying no selection. This is
consistent with the very low testis mutation frequency that is
individual. For the 19-year-old donor (59056-1), this selection
parameter is greater (see Discussion).
We examined the spatial distribution of spontaneous occur-
rences of the second most-common Apert syndrome mutation
(C758G) in pieces dissected from normal human testes. In older
individuals, we found some foci where the mutation frequency
was 3-4 orders of magnitude greater than most of the other
pieces. In conjunction with our modeling efforts, these data
allow us to reject the hot-spot model. As was the case for the
C755G mutations in the same gene (this work and ref. 6), the
observed high mutation frequency is not simply due to an
exceptionally high C to G transversion mutation rate per cell
The key insight to the hot-pot model is that for a 50- to
60-year-old man, the ratio of expected mutations in the adult
phase to the growth phase is ?500:1. (For a 19–23 year old, this
ratio is closer to 100:1.) In simulations with the mutation rates
per cell division set to match the observed mutation frequencies,
almost all of the mutations occur in the adult phase. Thus, these
simulated mutations are not clustered but are scattered uni-
formly throughout the testis; in simulations, the ‘‘mutation
jackpots’’ rarely occur. Two further observations argue against
the mutation hot-spot hypothesis, irrespective of modeling de-
tails. First, in our previous article (6), we carried out the same
experiment for a C to G transversion mutation at a neutral CpG
site (on chromosome 7). We found these mutations were uni-
formly spread throughout the testis and, unlike the Apert
syndrome disease sites, we could not reject our nonselection
model (p ? 0) for this neutral mutation case (as expected, the
overall mutation frequency was much lower than for the Apert
sites in that testis). Second, the younger testis donors have
substantially lower mutation frequencies than the older donors,
and they do not have significant mutation foci (except for C758G
in 59056-1). Apparently, the disease sites are different from the
sites of neutral mutation, and their mutation clusters are not
jackpots in the classical sense but grow in the adult phase.
An alternative to the mutation hot-spot hypothesis suggests
that high mutation frequencies can result if premeiotic diploid
cells experience a mutation that confers a selective advantage
over wild-type premeiotic cells. First, it was noted (5) that this
selection hypothesis might explain why ?99-fold more sporadic
Apert syndrome mutations (and the common achondroplasia
mutation; see below) arise in males than females (29, 30),
whereas neutral germ-line mutations (at many different sites)
show only an ?5-fold male bias (31). Second (12), rare Apert
syndrome patients born with two FGFR2 mutations were argued
to be much more likely to result from selection than from two
independent mutation events in the same germ-line cell. Third,
when semen from normal men heterozygous for a single-
nucleotide polymorphism tightly linked to the mutation site were
Table 2. Model parameters and goodness-of-fit P values
Hot-spot model (p ? 0)Selection model (p ? 0)
TestisAge Optimal ?, 95% C.I.GOF P value Optimal ?, 95% C.I.Optimal p, 95% C.I.
6.4 ? 10?7, (6.0–6.7) ? 10?7
9.0 ? 10?8, (8.0–10) ? 10?8
1.2 ? 10?7, (1.1–1.3) ? 10?7
1.5 ? 10?7, (1.4–1.6) ? 10?7
2.9 ? 10?9, (1.6–4.3) ? 10?9
4.1 ? 10?9, (1.2–6.4) ? 10?9
5.2 ? 10?11, (1.2–13) ? 10?11
5.6 ? 10?12, (1.1–10) ? 10?12
2.8 ? 10?11, (0.8–5.2) ? 10?11
1.4 ? 10?11, (0.6–6.6) ? 10?11
2.7 ? 10?9, (0.3–4.6) ? 10?9
4.3 ? 10?10, (1.3–40) ? 10?10
3.3 ? 10?7, (3.1–3.5) ? 10?7
5.9 ? 10?8, (5.6–6.2) ? 10?8
7.2 ? 10?7, (6.6–7.5) ? 10?7
2.1 ? 10?7, (2.0–2.2) ? 10?7
4.0 ? 10?9, (2.0–5.9) ? 10?9
1.1 ? 10?7, (0.9–1.2) ? 10?7
2.8 ? 10?11, (1.0–8.4) ? 10?11
1.2 ? 10?11, (0.4–3.2) ? 10?11
2.0 ? 10?11, (0.4–6.0) ? 10?11
4.8 ? 10?11, (1.4–11) ? 10?11
4.0 ? 10?9, (0.2–6.6) ? 10?9
3.6 ? 10?10, (2.3–7.0) ? 10?10
? is the mutation rate per cell division, which is less than the mutation frequency (see ref. 6). To correct the P values for multiple tests (Bonferroni), multiply
the number of tests performed by 24. For example, 0.004 becomes 0.096 and is not significant; in fact, the only uncorrected P values that remain significant
after correction are those ?10?6. p, probability of a symmetric division in the adult phase; GOF, goodness of fit.
www.pnas.org?cgi?doi?10.1073?pnas.0801267105Choi et al.
studied for the presence of C755G mutations, the results showed
what appeared to be a nonrandom distribution of mutations
among the two homologous chromosomes (ref. 4; see, however,
ref. 6). A truly nonrandom distribution could be explained if
positive selection acted on rare progenitor spermatogonia car-
rying a newly arisen C755G mutation. Finally, analysis of C755G
this site was not a mutation hot spot, but a selection model fit the
data (6). We now have shown that the same is true for an
independent mutation (C758G) that causes the same disease.
Modifying our model of germ-line development by incorpo-
rating a simple selection scheme led to predictions on mutation
frequency and the distribution of mutations in the testis, con-
sistent with our data. We model selection by proposing that
mutant SrAp cells occasionally divide symmetrically; the in-
ferred rate is ?1 every 100 divisions on average or around once
every 4 years. These divisions cause mutation clusters to grow
with time (as microscopic ‘‘tumors’’), thereby increasing the
overall testis mutation frequency. This growth could also explain
the paternal age effect (the observed exponential increase in
disease incidence with the age of the father), by a model (4, 6,
11) that differs from the classical explanation (1, 13–15). For the
older donors we studied, the inferred mutation rate per cell
division for the selection model would produce mutation fre-
quencies similar to the existing data on neutral transversion
mutations (9, 10), if the selection parameter p were set to zero
(so no selection, all adult phase divisions asymmetric). This
result suggests that the C755G and C758G mutations arise at
approximately the frequency expected for neutral transversion
For the 19-year-old donor (59056) and mutation C755G, the
inferred p parameter value is unrealistically high (0.055). If
age of the older donors, most of the testis would be mutated (6).
For modeling purposes, the difficulty is that, although the
mutation frequency for this testis is lower than for the older
testes, the observed mutation frequency is still greater than we
would expect if we were to insert the parameters inferred from
the older donors into the selection model. There is evidence that
there are some SrAp asymmetric divisions before age 13 (24),
and that the 16-day division rate we have assumed is constant
from age 13 until death slows with age (32). If one or both of
these modifications is included in a more complicated selection
model, there would be relatively more divisions early in an
individual’s life, and then the model could match the data on the
19 year old with a lower and therefore more realistic selection
We have also considered other modifications to the model. In
our previous article (6), we considered replication-independent
mutations, but it turned out this addition did not change our
conclusions. A new variant, which in principle could match the
data and for which the effect of the disease mutations is perhaps
less extreme, is to allow all adult SrAp to alternate between
symmetric divisions, asymmetric divisions, and sometimes not
dividing at all; in this model, the mutant SrAp would divide
symmetrically more often than the nonmutants. We have also
considered cell death; if all cells are equally likely to die, this
affects the parameter values of the models but not the conclu-
sions; if the mutant SrAp are less likely to die, this process would
be another form of selection, but because the total number of
SrAp are estimated to decrease by only 25% from age 35 to 65
(33), this reduction is not great enough to produce the observed
mutation clusters. Likewise, if the mutant SrAp cells were to
have a higher asymmetric division rate than the nonmutant cells,
this would be yet another form of selection, but the change in
division rate necessary to produce the intensity of the observed
mutation clusters is unrealistically high.
Is there any evidence for positive selection of mutant premei-
otic spermatogonia at other loci? Virtually all new achondro-
plasia (the most common form of dwarfism) mutations arise at
(G1138A) in the FGFR3 gene, suggesting that the high fre-
quency may also be explained by selection (5, 11). A recent study
on G1138A mutations in sperm from semen also included some
testis biopsies. In two individuals ?80 years old, the frequencies
in two independent biopsies from the same testis were not
concordant (34), just as our selection model would predict.
Some insight into how the FGFR2 and FGFR3 mutations
contribute to a selective advantage comes from noting that both
gene products are receptor tyrosine kinases and can influence
downstream members of the signal transduction pathway (for a
may be that some mutations in FGFR2 and FGFR3 (although
usually not the specific mutations that cause Apert syndrome or
achondroplasia) have been associated with certain cancers (37–
39). No published data yet exist on the specific molecular
processes that these mutations might influence in providing a
selective advantage to human spermatogonia carrying a new
Considering all of the published work (4–6, 11, 12) and our
present results, we suggest that positive selection can be a driving
force acting to increase the germ-line mutation frequency in
humans above that at which spontaneous nucleotide substitu-
tions arise. Germ-line selection of this kind has a root in
theoretical work by Hastings published almost 20 years ago
(40–42) on recessive deleterious mutations segregating in ani-
mal populations. Experimental literature on germ-line selection
in premeiotic diploid cells in animals is very sparse (43).
Interestingly, Hastings also realized that alleles conferring a
selective advantage in the germ line may be disadvantageous in
the adult and might lead to ‘‘mitotic drive’’ systems that increase
the mutational load of a population. Both Apert syndrome and
achondroplasia may be examples of such a system and others,
including those of medical interest, may also exist (see refs. 11,
13–15, 31). The testis dissection method can be useful in
examining this hypothesis at any locus in a species.
Source of Testes. Testes were obtained from the National Disease Research
University of Southern California. All samples were frozen within 12 h after
Testis and Epididymis Processing. Sperm cells were collected from the epidid-
ymal tail and adjacent vas deferens. For testis dissection (Fig. S1), the epidid-
ymis is removed and the testis is cut into six approximately equal-size slices at
right angles to the testis’ long axis. Each slice is divided into 32 approximately
and run along the long axis on the dorsal (epididymal) surface, etc. Details of
the dissections, DNA isolation, and quantitation are found in ref. 6. Variation
in the number of genomes per piece (see Table S1) results from variation in
slice and piece size because of testis shape.
C758G Mutation Frequency Assay. We modified the pyrophosphorolysis acti-
vated PCR (PAP) protocol (44). Our assay is almost identical to the C755G
mutation assay described in the methods and text S2 of ref. 6. Each C758G
reaction contained 20 mM Hepes (pH 7.35), 30 mM KCl, 50 ?M Na4PPi, 2 mM
MgCl2, 80 ?M of each dNTP; 160 nM of each primer; 2 ?M Rox; 0.2? Syber
green I; 0.04 unit/?l TMA31FS DNA polymerase (Roche); and 25,000 copies of
polymerase may be obtained from Thomas W. Myers (Roche Molecular Sys-
tems). The C758G-specific PAP primers were: 5?-CCCCACTCCTCCTTTCTTC-
CCTCTCTCCACCAGAGCGATGdd 3? and 5?-TTTGCCGGCAGTCCGGCTTGGAG-
GATGGGCCGGTGAGGCCdd3?. Cycling conditions were: initial denaturation 2
min, 94°C and 150 cycles of 6 s, 94°C and 40 s, 74°C. For sample PAP data, see
as described in ref. 6 except for the Hepes pH (7.35), the primer concentration
Choi et al.
July 22, 2008 ?
vol. 105 ?
no. 29 ?
(320 nM), the initial denaturation step (94°C), and the cycling conditions (125 Download full-text
cycles of 6 s at 94°C and 40 s at 77°C).
Mutation Counting Strategy. Ten reactions (250,000 total genomes) were used
to estimate the C758G mutation frequency for every testis piece. If ?5/10
reactions were positive, we took the number (after Poisson correction) as an
estimate of the mutation frequency. If five or more reactions were positive,
then the experiment was repeated by using dilutions until ?10/20 (or in some
cases ?25/40) were positive. Single mutant molecule reactions were distin-
guished from negative ones by the kinetics of fluorescence increase as a
function of cycle number and PCR product melting profile using quantitative
was the average of the frequencies of the pieces weighted by the number of
genomes in those pieces (see Table S1).
For each experiment, 20 negative controls each contained 25,000 human
blood genomes (Clontech). Twenty positive controls each had added an
Mimi Jabs, Mount Sinai School of Medicine, New York).
Quantitative Modeling and Testing. The details of the quantitative model, the
previous article (6), with one subtle difference (see SI Text). The hot-spot model
has one free parameter: the mutation rate per cell division (the number of
for each of the two mutations, we found the value of the mutation rate per cell
division, which maximizes the likelihood of the observed mutation frequency
(Table 2). We then performed a goodness-of-fit test; the statistic we used is the
testis. We simulated the model many times with the inferred optimal mutation
rates, so that for each testis and each mutation, there would be one million
data, and we could then compare the distribution of the statistic in the simula-
tions to the statistic for the data.
The selection model has two parameters. We infer the selection parameter
p and the mutation rate per cell division ? by fitting both the overall testis
mutation frequency and the minimum number of testis pieces, which contain
mutation, we quantitatively tested the selection model by simulating the
the data corresponding to the older donor’s testes at both mutations and
frequencies in the different color categories in Fig. 1; for both mutations for
testis 60832-1 and for the C758G mutation for testis 59056-1, because the
?20 ? 10?6. Separately for each testis and mutation, we then performed a ?2
test comparing these counts for the simulation and the data.
donors. We thank David Gelfand (formerly at Roche Molecular Systems) and
Thomas Myers for supplying Tma31FS and Tina Hu for other assistance. This
work was supported in part by grants from the National Institute of General
Medical Sciences (N.A. and P.C.) and the Ellison Medical Research Foundation
1. Vogel F, Motulsky AG (1997) Human Genetics: Problems and Approaches (Springer, Berlin).
2. Haldane JBS (1935) The rate of spontaneous mutation of a human gene. J Genet
3. Glaser RL, et al. (2003) The paternal-age effect in Apert syndrome is due, in part, to the
increased frequency of mutations in sperm. Am J Hum Genet 73:939–947.
4. Goriely A, McVean GA, Rojmyr M, Ingemarsson B, Wilkie AO (2003) Evidence for
selective advantage of pathogenic FGFR2 mutations in the male germ line. Science
the achondroplasia paternal age effect. Proc Natl Acad Sci USA 99:14952–14957.
6. Qin J, et al. (2007) The molecular anatomy of spontaneous germ line mutations in
human testes. PLoS Biol 5:e224.
7. Tolarova MM, Harris JA, Ordway DE, Vargervik K (1997) Birth prevalence, mutation
rate, sex ratio, parents’ age, and ethnicity in Apert syndrome. Am J Med Genet
8. Cohen MM, et al. (1992) Birth prevalence study of the Apert syndrome. Am J Med
9. Nachman MW, Crowell SL (2000) Estimate of the mutation rate per nucleotide in
humans. Genetics 156:297–304.
10. Kondrashov AS (2003) Direct estimates of human per nucleotide mutation rates at 20
loci causing Mendelian diseases. Hum Mutat 21:12–27.
complexities. J Radiat Res (Tokyo) 47 Suppl B:B75–B82.
12. Goriely A, et al. (2005) Gain-of-function amino acid substitutions drive positive selection of
FGFR2 mutations in human spermatogonia. Proc Natl Acad Sci USA 102:6051–6056.
13. Risch N, Reich EW, Wishnick MM, McCarthy JG (1987) Spontaneous mutation and
parental age in humans. Am J Hum Genet 41:218–248.
14. Glaser RL, Jabs EW (2004) Dear old dad. Sci Aging Knowledge Environ 2004:re1.
Nat Rev Genet 1:40–47.
17. Muenke M, Wilkie AO (2007) in Craniosynostosis Syndromes: The Online Metabolic
and Molecular Bases of Inherited Disease, www.ommbid.com/OMMBID/the_online_
Fertil Steril 17:705–721.
19. Clermont Y (1963) The cycle of the seminiferous epithelium in man. Am J Anat 112:35–51.
20. Heller CG, Clermont Y (1963) Spermatogenesis in man: An estimate of its duration.
21. Drost JB, Lee WR (1995) Biological basis of germline mutation: comparisons of spon-
taneous germline mutation rates among drosophila, mouse, and human. Environ Mol
time of spermatozoa through the human epididymis. Biol Reprod 39:812–817, 1988.
perspectives. Hum Reprod Update 12:275–282.
24. Nistal M, Paniagua R (1984) Testicular and Epididymal Pathology (Thieme-Stratton, New
25. Zhengwei Y, Wreford NG, Royce P, de Kretser DM, McLachlan RI (1998) Stereological evalu-
ation of human spermatogenesis after suppression by testosterone treatment: heteroge-
neous pattern of spermatogenic impairment. J Clin Endocrinol Metab 83:1284–1291.
26. Muller J, Skakkebaek NE (1992) The prenatal and postnatal development of the testis
Baillieres. Clin Endocrinol Metab 6:251–271.
27. Nagano M, Avarbock MR, Brinster RL (1999) Pattern and kinetics of mouse donor
spermatogonial stem cell colonization in recipient testes. Biol Reprod 60:1429–1436.
28. Morrison SJ, Kimble J (2006) Asymmetric and symmetric stem-cell divisions in devel-
opment and cancer. Nature 441:1068–1074.
29. Wilkin DJ, et al. (1998) Mutations in fibroblast growth-factor receptor 3 in sporadic
Hum Genet 63:711–716.
syndrome and Pfeiffer syndrome. Am J Hum Genet 66:768–777.
32. Kimura M, et al. (2003) Balance of apoptosis and proliferation of germ cells related to
spermatogenesis in aged men. J Androl 24:185–191.
spermatogonial stem cell. J Androl 8:64–68.
34. Dakouane Giudicelli M, et al. (2008) Increased achondroplasia mutation frequency with
advanced age and evidence for G1138A mosaicism in human testis biopsies. Fertil Steril
during embryonic development. Dev Biol 287:390–402.
36. Eswarakumar VP, Lax I, Schlessinger J (2005) Cellular signaling by fibroblast growth
factor receptors. Cytokine Growth Factor Rev 16:139–149.
growth factor receptor 3b. Carcinogenesis 27:740–747.
38. Hansen RM, Goriely A, Wall SA, Roberts IS, Wilkie AO (2005) Fibroblast growth factor
receptor 2, gain-of-function mutations, and tumourigenesis: investigating a potential
link. J Pathol 207:27–31.
39. Easton DF, et al. (2007) Genome-wide association study identifies novel breast cancer
susceptibility loci. Nature 447:1087–1093.
40. Hastings IM (1989) Potential germline competition in animals and its evolutionary
implications. Genetics 123:191–197.
41. Otto SP, Hastings IM (1998) Mutation and selection within the individual. Genetica
42. Hastings IM (1991) Germline selection: population genetic aspects of the sexual/
asexual life cycle. Genetics 129:1167–1176.
melanogaster. Proc Natl Acad Sci USA 98:11341–11346.
44. Liu Q, Sommer SS (2004) Detection of extremely rare alleles by bidirectional pyrophospho-
www.pnas.org?cgi?doi?10.1073?pnas.0801267105Choi et al.