A germ-line-selective advantage rather than an increased mutation rate can explain some unexpectedly common human disease mutations

Abstract

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 rejected. 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 individuals who inherit them have reduced fitness. In addition, we compared 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.

The frequency at which human germ-line nucleotide substitutions arise in each generation can be measured by analysis of sporadic cases of human autosomal dominant or sex-linked diseases (1, 2) or by DNA analysis of sperm (3–5). In a recent article (6), we described an alternative method to study germ-line mutations by dissecting the whole human testis and measuring the mutation frequency at a specific nucleotide site in individual pieces. This procedure allowed us to study the spatial distribution 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 cases result from a spontaneous fibroblast growth factor receptor 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 10⁻⁵ and 10⁻⁶ (7, 8). Surprisingly, >98% of Apert syndrome cases arise by transversion mutations at only two sites (C755G and C758G), and a single copy of either mutation is sufficient to cause the disease. The birth frequency of individuals 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 mutation hot-spot model (the nucleotide has a higher-than-average chance of undergoing a base substitution) at one of these sites (C755G). Positive selection for spermatogonial cells carrying 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).

Results

C758G Mutation Distribution in Individual Testes.

We studied the spatial distribution of de novo C758G mutations within six testes from donors who do not suffer from Apert syndrome. Four of the 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 genomes).

Table 1.

Summary data on six testes

Testis	Age	Epididymal sperm mutation frequency	Testis mutation frequency	Frequency range among testis pieces	No. pieces containing 95% of mutants*	Percentage of all testis genomes in those pieces
Mutation C758G
374–1	62 y	5.6 × 10−4	7.3 × 10−4	<4 × 10−6–0.043	12	8.2
374–2	62 y	3.1 × 10−6	1.0 × 10−4	<4 × 10−6–0.039	1	0.27
854–2	54 y	2.7 × 10−4	1.2 × 10−4	<4 × 10−6–0.011	7	3.6
59089–1	45 y	8.2 × 10−5	1.1 × 10−4	<4 × 10−6–0.005	6	4.5
60832–1	23 y	‡	6.5 × 10−7	<4 × 10−6–9.0 × 10−6	27	15.3
59056–1	19 y	2.0 × 10−6	5.8 × 10−7	<4 × 10−6–3.1 × 10−5	14	9.4
Mutation C755G
374–1†	62 y	4.5 × 10−4	3.8 × 10−4	<10−6–0.027	12	5.7
374–2†	62 y	3.9 × 10−5	6.7 × 10−5	<10−6–0.007	5	2.6
854–2†	54 y	1.1 × 10−4	6.8 × 10−4	<2 × 10−6–0.047	7	5.0
59089–1	45 y	9.1 × 10−5	1.6 × 10−4	<4 × 10−6–0.008	10	4.9
60832–1	23 y	‡	9.4 × 10−7	<4 × 10−6–9 × 10−6	35	20.1
59056–1	19 y	3.0 × 10−6	1.6 × 10−5	<4 × 10−6–0.004	20	17

*Based on a resolution of 192 pieces/testis.

^†Previously published data (6).

^‡Not enough sperm could be collected.

Fig. 1.

Distribution of mutants in dissected human testes. Each of the six testes are depicted twice, once in the C758G column and once in the C755G column. The donor/testis identification number sits between the data for each mutation type. Each testis is shown as six slices, each divided into the 32 pieces. 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).

Among the three older individuals (374, 854, and 59089), the average mutation frequency per testis ranged from 7.3 × 10⁻⁴ to 1.0 × 10⁻⁴ (Table 1). Each of the four testes is characterized by 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 older donors (Table 1). Moreover, perhaps, as would be expected 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 nucleotide 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 mutation C758G, the ratio is much greater at 32.3. This testis–mutation combination is unique in that >95% of the mutants are 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 microscopically, 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-extraction procedure.

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 number of individual testes examined. However, for example, the 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 whether a testis piece, which has a high mutation frequency for the one mutation, will also have a high mutation frequency for the other. We restrict our attention to those pieces with mutation frequencies >0.001 (the colors representing the four highest-frequency categories in Fig. 1) for either mutation. For five of the 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 performed 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, models the testis from zygote formation to puberty. In this 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 primordial 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 generations between 27 and 34. For all values, the conclusion was 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 descendents, after a few additional divisions, will produce sperm. 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 descendent mutant meiotic and postmeiotic cells, including sperm), unlike the clusters produced in the growth phase. Thus, mutations 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 the computer program are the same as in our previous article (6), 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 was substantially <5%, but this was not due to “hot spots,” because there were not any; rather, it was because, for these data, most testis pieces had zero observed mutants (Table S1).

Table 2.

Model parameters and goodness-of-fit P values

Testis	Age	Hot-spot model (p = 0)		Selection model (p > 0)
		Optimal λ, 95% C.I.	GOF P value	Optimal λ, 95% C.I.	Optimal p, 95% C.I.	χ2P value
Mutation C758G
374–1	62 y	6.4 × 10−7, (6.0–6.7) × 10−7	<10−6	5.2 × 10−11, (1.2–13) × 10−11	0.016, (0.010–0.019)	0.13
374–2	62 y	9.0 × 10−8, (8.0–10) × 10−8	<10−6	5.6 × 10−12, (1.1–10) × 10−12	0.012, (0.009–0.019)	0.30
854–2	54 y	1.2 × 10−7, (1.1–1.3) × 10−7	<10−6	2.8 × 10−11, (0.8–5.2) × 10−11	0.011, (0.010–0.021)	0.05
59089–1	45 y	1.5 × 10−7, (1.4–1.6) × 10−7	<10−6	1.4 × 10−11, (0.6–6.6) × 10−11	0.016, (0.013–0.025)	0.62
60832–1	23 y	2.9 × 10−9, (1.6–4.3) × 10−9	0.42	2.7 × 10−9, (0.3–4.6) × 10−9	0, (0–0.018)	0.36
59056–1	19 y	4.1 × 10−9, (1.2–6.4) × 10−9	0.004	4.3 × 10−10, (1.3–40) × 10−10	0.025, (0–0.040)	0.19
Mutation C755G
374–1	62 y	3.3 × 10−7, (3.1–3.5) × 10−7	<10−6	2.8 × 10−11, (1.0–8.4) × 10−11	0.011, (0.009–0.018)	0.14
374–2	62 y	5.9 × 10−8, (5.6–6.2) × 10−8	<10−6	1.2 × 10−11, (0.4–3.2) × 10−11	0.010, (0.008–0.018)	0.88
854–2	54 y	7.2 × 10−7, (6.6–7.5) × 10−7	<10−6	2.0 × 10−11, (0.4–6.0) × 10−11	0.016, (0.012–0.023)	0.93
59089–1	45 y	2.1 × 10−7, (2.0–2.2) × 10−7	<10−6	4.8 × 10−11, (1.4–11) × 10−11	0.015, (0.011–0.028)	0.38
60832–1	23 y	4.0 × 10−9, (2.0–5.9) × 10−9	0.28	4.0 × 10−9, (0.2–6.6) × 10−9	0, (0–0.018)	0.49
59056–1	19 y	1.1 × 10−7, (0.9–1.2) × 10−7	<10−6	3.6 × 10−10, (2.3–7.0) × 10−10	0.055, (0.040–0.070)	0.01

λ 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.

Germ-Line Selection.

In our previous article (6), we modified the model to incorporate selection. It is known from model organisms that stem cells can switch from an asymmetric to a symmetric division pattern and back again, and that such behavior 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 quantitatively test the selection model, we used the χ² test to compare simulations to the actual testis data (see Methods). After applying the Bonferroni correction for multiple tests, we could not reject the selection model for any of the testis–mutation combinations. 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 responsible for the hot-spot model also not being rejected for this individual. For the 19-year-old donor (59056-1), this selection parameter is greater (see Discussion).

Discussion

We examined the spatial distribution of spontaneous occurrences 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 division.

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 uniformly throughout the testis; in simulations, the “mutation jackpots” rarely occur. Two further observations argue against the mutation hot-spot hypothesis, irrespective of modeling details. 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 uniformly 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 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 carrying a newly arisen C755G mutation. Finally, analysis of C755G mutation data from dissected testis of older individuals indicated 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 incorporating a simple selection scheme led to predictions on mutation frequency and the distribution of mutations in the testis, consistent with our data. We model selection by proposing that mutant SrAp cells occasionally divide symmetrically; the inferred 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 frequencies 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 mutations.

For the 19-year-old donor (59056) and mutation C755G, the inferred p parameter value is unrealistically high (0.055). If the selection model is simulated with this p parameter then by the 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 parameter p.

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 conclusions; 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 premeiotic spermatogonia at other loci? Virtually all new achondroplasia (the most common form of dwarfism) mutations arise at a very high frequency in the male germ line at one nucleotide site (G1138A) in the FGFR3 gene, suggesting that the high frequency 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 more detailed discussion, see refs. 35 and 36). Especially relevant 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 mutation.

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 substitutions 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 animal 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.

Methods

Source of Testes.

Testes were obtained from the National Disease Research Interchange (NDRI) with the approval of the Institutional Review Board of the University of Southern California. All samples were frozen within 12 h after death.

Testis and Epididymis Processing.

Sperm cells were collected from the epididymal tail and adjacent vas deferens. For testis dissection (Fig. S1), the epididymis 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 equal-size pieces such that piece 1 from slices 1–6 are adjacent to one another 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 activated 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 μM Na₄PPi, 2 mM MgCl₂, 80 μM of each dNTP; 160 nM of each primer; 2 μM Rox; 0.2× Syber green I; 0.04 unit/μl TMA31FS DNA polymerase (Roche); and 25,000 copies of testis or epididymal sperm DNA molecules. Research samples of Tma31FS DNA polymerase may be obtained from Thomas W. Myers (Roche Molecular Systems). The C758G-specific PAP primers were: 5′-CCCCACTCCTCCTTTCTTCCCTCTCTCCACCAGAGCGATG_dd 3′ and 5′-TTTGCCGGCAGTCCGGCTTGGAGGATGGGCCGGTGAGGCC_dd 3′. 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 figure S1 of ref. 6. The C755G reagents and primers in this study were the same as described in ref. 6 except for the Hepes pH (7.35), the primer concentration (320 nM), the initial denaturation step (94°C), and the cycling conditions (125 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 distinguished from negative ones by the kinetics of fluorescence increase as a function of cycle number and PCR product melting profile using quantitative PCR. The estimate of the total testis mutation frequency (per million genomes) 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 average of 0.5 or 1 genome of Apert C758G or C755G DNA (kindly provided by Mimi Jabs, Mount Sinai School of Medicine, New York).

Quantitative Modeling and Testing.

The details of the quantitative model, the computer code, and the estimation and testing procedures are the same as in our 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 growth-phase generations and adult-phase generations have been set independent of the mutation data, as described in Results). Separately for each testis and 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 minimum number of testis pieces required to contain 95% of the mutants in 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 simulations with overall mutation frequency within 5% of the frequency for the data, and we could then compare the distribution of the statistic in the simulations 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 λ by fitting both the overall testis mutation frequency and the minimum number of testis pieces, which contain 95% of the mutant genomes (Table 2 and ref. 6). Separately for each testis and mutation, we quantitatively tested the selection model by simulating the selection model with the inferred optimal parameters. For the simulations and the data corresponding to the older donor's testes at both mutations and testis 59056-1 at mutation C755G, we counted the number of testis pieces with 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 mutation frequencies were so low, we counted the testis pieces with frequencies in the following categories: [0–5]×10⁻⁶, [5–10] ×10⁻⁶, [10–20] ×10⁻⁶, and >20 × 10⁻⁶. Separately for each testis and mutation, we then performed a χ² test comparing these counts for the simulation and the data.