Haploid selection within a single ejaculate increases offspring fitness

Significance

Diploid organisms produce haploid gametes for sexual reproduction, resulting in a biphasic life cycle. Although selection during the diploid phase is well understood, selection during the haploid gametic stage and its consequences are largely ignored despite its potential importance for fundamental evolutionary processes, including the rate of adaptation and inbreeding depression, as well as for applied research into fertilization technology. A current dogma assumes that in animals selection on the haploid gametic genotype is minimal. We examined the importance of haploid selection in the zebrafish and found strong fitness consequences of selection on sperm phenotype in the resulting offspring. Genomic data support the idea that these effects may well be the consequence of selection on the haploid sperm genotype.

Abstract

An inescapable consequence of sex in eukaryotes is the evolution of a biphasic life cycle with alternating diploid and haploid phases. The occurrence of selection during the haploid phase can have far-reaching consequences for fundamental evolutionary processes including the rate of adaptation, the extent of inbreeding depression, and the load of deleterious mutations, as well as for applied research into fertilization technology. Although haploid selection is well established in plants, current dogma assumes that in animals, intact fertile sperm within a single ejaculate are equivalent at siring viable offspring. Using the zebrafish Danio rerio, we show that selection on phenotypic variation among intact fertile sperm within an ejaculate affects offspring fitness. Longer-lived sperm sired embryos with increased survival and a reduced number of apoptotic cells, and adult male offspring exhibited higher fitness. The effect on embryo viability was carried over into the second generation without further selection and was equally strong in both sexes. Sperm pools selected by motile phenotypes differed genetically at numerous sites throughout the genome. Our findings clearly link within-ejaculate variation in sperm phenotype to offspring fitness and sperm genotype in a vertebrate and have major implications for adaptive evolution.

Sperm within an ejaculate exhibit remarkable phenotypic variation (1), but little is known about the causes and consequences of such variation and selection among sperm produced by one male [hereafter referred to as “sib sperm” (2, 3)]. The key reason for this lack of knowledge is the current assumption that performance of sperm produced by a male is under diploid control (4–6), a notion that is further supported by the apparent lack of association between the phenotypic variation among sib sperm and their genetic content (7, 8). Nevertheless, some empirical evidence shows that genes may be expressed at the haploid stages of spermatogenesis and that the transcripts of these genes are not always perfectly shared through cytoplasmic bridges among haploid spermatids (9, 10). Furthermore, the lack of perfect symmetry in sharing of transcripts among haploid cells suggests that phenotypic variation within an ejaculate may have a genetic or epigenetic basis and hence be under selection (11, 12).

Theory predicts that genetic/epigenetic variation among sib sperm may lead to competition between different sperm phenotypes for the fertilization of eggs and may translate into differential fitness effects in the offspring (3). In fact, two recent studies suggested a possible link between sperm phenotype and offspring phenotype: In a broadcast spawning ascidian, Styela plicata, longer-lived sperm sired offspring with higher early-life survival (13), and in the Atlantic salmon, Salmo salar, sperm with intermediate longevity sired faster-hatching offspring (14). However, no published study to date has separated sperm aging from the underlying genetic or epigenetic variation among sib sperm or has provided insights into the long-term fitness effects of variation in sperm phenotypes within a single ejaculate. Therefore our current understanding of the importance of selection at the gametic stage for Darwinian fitness continues to be incomplete.

Results and Discussion

Here we demonstrate that different cohorts of sperm phenotypes and genotypes, which exhibit varying levels of longevity and differentially affect offspring fitness, coexist within the ejaculate of a single male. We used the externally fertilizing zebrafish Danio rerio for a series of experiments using in vitro fertilizations (IVF) in which we selected on sperm phenotypes based on their longevity. Zebrafish gametes activate upon contact with water, and IVF allows precise control over the activation and fertilization of gametes as well as gamete numbers. Selection on sperm longevity was performed by experimentally manipulating the timing between sperm activation and fertilization. We divided the ejaculate of a male and the eggs of a female into two cohorts each and exposed each sperm cohort to one of two treatments. Sperm were activated with water; then, in the “short activation time” (SAT) treatment, one of the sperm cohorts was immediately added to one of the egg cohorts. In the “long activation time” (LAT) treatment, the sperm in the other cohort were held until about 50% were no longer motile, and then the sperm were added to the second egg cohort. Thus, the LAT treatment directly selected against short-lived sperm. So that the fertilization opportunity was equal in the two treatments, we doubled the amount of sperm present in the LAT treatment to compensate for the nonmotile sperm (SI Materials and Methods for more details). To avoid any effect of egg aging, eggs were used within 1 min after collection, and previously activated sperm from both treatments were added to each of the two egg clutches at the same moment.

Our first aim was to describe any association between variation in sperm longevity and offspring fitness and to estimate its importance (experiment 1 A and B). Using the split design described above, we performed IVF and measured fitness traits of the resulting offspring from early development to adulthood. In experiment 1A, we measured sperm longevity for every male and calculated the time until 50% of sperm were no longer motile. We evaluated the effect of sperm selection on early offspring survival in 57 families and found that offspring sired by LAT sperm exhibited a 7% increase in survival compared with offspring sired by SAT sperm (treatment: χ²₁ = 15.93; P < 0.0001, time: χ²₁ = 6.24, P = 0.012) (Fig. 1A). Moreover, when measuring sperm swimming velocity in one to three sons from each of 35 families (n = 108 sons), we observed that sons sired by LAT sperm produced significantly faster-swimming sperm than their brothers sired by SAT sperm (about 5 μm/s faster in LAT males at 10 s postactivation) [curvilinear velocity (VCL): treatment: χ²₁ = 14.55, P = 0.00013; time: χ²₁ =189.71, P < 0.0001; time²: χ²₁ = 145.89, P < 0.0001; treatment × time: χ²₁ = 29.66, P < 0.0001; treatment × time²: χ²₁ = 11.50, P = 0.0007] (Fig. 1C).

Fig. 1.

Effect of sperm selection on offspring survival and sperm traits. Selection on sperm longevity results in an increase in offspring survival (A and B) and in increased sperm swimming VCL (C and D) and sperm density (E) in the resulting male offspring. Dark-blue circles and solid lines represent SAT; violet-red triangles and dotted lines represent LAT. Mean values ± SEs are shown.

We repeated this experiment with a different selection protocol to verify the robustness of our results and to perform further fitness assays on offspring. Because the variation in sperm longevity among males observed in experiment 1A was relatively small, we simplified our protocol to standardize the time postactivation in the LAT treatment to 25 s; this change resulted in a reduction of motile sperm to ∼50% in the LAT treatment (experiment 1B). Using a total of 39 families, we found that offspring from the LAT treatment exhibited a 5% increase in viability at 24 h postfertilization (hpf) (treatment: χ²₁ = 30.86, P < 0.0001) (Fig. 1B), supporting our original findings. We measured sperm from three to five sons from each of 34 families in the two treatments (n = 264 sons). Sons resulting from the LAT treatment produced ejaculates that again exhibited faster-swimming sperm (on average about 3 μm faster than in males resulting from the SAT treatment) (treatment: χ²₁ = 3.93, P = 0.047; time: χ²₁ = 370.79, P < 0.0001; time²: χ²₁ = 678.18, P < 0.0001; time³: χ²₁ = 456.38, P < 0.0001) (Fig. 1D) and a 20% higher sperm density (χ²₁ = 728.8, P < 0.0001) (Fig. 1E) than their SAT brothers.

To understand further how embryo development may be affected by sperm selection during the first 24 hpf, we examined the occurrence of apoptotic cells (15), which are a potential indicator of embryonic fitness (16, 17), in embryos at the age of 8 hpf (experiment 4) (Fig. 2 A and B). We used a total of six pairs to perform split IVFs as described for experiment 1. More eggs exhibited apoptotic cells when fertilized by SAT sperm than when fertilized by LAT sperm (treatment: χ²₁ = 6.56, P = 0.010) (Fig. 2C). This difference may reflect either a general increase in apoptosis in SAT embryos or a shift in the timing of apoptosis events as part of normal development.

Fig. 2.

Differential cell apoptosis at 8 hpf in response to sperm selection. (A and B) Examples of different levels of apoptosis in eggs fertilized by SAT sperm (A) and by LAT sperm (B) resulting from differential numbers of apoptotic cells, which are marked with green fluorescent dye. (C) Significantly more eggs exhibited signs of cell apoptosis when fertilized by SAT sperm (dark-blue circle) than when fertilized by LAT sperm (violet-red triangle). Mean values ± SEs are shown.

We then measured the reproductive success of adult male and female offspring resulting from experiment 1B by assessing number and quality of offspring resulting from natural matings with nonexperimental fish by setting up pairs comprising one experimental fish and one nonexperimental fish of the opposite sex. Here we used a total of 26 families and two to four offspring of either sex in each family in both treatments (n = 202 offspring). We found no difference between LAT and SAT females in fertilization success (χ²₁ = 1.77, P = 0.18) (Fig. 3A) or in the total number of eggs produced (χ²₁ = 2.40, P = 0.12) (Fig. 3B). However, fertilization success was higher in LAT males by about 4% (χ²₁ = 31.87, P < 0.0001) (Fig. 3A), and females mated to LAT males produced about 20% more eggs (fertilized and unfertilized) than females mated to SAT males (χ²₁ = 17.19, P < 0.0001) (Fig. 3B). The difference in fertility in females mated to experimental males is likely induced by behavioral patterns based on sperm numbers available. In zebrafish, pairs spawn in bouts initiated by the male; females release several (5–20) eggs during each bout until the female has run out of eggs or the male stops courting (18). Furthermore, we found a higher survival rate in offspring of LAT females (treatment: χ²₁ = 4.43, P = 0.035) (Fig. 3C) but not males (treatment: χ²₁ = 0.28, P = 0.60) and a higher percentage of normal embryos among the offspring from matings between LAT offspring of both sexes and nonexperimental fish than among matings between SAT offspring of both sexes and nonexperimental fish (treatment: χ²₁ = 129.82, P < 0.0001; sex: χ²₁ = 20.00, P < 0.0001) (Fig. 3D). Our results show that fitness traits were strongly affected by sperm selection not only in the immediate offspring but also in the F2 generation when crossed with nonexperimental fish. The finding of sex differences in fitness effects—that LAT males have a clear fitness advantage, producing more offspring than SAT males, but LAT females do not have this advantage over SAT females—is intriguing and may provide a possible explanation for the maintenance of variation in sperm phenotypes.

Fig. 3.

The effect of sperm selection on reproductive success in male and female offspring. Although there was no difference in fertilization success or in the number of eggs in female offspring (A and B), males from LAT treatments fertilized more eggs (A), and females produced more eggs in matings with such males (B). (C and D) The resulting offspring were more viable in LAT females than in SAT females, whereas the difference between LAT and SAT males was not significant (C), and offspring of both sexes exhibited a higher percentage of normal embryos (D). Mean values ± SEs are shown. In A–C results are shown for males and females separately; empty symbols indicate females, filled symbols indicate males, dark-blue circles indicate SAT progeny, and violet-red triangles indicate LAT progeny; significant differences between treatments are indicated by a connecting bar; *P < 0.0001 (Tables S1–S3 for statistics with both sexes included). In D, results for sexes are combined.

Table S1.

Testing for difference in fertilization success in crosses between LAT and SAT offspring and wild-type fish, including both sexes in one model

Parameter	χ2	df	P
Treatment	0.245	1	0.621
Sex	177.33	1	<0.0001
Treatment × sex	15.63	1	<0.0001

Where the term “sex” was significant, we analyzed the two sexes separately.

Table S2.

Testing for difference in egg number in crosses between LAT and SAT offspring and wild-type fish, including both sexes in one model

Parameter	χ2	df	P
Treatment	0.19	1	0.66
Sex	464.19	1	<0.0001
Treatment × sex	10.81	1	0.001

Where the term “sex” was significant, we analyzed the two sexes separately.

Table S3.

Testing for difference in offspring survival in crosses between LAT and SAT offspring and wild-type fish, including both sexes in one model

Parameter	χ2	df	P
Treatment	3.04	1	0.08
Sex	17.29	1	<0.0001

Where the term “sex” was significant, we analyzed the two sexes separately.

Our LAT treatment resulted in a decrease in fertilization success by about 5% despite doubling the amount of sperm in experiments 1 A and B (Fig. S1; see SI Materials and Methods for details), but this difference in fertilization success is unlikely to cause the fitness differences observed between the two treatments for two reasons. First, decreasing fertilization success and resulting potential selection among eggs for sperm did not have any fitness effects on offspring in a similar setup (14). Second, in our outcrosses between experimental fish and wild-type fish, the fertilization success was about 4% higher in the LAT treatment than in the SAT treatment in males (Fig. 3A), i.e., opposite the pattern found in experiment 1 A and B. However, the survival rate was again higher in the offspring from LAT fish than from SAT fish (Fig. 3C). We therefore conclude that fertilization success has no impact on our results.

Fig. S1.

Fertilization success rates. Fertilization success was slightly lower in LAT than in SAT treatments in experiment 1A (A) and experiment 1B (B).

An immediate question arising from these results is whether the effects observed in experiment 1 A and B are the result of the selection of sperm cohorts differing in phenotype or are the result of sperm aging. Sperm aging, before or after ejaculation, may affect the epigenetic composition of the sperm (19) and/or the quality of sperm DNA by inducing deleterious mutations (20, 21) that may, in turn, affect the development of the resulting zygote. For experiment 2, we investigated the possibility of preejaculation sperm aging by collecting consecutive sperm subsamples, each containing 0.8 μL of ejaculate, until the male had no more sperm; thus the first subsample contained the oldest sperm, and the last sample contained the youngest (SI Materials and Methods for details). This procedure resulted in a maximum of three subsamples per male from 11 different males. We found no evidence that the subsample identity and hence preejaculation sperm age had any differential effect on offspring viability (subset effect: χ²₁ = 0.063, P = 0.97). In many animals, males continuously release unused sperm, apparently to avoid preejaculation sperm aging (22–24), and this continuous release may explain the lack of an effect.

To test for a possible postejaculation sperm-aging effect on offspring performance, we reduced the osmotic stress on sperm by increasing the ratio of Hank’s balanced salt solution (HBSS) to water during sperm activation before fertilization to extend sperm lifespan (experiment 3). We divided the ejaculate of a male and the clutch of eggs of a female (n = 23 pairs) and performed IVF as described above but activated the sperm 25 s or 50 s before fertilization. Offspring viability during the first 24 hpf did not differ between the treatments (25 s: dead = 6.6% ± 0.5 SD; 50 s: dead = 6.3% ± 0.6 SD; treatment: χ²₁ = 1.36, P = 0.24). Thus, postejaculation sperm aging has no impact on offspring performance in this system, and the increase in the viability of offspring sired by LAT sperm is not a result of sperm aging. We conclude that our experimental protocol allows selection on sperm cohorts within an ejaculate that differ in fertilization success and longevity. This conclusion is further supported by the observation that selection for long-lived sperm resulted in increased offspring fitness in every trait that we examined, the opposite of what would be predicted if degradation arising from sperm aging had occurred.

A possible mechanism underlying the observed differences between LAT and SAT treatments is a trade-off between sperm swimming speed and sperm longevity (25). Swimming speed is assumed to play a major role in fertilization success in external fertilizers (26), and if such a trade-off occurred in our system the fast, short-lived sperm could fertilize eggs in our SAT treatment, whereas the slow, longer-lived sperm could fertilize eggs in our LAT treatment. This assumption also would imply that slower, longer-lived sperm sire offspring with increased fitness. However, when looking for such a trade-off within the ejaculates of six males, we found no evidence for any significant association between these two traits when tracking individual sperm over time (Fig. S2; see SI Materials and Methods for details). Therefore an alternative and more likely scenario, independent of swimming speed, is that SAT offspring may be sired by both short-lived and long-lived sperm, whereas LAT offspring are sired only by long-lived sperm. Of course, the possibility that other traits may determine variation in fertilization success among sib sperm needs to be explored carefully.

Fig. S2.

We found no evidence for a trade-off between sperm swimming speed and longevity. We found no significant association between initial sperm swimming speed at 10 s postactivation and velocity at 25 s postactivation (slope = 0.30, t = 1.59, df = 19, P = 0.11) or between initial velocity and longevity (slope = 0.46, t = 1.75, df = 119, P = 0.083). Symbols and colors depict individual sperm belonging to the same male.

An open question is whether within-ejaculate sperm variation is based on genetic mechanisms. To test for a genetic difference between haploid sperm phenotypes, we performed in vitro assays to separate sperm within an ejaculate according to their ability to survive and cover a certain distance throughout their motile phase. We then examined allele frequencies at heterozygous paternal sites throughout the genome, comparing the separated pools in three different males. We placed a sample of the ejaculate of one male in the center of a 280-µL water droplet harbored in a concave microscope slide. The droplet was framed with a concentrated glucose solution to provide a dilution gradient attracting sperm toward the edges of the droplet (27). Upon contact with water, sperm were activated and dispersed within the water droplet, and longer-lived sperm were expected to reach the outer edge of the water droplet more frequently than short-lived sperm. Although this selection regime is not identical to the selection regime in the experiments described above, we know that longer-lived sperm cover longer distances (Fig. S3 and SI Materials and Methods for details), and hence sperm collected from the outer edges of the droplet will show phenotypic overlap with LAT sperm (Fig. S3). Sperm pools collected from the center will contain a mix of all sperm, including some nonmotile sperm, which would sire no offspring in our SAT treatment. The center and outer pools, each containing many thousands of sperm, as well as an untreated sperm pool and a finclip from each of the three males, were subjected to whole-genome sequencing to ∼60× coverage after a PCR-free library preparation to reduce bias in allelic ratios (Table S4). We mapped reads to the D. rerio Zv9 reference assembly, determined heterozygous paternal sites using reads from finclips, and then conducted statistical tests of sperm pool allele frequencies at these paternal sites using 400-kbp half-overlapping windows throughout the genome of each male. We checked for two possible sources of allele frequency bias. First, we checked for allele transmission bias from male to sperm by comparing allele counts in finclip reads and untreated sperm pool reads using allele-frequency likelihood ratio tests (LRTs) (28). Second, we checked for handling bias possibly introduced by the in vitro gradient assay by comparing allele counts in untreated sperm pool reads and reads from the center-selected sperm pool using LRTs (28). We found no systematic evidence of transmission bias (Fig. S4) or handling bias (Fig. S5) for two males; the read pool for unselected sperm from a third male (male 32) had an undetermined technical error and could not be used.

Fig. S3.

Association between sperm longevity and swimming distance. A significant positive correlation exists between sperm longevity and swimming distance (linear term: t = 4.97, df = 118, P < 0.0001, effect size Fisher’s z = 0.44, 95% CI: 0.27–0.62; quadratic term: t = −3.13, df = 118, P = 0.002, effect size Fisher’s z = 0.29, 95% CI: 0.11–0.46). Selecting on longer-lived (blue field) or far-swimming (red field) sperm results in a strong overlap in selection pressure indicated by the overlap of the two fields (purple field). Sperm from different males are shown as different colors and symbols.

Table S4.

Read number, mapping rate, proper pairing rate, and read depth for nuclear and mitochondrial DNA from all samples

Male ID	Sample	Reads	Mapping, %	Proper pair, %	nucDNA depth	mtDNA/nucDNA depth ratio
31	Finclip	189,601,184	99.1	92.1	65.1	74.1
31	Unselected	163,143,447	99.1	91.5	55.0	15.8
31	Center	272,541,486	99.1	92.1	92.5	17.2
31	Outer	179,569,603	99.0	92.1	60.6	18.0
32	Finclip	221,908,530	99.1	91.2	76.1	61.9
32	Unselected	235,713,440	99.2	91.9	80.6	20.9
32	Center	191,342,523	99.1	91.1	65.0	18.7
32	Outer	154,708,694	99.1	90.9	53.1	17.9
34	Finclip	224,220,018	99.2	91.2	77.9	81.8
34	Unselected	181,525,036	99.1	91.1	62.1	18.7
34	Center	199,626,918	99.1	91.0	68.5	20.0
34	Outer	228,218,827	99.2	91.0	77.9	20.8

Fig. S4.

Likelihood ratios comparing allele frequency estimates in the unselected sperm pool vs. finclip reads. Likelihood ratios are calculated following Lynch et al. (28) and are means for heterozygous paternal sites along nuclear chromosomes within half-overlapping 400-kbp windows containing at least one heterozygous site/10 kbp for male 31 (red line) and male 34 (green line). The read set from the unselected sperm pool for male 32 had an unidentified technical error and is not shown. The coefficient of variation of the summed read coverage of all four samples per male within each window (gray line) is scaled to range between 1 and 10 on the LRT scale.

Fig. S5.

Likelihood ratios comparing allele frequency estimates in center vs. unselected sperm pools. Windows, male colors, and scaled coverage coefficient of variation are as in Fig. S4.

We then tested for differences in allele frequency by comparing the allele counts in the reads from the two selected sperm pools (center and outer), using allele-frequency LRTs (28) with critical values set empirically as equal to or greater than the 99% quantile of the distribution of likelihood ratios for each male. We supplemented the LRTs with tests of significant allele-frequency skews using logarithm of the odds (LOD) scores calculated from binomial probabilities of allele counts in reads from each selected pool. Binomial tests could result in no allele-frequency skew in either pool, in a skew (binomial probability <0.01) in one pool only, or in opposed skews, i.e., alleles A and a being most frequent in different pools. As for LRTs, critical values for binomial tests were set empirically at equal to or greater than the 99% quantile of the LOD scores for each male (SI Materials and Methods for further methodological details). In contrast to the bias checks, we found differences in allele frequency between selected sperm pools throughout the genome (Fig. 4 and Fig. S6), although there was considerable variation among males and between tests. One male (male 31) showed consistently elevated LRTs in the long arm of chromosome 4 (Fig. S6), which is unusually repeat-rich (29) and did not feature in either of its bias comparisons (Figs. S4 and S5). We will not speculate on the functional basis of these results with respect to specific genes or genomic regions at this point, because we need a stronger dataset with more males to support such speculation. Nevertheless, there is clear evidence that genetic variation accompanies the phenotypic variation among selected sperm pools and that this variation is not the result of transmission or handling biases.

Fig. 4.

Genetic differences between selected sperm pools from three males. Each symbol aggregates allele frequency comparisons at heterozygous sites within half-overlapping 400-kbp windows containing at least one site/10 kbp and shows windows in which the given test value is within the 99% quantile of its distribution for each male. Upward-pointing triangles indicate allele frequency assessed by LRT. Downward-pointing triangles indicate binomial tests showing skewed and opposed allele frequencies in selected sperm pools, via binomial LOD scores. Vertical lines indicate binomial tests showing skewed allele frequencies in either the central or outer selected sperm pools, via binomial LOD scores. Color indicates male identity: red, male 31; blue, male 32; green, male 34.

Fig. S6.

Comparisons of allele frequency between the outer vs. center sperm pools via likelihood ratios and binomial tests. Windows are as described for Fig. S4. The likelihood ratio comparisons for male 32 are included here (blue line). Binomial test results for each window are also shown using dots and vertical lines; a dot indicates a window in which the mean LOD score for opposed allele frequencies was equal to or greater than the 99% quantile for that male; a vertical line indicates a window in which the absolute mean LOD score for sites with skewed allele frequencies was equal to or greater than the 99% quantile for that male; see SI Materials and Methods, Sperm sequencing, bioinformatics and allele frequency comparisons above for more details. Colors of dots and lines correspond to colors used for likelihood ratio traces in Figs. S4 and S5.

We provide clear evidence that variation in sperm produced by the same male in a single ejaculate has pronounced effects on several fitness-related traits throughout life and that this variation has a genetic basis. Selection on sperm within the ejaculate results in reduced occurrence of apoptotic cells during early development, more viable embryos, and more fit adult offspring. The sequenced sperm pools further suggest a link between sperm phenotype and sperm genotype. Such a link may have several nonmutually exclusive causes, and one possible explanation is variation in epistatic interactions and hence additive genetic effects of the different sperm haplotypes. This hypothesis provides a particularly plausible scenario for the variation in sites diverging among the sperm pools of the three males. Regardless of the exact genetic underpinning of our observations, our findings are likely to have major implications for key evolutionary processes including the rate of adaptation (30), the evolution of a sexually dimorphic recombination rate (31, 32), the load of deleterious mutations (33), and the extent of inbreeding depression (34). They also may account for hitherto unexplained patterns of non-Mendelian inheritance (35) and apparent discrepancies in observed mutation rates (36). In addition, our findings provide insights that are crucial for clinical and agricultural assisted-fertilization techniques such as IVF and intracellular sperm injection (ICSI). These techniques omit many if not all naturally occurring steps of within-ejaculate sperm selection, and the consequences of such omission need to be understood (1, 37). Future research therefore should focus on the consequences of gametic selection in a broad variety of taxa with both external and internal fertilization.

Materials and Methods

All experiments described here were performed in accordance with the guidelines and approved by the Swedish Board of Agriculture (Jordbruksverket approval number C341/11). For a detailed description of materials and methods, please see SI Materials and Methods.

In a first step we performed IVF experiments using the zebrafish D. rerio in which we split the male ejaculate and the female clutch of eggs into two halves. We exposed sperm to one of two treatments differing in the time from sperm activation to fertilization: SAT, 0 s; LAT, ∼25 s. We repeated this experiment twice using slightly different selection criteria for SAT and LAT. We monitored offspring fitness by assessing differences in cell apoptosis during early developmental stages, embryo survival, sperm swimming velocity, sperm density, and reproductive success.

We tested for preejaculation sperm aging by collecting three successive sperm samples from each male with the first sample containing the oldest sperm and the third containing the youngest sperm and tested for postejaculation sperm aging by delaying the time between sperm activation and fertilization by 25 s or 50 s. Using IVF and a split-clutch design we tested for differences in embryo survival. Finally, we selected sperm for their swimming phenotype by placing them in a droplet surrounded by a glucose ring to let them swim toward the edges. We collected sperm from the center and the edges of the droplet and sequenced sperm collected from each site, sperm from an unselected droplet, and a finclip from each male.

SI Materials and Methods

Study Species.

For all experiments, we used D. rerio zebrafish from the outbred wild-type AB strain. All AB fish were originally obtained from the Zebrafish International Resource Center and were maintained at the SciLifeLab zebrafish platform at Uppsala University (https://www.scilifelab.se/facilities/genomeengineeringzebrafish/) for up to two generations following a strict outbreeding regime. The fish were housed in 3-L tanks in a recirculating rack system (Z-Hab System; Aquatic Habitats) at 26.4 ± 1.4 °C and a 12:12 diurnal light cycle. They were fed two to three times/d with a mixture of dry food (Zeigler adult zebrafish diet; Aquatic Habitats) and Artemia (brine shrimp cysts; ZM Systems).

Experimental Procedures.

IVF.

Two weeks before the experiment the fish were kept at low densities of 6–10 fish in mixed-sex groups. One day before the experiments, the males were separated into unisexual groups and kept overnight. The experimental females were kept in small groups with a company male in a compartment separated by a mesh. Because oviposition is induced by light, each tank containing experimental females was covered by black cloth until use. Males and females were not fed for the last 20 h before the experiment to avoid fecal contamination of sperm and egg samples.

Males and females were anesthetized in 0.16 g/L tricaine methanesulfonate (MS222; Sigma-Aldrich), rinsed in fish water, and positioned on a moist sponge under a dissecting microscope (SMZ800; Nikon). The genital area was carefully dried with soft paper towel to avoid sperm activation. Males were stripped by gently squeezing in the cranio–caudal direction using a smooth-edged forceps, and the ejaculate was collected using a microcapillary aspirator tube assembly (for details see ref. 38). About 0.8 µL ejaculate was collected from each male and was transferred into 0.2-mL Eppendorf tubes containing 20 µL (experiment 1A) or 80 µL (experiment 1B) of HBSS. The ejaculates were kept on ice until fertilization.

Females were anesthetized in 0.16 g/L tricaine methanesulfonate (MS222; Sigma-Aldrich) and were rinsed in fish water; then the genital area was carefully dried with a soft paper towel to avoid egg activation. Females were stripped in 9-cm Petri dishes by gently squeezing the belly. Only clutches of 20–300 yellowish, translucent eggs were used for IVF. Eggs were used for fertilization within 1 min after stripping.

Experiment 1: Sperm-selection experiments.

Sperm samples were gently and thoroughly mixed with a Gilson pipette right before IVF. Both egg clutches and ejaculates were divided into two halves, and sperm were exposed to either LAT or SAT treatment. To avoid any bias between treatments caused by egg aging, eggs were fertilized within 1 min after collection, and sperm from both treatments were added to each egg subclutch at exactly the same moment. The exact IVF conditions differed slightly between experiment 1A and 1B and are described separately below.

Experiment 1A.

We assessed sperm longevity for every male individually and calculated halftime sperm longevity from these values for each male. To do so, we performed direct observations of the sperm and, using a stop watch, measured the time until 99% of sperm stopped forward movement. In the SAT treatment, 5 µL of sperm-HBSS from the same male was activated with 30 µL of water in a 1.5-mL Eppendorf tube and was transferred immediately onto an egg subclutch. For the LAT treatment, 10 µL of sperm-HBSS from the same male was activated with 30 µL of water in a 1.5-mL Eppendorf tube, and sperm were added to the other egg subclutch at sperm longevity halftime. We used 5 µL of sperm-HBSS in the SAT treatment and 10 µL in the LAT treatment to ensure that sperm density at fertilization was the same in both treatments.

Experiment 1B.

For the SAT treatment, 10 µL of sperm-HBSS was mixed with an extra 15 µL of HBSS and activated with 400 µL of water in a 1.5-mL Eppendorf tube, and sperm were added to one female subclutch immediately after activation. In the LAT treatment, 25 µL of sperm-HBSS from the same male was activated with 400 µL of water in a 1.5-mL Eppendorf tube and was transferred onto the other egg subclutch 25 s after activation. This time was identified as approximating a 50% decline in motile sperm and is based on the mean of the measurements obtained in experiment 1A. We used 10 µL of sperm-HBSS in the SAT treatment and 25 µL in the LAT treatment to ensure that sperm density at fertilization was the same in both treatments. In addition, we performed an experiment to determine the fertilization success over time and verified that the fertilization success at 25 s post activation is about 50%.

In both experiments 1A and 1B, LAT and SAT sperm were transferred simultaneously onto the egg subclutches to treat eggs in both treatments equally. Eggs and sperm were gently mixed with a soft brush and were left in contact for 1.5 min for fertilization to occur. Then eggs were transferred into a 9-cm Petri dish with Methylene blue solution (an antifungal agent) and were kept in a 28 °C incubator. All in vitro fertilizations were performed on a hot plate (Minitube HT50) at 28.5 °C. Fertilization time was recorded.

Fertilization success and offspring survival.

Eggs were checked 2–4 h hpf to assess the fertilization success. Offspring survival was recorded at 24 hpf and in experiment 1A also was assessed at 144 hpf [6 d postfertilization (dpf)]. After being checked for survival, up to 70 embryos were placed in a 15-cm Petri dish with Methylene blue solution I until the age of 6 dpf and then were transferred into 3-L tanks in the general system for rearing to adulthood. Larval densities in the Petri dish were consistent for each family across treatments.

Sperm swimming velocity.

To measure sperm velocity, males were kept in groups of brothers without females overnight before the measurements. Males were stripped as described above to obtain 0.7–0.8 μL of ejaculate per male, and the ejaculate was transferred into a 0.2-mL PCR tube containing 40 μL of HBSS. Ejaculates were kept on ice for a maximum of 15 min until analysis. For the actual analysis, 1 μL of ejaculate/HBSS was activated with 3 μL of fish water (28 °C) placed on a two-cell 20-μL pocket slide (2X-Cel Chamber; Hamilton Thorne Biosciences). Sperm velocity was measured with a CASA system (ISAS; Proiser, R+D, S.L.). We recorded sperm movement using a brightfield microscope (UOP UB203i trinocular microscope; Proiser) at 100× magnification and a black and white video camera (782M monochrome CCD progressive camera; Proiser). The records were analyzed using ISAS v1 software (Proiser) with the following settings: frame rate: 50 frames/s; frames used: 50; particle area: 5–50 μm²; threshold measurements for VCL: slow, 10–45 μm/s; medium, 45–100 μm/s; rapid, >100 μm/s. Sperm velocity was recorded every 10 s after sperm activation until 95% of sperm stopped moving. We analyzed 301 (±134 SD) motile sperm from each male in experiment 1A and 384 (±151 SD) motile sperm from each male in experiment 1B at 10 s postactivation (the numbers of motile sperm declined over time).

Individual sperm tracking over time.

We collected ejaculates from six males for tracking of individual sperm within the ejaculate over time. Ejaculates were prepared as described for measurements of sperm swimming velocity, and the same sperm:water ratio was used for the activation of sperm during the video recording. Sperm were viewed with an Olympus BX41 microscope (200× magnification) and were video recorded using a Nikon DS-Vi 1 camera and NIS elements software until they stopped swimming. All sperm that did not leave the field of view throughout their entire motile lifespan were included in the analyses (n = 19 ± 10 sperm per male). For the actual sperm tracking, we used the software Image J Fiji and R. The data on sperm swimming longevity and sperm swimming distance used in Fig. S3 were obtained from this assay.

Reproductive output of F1 offspring.

For assessment of reproductive success, males and females from LAT and SAT treatments were paired with nonexperimental outbred fish. They were kept in pairs separated by a plastic divider in a 1.5-L breeding tank overnight. The divider was removed at 8.30 AM, and the eggs were collected 2–3.5 h later. Eggs were counted and transferred into a Petri dish. At 4 hpf, we checked fertilization success, and at 24 hpf we counted dead embryos. We counted abnormal embryos at 6 dpf.

Experiment 2: Preejaculation sperm aging.

Ejaculates were obtained from males and stored as described above. We took three consecutive samples (∼0.8 μL each) from each male. Each sample was transferred into 80 μL HBSS and stored on ice until use. Eggs from one female were split into three equal parts and fertilized with one of the consecutive samples. Fertilization success at 2 hpf and offspring survival at 24 hpf were monitored.

One spermatogenic cycle in zebrafish lasts for about 6 d as shown by BrdU labeling and continuous sampling at different time points after BrdU exposure (39). This study shows that although the first mature sperm occur in the tubular lumen after about 5 d, they occur in the efferent duct only after 6–7 d. Therefore we can be certain that in the efferent duct only sperm at least 6 d old will be sampled. Because we collected three subsamples from each male, the first sample will contain only the most mature sperm stored in the efferent duct, and the last sample will contain younger sperm coming from the tubular lumen of the testes.

Experiment 3: Postejaculation sperm aging.

Ejaculates were obtained from males and stored in HBSS as described above. The ejaculate of one male was divided into two equal parts, and each part was exposed to one of two treatments. For the LAT treatment, as described above, we used 15 μL of sperm/HBSS and 30 μL of HBSS and waited for 25 s before adding the mixture to the eggs. In the aging treatment, we used 45 μL of sperm/HBSS and waited for 50 s before adding the solution to the eggs. For the IVF, we used split half clutches from each treatment and added 200 μL of fish tank water to activate sperm and eggs.

Experiment 4: Apoptosis assay.

Embryos were collected 8 hpf and were processed following the protocol described in Sorrells et al. (15) to visualize apoptotic cells. The total number of eggs and the number of eggs exhibiting apoptotic cells were counted.

In vitro sperm-selection assays.

To select sperm phenotypes for subsequent sequencing, we collected ∼1.8 µL of ejaculate from three anesthetized nonexperimental AB males and transferred it into 0.2-mL Eppendorf tubes containing 80 µL of HBSS on ice (as described above). For the actual selection treatment, a 280-µL water droplet was placed on a concave microscope slide, and the concave edges were lined with a continuous ring of 125% glucose (20 μL) to form a concentration gradient allowing the sperm to swim straight and toward the outer edges. Within 15 min of sampling, the ejaculate tubes were centrifuged at 600 × g for 1 min at 4 °C. After the excess supernatant was discarded, a <10-µL gently homogenized volume of the pellet plus supernatant was carefully pipetted into the center bottom of the water droplet with a 10-µL tip, with care taken not to disturb the gradient setup. After 3.5 min, sperm cells within a radius of 3 mm from the center of the droplet were collected into a 1.5-mL tube using a fine Eppendorf Microloader tip, and sperm cells around the outer edges (8 mm from the center of inoculation) were collected into another tube. In addition, we collected an unselected sperm sample and a finclip from each of the three males to use as a control and a reference for allele frequencies in the two selected pools. The unselected sperm sample and the finclip were not centrifuged but were transferred directly to the lysis step. All samples were lysed overnight with Proteinase K and DNeasy Blood & Tissue Kit buffers (Qiagen), and the DNA was extracted with a standard phenol:chloroform:isoamyl alcohol (PCI, 25:24:1) (Sigma-Aldrich) protocol and eluted in 35 µL of 1× Tris-EDTA buffer (Thermo-Fisher Scientific).

Statistical analysis.

All analyses were conducted using packages developed for the software R v. 3.3.2 (40).

Data with binomial (e.g., offspring viability, fertilization success, abnormal embryos) or Poisson (e.g., number of eggs) error distribution were analyzed running the generalized linear mixed effect models (glmer) function in the package lme4 (41) for R software with a binomial error distribution and a logit link function. All data with Gaussian error distribution (e.g., sperm swimming velocity) were analyzed using the linear mixed effect models (lmer) function in the package lme4 for R software. In all models, we included family as a random effect. The significance of the fitted model was assessed using ANOVA with type III sums of squares tested with an analysis of deviance on a χ² distribution using the R package car (42).

Tests were performed by removing model terms (backward selection), starting with interaction and highest-order terms (e.g., treatment × time³ for the cubic curvature of the polynomial). This reduced model was tested against the full model, using LTRs, with twice the difference in log-likelihoods assumed to follow a χ² distribution. In the final model, only terms whose elimination from the model did not enhance the model fit (i.e., with P < 0.05) were retained.

To test for treatment differences in offspring reproductive success, we performed models including treatment and sex as fixed effect and tested for an interaction between the two. If the interaction was significant, we also performed separate models for each sex to assess the treatment effects in each sex. If the interaction was nonsignificant, it was excluded from the final model.

Sperm sequencing, bioinformatics, and allele frequency comparisons.

The DNA extracted from sperm pools and the finclip were prepped using the Illumina PCR-free library (TruSeq DNA PCR-Free Library Prep; Illumina) for a standard insert size (∼350 bp). Whole-genome paired-end sequencing on both samples was performed on an Illumina HiSeq platform with a HiSeq PE Cluster Kit v4 in high-output mode to a coverage of 60× per pool. After adapter removal and quality trimming with Trimmomatic 0.32 (43), the sequencing resulted in an average of 203.5 × 10⁶ reads per sample. Reads were aligned to the D. rerio Zv9 reference assembly with Burrows–Wheeler aligner (BWA) 0.7.12 bwa-mem (44), duplicate reads were marked with Picard (broadinstitute.github.io/picard), and reads for each sample were realigned with GATK 3.5.0 (45). Mean mapping and proper-pairing rates were 99.1% and 91.4%, respectively. Average read depth ± SD was 69.5 ± 11.6 per nuclear chromosomal site and 2,288.6 ± 1,876.9 per mtDNA site. The finclip samples contained a greater fraction of mtDNA reads (mtDNA/nucDNA ratio 72.6 ± 10.2) than did any of the sperm pools (ratio 18.7 ± 1.7); otherwise, pools did not differ systematically (Table S4).

Alignments of reads from the finclips were used to determine heterozygous paternal sites for each male using GATK 3.5.0 HaplotypeCaller, GenotypeGVCFs, and SelectVariants (46). Alignments of reads from unselected, center, and outer sperm pools were used to generate base count profiles for each pool after excluding duplicate reads and low-quality bases and allowing for sites with deep coverage [SAMtools 1.3 options –q1 –Q13 –d100000 (www.htslib.org/), profiles generated with pileup2pro2.pl (https://github.com/douglasgscofield/gameteUtils); other scripts mentioned here can be found at this same repository]. Base count profiles were used to generate allele frequency estimators and likelihood ratios using the methods of Lynch et al. (28) as implemented in the script hetPoolLikelihoods.pl, excluding sites with >2× mean coverage following the recommendation of Lynch et al. (28) (mean coverage was 263.3, 266.4, and 277.6 for males 31, 32, 34, respectively, based on base counts at heterozygous sites on chr1 summed across all read pools). Base count profiles were also used for binomial tests of allele frequency skews, as implemented in the script hetTwoPoolTest.pl, at all paternal heterozygous sites for which the binomial probability of observed allele counts in the unselected sperm pool given equal expected frequencies was Binom(unselected) ≥0.01. For example, at the mean base count of 49× for chr1 in the unselected sperm pool from male 31, this criterion corresponds to a minimum minor allele depth of 16. For all such sites, binomial probabilities were calculated for central and outer sperm pool base counts, and results were given one of three classifications: (i) opposed, with binomial probabilities <0.01 in both pools and allele A most common in one pool with allele a most common in the other pool; (ii) skewed, with binomial probability <0.01 in one pool but not in both, regardless of major allele identity; or (iii) neither. For each opposed or skewed site, LOD scores were calculated from binomial probabilities using the subtraction of logs if central and outer pools shared the same major allele and addition of logs if they did not. Mean likelihood ratios and LOD scores were summarized within 400-kbp half-overlapping windows using the script windowify.pl, with a paternal heterozygous site included if the total coverage across all read pools was ≥40× and a window included if it contained at least one heterozygous site/10 kbp. Critical values for likelihood ratio and binomial tests were the ≥99% quantiles of the distributions of LRT and LOD scores across all windows for each male. Critical values for comparisons of central and outer sperm pools for the 99% quantile for males 31, 32 and 34, respectively, are 3.00, 1.65, and 1.57 for LRTs; 8.79, 5.58, and 5.50 for LOD scores of binomial tests for opposing allele frequencies; 4.50, 3.07, and 3.50 for LOD scores of skew toward the outer pool; and 4.80, 3.44, and 3.12 for skew toward the central pool.

Ethics.

All experiments described here were performed in accordance with the guidelines and approved by the Swedish Board of Agriculture (Jordbruksverket approval number C341/11).

SI Results and Discussion

Fertilization success was slightly but significantly lower in the LAT treatment than in the SAT treatment in experiment 1A (treatment: χ²_1, = 5.15, P = 0.023) (Fig. S1A) and also in experiment 1B (treatment: χ²_1, = 9.08, P = 0.003) (Fig. S1B). The reason for the significant difference in fertilization success between SAT and LAT is largely a statistical problem. The fertilization success in SAT treatment likely reflects the maximum possible success because of the relative high sperm density used in our assays. Therefore the fertilization success in the LAT treatment can be only an equally high or lower than, but not higher than, the fertilization success in the SAT treatment.