Characterizing TE content in Daphnia

The relative abundance of TEs across the nine D. magna genotypes is similar (Fig 2B and Tables 1 and S3 and S4). In Daphnia, LTR retrotransposons are the most common type of TEs, with the Gypsy superfamily being the most abundant (Table 1). All other categories of TEs (DNA transposons, Long and Short Interspersed Nuclear Elements [LINEs and SINEs], and rolling circle [RCs]) constitute less than 2% of the genome, although DNA transposons are still highly diverse with 18 different families represented (S5 Table). Although abundance is consistent within D. magna when comparing across genotypes, overall abundance and the abundance of individual families differed between D. magna and its congener, D. pulex (D. pulex > D. magna; t₈ = -14.2, P < 0.0001; Fig 2B and S6 Table), with 7 and 19 families of DNA transposons being specific to D. magna and D. pulex, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Abundance of TE types by family or superfamily for Daphnia magna (averaged across nine genotypes) and D. pulex (PA42 [PRJNA307976]).

https://doi.org/10.1371/journal.pgen.1009827.t001

It is important to note, TE abundance can be measured in two ways—repeat masking a genome assembly with a TE library or mapping short reads to a TE library to use depth of coverage as an estimate of abundance. In fact, estimates of overall TE content in D. magna differ by more than a factor of two using these two methods (6% using repeat masking and 16% using read mapping; S4 Table), which is likely because repeat masking is more sensitive to the quality of the assembly. We recommend using a read-mapping approach for accuracy, however repeat masking is more common and provides the opportunity for inter- as well as intraspecific comparisons here; see S2 Text, S2 Text, and S5 Table for estimates using both methods.

Variation in TE activity over long time periods

Despite the consistency in terms of abundance, we quantified TE insertion polymorphisms (TIPs) among the 9 genotypes of D. magna sampled and were able to clearly distinguish genotypes based on their population-of-origin using principal components analysis (PCA; Fig 2C), regardless of reference genome used (S1 Fig). Depending on the assembly used as reference (see S2 Text), we identified between 1442 and 1903 TE sites, of which 13% to 16% were polymorphic across the 9 genotypes (S7 Table) and which, by k-means clustering, always revealed non-overlapping clusters corresponding to their population-of-origin (S8 Table). On average, we find 19% of TIPs are specific to a single genotype (i.e., singletons) and an additional 29% of TIPs are specific to a single population (Figs 2D and S2 and S3, and S9 and S10 Tables). Whether a particular position in the genome is occupied by a TE is determined by events at multiple levels: the chromosome level (e.g., gains/losses due to insertions, deletions, or gene conversion events,) and/or at the individual/population level (e.g., frequency of sexual reproduction or the strength of selection against new insertions). An additional interpretation of an excess of singletons is that the TE family is, or has recently been, active.

Another indicator of recent activity is low levels of mean pairwise divergence (MPD) among copies belonging to a given TE family because new copies have not yet accumulated point mutations. The range of MPDs across TE families was 15–31%, with SINE elements having the lowest values (S4–S15 Figs). Surprisingly, we observed higher MPDs in TE families that were currently active in our MA experiments (~21% for active families and ~19% for inactive; S2 Text and S11 Table). An alternative explanation for high MPDs is a higher base substitution mutation rate, which has been reported for D. magna (greater than D. pulex; [29]). While we observed interspecific differences in MPDs across TE families between the two species, they were not consistently higher in D. magna as one would predict (S12 Table), nor did they correlate with known intraspecific variation in base substitution mutation rates within this species (S16 Fig; ρ = -0.66, t₇ = -2.3, P = 0.055).

Estimated rates of TE loss and gain using mutation accumulation experiments

We used mutation accumulation (MA) experiments initiated from each of the 9 genotypes of D. magna from each of the three populations to estimate overall (Tables 2 and S13) and family-specific TE mutation rates (Table 3). Using whole genome sequence (WGS) data from the MA lines, we detected 67 gain and 28 loss mutations; S14 Table shows the location and read support for each event. Rates of gain across MA lines ranged from 0 to 22.6 x 10⁻⁵ per copy per generation with a mean rate of 1.39 x 10⁻⁵ /copy/gen (95% CI: 0.41 x 10⁻⁵–2.66 x 10⁻⁵) and loss rates ranged from 0 to 31.8 x 10⁻⁵ /copy/gen with a mean of 1.70 x 10⁻⁵ /copy/gen (95% CI: 0.53 x 10⁻⁵–3.23 x 10⁻⁵; S15 Table). Looking across genotypes, averaging across rates for all TE families (with non-zero copy numbers in all genotypes), it is clear that some genotypes have a bias towards gains while others exhibit mainly losses (Fig 3A and S16 Table). To test for a population effect, we fit a binomial mixed effects model (Fig 3B; for gains, χ² = 5.9, df = 2, p = 0.0514 and losses χ² = 12.1, df = 2, p = 0.0024). Post-hoc Tukey HSD tests reveals that Israel genotypes had lower gain rates than Finland genotypes (p = 0.039) and Germany genotypes had greater losses than Israel genotypes (p = 0.0005; S17 Table). In addition to the gain and loss rates, we also calculated a net mutation rate for each genotype (S16 Table) and for each active TE family (S18 Table). These rates range from negative (e.g., P elements in genotype IB are decreasing at a rate of -7.44 x 10⁻⁴ per copy per generation) to positive (e.g., Penelope elements increasing at a rate of 9.06 x 10⁻⁴ /copy/gen in IC), and can even vary for the same TE family among genotypes (e.g., -2.22 and 4.06 x 10⁻⁴ /copy/gen for Gypsy elements in GA and FC, respectively).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. TE gain and loss rates for D. magna MA lines.

Gain and loss rates (per copy per generation) for each (A) genotype and (B) population of D. magna averaged across all TE families. Mean rates for MA lines from Finland, Germany and Israel represented in gold, green and blue, respectively, and are provided, along with estimates of 95% CI, in S17 Table. In (A), gain rates for FB, IA and loss rate for FB are zero.

https://doi.org/10.1371/journal.pgen.1009827.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Number of events and mean TE mutation rates (per copy per generation, including 95% confidence intervals [CI]) for gains and losses based on whole genome sequence data from 66 Daphnia magna mutation accumulation and extant control lines descended from 9 starting genotypes collected from Finland, Germany, and Israel.

https://doi.org/10.1371/journal.pgen.1009827.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Number of events and rates (plus 95% confidence intervals [CI]) of gain and loss (per copy per generation) for each TE superfamily in which events were observed averaged across all MA lines.

Gains and losses based on whole genome sequence data from Daphnia magna mutation accumulation lines descended from 9 starting genotypes collected from Finland, Germany, and Israel.

https://doi.org/10.1371/journal.pgen.1009827.t003

When selection was not minimized (i.e., in the extant control lineages maintained in large populations in parallel to the MA lines), we only detected 2 gain and 17 loss mutations (Tables 2, S19, and S20). Fitting binomial mixed-effects models, we found that EC lines had significantly lower gain rates (χ² = 27.9, df = 1, P < 0.001) and significantly lower loss rates (χ² = 10.5, df = 1, P = 0.0012) compared to MA lines, revealing the deleterious effect of TE activity. Furthermore, gain rates in MA lines were 695x higher than in EC lines, compared to loss rates which were only 7.4x higher in MA lines, suggesting that TE gains are much more deleterious than losses (Table 2).

Validation methods

Rather than perform PCR validation to gauge the sensitivity of our methods, given that each event was of an unknown length, we performed simulations to estimate the false discovery and false omission rate (FDR and FOR) for the four cases of TE events that can occur (Fig 1 and S21 Table). FDRs were relatively low (< 3%) for all four types of mutations (S2 Text and S21 Table), and neither FDRs or FORs varied greatly for TEs of different lengths or for different mutational events (S22 Table). Mutation rates for each type of event in the MA and EC lines adjusted for FDRs can be found in S13 Table. Notably, the fact that the four cases of events are not equally likely (most gains were novel (0 → 1 [n = 62/67]) and most losses were at previously heterozygous sites (1 → 0 [n = 26/28]) is potentially revealing about what proximal mechanisms explain the bulk of TE proliferation and loss (see Discussion). It is important to note, our rate estimates for TE activity likely represent a lower bound. This is, in part, because our analyses focus only on those TEs that could be classified as belonging to one of the five major groups of known TEs (rates for all TEs, classified and unknown, are presented in S23 and S24 Tables).

TE mutation rates are not correlated with other types of mutation rates

Overall, TE mutation rates in D. magna vary intraspecifically among genotypes (Fig 3A) mirroring the high levels of intraspecific variation observed in other mutation rate estimates for this species (see [28, 29]). In terms of frequency per site, TE mutations are intermediate among the other types of mutation examined so far in D. magna, (i.e., microsatellite mutation rates are much higher (~10⁻²) and nuclear and mtDNA base substitution rates are much lower (~10⁻⁸ and ~10⁻⁷, respectively), on a per site per generation basis). As expected, we observe more events in higher copy number families (S17A Fig). We looked at the relationship between rates of TE gain and loss (and net rates) and the proportion of the genome that is TEs in each genotype and found no correlation (S25 Table), nor do TE rates correlate with base substitution mutation rates (Fig 4A and S25 Table). The only correlation with other mutational processes is between TE mutation rates and gene conversion rates (when plotting only rates for TE events that are likely to be caused by gene conversions [1→0 TE losses and 1→2 TE gains]; ρ = 0.83, t⁷ = 3.91, P = 0.0058; Fig 4B), although even this predicted correlation is driven largely by one genotype (GB) with high estimates for both rates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The relationship between TE, base substitution, and gene conversion mutation rates in D. magna MA lines.

(A) Base substitution rates (per bp per generation) plotted against TE mutation rates (per copy per generation). Circles represent the sum of all TE gains and losses, triangles represent only 0→1 TE gains. (B) Gene conversion rates are plotted against TE events that could be caused by gene conversion (the sum of 1→0 TE losses and 1→2 TE gain rates; shown as squares). Points in gold, green, and blue represent rate estimates for genotypes collected from Finland, Germany and Israel, respectively. Base substitution and gene conversion rates are from [29].

https://doi.org/10.1371/journal.pgen.1009827.g004

Long-term patterns of TE accumulation do not correspond to short-term mutation rates

Overall, TE content, in terms of abundance, is very similar across genotypes from the three populations sampled for this study (Fig 2B). Elements from the Gypsy superfamily of LTRs (Class 1) are the most numerous, as has been reported in the congener, D. pulex (Rho et al. 2010), which has more TEs overall than D. magna (Table 1) even though D. magna has a larger genome (as measured by flow cytometry, D. magna = 0.30 pg and D. pulex = 0.23 pg [33]). Despite these similarities in patterns of TE abundance, patterns of insertion site polymorphism (differences among individuals in terms of which specific sites are occupied by TEs of a given family) make all three populations readily distinguishable (Fig 2C and S8 Table), which begs the question—how much do mutation rates for TEs differ intraspecifically in Daphnia?

Based on over 100 observed events in our multi-year MA experiments, we were able to estimate rates of gain and loss for each type of TE mutation (Table 2). Rates of gain and loss in D. magna are similar (1.4 and 1.7 x 10⁻⁵ per copy per generation, respectively; Table 2), but they vary widely among genotypes and populations (Fig 3A and 3B) and among TE families (Table 3). The majority of the gains observed are novel gains (0 → 1 gains; Fig 1), most likely resulting from insertions of TEs either excised from elsewhere in the genome (in the case of cut-and-paste elements) or retrotransposed (in the case of Class I elements, such as Gypsy), rather than 1 → 2 gains which can result from homolog-dependent DNA repair [50]. The majority of loss events were at positions that were initially heterozygous (1 → 0), again a pattern expected based on mechanism since both DNA repair and gene conversion events could “reconstitute” a TE lost due to excision or deletion at an ancestrally homozygous site. A genome-wide assay of TE mutation rates in Drosophila showed insertions far outnumber deletions, but in flies the per copy per generation rates differ significantly, with insertions higher (~10⁻⁹) than deletion rates (~10⁻¹⁰), and much lower rates overall compared to those observed here [25].

Little is known about intraspecific variation in TE mutation rates in other animal species, even though there have been several large-scale studies of their polymorphism (e.g., [51–53]). Among D. magna genotypes, rates ranged from a high gain bias in one genotype from Finland (FA; 5.3 x 10⁻⁵ per copy per generation) to a deletion bias in one genotype from Germany (GB; -5.5 x 10⁻⁵ per copy per generation; S16 Table), with the highest number of events overall occurring in a single genotype (FC; S16 Table) in a single family (Gypsy; n = 51; S18 Table). Looking across families of TEs, populations are distinct in their rates, with Finland exhibiting higher rates of gain overall, Germany exhibiting high rates of loss overall, and Israel exhibiting gains and losses with almost equal frequency resulting in the lowest net rates overall (Fig 3B and S16 and S17 Tables). Thus, while genotype-specific rates of mutation surely introduce variable levels of TE-related genetic variation in these lineages, evolutionary forces acting at the population-level likely explain the consistent overall abundance of TEs (Fig 2B) and distinctive patterns of insertion site polymorphism (Fig 2C).

Ultimately, the lack of correspondence between the variable mutation rates and the consistent patterns of TE accumulation across the 9 genotypes suggests natural selection may prevent TEs from over-running the genome long-term. Evidence in support of selection against TE activity from this study was our observation of much lower rates in control lines (where lineages were maintained in large population sizes) compared to MA lines (where selection is minimized by propagating lines via single-progeny descent), suggesting that TE mutations, especially gains, are highly deleterious (Table 2). Early papers on rates of TE activity posited that high copy number families might even evolve lower transposition rates because of the deleterious effects of TE insertions (much like parasites evolve to be less virulent; [54]), however the relationship we observe between per copy per generation rates of mutation and abundance in the genome observed is weak (S17A Fig), with no clear downward trend even for high copy number families (S17B Fig) or with rates of gain (S17C Fig).

Looking specifically at the most abundant family with the most mutation events, Gypsy, we see rates of gain and loss can vary greatly among genotypes (Fig 5A) and, in this case, the variation is reflected in the long-term patterns of insertion site polymorphism (Fig 5B) and abundance (Fig 5C). While the patterns reflect a mixture of active and inactive elements, some population-specific trends, which have been reported previously for Gypsy elements [55], are notable. Specifically, high rates of gain in Finnish genotypes could explain a non-significant trend in terms of singletons (excess in Finland [n = 13] compared to Germany [n = 5] or Israel [n = 7]; G = 4.01, df = 2, P = 0.13) or the higher percent abundance of Gypsy elements in Finnish clones compared to Germany (Fig 5C). Future studies with additional genotypes and populations and a longer mutation accumulation experiment will be necessary to determine if the patterns of TE accumulation reflect the mutational variation, as suggested by the data for this large TE family, or if evolutionary forces mute the variation introduced by TE movement, as observed when looking across all families of elements.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Rates of insertion site polymorphism and abundance for Gypsy family TEs in 9 genotypes of D. magna from Finland (gold), Germany (green), or Israel (blue).

(A) Mean gain and loss rates (per copy per generation) for each genotype. Gain rates for FB, IA, IB and loss rates for FB, IB are zero. (B) Colored bars indicate the number of genotype-specific polymorphic sites (singletons; x = 1) or population-specific sites (when x = 2 and x = 3), grey bars represent sites where elements are shared across populations (reference genome used for this analysis was FASC). (C) Percent abundance in the genome for each genotype estimated using RepeatMasker.

https://doi.org/10.1371/journal.pgen.1009827.g005

Rates of TE gain and loss do not correlate with other mutation rates

Base substitution mutation rates (bsMRs) are the most frequently estimated, and are used broadly in models and discussion of the mutation rate in evolutionary biology. Although they are the most commonly studied, bsMRs are not necessarily representative of mutation rates for other categories of mutation, nor are they likely to generate they greatest amount of genetic variation [56]. Microsatellites are known to be highly mutable (reviewed in [57]) and the average genome-wide rates of mutation at these loci in these genotypes of D. magna are several orders of magnitude higher (~10⁻²; [28]) than the TE mutation rates we report here (~10⁻⁵). The bsMRs we reported for D. magna were the highest and most variable direct estimates reported in animals so far using an MA approach (~10⁻⁷ and ~10⁻⁹ for the mtDNA and nucleus, respectively; [29]), but are also several orders of magnitude lower than the overall TE mutation rates we report. Evolutionary theory aimed at explaining how mutation rates evolve does not specify mutation types, however, thus we would expect that lineages with relatively high rates of mutation in one category would have high mutation rates for other types of mutation as well. The data do not support this prediction, as there is no correlation between TE mutation rates and bsMRs across the 9 genotypes (Fig 4A and S25 Table). Rates of gene conversion, however, do positively correlate with TE rates when based on those events that can be produced by gene conversion as predicted (Figs 1 and 4B and S25 Table).

While there is no positive linear correlation among mutation rates for different types of mutations (comparing TEs and base substitutions) across all 9 genotypes (Fig 4A), it is interesting to note that, in our MA experiment, genotypes from Finland have the highest rates of TE gain (and gains are more deleterious than losses), the highest rates of microsatellite deletions [28], the highest rates of base substitution among the three populations assayed [29], and the highest rates of mutations causing structural variation (e.g., insertions and deletions) [30]. These commonalities among our direct estimates based on rearing animals in a common laboratory environment point to the historical selection regime due to population genetic constraints or the frequency of recombination, rather than mutagens in the atmosphere, as an explanation for higher rates of deleterious mutation in the Finnish genotypes. Alternatively, this pattern could result if selection on DNA repair mechanisms, as opposed to the mechanisms causing mutations, is more influential. In contrast, genotypes from Israel consistently exhibit the lowest net rates of TE mutation, microsatellite mutation, and base substitutions of the three populations assayed.

Study system

The D. magna genotypes used in this experiment were provided by Dieter Ebert and are part of a collection of samples from across the species range. Genotypes were selected from populations along a latitudinal gradient (Finland, Germany, and Israel) in order to sample individuals originating from a broad range of environments. Different maximum and mean temperatures and photoperiods (S1 Table), both of which can also result in fluctuating habitat sizes [66], are represented along the gradient.

Experimental design

Three genotypes from each of three populations (Finland, Germany, and Israel) were used to initiate laboratory stocks. From these lab stocks, starting controls (SCs) were selected (immediate descendants of which were frozen and sequenced) for each of the 9 genotypes. From the SCs, mutation accumulation (MA) lines (n = 5–12 per genotype; total of 66) and large population controls (extant controls [ECs]; n = 2 per genotype; total of 18) were initiated and propagated in parallel. Tissue from each line (MAs and ECs) was frozen after the mutation accumulation period; the average number of generations across MA lines was 12 and the experiment ran for approximately 30 months in total (S26 Table; see S1 Text for additional details).

The MA and EC lines from each genotype were maintained as single individuals or large populations in 250 mL beakers containing 175–200 mL or 3.5 L jars containing 3 L of Aachener Daphnien Medium (ADaM [67]), respectively. All lines were maintained under a constant photoperiod (16L:8D) and temperature (18°C), and fed the unicellular green alga Scenedesmus obliquus (2–3 times per week ad libitum). While selection is permitted to act in the large population ECs, the single-progeny descent used to propagate the MA lines maximizes chance and minimizes selection, and thus allows for the accumulation of mutations. The experimental protocols used here have been described previously [28, 29].

DNA extraction and sequencing

At the end of the mutation accumulation period, the 9 SCs, 66 MA lines, and 18 ECs were sequenced (Illumina) to assess the TE content in the original genotypes (SCs), to quantify TE mutation rates (MA lines), and to compare to laboratory-reared lines where selection is not minimized (ECs). Five asexually-produced clonal individuals from each SC line, all derived MA lines, and the extant control lines were flash frozen for DNA extractions (see S1 Text for details). Libraries were used to generate approximately 50x depth of coverage genome-wide for each sample. Paired-reads from SC lines were then used to construct reference-guided assembles for each of the 9 genotypes (see S2 and S17 Tables for genome assembly statistics and S1 Text for assembly methods).

Characterizing TE content

A custom D. magna TE consensus library was created from a concatenated file of the 9 reference-guided assemblies from the SC for each genotype using RepeatModeler v1.0.11 [68] and used to mask each assembly using the slow search setting of RepeatMasker v4.1.0 [39]. We clustered elements in the TE library that exhibited ≥ 98% nucleotide identity over their full length to a longer sequence in the library using cd-hit-est v4.8.1 [69], yielding a non-redundant TE library containing full and partial TE copies (S1 Data). The non-redundant TE library was then used to determine the abundance, length, percent occupancy, insertion site polymorphism, and pairwise divergence for all categorized TEs in each assembly (see S1 Text for details), and in some cases analyses were performed using both categorized and ‘unknown’ TEs. To compare TE abundance and diversity to the congener D. pulex, we utilized the publicly available reference assembly PA42 (https://www.ncbi.nlm.nih.gov/bioproject/307976). The quality of our D. magna assemblies were similar to that of the D. pulex assembly (S2 Table).

TE mutation rate estimation in MA lines

We used TEFLoN v0.4 [25] to identify active TEs in the MA lines (see S1 Text for details). There are two types of TE gain mutations (0→1 and 1→2) and two types of TE loss mutations (2→1 and 1→0) that can be observed based on whether the ancestor (SC) was homozygous, heterozygous or lacked a TE (an “absence allele”) at a given site relative to the status in the descendant MA line (e.g., if the SC was heterozygous and experienced a gain, it would be classified as a 1→2 gain event in the MA line; Fig 1). Our ability to detect these different events is not uniform, however, thus we used a series of filtering steps and simulations to assess the support for each observed event and to assess the sensitivity of our methods (see S1 Text). Family-specific mutation rates for each of the four mutation types were calculated using N_m / (N_SC*G), where N_m represents that number of sites that experienced a particular mutation event, N_SC represents the initial copy number of that TE family in the SC line, and G represents the number of MA generations. For a full description of our estimates of our false discovery and false omission rates and our simulations, see the S1 Text.

Statistical analyses

Statistical analyses were performed in R [70]. Family-specific TE mutation rates for a particular genotype was estimated by averaging across MA lines. Rates of a particular mutation type (0→1 gain, 1→2 gain, 1→0 loss, 2→1 loss) of an MA line were estimated by averaging that rate across all TE families. Rates of a particular mutation type for a genotype were estimated by averaging that rate across MA lines. Confidence intervals for mutation rates were estimated by bootstrapping across MA lines 10000 times. Details on all statistical test are included in S1 Text and all code for data processing and analysis is available at https://github.com/EddieKHHo/DaphiaMagna_MA_TE.