GWAS reveals determinants of mobilization rate and dynamics of an active endogenous retrovirus of cattle

Tang, Lijing; Swedlund, Benjamin; Dupont, Sébastien; Harland, Chad; Costa Monteiro Moreira, Gabriel; Durkin, Keith; Artesi, Maria; Mullaart, Eric; Sartelet, Arnaud; Karim, Latifa; Coppieters, Wouter; Georges, Michel; Charlier, Carole

doi:10.1038/s41467-024-46434-1

Download PDF

Article
Open access
Published: 09 March 2024

GWAS reveals determinants of mobilization rate and dynamics of an active endogenous retrovirus of cattle

Nature Communications volume 15, Article number: 2154 (2024) Cite this article

1319 Accesses
6 Altmetric
Metrics details

Subjects

Abstract

Five to ten percent of mammalian genomes is occupied by multiple clades of endogenous retroviruses (ERVs), that may count thousands of members. New ERV clades arise by retroviral infection of the germline followed by expansion by reinfection and/or retrotransposition. ERV mobilization is a source of deleterious variation, driving the emergence of ERV silencing mechanisms, leaving “DNA fossils”. Here we show that the ERVK[2-1-LTR] clade is still active in the bovine and a source of disease-causing alleles. We develop a method to measure the rate of ERVK[2-1-LTR] mobilization, finding an average of 1 per ~150 sperm cells, with >10-fold difference between animals. We perform a genome-wide association study and identify eight loci affecting ERVK[2-1-LTR] mobilization. We provide evidence that polymorphic ERVK[2-1-LTR] elements in four of these loci cause the association. We generate a catalogue of full length ERVK[2-1-LTR] elements, and show that it comprises 15% of C-type autonomous elements, and 85% of D-type non-autonomous elements lacking functional genes. We show that >25% of the variance of mobilization rate is determined by the number of C-type elements, yet that de novo insertions are dominated by D-type elements. We propose that D-type elements act as parasite-of-parasite gene drives that may contribute to the observed demise of ERV elements.

Contrasting segregation patterns among endogenous retroviruses across the koala population

Article Open access 21 March 2024

Evolution of the rodent Trim5 cluster is marked by divergent paralogous expansions and independent acquisitions of TrimCyp fusions

Article Open access 02 August 2019

Adaptive expansion of ERVK solo-LTRs is associated with Passeriformes speciation events

Article Open access 11 April 2024

Introduction

Half of mammalian genomes is composed of interspersed repeats, including endogenous retroviruses (ERVs) that occupy ~5–10% of genome space¹. ERVs derive from retroviral infection of the germline enabling vertical viral transmission. Such retroviral endogenization may lead to the progressive expansion of a clade of ERV elements that may count tens to thousands of members, by an ERV-encoded reverse transcriptase (RT)-dependent copy-paste mechanism. At first, this process entails ERV-encoded envelope (ENV)-associated budding of viral particles followed by germline reinfection. Subsequently, the ERV may sometimes at least in part forgo the extracellular phase yet continue to multiply by more efficient intracellular retro-transposition. In mice, such a transition from IAPE to IAP elements has been shown to result from the combined acquisition of a novel GAG addressing signal and loss of ENV. Expansion of endogenized ERV clades may perdure for hundreds of thousands to millions of years, well after extinction of the initiating exogenous retrovirus. Further expansion of ERV clades is eventually curtailed by the accumulation of ERV-disrupting mutations and the acquisition - by the host - of specific ERV-silencing mechanisms including targeted DNA and chromatin epigenetic modifications, piRNAs, and so-called restriction factors targeting various steps of the ERV life cycle, some of which are co-opted ERV genes. ERV endogenization has been caught in the act at least six times, including in poultry, sheep/goat, mice, cat and koala, indicating that it is a common phenomenon^2,3,4,5,6.

As for other transposable elements, de novo ERV transposition events in the germline increase the mutational load of populations including by causing disease. Deleterious ERVs are hence subject to purifying selection. Conversely, ERV elements provide a substrate for the emergence of new functionalities such as novel cis-acting regulatory elements^4,7 and even new genes. As an example, the syncytin genes, which are essential for placentation, derive from ancient ERVs. Also, ERV elements may condition the host’s susceptibility to exogenous retroviruses, including by providing protection. Hence, ERVs are important drivers of genomic innovation^2,3,4,5,6.

As ERVs settle, expand and then regress in a species’ genome, the rate of ERV mobilization is bound to be an evolving phenotype. Presently, ERVs are largely silent in the human germline⁸, while several clades are still active in the mouse in which they account for a sizeable fraction of deleterious mutations⁹. To what extent the ERV mobilization rate varies between individuals within species during this process, and what the underlying determinants of this variation may be, remains largely unexplored. Modifiers of epigenetic modification and expression of specific ERV elements have been mapped in mice and polymorphisms in KRAB zinc-finger proteins nominated as plausible candidates¹⁰, yet whether these also affect transposition rate is not known.

We herein take advantage of unique features of domestic cattle, and of advances enabled by combining next-generation sequencing (NGS) with CRISPR/Cas9, to perform genetic analyses of interindividual variation in ERV mobilization rate in this species.

Results

The insertion of an ERVK[2-1-LTR] element in exon 5 of the APOB gene causes cholesterol deficiency in cattle

A new autosomal recessive defect, referred to as cholesterol deficiency (CD), emerged in Holstein-Friesian dairy cattle around 2015 (OMIA:001965-9913)^11,12,13. Autozygosity mapping positioned the corresponding locus on chromosome 11¹¹. We and others sequenced the whole genome of acknowledged carriers and showed that CD is caused by the insertion of an ERV element in exon 5 of the apolipoprotein B (APOB) gene^12,13,14. We PCR-amplified and sequenced the corresponding ERV element and showed that (i) it is ~6.8 Kb long, (ii) belongs to the [2-1-LTR] subgroup of ERVK elements^15,16, and (iii) does not contain full-length open reading frames (ORFs) for either the GAG, PRO, POL or ENV genes and is therefore non-autonomous. RNA-Seq analysis of liver of an affected animal showed complete transcriptional shutdown in the ERV’s 5’LTR, truncating ~ 97% of the APOB ORF (Fig. 1a, b; Supplementary Fig. 1). All carrier animals trace back to Maughlin Storm, a sire that was popular in the 1990-ies, suggesting that the birth of the CD mutation might be a recent event and, hence, that ERVK[2-1-LTR] might still be mobile in cattle.

**Fig. 1: The insertion of an ERVK[2-1-LTR] element in the *APOB* gene causes cholesterol deficiency in Holstein-Friesian cattle.**

ERVK[2-1-LTR] elements are active in the bovine germline

We previously generated the Holstein-Friesian Damona pedigree to study de novo mutations in the bovine germline. It comprises 743 animals constituting 127 (overlapping) three generation pedigrees including at least two parents (sire and dam), one offspring and ≥5 grand-offspring. The whole genome of parents and offspring was sequenced at average 26-fold depth, while that of the grand-offspring was sequenced at average 10-fold depth (Methods). We developed LocaTER (Localization of Transposable Elements and Repeats), a bioinformatic pipeline for the detection of transposable element insertions that are not present in the reference genome (Methods). LocaTER detected 1222 ERV insertions that are polymorphic in Holstein-Friesian. Repbase^15,16 reports four main groups of ERVs: ERVL-MaLR, ERVL, ERV1 and ERVK, jointly accounting for ~4.7% of genome space. While ERVL (ERVL-MaLR + ERVL) are the most abundant group in the reference genome (55.5% of genome space), followed by ERV1 (29.3%) and ERVK (15.2%), ERVK includes the most abundant amongst polymorphic ERV elements (81.5% of polymorphic elements), followed by ERV1 (15.6%) and ERVL (2.9%). This is compatible (assuming approximately equal element size) with ERVK being younger and still active (Fig. 2a). Repbase reports 33 subgroups of ERVK, of which four are overrepresented amongst polymorphic ERVK elements, including ERVK[2-1-LTR] (20.2% of ERVK space, 26.6% of polymorphic ERVK elements) and BTLTR1B (10.2% of ERVK space, 27.1% of polymorphic ERVK elements) (Supplementary Fig. 2a).

**Fig. 2: Detection of polymorphic ERVs and de novo ERV mobilization events in the *Damona* pedigree.**

Nine hundred and four ERV elements mapped to intergenic regions (expected: 611), 109 within genes in sense orientation (expected: 159), and 209 within genes in antisense orientation (expected: 159) (p < 10⁻⁶), supporting purifying selection against genic sense insertions (Fig. 2b). These trends did not differ significantly between the ERVK and non ERVK groups (p_{genic vs intergenic} = 0.61; p_{sense vs antisense} = 0.42) (Supplementary Fig. 2b, c). Allelic frequency distribution was mildly shifted towards lower values for genic versus intergenic (p = 0.06), but not for genic sense versus antisense insertions (p = 0.99) (Supplementary Fig. 2d, e; Supplementary Data 1).

Most interestingly, we detected five ERV insertions that were present in an offspring, transmitted to grand-offspring, but absent in both sire and dam, which we considered to be de novo mobilization events (Fig. 2c). Linkage analysis in the grand-offspring indicated that four of these occurred in the germline of a sire, and one in the germline of a dam. Intriguingly, three of the four male mobilizations occurred in the germline of the same bull, of which two in the same sperm cell. We PCR-amplified and sequenced the five de novo insertions. All five measured ~6.8 Kb and their sequence was very similar to the non-autonomous ERVK[2-1-LTR] insertion in the APOB gene (Fig. 2d; Supplementary Fig. 3).

These results confirm that ERVK[2-1-LTR] are presently active in bovine, at a rate of ~ one event per 51 gametes (or ~ one event per 126 gametes when ignoring the exceptional sire), and suggest that the mobilization rate may differ between individuals. Mining of the whole genome sequence of the bull transmitting three ERVK[2-1-LTR] de novo insertions did not reveal striking anomalies in 38 genes that have been connected with control of ERV mobilization¹⁷ (Supplementary Data 2).

Developing a method to measure ERVK[2-1-LTR] mobilization rate in germline and soma

We adapted Pooled CRISPR Inverse PCR (PCIP)¹⁸ to measure the rate of ERVK[2-1-LTR] mobilization in a sample of genomic DNA. As for PCIP, we used CRISPR/Cas9 to augment the efficacy of inverse PCR, yet used Illumina short read sequencing instead of Oxford Nanopore long read sequencing to increase coverage and accuracy (Fig. 3a). Noteworthy features of the method are: (i) the realization of technical replicates by targeting both 5’ and 3’ LTR, (ii) the ability to recognize PCR duplicates on the basis of shared shearing sites (SS)(if two sperm cells carry the same de novo ERVK[2-1-LTR] insertion, the probability that the two corresponding DNA molecules are broken at the exact same position is assumed to be very low; shared SS are therefore assumed to correspond to PCR duplicates), and (iii) the ability to express the number of detected de novo mobilization events as a function of the effective number of explored genomes (using inherited ERVK[2-1-LTR] elements as internal controls, see Methods). Of note, ERVK[2-1-LTR] with polymorphisms in the CRISPR target sites may escape detection by PCIP, an issue which is at least partially mitigated by targeting both 5’ and 3’ proviral ends. The correlation between 5’ and 3’ rate estimates was 0.68, and 0.85 between averages of 5’ and 3’ estimates for biological replicates, testifying for the accuracy of the approach (Supplementary Fig. 4a, b; Supplementary Data 3 & 4).

**Fig. 3: Measuring the de novo rate of ERVK[2-1-LTR] mobilization using an adaptation of Pooled CRISPR inverse PCR sequencing (PCIP-Seq).**

The ERVK[2-1-LTR] mobilization rate in the male germline varies between individuals

We first obtained pairs of semen samples from 10 Belgian Blue (BB) bulls, collected at ≥ 7 years interval. We applied our method to these samples and computed the correlation between ERVK[2-1-LTR] mobilization rate (5’ and 3’ average) estimated at young and old age. The correlation, corresponding to the so-called repeatability of the trait (upper bound of the heritability), was 0.76 (Fig. 3b). This implies that the ERVK[2-1-LTR] mobilization rate (i) varies between animals, and (ii) is stable over long periods of time within animal. There was no evidence of an effect of age on ERVK[2-1-LTR] mobilization rate. We detected 260 de novo ERVK[2-1-LTR] mobilization events in these experiments. Their median “dosage” in sperm DNA was 0.00023, or one copy per 4318 sperm cells (range: 0.00018–0.00265). This indicates that the level of mosaicism for these de novo insertions is low, and hence that they mostly occur at late stages of spermatogenesis. Yet, we observed 38 cases where the same insertion was captured both in the young and old samples. This suggests that the corresponding insertions are present in spermatogonial stem cells that maintain the capacity to give rise to spermatogenic waves during the entire life of the animal (Supplementary Data 5).

We then applied the method to semen samples from 430 Belgian Blue (BB) bulls. We captured a total of three fixed ERVK[2-1-LTR] elements (i.e., detected in all studied animals), 306 polymorphic ones (i.e. detected in some but not all animals), and 3669 de novo ERVK[2-1-LTR] insertions in this experiment (Supplementary Data 4 & 6). De novo ERVK[2-1-LTR] elements tended to preferentially insert near telomeric ends and in GC-rich regions (even after correcting for distance from chromosome end). Insertion sites were characterized by a 6-bp duplication and an 8 bp pseudo-palindromic motif¹⁹ (Supplementary Fig. 5). They resided in genic regions more often than expected by chance (p = 0.0004), and – in those – equally often in sense as in antisense orientation (p = 0.40). We observed a paucity of de novo insertions on chromosome X given its size, which is likely due to hemizygosity in males (Supplementary Fig. 6). The estimated ERVK[2-1-LTR] mobilization rate averaged one per 146 sperm cells, yet ranging from zero to one in 25 (Fig. 3c, Supplementary Data 4). There was no evidence of an effect of the bulls’ inbreeding coefficient on ERVK[2-1-LTR] mobilization rate (Supplementary Fig. 4c).

GWAS identifies eight loci affecting ERVK[2-1-LTR] mobilization rate

In order to investigate whether the variance in inter-individual transposition rate has a genetic basis, we determined the whole genome sequence (average 40-fold depth) of 40 of the 430 BB bulls and genotyped them at ~14 million variant positions using GATK²⁰. The remaining 390 bulls were genotyped using a 50 K medium density Illumina array and genotypes augmented to whole genome by imputation. We kept ~10 million variants with minor allele frequency (MAF) ≥ 0.02, together with the PCIP-deduced genotypes (+/+, + /ERV, ERV/ERV) for 87 polymorphic ERVK[2-1-LTR] elements (MAF ≥ 0.02), for further analyses on the complete dataset (Supplementary Data 7). Of note, 80 of 306 polymorphic ERVK[2-1-LTR] elements were singletons, 163 had a population frequency < 0.05 (rare non-singletons), and 63 a frequency ≥ 0.05 (common). We conducted a GWAS for ERVK[2-1-LTR] mobilization rate using a linear model including a fixed variant dosage effect and a random polygenic effect to correct for stratification. We obtained a major, genome-wide significant signal (-log(p) = 29.7) on chromosome 19 (Fig. 4a). We fitted this effect as covariate in the model and repeated the genome scan. This revealed seven additional genome-wide significant (or near) (-log(p) = 7) signals (Fig. 4b).

**Fig. 4: GWAS for the rate of de novo ERVK[2-1-LTR] mobilization in the male germline of cattle.**

Strikingly, a polymorphic ERVK[2-1-LTR] element was either the top variant (chromosome 19) or in high linkage disequilibrium (LD) with it (r² ≥ 0.62; chromosomes 2, 4 and 21) for four of the eight loci (Fig. 4c, Supplementary Fig. 7), a coincidence which is very unlikely to have occurred by chance alone (p < 10⁻⁸; see Methods).

While it may have been expected that inherited ERVK[2-1-LTR] elements promote ERV mobilization rate, this finding raised the question of what differentiates the ERVK[2-1-LTR] elements in the four associated loci from other polymorphic ERVK[2-1-LTR] elements.

Genome-wide landscape of polymorphic ERVK[2-1-LTR] elements in Belgian Blue cattle

This question prompted us to extensively characterize the polymorphic ERVK[2-1-LTR] repertoire in BB cattle. As mentioned before, PCIP coamplified a total of 309 endogenized ERVK[2-1-LTR] elements from the semen samples of the 430 BB bulls. We successfully PCR-amplified and completely sequenced 221 of these (Supplementary Data 8). This revealed two subclades, differing by poorly alignable central fragments of 6.2 (C-type) or 2.7 Kb (D-type), and accounting respectively for 15% and 85% of PCIP-detected ERVK[2-1-LTR] elements (Fig. 5a and Supplementary Fig. 8a). Intact ORFs for GAG, PRO, POL and ENV were present for ~50% of C-type members, hence assumed to be autonomous or Competent, while being absent for all D-type members, hence assumed to be non-autonomous or Defective (Supplementary Fig. 9).

**Fig. 5: Catalogue of polymorphic ERVK[2-1-LTR] elements segregating in the Belgian Blue Cattle population.**

We detected 216 substitutions (i.e. differences between ERVK[2-1-LTR] elements) within the C-specific fragment (of which 60 singletons), 187 substitutions within the D-specific fragment (45 singletons), and 391 substitutions in the 5’ and 3’ flanking regions (123 singletons). There were ~6 times more CpG to TpG and CpG to CpA substitutions than expected, reflecting the increased mutability of methylated cytosines. There were more G-to-A substitutions than C-to-T substitutions at non-CpG sites (p = 0.0007), supporting strand-specific APOBEC3-dependent G-to-A editing²¹ (Fig. 5b, c). The latter was substantiated by the observation of clustered G-to-A singletons for several endogenized ERVK[2-1-LTR] elements (Supplementary Fig. 10a).

We designed probes targeting the insertion sites of 291 polymorphic ERVK[2-1-LTR] elements, added them to Illumina 50 K arrays, and successfully re-genotyped the 430 bulls for 193 loci. Striking discrepancies between the array and PCIP genotypes were observed for at least five loci. Examination of the whole genome sequences (WGS) of the 40 bulls, revealed that this was due to the segregation of a third solo-LTR allele (in addition to “+” and full-length ERV, where “+” corresponds to the ancestral or wild-type allele) for these loci. Further examination of the WGS data with LocaTER output indicated that there exist at least 100 additional polymorphic ERVK[2-1-LTR] elements for which the only two segregating alleles are solo-LTR and “+”, hence never captured by PCIP. The presumed solo-LTR status was confirmed by PCR for 84 tested elements, indicating that no important ERVK[2-1-LTR] subclade (other than C and D) was missed by PCIP (Supplementary Data 9). The distribution of allele frequencies in the 40 sequenced bulls was slightly shifted towards higher values for solo-LTRs when compared to C- and D-type (p = 1.4 × 10⁻⁴ and 2.0 × 10⁻⁸), supporting their older age as expected (Supplementary Fig. 11).

It is well established that recombination between ERV genomes can occur during reverse transcription of the pseudodiploid genomic RNAs (gRNAs)²². We examined the haplotype patterns of the 221 ERVK[2-1-LTR] elements to search for evidence of recombination. When a new variant arises by mutation (f.i. + becomes M), it is found in “complete” association (D' = 1) with the variants (f.i. V₁) constituting the ERV element upon which it occurred: only three of the four possible haplotypes exist across all elements (V₁-M, V₁-+ and +-+). The appearance of the fourth haplotype (+-M) requires a recombination between a gRNA with V₁-M haplotype and one with +-+ haplotype. We first conducted this “four haplotypes” test for all pairs of variants separately for the C and D subclades. There was ample evidence for past recombinations having occurred throughout the ERVK[2-1-LTR] genome both within the C and within the D subclades, indicating that gRNA originating from distinct ERVK[2-1-LTR] elements (albeit from the same subclade) can be co-packaged in the same virus-like particle (VLP) and recombine (Fig. 5d). We then looked for evidence of recombination between C- and D-type ERVK[2-1-LTR] elements. We did this by examining the linkage disequilibrium (LD) between C vs D genotype and all variants in the 5’ and 3’ flanking segments. Here also, there was ample evidence for past recombinations between C-and D-type gRNA indicating that these can also be co-packaged in VLP (Fig. 5e and Supplementary Fig. 10b). LD with C/D genotype was nearly perfect (r² ~ 1) for a set of variants immediately flanking the subclade-specific segments, suggesting that recombination between C and D gRNAs might be hampered in close proximity to the point of C-D divergence, leading to the progressive displacement of the C-D boundaries in a manner reminiscent of the “attrition” observed at mammalian pseudo-autosomal boundaries²³ (Fig. 5e and Supplementary Fig. 8a).

The consensus GAG, POL (reverse transcriptase domain) and ENV (transmembrane domain) sequences of C-type elements each cluster with betaretroviridae (Supplementary Fig. 12a). The ENV protein encompasses a surface unit (SU), proteolytic processing site, and transmembrane unit (TM) with fusion peptide, CX₇C cysteine motif, transmembrane region (TR) and cytoplasmic tail (CT) domains, typical of beta- and lenti-retroviruses²⁴. The amino-terminal end of the GAG protein was predicted by Myristoylator²⁵ to correspond to a myristoylation site (high confidence score of 0.98) (Supplementary Fig. 12b). Taken together this suggests that ERVK[2-1-LTR] mobilize by inter-cellular reinfection²⁶. The consensus primer binding site (PBS), shared by C-type and D-type elements (Supplementary Data 8), presents striking complementarities with the 3’ extremities of the two bovine lysine tRNAs (CTT and TTT codons, respectively), supporting the ERVK denomination (Supplementary Fig. 12c). Observed mismatches might reflect the required balance between adequate tRNA priming for effective reverse transcription yet reduced complementarity to tRNA fragments mediating silencing²⁷.

The number of inherited competent ERVK[2-1-LTR] elements explains > 25% of interindividual variation in male mobilization rate

Analysis of the ERVK[2-1-LTR] catalogue showed that the four ERV elements in the GWAS peaks were all of C-type (p = 0.0008). This suggests that the number of competent ERVK[2-1-LTR] elements in the genome is a major determinant of their mobilization rate. To more accurately evaluate the effect of the number of inherited C-type members on ERVK[2-1-LTR] mobilization rate, we compiled this number for the 430 BB bulls considering not only the four loci identified by GWAS, but all other (rarer and monomorphic) competent elements as well. It ranged from 8 to 19. We tested its effect on ERVK[2-1-LTR] mobilization rate using a linear model. It was highly significant (p ≤ 2.2 × 10⁻¹⁶) and explained 24% of the variation for this trait (Fig. 6a). In striking contrast, the number of non-competent D-type elements, which ranged from 7 to 31, had no effect at all on ERVK[2-1-LTR] mobilization rate (p = 0.87; Fig. 6c).

**Fig. 6: Effect of the number of inherited ERVK[2-1-LTR] elements in the genome on the de novo mobilization rate in the male germline.**

Twenty-five Competent ERVK[2-1-LTR] elements harbor coding variants in at least one of the GAG, PRO, POL and/or ENV ORFs. This includes a stop codon at amino acid position 13 of the POL gene that is carried by the ERVK[2-1-LTR] element coinciding with the GWAS peak on chromosome 4 (Supplementary Data 10). We repeated the analysis of the effect of the number of inherited C elements on the ERVK[2-1-LTR] mobilization rate, considering only “intact” C elements (all are polymorphic). The number of intact C elements in the genome ranged from 0 to 6. The effect on mobilization rate became even stronger, now explaining 34% of the trait variance (Fig. 6b). To more accurately evaluate the effect of the coding variants on the de novo mobilization rate, we tested the effect of the number of such elements (with ORF-disrupting mutation in GAG, PRO, POL or ENV) in the genome (0, 1 or 2), conditional on having 0, 1, 2 or 3 copies of intact Competent elements (without ORF-disrupting mutation in GAG, PRO, POL and ENV). In the absence of any intact Competent element, increasing the number of copies of ERVK[2-1-LTR] elements carrying a coding variant did not increase the mobilization rate. However, if the genome additionally harbored one or more intact Competent ERVK[2-1-LTR] elements, the copies with coding variants affected the de novo transposition rate (p = 1.5 × 10⁻⁴), pointing towards a possible (p = 0.11) epistatic interaction between mutated and intact C-type elements (Supplementary Fig. 13).

Preferential de novo mobilization of defective ERVK[2-1-LTR] elements

The role of competent ERVK[2-1-LTR] elements in driving mobilization rate contrasts with the observation that the ERVK[2-1-LTR] element disrupting the APOB gene, and the five de novo mobilized ERVK[2-1-LTR] elements detected in the Damona pedigree are all of non-competent D-type. To verify whether this was mere coincidence, we selected three BB bulls for which the constitutive C- and D-type ERVK[2-1-LTR] elements could be best discriminated based on the sequence tags obtained by PCIP (Fig. 3a and Supplementary Data 11, 12). The three bulls harbored 26 (bull 1), 34 (bull 2) and 33 (bull 3) inherited (PCIP-detectable, i.e. ignoring solo-LTRs) ERVK[2-1-LTR] elements in their genome, of which 12, 11 and 13 of C-type. PCIP tag information discriminated 13, 17 and 17 haplotypes, of which 2, 2, and 2 included C and D elements, while all others (11, 15, and 15) were either pure C or pure D (Fig. 7a–d). We performed 9, 6 and 9 PCIP experiments per bull, revealing respectively 178, 73, and 147 de novo insertions. We estimated the proportional contribution of the different inherited ERVK[2-1-LTR] elements to the de novo insertions by expectation-maximization (EM)(Methods). In all three bulls, D elements accounted for a significantly (p ≤ 0.01) higher proportion of de novo insertions than of inherited D-type ERVK[2-1-LTR] elements (Fig. 7e–g). This suggests that D elements are able to outcompete C elements despite the fact that the latter are driving de novo mobilization supposedly by trans-complementation²⁸.

**Fig. 7: Preferential mobilization of D-type ERVK[2-1-LTR] elements.**

Strikingly, distinct haplotypes are not contributing equally to de novo events (Fig. 7b–d). For example, the D-only A23 and A33 haplotypes are jointly contributing 64.3% of de novo events while they are accounting for only 10.7% of inherited ERVK[2-1-LTR] elements. Also, the mixed A41 haplotype contributes 31.9% of de novo insertions, although it accounts for only 13.9% of inherited ERVK[2-1-LTR] elements. Of note, the A41 haplotype contributes 95.8% of putative C-type de novo insertions, while only accounting for 50% of constitutive C-type ERVK[2-1-LTR] alleles. Taken together, and even if we conservatively consider that all A41 de novo insertions are of C-type, A23 and A33 D-type de novo insertions are outcompeting A41 ones 2.6 to 1. It is noteworthy that we also observed five de novo insertions whose PCIP-tags didn’t match any of the inherited ERVK[2-1-LTR] elements of the cognate bull, yet could have arisen by recombination.

Amongst the 398 de novo insertions detected in the three bulls, 304 were sampled once, 63 twice, 17 three times, 9 four times, and 5 more than four times (Fig. 8a and Supplementary Data 13). The frequency distribution of resampling rate informs about the developmental window during which de novo ERVK[2-1-LTR] mobilization is most likely to occur (Methods). We compared the real distribution, with the distribution obtained by simulations conducted under various timing scenarios, yet matching the real data for de novo insertion frequency (average number of de novo insertions per sperm cell) and number of explored haploid genomes. The results are compatible with de novo mobilization occurring during a window of ~ 5–9 consecutive cell divisions during the second half (in terms of number of cell division) of spermatogenesis, which may coincide with a period of reduced genome methylation²⁹ (Fig. 8a, b, Supplementary Fig. 14 and Supplementary Data 13). This is in striking contrast with de novo point mutations of which at least 30% were shown to occur during early cleavage embryonic cell divisions in bulls³⁰.

**Fig. 8: Determining the developmental window during which de novo ERVK mobilization occurs from the frequency distribution of resampling rate.**

Self-regulation of the number of ERVK[2-1-LTR] elements in the genome?

It is generally assumed that, after initial expansion, families of ERV elements progressively regress as host defense mechanisms emerge. We expected that GWAS might reveal components of these host defense systems, such as clusters of polymorphic piRNA targeting young ERVs. Yet, we could not identify obvious candidate defensive genes in the vicinity of the four associated loci without ERVK[2-1-LTR] elements (Supplementary Fig. 7). Moreover, the minor alleles (and hence more likely derived allele) in these four loci were systematically increasing rather than decreasing mobilization rate. These findings leave open the question of the nature of ERVK[2-1-LTR] neutralizing mechanisms.

The observation that D-type ERVK[2-1-LTR] elements outcompete C-type ones, suggests an alternative mechanism that may contribute to the demise of ERV families. Indeed, as the number of defective ERV elements takes over the family at the expense of competent copies, which are themselves undergoing the assault of de novo mutations, the family may eventually lose all its competent elements and hence die out. We performed simulations to explore this hypothesis³¹. We assumed (i) that the mobilization rate in the germline of an individual is increasing with the number of competent elements (as observed), (ii) that elements undergo mutations that can only render them defective, but can either increase or decrease their affinity for the mobilization machinery, and (iii) that fitness is decreasing with or independent of the number of ERV elements in the genome (Methods). Under various combinations of parameter values, we systematically observed that defective elements with higher affinity for the mobilization machinery would emerge by mutation, and rapidly displace the competent elements, precluding further expansion of the family. From thereon, the impact on fitness determined the speed at which the family was eventually eliminated from the population (Fig. 9 and Supplementary Fig. 15).

**Fig. 9: Possible contribution of defective yet high affinity elements to the demise of ERV families.**

Discussion

The polymorphic ERVK[2-1-LTR] clade occupies ~400 loci in the bovine genome

In this work, we have performed an in-depth characterization of a specific clade of bovine ERVs referred to by Repbase¹⁵ as ERVK[2-1-LTR]. By combining whole genome and targeted sequencing, we show that members of this clade occupy ~400 loci in BB cattle, and that a large fraction of those are polymorphic in this population. We show that ERVK[2-1-LTR] elements come in three “morphs”: solo-LTR, C(ompetent)-type elements and D(effective)-type elements. The majority of loci are biallelic, i.e. characterized by the wild-type “+” allele and one “ERV” allele which can be of C- (~9% of loci), D- (~47%), or solo-LTR-type (~41%). A minority of loci are triallelic (~2%), characterized by the cosegregation of “+” allele, a full-length (C- or D-type) ERV allele, and the derived solo-LTR allele (Supplementary Fig. 11). Of note, solo-LTRs are expected to be older than their full-length counterparts. Accordingly, their frequency spectrum is shifted upwards. The estimated proportions of C-, D-, and solo-LTR-type loci therefore change with sample size: as sample size increases, more new C- and D-type than solo-LTR-type loci are uncovered, hence their proportion is increasing at the expense of the solo-LTR class.

The ERVK[2-1-LTR] clade derives from ancient endogenization of an unknown exogenous retrovirus

That a high proportion of ERVK[2-1-LTR] loci are polymorphic suggests that their colonization of the bovine genome is a relatively recent event. However, contrary to recent episodes of ERV endogenization reported in poultry, mice, sheep/goat, koala and cat⁵, the identity of the exogenous retrovirus at the origin of the bovine ERVK[2-1-LTR] family remains unknown and may be extinct. Sequences with similarities ≥ 95% over ≥ 95% of the full length of the chromosome 19 C-type element exist in the reference genomes of other domesticated taurine (Bos taurus) and indicine breeds (Bos indicus), gaur (Bos gaurus), gayal (Bos frontalis), domestic yak (Bos grunniens), wild yak (Bos mutus), bison (Bison bison), water buffalo (Bubalus bubalis), and African buffalo (Syncerus caffer). A marked drop in similarity is observed when querying the genomes of caprinae, including sheep (Ovis aries) and goat (Capra hircus) (Supplementary Data 14). This suggests that ERVK[2-1-LTR] endogenization may have occurred in an ancestor of Bovinae, i.e. ~15 million years ago. To gain additional insights in the history of the ERVK[2-1-LTR] clade we compared the shared GAG and ENV sequences between C- and D-type elements. We restricted the analysis to regions immediately flanking the boundaries of the C- and D-specific segments and presenting little or no evidence of recombination (Supplementary Fig. 8). The average pairwise difference between C- and D-type elements was 38.4 in 250 base-pairs. Assuming a de novo mutation rate of 1 × 10⁻⁸ base pairs per generation³⁰, this would correspond to ~15 million generations or ~45 million years. However, this figure doesn’t account for the extra mutations introduced during the undetermined number of reputedly low-fidelity reverse transcription steps that separate the different C- and D-type elements considered in this analysis. It is therefore bound to be overestimated, possibly grossly. We also analyzed the sequence divergence between the 5’ and 3’ LTRs of all C- and D-type elements. Upon creation of a new insertion, 5’ and 3’LTR have identical sequence. Observed differences therefore reflect the accumulation of de novo mutations and therefore presumably provide information about the time of insertion. The full-length ERVK[2-1-LTR] element with the most divergent 5’ and 3’LTRs (98.3% similarity over 1287 base pairs) is a C-type element on chromosome 6, present in the bovine reference genome and fixed in the Belgian Blue cattle population (Supplementary Data 8). Assuming a mutation rate of 1 × 10⁻⁸ base pairs per generation, this level of divergence is expected to accrue over ~ 622,000 generations or ~1.9 million years, hence supporting a more recent time of primordial endogenization of the ERVK[2-1-LTR] clade than the two previous estimates. It should be noted, however, that the eldest ERVK[2-1-LTR] elements, whose 5’ and 3’LTR comparison would best inform about the time of origin of the clade, are more likely to have been reduced to solo-LTRs. This third figure is therefore liable to be an underestimate.

ERVK[2-1-LTR] elements are still actively mobilizing by reinfection of the bovine germline

Approximately 50% of C-type elements have intact ORF for GAG, PRO, POL and ENV, suggesting that ERVK[2-1-LTR] might still be active in the bovine germline. This prediction was further supported by the discovery that the insertion of an ERVK[2-1-LTR] element in the APOB gene underpins cholesterol deficiency in HF cattle. In this work, we unambiguously demonstrate that ERVK[2-1-LTR] are indeed still active in the bovine male and female germline, first by identifying five de novo mobilization events in whole genome sequenced three-generation pedigrees, and secondly by capturing thousands of de novo insertions in sperm cells by means of PCIP¹⁸. By analyzing the degree of mosaicism of captured de novo events, we show that ERVK[2-1-LTR] mobilization occurs in late spermatogenesis, yet can affect stem cells that persist throughout the entire life of the animal. The timing of mobilization could coincide with a phase of spermatogenesis during which genome methylation is at its lowest point^17,29. The observation of an intact ENV ORF in a large proportion of C-type elements, combined with in silico prediction that the matrix domain (MA) of the GAG protein of C-type elements is a target for N-terminal myristoylation (Supplementary Fig. 12), strongly suggests that ERVK[2-1-LTR] elements still multiply by within-host, intercellular reinfection rather than by the supposedly more effective intracellular retro-transposition route. Which cells produce the viral-like particles (somatic or germline cells), which membrane receptor viral-like particles recognize in the recipient cells, and which reinfection path is used (free particles, virological synapses or microtubule-mediated transport) remains unknown.

The rate of ERVK[2-1-LTR] mobilization is determined by the number of inherited C-type elements

On average, ERVK[2-1-LTR] elements mobilize in the male germline at a rate of ~ one in 150 sperm cells. Yet, this rate varies at least ten-fold between individuals, while remaining remarkably constant over time for a given animal (76% repeatability). We show in this work that the individual ERVK[2-1-LTR] mobilization rate of bulls is determined, to a large extent (r² ≈ 0.25), by the number of inherited C-type elements. Intact C-type elements (i.e. with uninterrupted ORF for all four GAG, PRO, POL and ENV genes) have the most consistent effect on mobilization rate (Fig. 6b). However, even intact C elements differ in their effect. For example, an extra copy of the chr19:50466809 ERVK[2-1-LTR] element increases the mobilization rate by 5 events per 1000 sperm cells, while the effects of the chr2:129626969 and chr21:19469667 elements are approximately half or 2.5 events per 1000 sperm cells (Supplementary Fig. 7b). Non-intact C-type elements appear to have genuine, albeit more modest effects on mobilization rate, provided that there is at least one intact C-type element in the genome (Supplementary Fig. 13). Thus, the effect of the number of non-intact C-type elements on mobilization rate depends on the genotype of the animal for intact C-type elements, i.e. there is an epistatic interaction between intact and non-intact C elements with regards to their effect on mobilization rate. This probably indicates that the concentrations of at least some of the four gene products remain a limiting factor in the mobilization process.

De novo mobilizations are dominated by trans-complemented D-type insertions

We further show that – despite the driving role of C-type ERVK[2-1-LTR] elements – de novo mobilization events are dominated by the insertions of specific D elements (Fig. 7). We assume that this reflects (i) the occurrence of trans-complementation between C- and D-type elements, and (ii) a higher affinity of specific D- over C-type elements for the mobilization machinery. We don’t know at this point what the molecular bases of the trans-complementation and differential affinity may be. However, the evidence for pervasive recombination between ERVK[2-1-LTR] elements, including between C- and D-type elements, suggests that C- and D-type gRNAs can pair in the pseudodiploid virus-like-particles and generate recombinant extrachromosomal DNA (ecDNA) molecules²². That D-type insertions are able to outnumber C-type insertions could either be due to the fact that D-type gRNAs are more abundant in cells than their C-type counterparts (f.i. because some of them are transcribed at a higher rate or are more stable), or because they are more effective in utilizing the C-type provided machinery to generate ecDNA³². Another intriguing possibility would be that “heterozygous” virus-like particles harboring a C- and a D-type gRNA would preferentially produce D-type double-stranded extracellular DNA. Trans-complementation between mobile elements has been extensively documented before²⁸. Examples of complementation in trans between morphs of the same ERV clade include ETn by MusD^33,34 and IAP I∆1 by IAP^35,36 in mice, RecKoRV by KoRVA in koala^37,38, and possibly Type 1 by Type 2 HERV-K in human³⁹. Strikingly, the bovine ERVK[2-1-LTR] C/D pair, murine MusD/ETn pair, and koala KoRVA/RecKoRV pair share a pattern of swapping of a central segment of the competent ERV element with an old piece of retroelement, yet conservation of flanking sequences encompassing portions of the GAG and ENV (ERVK[2-1-LTR] and KoRVA) or GAG and POL (MusD) genes (in addition to the LTRs), which may be suggestive of cis-effects on the effectiveness of trans-complementation as reported for IAP I∆1³⁵.

D-type elements may act as parasite-of-parasite gene drives

We expected that GWAS for ERVK[2-1-LTR] mobilization rate would reveal emerging components of the host silencing machinery. Yet, there was no obvious evidence for genes participating in such mechanisms in any of the eight detected association peaks. In fact, the minor allele for all top variants (and hence more likely derived allele) was always increasing (rather than decreasing) mobilization rate. This probably indicates the ERVK[2-1-LTR] endogenization is too recent, and that specific silencing mechanisms are yet to evolve in the bovine population. However, our results reveal another mechanism that may precipitate the demise of ERV families. Indeed, D-type elements may act as parasite-of-parasite gene drives that may cause the spontaneous implosion of the ERVK[2-1-LTR] clade.

Possible phenotypic effects of ERVK[2-1-LTR] activity

We show that ERVK[2-1-LTR] mobilization generates deleterious mutations underpinning genetic defects in the cattle population. As shown by the outcome of knocking down host defense mechanisms against transposable elements¹⁷, excessive ERV mobilization rates may compromise fertility, which is an important concern in livestock breeding. To what extent differences in ERVK[2-1-LTR] mobilization rate correlate with fertility in cattle can now be addressed. Also, whether ERVK[2-1-LTR] mobilization affects the transcriptome and contributes to beneficial variation that can be exploited in breeding programs is another interesting question to pursue.

Methods

Ethics approval

We used biological materials provided by a breeding program, and the biological materials were collected by veterinarians as a part of routine animal breeding activities, not for experiments. All sperm straws were provided by commercial breeding companies. Hence, our work does not involve animal experiments.

The Damona pedigree for the detection of de novo ERVK mobilizations

The Damona pedigree comprises 743 Dutch Holstein-Friesian cattle assigned to 127 three generation pedigrees including a sire, a dam, an offspring, an average of 8 sibs of the offspring (range: 0–17), and an average of 5 grand-offspring (range:1–11). Moreover, 1.4 grandparents per pedigree are available on average as well (5 pedigrees with 4 grandparents, 16 with 3, 38 with 2, 34 with 1, 34 with 0). The 127 pedigrees overlap: for instance, an animal can be offspring in one pedigree and parent in another. All animals from the Damona pedigree were whole genome sequenced (females: blood DNA; males: 24% blood, 76% sperm DNA). The average sequence depth is 26x for sire-dam-offspring trios, 17x for sibs, 10x for grand-offspring and 27.3x for grand-parents.

LocaTER

The LocaTER pipeline scans individual, whole genome sequences (short reads) for insertions of queried interspersed repeats (including ERV elements) that are present in the newly sequenced genome but not in the reference genome to which the reads are mapped (Supplementary Fig. 16). It searches for two features in the sequence data that are characteristic of such insertions. The first is an above normal concentration of discordant paired reads (i.e., the paired ends map to different genome locations) that assort in two adjacent sets. Set A comprises paired reads mapping respectively to the sense strand of the candidate location and to one end of the queried type of repeat (f.i. ERV elements), followed (in the 5’ to 3’ direction on the reference genome) by set B, comprising paired reads mapping respectively to the anti-sense strand of the candidate location and to the other end of the queried repeat. The second characteristic is an above normal concentration of soft- and hard-clipped reads, with (nearly) identical clip position located between sets A and B, and mapping the actual insertion site of the repeat. LocaTER also attempts to determine the genotype of the individual for the corresponding insertion, i.e. heterozygous or homozygous. Predictions of LocaTER were manually checked (using IGV) for a third characteristic of ERV insertions, namely a short, local target site duplication. A more detailed description of the functioning of LocaTER is provided in Supplementary Methods. The LocaTER pipeline is available from https://github.com/Lijingtangbo/TE-rate_manuscript.

Adapting PCIP to quantitatively estimate the rate of TE mobilization

Molecular biology

The modified PCIP reaction comprises six steps (Fig. 3a): (i) using 500 ng of genomic DNA as starting material, fragments containing ERVK[2-1-LTR] sequences were cleaved using a pair of single guide RNAs (Integrated DNA Technologies) targeting sequences at 429 and 374 bp from the 5’LTR and 3’ LTR, respectively (Supplementary Data 15), and S.pyogenes Cas9 (New England Biolabs, M0386S). Of note, these sequences are the same for the C- and D-type elements. (ii) The digested DNA was further mechanically sheared to ~ 3 Kb using a Megaruptor-1 (Diagenode), end-repaired using NEBNext EndRepair Module (New England Biolabs, E6050L), and purified with Agencourt AMPure XP beads (Beckman Coulter, A63881). (iii) The resulting DNA fragments were circularized using T4 DNA Ligase (New England Biolabs, M0202L), residual linear fragments eliminated with Plasmid-Safe-ATP-Dependent DNase (Epicentre, E3110K), purified, and reaction products split in two aliquots. (iv) Circular molecules encompassing ERVK[2-1-LTR] sequences were reopened with S. pyogenes Cas9 using distinct single guide RNAs (Integrated DNA Technologies) for the two aliquots (Supplementary Data 15), and purified. (v) ERVK[2-1-LTR] encompassing linear fragments were inverse PCR amplified using aliquot-specific primer pairs (Supplementary Data 15) with LongAmp Taq DNA Polymerase (New England Biolabs, M0533S), and purified. (vi) The purified amplicons were mechanically sheared to ~350 bp using a Bioruptor-pico (Diagenode), sequencing libraries generated using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, E7645L), and indexed libraries pooled and sequenced on a Novaseq 6000 sequencer (Illumina) targeting ~8 million 150 bp paired-end reads per library.

Data processing

The ensuing sequence reads were demultiplexed, quality assessed using fastQC (https://github.com/s-andrews/FastQC), adapter sequences trimmed using Cutadapt (https://github.com/marcelm/cutadapt), and trimmed reads mapped to the bovine reference genome using BWA-MEM40 and converted to BAM format using SAMtools41. Using a custom-made python script, we first identified clipped reads using CIGAR information. We selected clipped reads with mapping quality ≥40 and a minimum of 10 clipped bases. We then mapped the clipped reads to the segments of the ERVK[2-1-LTR] genome corresponding to the ERV-Tags in Fig. 3 (1 and 2 for the 5’LTR libraries and 3 and 4 for the 3’LTR libraries). We demanded an alignment score ≥0.6 to declare a hit, and labelled the read as either an insertion site (IS) or a shearing site (SS) read. When possible, we extended the alignment with ERVK[2-1-LTR] into non-clipped bases to refine the positions of the SS and IS. We then merged SS and IS site with same “breakpoint”, thereby identifying candidate SS and IS supported by multiple concordant reads. We then paired IS with their cognate SS. The pairing was based on orientation (f.i. a 5’ SS should be located upstream of the 5’ IS for an ERV element in “sense” orientation, and downstream of the 5’ IS for an ERV in “antisense” orientation; Fig. 3) and distance (the maximum distance between IS and SS was set at 5 Kb). One IS can be paired with one SS (typical for non-mosaic de novo insertions) or with multiple SS (typical for mosaic de novo insertions or inherited ERVK[2-1-LTR] elements). Additional filters were then applied to select candidate de novo insertions for manual inspection, including a distance of more than 3 Kb from constitutive ERVK[2-1-LTR] elements, and only observed in one animal.

Data normalization

The above-mentioned procedure yields, for each individual s of T, the number of shearing sites for each i of 123 (5’LTR) or 189 (3’LTR) selected “constitutive” ERVK[2-1-LTR] elements (SSC_is), as well as (at least one) shearing site for N_s de novo ERVK[2-1-LTR] insertions. The SSC_is’s were modeled (with the R lm() function) as \({{SSC}}_{{is}}={{ERV}}_{i}+{{ID}}_{s}+{\varepsilon }_{{is}}\), where ERV_i is the effect of locus i (not all ERVK[2-1-LTR] elements undergo PCIP as effectively), ID_s is the effect of individual s (we don’t engage exactly the same amount of DNA and of same quality in the PCIP reaction for all samples), and ε_is is the error term. We determine the values of ERV_i and ID_s that minimize the error sum of squares. Twice the ensuing ID_s (corresponding to the number of effectively captured haploid genomes for individual s) is used as normalization factor to estimate the individual-specific ERVK[2-1-LTR] mobilization rate as \({{TR}}_{S}={N}_{S}/{2{ID}}_{s}\).

Software

Analysis of the sequence reads was conducted using a mixed Python/R pipeline that uses raw sequence data as input and detects both constitutive and de novo ERV integration sites automatically. The pipeline consists of four modules: the mapping module, junction reads annotation module, clustering module and normalization module. It is available from https://github.com/Lijingtangbo/TE-rate_manuscript.

GWAS for ERVK[2-1-LTR] mobilization rate

SNP/Indel genotyping and imputation

Genomic DNA was extracted from sperm using standard procedures. The genome of 40 of 430 Belgian Blue bulls was whole genome-sequenced using Illumina S4 chemistry and a Novaseq 6000 instrument at average sequence depth of 40, and SNPs/Indels detected and genotyped using GATK²⁰. The remaining 390 animals were genotyped using Illumina’s medium density (~55 K variants) SNP genotyping arrays. For the latter, genotype information was augmented by imputation in two steps (first to high density using 890 Belgian Blue animals genotyped with Illumina’s high density (~770,000 variants) as reference, and then to whole genome using sequenced Belgian Blue animals as reference) using Shapit4⁴⁰ for phasing, followed by Minimac4⁴¹ for actual imputation. We kept 10,875,490 variants with minor allele frequency over 2% and imputation accuracy over 90% for GWAS. Four hundred fifteen of the 430 utilized Belgian Blue animals were homozygous for the nt821(del11) mutation in the myostatin (MSTN) (gene causing double-muscling in homozygotes⁴²), eight heterozygous, and seven homozygous wild-type. MSTN genotype had no effect on ERVK[2-1-LTR] mobilization rate.

Genotyping at 309 constitutive ERVK[2-1-LTR] loci

We applied the modified PCIP procedure to sperm DNA of all 430 Belgian Blue bulls. Three hundred and nine constitutive ERVK[2-1-LTR] loci were identified as loci marked by high numbers of clustered shearing sites (SS in Fig. 3a) relative to the number of explored haploid genomes (see description of PCIP method above) in at least some animals. Animals were genotyped for the corresponding ERVK[2-1-LTR] loci based on the within-locus (to account for the difference in PCIP efficiency for different ERVK[2-1-LTR]) distribution of the ratio between the number of observed SS and number of explored haploid genomes.

GWAS

We conducted GWAS with GEMMA⁴³ using TR_S as phenotype. The model included a fixed regression on variant dosage (additive effect), as well as a random polygenic effect. Estimating the polygenic effects requires the additive genetic relationship matrix, which was computed from marker data using option “-gk1” in GEMMA. In a second round of GWAS, dosage of the chromosome 19 ERVK[2-1-LTR] locus (i.e. number of alleles with the ERVK[2-1-LTR] insertion) was added as an extra covariate in the model. Linkage disequilibrium in the Belgian Blue population results in the fact that the ~11 million variants behave as ~ 500,000 independent tests⁴⁴, yielding a genome-wide significance threshold of 0.05/500,000 = 10⁻⁷.

Testing the effect of inbreeding on mobilization rate

The proportion of autosomal SNPs (MAF ≥ 0.1) with homozygous genotype was used as proxy for a bull’s coefficient of inbreeding. Its effect on mobilization rate was estimated by linear regression using the lm() R function.

Probability of coincidence of polymorphic ERVK[2-1-LTR] elements and association peaks

For four of the eight association peaks identified by GWAS, an ERVK[2-1-LTR] element was either the lead variant (1x), or in high LD (r² ≥ 0.62) with the lead variant (3x). The eight corresponding lead variants were characterized by a minor allele frequency (MAF) in the 430 bulls which we denote f_i. To estimate the probability to observe this level of coincidence or more, fortuitously, we determined, for each lead variant i, the proportion of variants (across the entire genome) with \({f}_{i}-0.025 < {MAF} < {f}_{i}+0.025\) (0.05 bin) that would be in high LD (r² ≥ 0.62) with anyone of the 87 polymorphic ERVK[2-1-LTR] elements with MAF ≥ 0.02: p_i. We then sampled one object in each of eight “urns” (with respective probability of success = p_i), repeated this process 10⁸ times, and counted how often we obtained 4 or more “successes”. The probability to obtain 1, 2, 3 and > 3 successes was 0.009, 3.9 × 10⁻⁵, 7.0 × 10⁻⁸, and zero.

Establishing a catalogue of ERVK[2-1-LTR] elements in BB cattle

Reference sequences for C- and D-type ERVK[2-1-LTR]

We used ERVK[2-1-LTR] present in the bovine reference genome (ARS-UCD1.2 genome assembly) as reference (C-type: chr18-59909149-59919759; D-type: chr14-70257004-70263970). We identified the open reading frames (ORFs) corresponding to the GAG, PRO, POL and ENV genes, bounded on the amino-terminal end by the first ATG for GAG, POL, ENV genes, and by the first non-stop codon in the PRO ORF (Supplementary Fig. 9). The corresponding protein sequences were blasted against ‘viruses’ (taxid: 10239) non-redundant protein sequences to annotate protein domains.

Determining the full-length sequence of 221 constitutive ERVK[2-1-LTR] elements

We aimed at amplifying the full-length of all constitutive ERVK[2-1-LTR] detected by PCIP (see above). This was done by amplifying each ERVK[2-1-LTR] element as two overlapping fragments, jointly spanning its full length. Each primer pair (for the 5’ half and 3’ half of the element, respectively) comprised one primer targeting flanking sequences, and one primer in the body of the ERVK[2-1-LTR] element. As we did not know in advance whether a targeted ERVK[2-1-LTR] element was of C- or D-type, we tested each flanking primer with C-type and D-type specific “ERVK body” primers. We performed long-range PCR using the LongAmp Hot Start Taq 2× Master Mix (New England Biolabs, M0533S) and primers listed in Supplementary Data 15. We successfully amplified both left and right halves for 221 ERVK[2-1-LTR] elements, and one half (either left or right) for an additional 74 for a total of 295. This indicates that there are no other important ERVK[2-1-LTR] elements other than the ~ 6.8 Kb D-type and the ~10 Kb C-type elements described in this work. The left and right amplicons of the 221 ERVK[2-1-LTR] elements were mechanically sheared to ~ 500 bp using a Bioruptor-pico (Diagenode), sequencing libraries generated using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, E7645L) with five cycles of PCR to amplify the adapter-ligated fragments. The quality of the libraries was checked using QIAxcel Advanced System (QIAGEN) and a qPCR. Indexed libraries were pooled and sequenced on a Miseq sequencer (Illumina) targeting 200× coverage. Resulting reads were aligned to cognate (i.e. C- or D- type, 5’ half or 3’ half reference) ERVK[2-1-LTR] references using BWA-MEM⁴². We further checked, for each library, whether the corresponding flanking sequences were mapping to the expected genomic coordinates. Variant sites with respect to the reference sequence were annotated with BCFtools⁴⁵ and manually curated using IGV⁴⁶.

Identifying ERVK[2-1-LTR] loci with solo-LTR alleles

We designed probes specific for the + and ERV allele, respectively, for 291 polymorphic ERVK[2-1-LTR] elements, added them to a medium density (MD) SNP genotyping arrays (Illumina), and re-genotyped the 430 bulls. This yielded usable genotypes for 193 ERVK[2-1-LTR] elements. For at least five loci, bulls were genotyped as ERV/+ with the array, while appearing +/+ by PCIP. This suggested the co-occurrence of a solo-LTR allele for these loci, and this hypothesis was confirmed by PCR amplification of the predicted solo-LTR in at least one bull for each one of the five loci. Confronting the results of PCIP-based genotyping of the 40 whole genome sequenced bulls with the results of the analysis of their genome with LocaTER revealed 100 ERVK[2-1-LTR] insertions detected by LocaTER but never by PCIP. We hypothesized that only solo-LTR alleles would still be segregating in the Belgian Blue population at these loci (i.e. the full-length ERV allele would have been lost). We confirmed the solo-LTR status by PCR amplification and sequencing in at least one bull for 84 of these loci.

Estimating the proportional contribution of PCIP-tag-defined haplotypes of constitutive ERVK[2-1-LTR] elements to de novo insertions

PCIP yields two sequence tags of the captured ERVK[2-1-LTR] element on the 5’-side, and two on the 3’-side (Fig. 3a). Polymorphisms in these tags (amongst the 298 endogenized ERVK[2-1-LTR] elements) allow to distinguish 28 5’-side combinations and 39 3’-side combinations, which assort into 77 5’-3’ combinations or haplotypes, referred to as A1, A2, … A77 in Fig. 7. The number of endogenized ERVK[2-1-LTR] elements sharing the same haplotype ranges from 1 to 76. PCIP typically generates 5’-side or 3’-side tag information for de novo insertions, exceptionally (~4%) both (5’-3’). Also, PCIP may only provide information about one (of the two) 5’- or 3’-side tags. Thus, the assignment of a de novo ERVK[2-1-LTR] insertion to a specific existing haplotype is often ambiguous: multiple haplotypes remain possible given the available tag information. To nevertheless be able to accurately estimate the contribution of each of the haplotypes present in the genome of a bull (i.e. the A41, A44, A37’s, etc. in Fig. 7) to the captured de novo insertions, we used an estimation-maximization (EM) approach. We started assuming uniform contributions of each endogenized haplotype. As an example, the 15 constitutive ERVK[2-1-LTR] elements inherited by bull A (Fig. 7b) fall into 13 haplotypes (A44, A41, A37, A63, A76, A77, A75, A36, A15, A23, A33, A52, A71). The 13 starting contributions were therefore set at 1/13 ( = 0.077). If a de novo insertion “Z” is compatible with, say haplotype A41 and A44, it is assigned for half to A41 (coefficient 0.5 = 0.077⁄(0.077 + 0.077)) and for half to A44 (coefficient 0.5). This process is repeated for all de novo insertions, and the contribution of each haplotype computed from the sum of coefficients across all de novo insertions (divided by the number of insertions). After this first iteration, the contributions will have shifted from uniformity. For instance, to \({p}_{A41}\) for A41 and \({p}_{A44}\) for A44. The coefficients, for insertion “Z”, now become \({p}_{A41}/\left({p}_{A41}+{p}_{A44}\right)\) for A44 and \({p}_{A44}/\left({p}_{A41}+{p}_{A44}\right)\) for A41. The same process is repeated over all insertions. New contributions are then computed from the sum of updated coefficients across de novo insertions. The process is repeated until convergence (~ 20 iterations).

Simulating frequency distributions of resampling rate (1,2, 3,.. times) of de novo insertions matching experimental data

When, during the development of the germline, the ERVK[2-1-LTR] elements are mobilized is an important question. The degree of mosaicism of de novo ERVK[2-1-LTR] insertions provides information about the timing of the mobilization event. If a de novo insertion has a “dosage” of 50% (i.e. detected once every two studied haploid genomes), it supposedly has occurred at the embryonic one cell stage, if 25% it supposedly has occurred at the two-cell stage, etc. If the dosage is ~ 1/1000 and assuming that there are ~ 1 million spermatogonial stem cells⁴⁷, it suggests that it has occurred ten cell divisions before the spermatogonial stem cell stage, yielding a “clone” of ~ 1000 (2¹⁰ = 1024) spermatogonial stem cells sharing the same insertion. If the dosage is ~ 1/1,000,000, it suggests that the mobilization event occurred during the very last cell division(s). A problem is that we only explored between 5,000 and 10,000 haploid genomes for the three bulls for which multiple (6 to 9) PCIP experiments were conducted. Thus, we cannot accurately measure dosages below ~ 1/5,000. Any de novo insertion that is captured once will have an estimated dosage of 1/g, where g is the number of effectively studied haploid genomes (see above). The actual dosage may be much lower; it could just be that this insertion was “by chance” one of those captured out of a very large pool of very rare de novo insertions. Of note, contrary to dosage for individual de novo insertions, which is poorly estimated if low, the number of de novo insertions per explored haploid genome is estimated in an unbiased fashion. Let us call the number of distinct de novo insertions captured (at least once) u, and the total number of captured de novo insertions (i.e. accounting for repeated capture) t. The number of de novo insertions per explored haploid genome is t/g. It was 0.030, 0.019 and 0.024 for bulls 1, 2 and 3, respectively. This is the phenotype that was used for the GWAS (see above).

There is, however, information to determine at which time during development de novo insertions occur, in the frequency distribution of insertions that are re-captured 1, 2, 3, … times. If all observed de novo insertions are only captured once (t = u), it means that there is a large pool of very rare insertions that must all have occurred at a very late stage of spermatogenesis. If most de novo insertions are recaptured multiple times (t ≫ u), it means that de novo insertions occur at earlier stages of spermatogenesis. Thus, there is information about the developmental timing of de novo mobilization in the frequency distribution of the number of de novo insertions captured 1x, 2x, 3x, etc (\({f}_{1},{f}_{2},{f}_{3},\ldots\)). The corresponding distributions are shown in Supplementary Fig. 14 for the three bulls.

To gain insights in the developmental windows during which ERVK[2-1-LTR] mobilization takes place, we performed simulations under various scenarios to find those yielding recapturing frequency distributions that matched the real data best. We assumed a very simple model of gametogenesis in which 1,048,576 spermatogonial stem cells (=2²⁰) would derive from 20 successive binary cell divisions starting from a single precursor cell. A de novo insertion occurring at cell generation 1, would be characterized by a dosage per haploid genome (in sperm) of 0.5. One occurring at cell generation 10, would have a dosage of 0.00097, and at generation 21, a dosage of 4.8 × 10⁻⁷. We defined windows of cell generations during which de novo mobilization may occur. Windows were centered around one specific generation (focal generation), that was most mobilization prone. Windows could be narrow (f.i. occurring in one generation only) or broad (up to 9 generations centered on the focal one). For windows encompassing multiple generations, the mutability of cells in a given generation was modelled using a binomial distribution, i.e., the mutability was highest for the focal generation and decreased with increasing distance from the focal generation. Given a choice of multi-generational window, the probability for a mobilization event to have occurred at one of the spanned generations was not only a function of the mutability of that generation, but also of the number of cells in that generation (which doubles at every additional generation). Having chosen a given window, we simulated de novo mobilization events, spread across the qualifying generations according to their relative probabilities, until the sum of the corresponding dosages corresponded to the number of de novo insertions per explored haploid genome (i.e. \(t/g\)) for the studied bull (1,2 or 3). Every in silico generated insertion corresponded to a ball in an urn with a sampling probability corresponding to its dosage. A ball corresponding to an insertion free genome was added to the urn with a sampling probability of 1-\(t/g\). We then sampled g balls from this urn with replacement, using the corresponding vector of sampling probabilities, where g is the actual number of studied haploid genomes for the three bulls (7690, 5386 and 9178). We then counted the number of non-insertion free balls (i.e. in silico insertions) that were sampled 1x, 2x, 3x, 4x and >4x. We performed 50 simulations per window and compiled the average number of observations for each category (1x, 2x, 3x, 4x and >4x). We compared this distribution with the corresponding real distribution (actual observations for bulls 1, 2 and 3). The quality of the match was quantified as the sum of the differences (absolute value) between simulated and real number of observations. Such simulations were conducted for window sizes ranging from 1 to 9, and sequentially considering every one of the 21 generations as focal one. Simulations were conducted separately by bull. As the results were very consistent for the three bulls, we summed the sum of the differences across bulls and selected the window that minimized the overall differences between simulated and real numbers across the three bulls (Supplementary Data 13). The corresponding window, as well as comparison between real and simulated frequency distributions, are shown in Fig. 8a, b.

Simulating the evolution of ERV families in a panmictic population

We simulated the evolution of a panmictic population of constant size N over G generations. The next generation (\({g}_{i+1}\)) was obtained from the previous one (g_i) by sampling N times (with replacement) two parents (i.e. sex was not considered). The probability to become a parent was affected by the fitness of the individual (f_i, range: 0 – 1). The fitness of the individual was determined by the total number of ERV elements in its genome (t_i), and was modelled as: \({f}_{i}=1-{{t}_{i}}^{p}/{{t}_{\max }}^{p}\). t_max is the maximum number of tolerated ERV elements in the genome, and was set at 250 (if t_i ≥ 250, f_i equals 0). The stringency of purifying selection was adjusted using p: p = 1 (“linear”) > p = 2 (“quadratic”) > p = 3 (“cubic”). In addition, we considered a scenario where fitness was not affected by the number of ERV elements (“none”; f = 1). At generation g₁, the population was “seeded” with one ERV element in the genome of a proportion s of individuals. These initiating ERV elements were considered to be competent (i.e. drive mobilization by producing the needed machinery), and have an affinity (a_ij) of 1 for this machinery. Once the parents of an offspring were selected, we simulated Mendelian transmission of the parental ERV elements to the offspring assuming that each parental ERV has a 50% chance to be inherited by the offspring. In addition, we allowed for de novo ERV mobilization in the gametes. Every parental ERV had a certain probability (q_ij) to generate a new copy of itself in the gamete inherited by the offspring: \({q}_{{ij}}={r}_{i}\times {a}_{{ij}}/{\sum }_{j=1}^{{t}_{i}}{a}_{{ij}}\). In this, r_i is the de novo mobilization rate of parent i. It was determined by the number of competent ERV elements (c_i) in the parent’s genome as follows: \({r}_{i}=\left(1-{e}^{-0.1{c}_{i}}\right)\times {r}_{\max }\), where \({r}_{\max }\) is the highest possible mobilization rate set at 0.05. Once the offspring and its ERV elements were generated, the ERV elements were allowed to undergo mutation at a rate μ. We tested 0.0001 and 0.001 as values for μ. When an ERV element underwent mutation, (i) it would either become defective (if it was competent before) or stay defective (if it was already defective before), and – if the model allowed - (ii) its affinity (a_ij) might change (unless its affinity was already 0 before mutation). This was accomplished by sampling a z value from N(0,1) and adding it to the affinity of the ERV element prior to mutation. If the sum was negative, a_ij was set at 0. The corresponding script was written in Perl and is made available in https://github.com/Lijingtangbo/TE-rate_manuscript.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

All WGS data of the 743 Dutch Holstein Friesian animals are deposited in European Nucleotide Archive under accession PRJEB53518. WGS data for 40 Belgian blue bulls are deposited in European Nucleotide Archive under accession PRJEB64406. PCIP sequencing data of 471 individual experiments generated in this study are available at European Nucleotide Archive under accession PRJEB64406. Complete sequences of the ERVK[2-1-LTR] inserted in APOB, five de novo insertions and 221 polymorphic ERVK[2-1-LTR]s are available from https://doi.org/10.5281/zenodo.7936181. The VCF file used for GWAS is available from https://doi.org/10.5281/zenodo.7937231. Source data are provided as a Source Data file. Source data are provided with this paper.

Code availability

All scripts used in this study are available at GitHub (https://github.com/Lijingtangbo/TE-rate_manuscript) and at Zenodo (https://doi.org/10.5281/zenodo.10630335)^48,49,50.

References

Osmanski, A. B. et al. Insights into mammalian TE diversity via the curation of 248 mammalian genome assemblies. Science 380, eabn1430 (2023).
Article CAS PubMed Google Scholar
Dewannieux, M. & Heidmann, T. Endogenous retroviruses: Acquisition, amplification and taming of genome invaders. Curr. Opin. Virol. 3, 646–656 (2013).
Article CAS PubMed Google Scholar
Mager, D. L. & Stoye, J. P. Mammalian endogenous retroviruses. In Mobile DNA III 1079–1100 (2015).
Johnson, W. E. Origins and evolutionary consequences of ancient endogenous retroviruses. Nat. Rev. Microbiol. 17, 355–370 (2019).
Article CAS PubMed Google Scholar
Chiu, E. S. & Vandewoude, S. Endogenous Retroviruses Drive Resistance and Promotion of Exogenous Retroviral Homologs. Annu. Rev. Anim. Biosci. 9, 225–248 (2021).
Article CAS PubMed Google Scholar
Blanco-Melo, D., Gifford, R. J. & Bieniasz, P. D. Co-option of an endogenous retrovirus envelope for host defense in hominid ancestors. Elife 6, 1–19 (2017).
Article Google Scholar
Lindblad-Toh, K. et al. A high-resolution map of human evolutionary constraint using 29 mammals. Nature 478, 476–482 (2011).
Article CAS PubMed PubMed Central Google Scholar
Feusier, J. et al. Pedigree-based estimation of human mobile element retrotransposition rates. Genome Res. 29, 1567–1577 (2019).
Article CAS PubMed PubMed Central Google Scholar
Zhang, Y., Maksakova, I. A., Gagnier, L., Van De Lagemaat, L. N. & Mager, D. L. Genome-wide assessments reveal extremely high levels of polymorphism of two active families of mouse endogenous retroviral elements. PLoS Genet. 4, e1000007 (2008).
Article PubMed PubMed Central Google Scholar
Elmer, J. L. & Ferguson-Smith, A. C. Strain-specific epigenetic regulation of endogenous retroviruses: The role of Trans-acting modifiers. Viruses 12, 1–21 (2020).
Article Google Scholar
Kipp, S. et al. A new Holstein Haplotype affecting calf survival. Interbull Bull. 49, 49–53 (2015).
Google Scholar
Becker, D. et al. Allele-biased expression of the bovine APOB gene associated with the cholesterol deficiency defect suggests cis-regulatory enhancer effects of the LTR retrotransposon insertion. Sci. Rep. 12, 1–14 (2022).
Article Google Scholar
Menzi, F. et al. A transposable element insertion in APOB causes cholesterol deficiency in Holstein cattle. Anim. Genet. 47, 253–257 (2016).
Article CAS PubMed PubMed Central Google Scholar
Schütz, E. et al. Correction: The Holstein Friesian Lethal Haplotype 5 (HH5) results from a complete deletion of TBF1M and Cholesterol Deficiency (CDH) from an ERV-(LTR) insertion into the coding region of APOB. PLoS One 11, 1–15 (2016).
Google Scholar
Bao, W., Kojima, K. K. & Kohany, O. Repbase Update, a database of repetitive elements in eukaryotic genomes. Mob. DNA 6, 11 (2015).
Article PubMed PubMed Central Google Scholar
Rosenkranz, D. piRNA cluster database: A web resource for piRNA producing loci. Nucleic Acids Res. 44, D223–D230 (2016).
Article CAS PubMed Google Scholar
Zamudio, N. & Bourc’His, D. Transposable elements in the mammalian germline: A comfortable niche or a deadly trap. Heredity (Edinb.). 105, 92–104 (2010).
Article CAS PubMed Google Scholar
Artesi, M. et al. PCIP-seq: simultaneous sequencing of integrated viral genomes and their insertion sites with long reads. Genome Biol. 22, 1–24 (2021).
Article Google Scholar
Kirk, P. D. W., Huvet, M., Melamed, A., Maertens, G. N. & Bangham, C. R. M. Retroviruses integrate into a shared, non-palindromic DNA motif. Nat. Microbiol. 2, 16212 (2016).
Article CAS PubMed PubMed Central Google Scholar
Van der Auwera, G. A. et al. From FastQ Data to High-Confidence Variant Calls: The Genome Analysis Toolkit Best Practices Pipeline. Curr. Protoc. Bioinforma. 43, 11.10.1–11.10.33 (2013).
Google Scholar
Ito, J., Gifford, R. J. & Sato, K. Retroviruses drive the rapid evolution of mammalian APOBEC3 genes. Proc. Natl Acad. Sci. USA. 117, 610–618 (2020).
Article ADS CAS PubMed Google Scholar
Drost, H. G., Sanchez, D. H. & Eyre-Walker, A. Becoming a Selfish Clan: Recombination Associated to Reverse-Transcription in LTR Retrotransposons. Genome Biol. Evol. 11, 3382–3392 (2019).
Article CAS PubMed PubMed Central Google Scholar
Van Laere, A. S., Coppieters, W. & Georges, M. Characterization of the bovine pseudoautosomal boundary: Documenting the evolutionary history of mammalian sex chromosomes. Genome Res. 18, 1884–1895 (2008).
Article PubMed PubMed Central Google Scholar
Greenwood, A. D., Ishida, Y., O’Brien, S. P., Roca, A. L. & Eiden, M. V. Transmission, Evolution, and Endogenization: Lessons Learned from Recent Retroviral Invasions. Microbiol. Mol. Biol. Rev. 82, e00044–17 (2017).
PubMed PubMed Central Google Scholar
Bologna, G., Yvon, C., Duvaud, S. & Veuthey, A. L. N-terminal myristoylation predictions by ensembles of neural networks. Proteomics 4, 1626–1632 (2004).
Article CAS PubMed Google Scholar
Ribet, D. et al. An infectious progenitor for the murine IAP retrotransposon: Emergence of an intracellular genetic parasite from an ancient retrovirus. Genome Res. 18, 597–609 (2008).
Article CAS PubMed PubMed Central Google Scholar
Cullen, H. & Schorn, A. J. Endogenous Retroviruses Walk a Fine Line between Priming and Silencing. Viruses 12, 792 (2020).
Article CAS PubMed PubMed Central Google Scholar
Bannert, N. & Kurth, R. The evolutionary dynamics of human endogenous retroviral families. Annu. Rev. Genomics Hum. Genet. 7, 149–173 (2006).
Article CAS PubMed Google Scholar
Saitou, M. & Miyauchi, H. Gametogenesis from Pluripotent Stem Cells. Cell Stem Cell 18, 721–735 (2016).
Article CAS PubMed Google Scholar
Harland, C. et al. Frequency of mosaicism points towards mutation-prone early cleavage cell divisions in cattle. bioRxiv 79863. https://doi.org/10.1101/079863 (2017).
Katzourakis, A., Rambaut, A. & Pybus, O. G. The evolutionary dynamics of endogenous retroviruses. Trends Microbiol. 13, 463–468 (2005).
Article CAS PubMed Google Scholar
Onafuwa-Nuga, A. & Telesnitsky, A. The Remarkable Frequency of Human Immunodeficiency Virus Type 1 Genetic Recombination. Microbiol. Mol. Biol. Rev. 73, 451–480 (2009).
Article CAS PubMed PubMed Central Google Scholar
Mager, D. L., Freeman, J. D. & Novel Mouse Type, D. Endogenous Proviruses and ETn Elements Share Long Terminal Repeat and Internal Sequences. J. Virol. 74, 7221–7229 (2000).
Article CAS PubMed PubMed Central Google Scholar
Ribet, D., Dewannieux, M. & Heidmann, T. An active murine transposon family pair: Retrotransposition of ‘master’ MusD copies and ETn trans-mobilization. Genome Res. 14, 2261–2267 (2004).
Article CAS PubMed PubMed Central Google Scholar
Saito, E. S., Keng, V. W., Takeda, J. & Horie, K. Translation from nonautonomous type IAP retrotransposon is a critical determinant of transposition activity: Implication for retrotransposon-mediated genome evolution. Genome Res. 18, 859–868 (2008).
Article CAS PubMed PubMed Central Google Scholar
Dewannieux, M., Dupressoir, A., Harper, F., Pierron, G. & Heidmann, T. Identification of autonomous IAP LTR retrotransposons mobile in mammalian cells. Nat. Genet. 36, 534–539 (2004).
Article CAS PubMed Google Scholar
Löber, U. et al. Degradation and remobilization of endogenous retroviruses by recombination during the earliest stages of a germ-line invasion. Proc. Natl Acad. Sci. 115, 8609–8614 (2018).
Article ADS PubMed PubMed Central Google Scholar
Tarlinton, R. E. et al. Differential and defective transcription of koala retrovirus indicates the complexity of host and virus evolution. J. Gen. Virol. 103, (2022).
Costas, J. Evolutionary Dynamics of the Human Endogenous Retrovirus Family HERV-K Inferred from Full-Length Proviral Genomes. J. Mol. Evol. 53, 237–243 (2001).
Article ADS CAS PubMed Google Scholar
Delaneau, O., Zagury, J. F., Robinson, M. R., Marchini, J. L. & Dermitzakis, E. T. Accurate, scalable and integrative haplotype estimation. Nat. Commun. 10, 24–29 (2019).
Article Google Scholar
Das, S. et al. Next-generation genotype imputation service and methods. Nat. Genet. 48, 1284–1287 (2016).
Article CAS PubMed PubMed Central Google Scholar
Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
Article CAS PubMed PubMed Central Google Scholar
Zhou, X. & Stephens, M. Genome-wide efficient mixed-model analysis for association studies. Nat. Genet. 44, 821–824 (2012).
Article CAS PubMed PubMed Central Google Scholar
Gualdrón Duarte, J. L. et al. Sequenced-based GWAS for linear classification traits in Belgian Blue beef cattle reveals new coding variants in genes regulating body size in mammals. Genet. Sel. Evol. 55, 83 (2023).
Article PubMed PubMed Central Google Scholar
Danecek, P. et al. Twelve years of SAMtools and BCFtools. Gigascience 10, 1–4 (2021).
Article CAS Google Scholar
Thorvaldsdóttir, H., Robinson, J. T. & Mesirov, J. P. Integrative Genomics Viewer (IGV): High-performance genomics data visualization and exploration. Brief. Bioinform. 14, 178–192 (2013).
Article PubMed Google Scholar
Mamsen, L. S., Lutterodt, M. C., Andersen, E. W., Byskov, A. G. & Andersen, C. Y. Germ cell numbers in human embryonic and fetal gonads during the first two trimesters of pregnancy: Analysis of six published studies. Hum. Reprod. 26, 2140–2145 (2011).
Article PubMed Google Scholar
Tang, L. GWAS reveals determinants of mobilization rate and dynamics of an active endogenous retrovirus of cattle. https://doi.org/10.5281/zenodo.10630335 (2024).
Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).
Article PubMed PubMed Central Google Scholar
Grobet, L. et al. A deletion in the bovine myostatin gene causes the double-muscled phenotype in cattle. Nat. Genet. 17, 71–74 (1997).
Article CAS PubMed Google Scholar

Download references

Acknowledgements

This project was conducted with funding from the ERC (ERC AdG-GA323030 to M.G., Damona project), the Walloon Region (DOG6 to M.G., Causel project), the University of Liège (FSR to C.C., RetroBlue project) and the Fund for Scientific Research in Belgium (F.R.S.-FNRS, PDR to C.C., TE-rate project); L.T. was supported by a Scholarship of the Chinese Scholarship Council (CSC); C.H.’s PhD fellowship was funded by Livestock Improvement Corporation (LIC, Hamilton, New Zealand); G.C.M.M. is post-doctoral research assistant of the H2020 EU project BovReg (GA815668); C.C. is senior research associate from the Fund for Scientific Research in Belgium (F.R.S.-FNRS). We are grateful to CRV (Arnhem, the Netherlands) for providing us with biological material for the Damona pedigree, to Inovéo (Ciney, Belgium) and FABROCA (Porcheresse, Belgium) artificial insemination centers for contributing Belgian blue sperm straws.

Author information

Authors and Affiliations

Unit of Animal Genomics, GIGA & Faculty of Veterinary Medicine, University of Liège, Liège, Belgium
Lijing Tang, Benjamin Swedlund, Sébastien Dupont, Chad Harland, Gabriel Costa Monteiro Moreira, Keith Durkin, Maria Artesi, Arnaud Sartelet, Latifa Karim, Wouter Coppieters, Michel Georges & Carole Charlier
Keck School of Medicine, University of Southern California, Los Angeles, USA
Benjamin Swedlund
Livestock Improvement Corporation, Hamilton, New Zealand
Chad Harland
Laboratory of Human Genetics, GIGA & Faculty of Medicine, University of Liège, Liège, Belgium
Keith Durkin & Maria Artesi
CRV, Arnhem, The Netherlands
Eric Mullaart
Comparative Veterinary Medicine, FARAH & Faculty of Veterinary Medicine, University of Liège, Liège, Belgium
Arnaud Sartelet
Genomics core facility, GIGA, University of Liège, Liège, Belgium
Latifa Karim & Wouter Coppieters

Authors

Lijing Tang
View author publications
You can also search for this author in PubMed Google Scholar
Benjamin Swedlund
View author publications
You can also search for this author in PubMed Google Scholar
Sébastien Dupont
View author publications
You can also search for this author in PubMed Google Scholar
Chad Harland
View author publications
You can also search for this author in PubMed Google Scholar
Gabriel Costa Monteiro Moreira
View author publications
You can also search for this author in PubMed Google Scholar
Keith Durkin
View author publications
You can also search for this author in PubMed Google Scholar
Maria Artesi
View author publications
You can also search for this author in PubMed Google Scholar
Eric Mullaart
View author publications
You can also search for this author in PubMed Google Scholar
Arnaud Sartelet
View author publications
You can also search for this author in PubMed Google Scholar
Latifa Karim
View author publications
You can also search for this author in PubMed Google Scholar
Wouter Coppieters
View author publications
You can also search for this author in PubMed Google Scholar
Michel Georges
View author publications
You can also search for this author in PubMed Google Scholar
Carole Charlier
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

C.C. and M.G. conceived and supervised the study; L.T. and S.D. performed the wet-laboratory experiments; L.T. and B.S. analysed the data and conducted all bioinformatics analyses; C.H. developed the LocaTER pipeline; G.C.M.M. mapped the WGS to the ARS-UCD1.2 reference genome; K.D. and M.A. gave technical support and conceptual advice; E.M. and A.S. provided pedigree data and biological material; W.C. and L.K. supervised sequencing data generation; M.G., C.C., L.T., and B.S. wrote the manuscript. All authors contributed to the final version of the paper.

Corresponding authors

Correspondence to Lijing Tang, Michel Georges or Carole Charlier.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Stanislaw Kaminski and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Peer Review File

Reporting summary

Description of Additional Supplementary Files

Supplementary Data 1-15

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Tang, L., Swedlund, B., Dupont, S. et al. GWAS reveals determinants of mobilization rate and dynamics of an active endogenous retrovirus of cattle. Nat Commun 15, 2154 (2024). https://doi.org/10.1038/s41467-024-46434-1

Download citation

Received: 29 June 2023
Accepted: 22 February 2024
Published: 09 March 2024
DOI: https://doi.org/10.1038/s41467-024-46434-1

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Subjects

Abstract

Similar content being viewed by others

Introduction

Results

The insertion of an ERVK[2-1-LTR] element in exon 5 of the APOB gene causes cholesterol deficiency in cattle

ERVK[2-1-LTR] elements are active in the bovine germline

Developing a method to measure ERVK[2-1-LTR] mobilization rate in germline and soma

The ERVK[2-1-LTR] mobilization rate in the male germline varies between individuals

GWAS identifies eight loci affecting ERVK[2-1-LTR] mobilization rate

Genome-wide landscape of polymorphic ERVK[2-1-LTR] elements in Belgian Blue cattle

The number of inherited competent ERVK[2-1-LTR] elements explains > 25% of interindividual variation in male mobilization rate

Preferential de novo mobilization of defective ERVK[2-1-LTR] elements

Self-regulation of the number of ERVK[2-1-LTR] elements in the genome?

Discussion

The polymorphic ERVK[2-1-LTR] clade occupies ~400 loci in the bovine genome

The ERVK[2-1-LTR] clade derives from ancient endogenization of an unknown exogenous retrovirus

ERVK[2-1-LTR] elements are still actively mobilizing by reinfection of the bovine germline

The rate of ERVK[2-1-LTR] mobilization is determined by the number of inherited C-type elements

De novo mobilizations are dominated by trans-complemented D-type insertions

D-type elements may act as parasite-of-parasite gene drives

Possible phenotypic effects of ERVK[2-1-LTR] activity

Methods

Ethics approval

The Damona pedigree for the detection of de novo ERVK mobilizations

LocaTER

Adapting PCIP to quantitatively estimate the rate of TE mobilization

Molecular biology

Data processing

Data normalization

Software

GWAS for ERVK[2-1-LTR] mobilization rate

SNP/Indel genotyping and imputation

Genotyping at 309 constitutive ERVK[2-1-LTR] loci

GWAS

Testing the effect of inbreeding on mobilization rate

Probability of coincidence of polymorphic ERVK[2-1-LTR] elements and association peaks

Establishing a catalogue of ERVK[2-1-LTR] elements in BB cattle

Reference sequences for C- and D-type ERVK[2-1-LTR]

Determining the full-length sequence of 221 constitutive ERVK[2-1-LTR] elements

Identifying ERVK[2-1-LTR] loci with solo-LTR alleles

Estimating the proportional contribution of PCIP-tag-defined haplotypes of constitutive ERVK[2-1-LTR] elements to de novo insertions

Simulating frequency distributions of resampling rate (1,2, 3,.. times) of de novo insertions matching experimental data

Simulating the evolution of ERV families in a panmictic population

Reporting summary

Data availability

Code availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding authors

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Comments

Search

Quick links