Host range expansion of asexual parasite can be explained by loss of adaptions in Muller’s Ratchet

Abstract

Sexual recombination is a hallmark of eukaryotic evolution. Without recombination, asexual eukaryotes should succumb to deleterious mutations and more rapidly evolving pathogens. Giardia duodenalis, a parasitic protist, sits within one of the earliest-branching eukaryotic lineages and has no known sexual stage. Whether Giardia are ‘ancient asexuals’ has been long explored but is unresolved. Here, we find clear evidence of sex in Giardia and also discover an asexual sublineage that has a broader host range than its sexual ancestor. This asexual lineage is not ancient, and is accumulating deleterious mutations. Unlike its sexual counterparts, its genetic variation lacks the signatures of selection and Red Queen coevolution. We propose a new hypothesis that explains how a mutational meltdown during Muller’s Ratchet might enable asexual pathogens to expand their host ranges transiently. Fittingly, our results suggest that Giardia is not the last exception to, but rather further evidence of, the essentiality of eukaryotic sex.

Sexual reproduction is thought to be essential for long-term survival of eukaryotes. This study shows that Giardia, once suspected to be anciently asexual, retains evidence of sex while a newly derived asexual lineage is accumulating mutations and expanding its host range.

Introduction

Sex drives eukaryotic survival. However, its ubiquity has long been the subject of intense research^1–3. Sex is costly, particularly in dimorphic species⁴. However, sexual recombination may disrupt beneficial allelic combinations a phenomenon known as ‘recombinational load’^5,6. Clonal offshoots of sexual species often have a significant, albeit short-lived, growth advantage when expanding into new niches⁷. Yet, sex predominates eukaryotic biology. Its benefits must outweigh its costs.

Sex allows eukaryotes to generate diversity through recombination and share advantageous mutations⁸. The Red Queen hypothesis posits that sex is essential for eukaryotic survival in the coevolutionary arms race against rapidly evolving pathogens⁹. Sexual recombination also allows eukaryotes to escape Muller’s Ratchet¹⁰. Indeed, the Red Queen hypothesis and Muller’s Ratchet are leading explanations for the paucity of asexual eukaryotes^8,11,12. Put simply, asexual eukaryotes do not last long.

As with many universal rules, the rulebreakers are the most interesting. Despite the ubiquity of sex, asexual eukaryotes exist, and many are highly successful. Surprisingly, most asexual parasites display wider host ranges than their closest sexual relatives¹³. This seems counterintuitive given that an inability to recombine should disadvantage asexuals by restricting their diversity. The persistence of ancient, asexual eukaryotes is intriguing¹⁴ and considered a ‘great biological scandal’¹⁵ that raises compelling questions. Are ancient, asexual eukaryotes really asexual and ancient? Why do asexual parasites have a broader host range?

Many ‘ancient asexuals’, including oribatid mites¹⁶, darwinulid ostracods¹⁷ and bdelloid rotifers¹⁸ cheat asexuality through parasexuality, ameiotic recombination and automixis, or a reduced mutation rate^19,20. Others are not ancient but arose as recent, short-lived offshoots from a parasexual²¹ or sexual²² trunk. Fittingly, one of the earliest diverging eukaryotic branches, the Excavata²³, includes one of the few unresolved ‘ancient asexual scandals’, Giardia duodenalis²⁴. Giardia are binucleate tetraploids²⁵, important gastrointestinal parasites²⁶ and long explored as ancient eukaryotes^27,28. Many studies sought evidence of sex in G. duodenalis^24,29,30 with conflicting results. No sexual stage has been confirmed for Giardia. G. duodenalis encodes homologs of meiotic genes³¹ and may undergo recombination^32,33, but appears to evolve through predominant clonal evolution (PCE)³⁴. Some G. duodenalis lineages, called ‘assemblages’ and ‘subassemblages’³⁵, differ in allelic divergence, i.e. allelic sequence heterogeneity (ASH)³⁶; possibly due to differing clonally-driven selective sweeps, or differences in their frequency or capacity to recombine^37,38.

Here, we provide evolutionary genomic evidence for sex in the human-specific G. duodenalis subassemblage AII, including a suite of putative virulence genes evolving through Red Queen dynamics. But, we find that zoonotic G. duodenalis subassemblage AI is asexual, albeit not ancient. AI accumulates mutations, yet despite this essential fuel for evolution, lacks signatures of positive selection and Red Queen coevolution. Nevertheless, asexual AI infects more hosts than its sexual AII. We posit a model for how asexual parasites, including G. duodenalis AI, temporarily survive by expanding their host range. We hypothesise that due to inefficient natural selection in the non-recombining asexual, multiple disruptive mutations can accumulate in a single genotype. Such mutation accumulation during Muller’s Ratchet disrupts host-specific adaptations that would have triggered immune recognition in non-hosts. Host-expansion may provide a temporary escape, but the asexual parasite is ultimately driven to extinction by Muller’s Ratchet.

Results

We analysed 107 G. duodenalis genomes (AI = 62 and AII = 45), including 65 newly sequenced isolates (Supplementary Data 1). Six likely mixed infections (Wright’s inbreeding coefficient, F < 0.8) were removed (Supplementary Fig. 1). The remaining 101 formed two major clades (Fig. 1 and Supplementary Figs. 2 and 3a) representing subassemblages AI and AII, with an intermediary cluster from cats, and one water sample. Within these clades, 61 AI isolates were identified across six species (human, cat, pig, beaver, sheep, and dog) from 10 locations, while 40 AII isolates were predominantly found in humans, with a single isolate detected in a beaver, spanning a broader range of 15 geographic locations (Supplementary Data 1).

Fig. 1. Global population structure of Giardia duodenalis Assemblage A illustrating three genetic clusters with different host-specificities.

a Maximum likelihood phylogeny based on whole genome SNPs and 1000 bootstrap replications. Sub-assemblages are represented by solid squares at the tips of the branches, while sources of infection are represented by solid circles at the branch nodes. b Principal Component Analyses (PCA) based on whole genome SNPs. c STRUCTURE plot for the best K = 3. Source data are provided as a Source Data Fig. 1.

Comparative genomics of Giardia duodenalis AI and AII

We identified 184,161 unique single nucleotide polymorphisms (SNPs) among all AI and AII isolates, compared to the reference G. duodenalis WB (AI) genome³⁹. AI and AII were separated by a median of 144,136 SNPs (1.22% of all nucleotides) and display a 100-fold difference (Bartlett’s K-squared = 9543, DF = 1, p-value < 2.2e–16) in within-group variability (Supplementary Fig. 3b): AI isolates differed by 0.0025% (median 292 SNPs), with AII displaying two orders of magnitude greater variability (0.24% variation at a median 27,881 SNPs per genome. In the absence of segregation, the independent accumulation of mutations resulted in the divergence of the two allelic gene copies, i.e. ASH. As a proportion of total SNPs, ASH is significantly more common in AI (Table 1 and Supplementary Fig. 4; T-test, t = 63.372, DF = 54, p-value < 2.2e–16). Furthermore, AI clusters (Fig. 2a) in a ‘near clade’³⁴ topology characteristic of PCE³⁴. In contrast, no nested topology was evident within AII (Fig. 2b). These results are consistent with AI isolates being asexual, accumulating ASH due to the Meselson effect²⁰, but with lower sequence diversity, possibly due to clonal interference, selective sweeps, and/or reduced effective population size (N_e). In summary, the wider geographic distribution, higher order of genetic variability, and deviation from the PCE model of the sexual lineage, along with the absence of the Meselson effect, suggests that AI is an asexual lineage that evolved recently from the sexual lineage AII.

Table 1.

Reporting variation within isolates (SNP counts in AI and AII isolates), within sub-assemblages (pairwise comparison of isolates within AI and AII) and between sub-assemblages (pairwise comparison of isolates between AI and AII)

	AI	AII	Within AI	Within AII	Between AI and AII
Median	333	15,075	292	27,881	144,136
Min	176	7526	37	2620	133,013
Max	652	22,681	802	38,230	148,428

Fig. 2. Evaluating predominant clonal evolution (PCE) and recombination in Giardia duodenalis Assemblage A.

a Maximum likelihood phylogenetic tree of AI isolates demonstrates ‘near-clading,’ one of the PCE features that represents discrete genetic subdivisions within AI sub-assemblage. b Maximum likelihood phylogenetic tree of AII isolates does not demonstrate ‘near-clading’ within the sub-assemblage. c Splitstree network of AI isolates displaying lack of reticulations. d Splitstree illustrating reticulations determined by potential recombination within the network of AII isolates. The numbers on the edges of Splitrstee network illustrates bootstrap values (c, d). e, f A Densitree illustrating a consensus of 10⁷ trees highlighted in dark blue. e Due to the low genetic variation within AI, resolving ancestral nodes was impossible, resulting in the uncertainty highlighted in dark green. f Within the AII consensus tree, red and green lines indicate potential recombination among isolates. g, h Linkage disequilibrium (LD) decay plots for AI and AII, respectively. Each point represents the LD value (r²) on Y-axis for a pair of SNPs plotted against the genomic distance between them on X-axis. g Lack of LD decay among AI isolates. h Rapid LD decay is observed among AII isolates. Source data are provided as a Source Data Fig. 2.

Recombination within Giardia sub-assemblages AI and AII

Splitstree analysis indicated little recombination within AI (Fig. 2c). Densitree (Fig. 2e) could not fully resolve the AI tree, likely due to its low genetic variation. However, the absence of linkage decay among AI SNPs (Fig. 2g) and lack of recombination events identified by RDP4 or HybridCheck (Fig. 3a) suggests AI does not recombine. Recombination results in genetic introgression from a donor lineage causing chromosomal mosaicism, with distinct phylogenetic signals across the chromosome⁴⁰. Conversely, a lack of recombination results in a single phylogenetic signal, which is what we observed within AI (Fig. 3c and Supplementary Fig. 5a).

Fig. 3. Chromosomal mosaicism and hybridisation within chromosome 1 of Giardia duodenalis Assemblage A.

a Comparison of pairwise sequence similarity of AI isolates in HybridCheck analysis showing an absence of introgression. b Multiple peaks of high sequence similarity among a recombinant triplet from AII, indicating multiple events of introgression and representing a mosaic-like chromosome in AII sub-assemblage. c, d Genome-wide sliding window analysis of phylogenetic distribution of three major clades (as highlighted in Fig. 2e, f) in AI and AII, respectively. Top panel illustrates all possible topologies. Bottom panel represents distribution of weights of the three topologies across chromosomes. e, f Distribution of minor allele frequencies (MAF) where allelic ratios of 1:3 and 1:1 are indicated in AI and AII, respectively. Source data are provided as a Source Data Fig. 3.

However, Giardia spp. may engage also in ameiotic intranuclear recombination and diplomixis²⁵. A high rate of ameiotic recombination should generate SNPs with a modal minor allele frequency of ~0.5, consistent with heterogeneity between but homogenisation within nuclei⁴¹. To the contrary, in AI the modal minor allele frequency is ~0.25 (Fig. 3e), consistent with one chromosome within one nucleus having the divergent allele and no interchromosomal recombination. Moreover, if AI is asexual, its divergent allelic haplotypes should evolve independently via the Meselson effect⁴². We see clear evidence of a Meselson effect among AI isolates (Fig. 4a and Supplementary Figs. 6 and 7).

Fig. 4. Allelic divergence and parallel evolutionary analyses between chromosomal haplotypes of Giardia duodenalis Assemblage A.

a, b Haplotype specific maximum likelihood phylogenetic analysis based on whole genome SNPs of chromosome 4. Blue and orange labels represent haplotype 1 and haplotype 2, respectively. a Phylogenetic tree showing the Meselson effect between AI haplotypes, as evidenced by the deep divergence of homologous haplotypes. b Phylogenetic tree showing absence of Meselson effect among AII isolates, consistent with the homogenising effect of recombination. c, d Comparison of the phylogenetic topologies of both haplotypes of chromosome 4 for the AI and AII isolates, respectively. The blue lines connect the same isolates in the haplotype phylogenies. The crossing lines highlight the dissimilarity in topologies of haplotypes in AI (in panel c), whilst the parallel lines show the broad similarity in the topology of haplotypes in AII (in panel d). Source data are provided as a Source Data Fig. 4.

In contrast, the Splitstree network (Fig. 2d), Densitree (Fig. 2f), and the rapid linkage decay among intrachromosomal AII SNPs (Fig. 2h) indicate frequent recombination in AII. Indeed, we identified numerous recombination events in AII (Fig. 3b and Supplementary Data 8), distinct chromosomal mosaicism (Fig. 3d), but found no evidence of a Meselson Effect (Fig. 4b, d and Supplementary Fig. 9) and no modal minor allele peak of ~0.25 for ASH sites in AII (Fig. 3f and Supplementary Fig. 8).

Red Queen evolution in AII drives divergence from AI

We wondered whether differences in sexuality between AI and AII alter the evolutionary model that drives the adaptation of genes involved in antagonistic coevolution with the host. We examined the signatures consistent with an Arms Race, i.e. genes with low genetic diversity (π) under positive selection (Ka/Ks > 1), and signatures consistent with a Red Queen Dynamics model, i.e. genes with high π under balancing selection (Tajima’s D > 0), as a hallmark of sexual recombination⁴³. We hypothesised that without sexual recombination, selection would be much less efficient in AI, and this lineage would lack the signatures of Red Queen coevolution.

We found no AI genes with a signature consistent with an Arms Race, or a Red Queen Dynamics model. We identified two AII genes, GL50803_0016808 (a PAP2 superfamily phosphatase) and GL50803_0041227 (a variant-specific surface protein), that may evolve through an Arms Race model. In addition, 138 AII genes met our stringent criteria for a Red Queen Dynamics model (Supplementary Fig. 12 and Supplementary Data 3). Notably, these 138 genes are highly divergent (Dxy) between AI and AII relative to all other genes (T-test: t = −7.5723, DF = 153, p-value < 2.2e–16) or genes of similarly ranked diversity but under neutral selection (T-test: t = −5.6016, DF = 247, p-value < 2.2e–16). This suggests AII genes evolve through Red Queen Dynamics, driving critical biological differences between AI and AII.

We explored the functions of these 138 Red Queen Dynamics (RQ) genes. Sixty-four are hypothetical proteins. The remainder have diverse functions but are enriched for transmembrane domains relative to the overall genome (binomial test, p = 0.01036). This suggests their enriched expression on Giardia’s cell surface, consistent with a potential role in host-parasite interactions. Supporting this, the 138 genes also have a significantly higher B-cell epitope prediction score (p-value < 2.2e–16) than all other G. duodenalis genes, suggesting that host immune responses may play some role in their evolution.

We asked whether the 138 RQ genes could be virulence genes. Virulence genes often cluster in specific locations within the genome (e.g. telomeres or sub-telomeres) and can be recombination hotspots^44,45. Indeed, the 138 RQ genes are non-randomly distributed in the AII genome (Runs Test: p-value < 2.2e–16) and are enriched (Chi-sq = 16.63, DF = 1, p-value = 4.5421e–05) in telomeres or sub-telomeres (Supplementary Fig. 13). In addition, they are enriched within AII recombinant breakpoints relative to other AII genes (Chi-sq = 29.192, DF = 1, p-value = 6.555e–08). This suggests that RQ genes contribute to ‘genomic islands’ under selection for sharing among AII isolates, possibly conferring an advantage for host-parasite interactions. Such genes should be high-priority candidates for functional studies.

Is Giardia AI a successful ancient, asexual eukaryotic lineage?

Our analyses provide compelling evidence that G. duodenalis AI is asexual. However, multiple lines of evidence suggest that AI is not an ancient lineage. First, phylogenetic analyses indicate that AI and AII diverged relatively recently, with divergence time estimates ranging from ~600 to 2000 years ago based on SNAPP⁴⁶ and TempEst⁴⁷ analyses, respectively. Moreover, recent phylogenetic and comparative genomic studies indicate that AI is a derived clade within the assemblage A, originating after the divergence of earlier lineages such as AIII^48,49. Second, Giardia assemblages including AI, AII, B, and E (https://giardiadb.org/giardiadb/app) share a conserved suite of meiosis-specific genes (e.g. Spo11, Hop1, Rad51)³¹ that are transcriptionally active⁵⁰, suggesting that sexual reproduction is ancestral across G. duodenalis. The most parsimonious interpretation of these findings is that AI descended from a sexually reproducing ancestor rather than representing an ancient asexual lineage. Third, as outlined above, AI isolates exhibit extremely low genetic divergence (0.0025% variation) compared to AII isolates (0.24% variation), supporting that AI is a recently derived lineage with limited time for mutation accumulation. In addition, phylogenetic trees (Fig. 1a and 2a,b) reveal that AI forms a well-supported, short-branched monophyletic group, consistent with recent divergence, whereas the deeper coalescence times and greater internal diversity observed in AII further strengthen the evidence for the recent origin of the AI asexual lineage. Noting that sexual recombination likely evolved once in eukaryotes⁵¹, AI appears to have emerged recently as an asexual population from a sexual AII trunk.

Next, we asked whether AI is successful, noting it infects a broader host range than AII⁵². Reconstructing the demographic histories each sub-assemblage indicates a substantial constriction in AI’s effective population size (N_e) over the last ~200 years compared with AII (Supplementary Fig. 14). This pattern is consistent with a recent clonal expansion followed by a contraction, rather than long-term evolutionary stability.

Finally, we asked whether AI displays a genomic signature consistent with Muller’s Ratchet, indicated by the accumulation of deleterious mutations that cannot be expelled through recombination^10,53. Although defining the phenotypic impact of each SNP in either assemblage is not feasible, the ratio of non-synonymous to synonymous mutations is a commonly used proxy^8,54,55. SNPEff^[ 56 suggests AI has a greater (Chi-sq = 21.65, DF = 1, p-value = 3.268e–6) non-synonymous:synonymous SNP ratio (0.98) than AII (0.59), indicating a higher rate of deleterious mutations. Furthermore, we examined the Meselson effect specifically using only synonymous mutations (Supplementary Fig. 7). The presence of the Meselson effect in AI, even when restricted to synonymous sites, suggests that mutations accumulate indiscriminately across the genome, not just at sites under selective pressure. This implies that the elevated non-synonymous:synonymous ratio in AI is not driven by a reduced synonymous rate but instead reflects reduced purifying selection due to the absence of recombination. In contrast, we did not observe the Meselson effect in AII (Supplementary Fig. 7), reinforcing the role of recombination in mitigating the accumulation of both synonymous and nonsynonymous mutations in sexual lineages.

Additionally, to investigate the accumulation dynamics of nonsynonymous mutations, we examined the non-synonymous:synonymous ratio as a function of the number of samples sharing each SNP (Supplementary Fig. X). In AI, we observed a clear increase in the non-synonymous:synonymous ratio with broader SNP sharing, suggesting that nonsynonymous mutations are not efficiently purged and instead spread through the population. This pattern is consistent with reduced purifying selection. Conversely, in AII, the non-synonymous:synonymous ratio decreased with broader SNP sharing, indicating more effective purging of deleterious mutations through recombination. These findings align with theoretical expectations under asexual reproduction and support our earlier interpretations of the Meselson effect. Thus, we conclude that AI represents a recently evolved asexual lineage experiencing mutational meltdown.

Although the elevated non-synonymous:synonymous ratio does not itself prove that nonsynonymous mutations are deleterious, nonsynonymous mutations are well-established to be more likely functionally impactful, whether beneficial or deleterious than synonymous changes. In populations such as AI that lack recombination, purifying selection becomes less efficient, increasing the probability that slightly deleterious mutations persist⁵⁷. This is especially relevant in populations with smaller effective sizes, where a larger proportion of mutations fall into the “nearly neutral” range.

We also recognise that non-synonymous:synonymous and related ratios can be highly stochastic over short evolutionary timescales, even under neutrality⁵⁸. Furthermore, as discussed by Kryazhimskiy & Plotkin, non-synonymous:synonymous ratios can be relatively insensitive to small selection coefficients over short timescales in population genetic data⁵⁹. These factors underscore the complexity of interpreting non-synonymous:synonymous ratios in recently diverged lineages like AI and AII.

Finally, our findings are consistent with theoretical expectations from Lynch et al.⁶⁰, who proposed that obligate asexual lineages are vulnerable to ‘mutational meltdown’ a process in which deleterious mutations accumulate due to the absence of recombination, eventually reducing fitness and effective population size, and leading toward extinction. Prior studies have also shown that the loss of sexual reproduction is often, though not universally, associated with elevated non-synonymous:synonymous ratios and increased genetic load⁶¹. In line with this framework, our analyses reveal patterns consistent with the early stages of mutational meltdown in the recently derived asexual AI lineage.

Increased pseudogenisation in AI with asexuality

Asexuals exhibit increased rates of pseudogenisation through stop codon mutations^62–64. We identified five AI genes encoding alleles with premature stop codons (Supplementary Data 4) versus none in AII (Binomial Test: p-value = 0.00030). Interestingly, pseudogenised AI alleles occurred at a modal within isolate AF of 0.25, indicating most occurred recently on one chromosome within each tetraploid cell. Pseudogenised alleles were common among AI isolates, with three having an allele frequency >30%, which suggests severe genetic drift or positive selection⁶⁵. We cannot exclude drift because the frequency of the pseudogenised variants that evolved in AI does not fall outside the range of observed frequency changes (mean 0.155 (5–95%CI = 0.011–0.833)). Notably, Splitstree analyses of the AI stop codon alleles are highly shared among isolates, but their overall pattern is not explained by phylogenetic distance (Supplementary Fig. 15a, b). Rather, it appears that stop codons have evolved multiple times in separate AI sublineages. This suggests that the pseudogenisation of certain genes may provide a unique selective advantage in AI.

The most prevalent Loss of Function (LoF) mutation (AF = 0.25) occurred in Argonaute (AGO) alleles that, despite losing 94% of the open reading frame (ORF), occur in 93% of AI but no AII isolates. This suggests a substantial selective advantage in AI specifically. We asked if pseudogenised AGO might hitchhike alongside an (unidentified) beneficial variant, or whether it was present in the origin genotype of AI. We therefore mapped each AGO mutation giving rise to a truncated allele to the overall AI haplotype network (Supplementary Fig. 15c). This indicated two independent LoF events. Both occur deep within AI at AGO codon position 56, leading to a highly truncated ORF. One resulted from the mutation from CCCCG in G. duodenalis WB to CCCCA* and is shared among 58% of AI isolates in our study. The other appears to include multiple sequential mutations from a basal (functional) CCATG to a CCATA* and then CCATC*A. Upon emergence, each AGO LoF mutation is retained with subsequent radiation of each clade.

Recombination between AI and AII

A previous study⁶⁶ reported little evidence of recombination between G. duodenalis AI and AII and proposed them as distinct species. We revisited this using our larger sample size and identified numerous apparent recombination events (Supplementary Data 5). Interestingly, these supported introgression of AI into AII (Fig. 5a and Supplementary Data 5). We also examined the small clade of cat-derived isolates clustered as an intermediary between AI and AII (Fig. 1). This cluster is consistent with a previously described ‘cat-associated’ hybrid clade thought to have evolved by recombination with AI and AII⁶⁷. A multilocus phylogeny inferred using the orthologous gene sequences from the latter study⁶⁷ confirmed this to be the same clade and identified G. duodenalis subassemblage AIII as its nearest neighbour (Supplementary Fig. 16a). RDP4 identified numerous recombinant breakpoints between the ‘cat-derived’ background and AI or AII isolates (Supplementary Data 6). These findings are puzzling. If AI is asexual, how can it engage in recombination? Although we found no evidence of recombination within AI, we further explored the AI-associated recombination with AII and the ‘cat-derived hybrid’ clade.

Fig. 5. Gene conversion between Giardia duodenalis AI and AII.

a The HYBRIDCHECK plot illustrates introgression between AI and AII sub-assemblages, as shown by pairwise sequence alignment of randomly selected minor (donor), major (recipient) and recombinant (hybrid) sequences detected by the RDP4 program on chromosome 1 (Supplementary Data 5)¹⁰⁵. Top panel illustrates the visualisation of sequence similarity between introgressing isolates using RBG color triangle. Chromosome-wide stretches of yellow in the top panel and yellow peaks in the bottom panel, shared between the minor parent (GduoAI-9) and recombinant (CZ2), illustrate the sharing of polymorphisms and the genomic signature of introgression between AI and AII at chromosome 1. b Splitstree network of breakpoint events between AI and AII isolates from chromosome 1 detecting lack of reticulation. c Comparing GC content of genes present in gene conversion events, true recombination events and rest of the genome. Data are presented as median values with interquartile range (IQR). Gene conversion: median = 0.541 [IQR: 0.493–0.602], n = 89; recombination: median = 0.487 [IQR: 0.470–0.513], n = 808; other: median = 0.488 [IQR: 0.469–0.514], n = 4879. Sample sizes (n) refer to the number of genes classified as participating in gene conversion, recombination events, or neither. Statistical significance was assessed using a two-sided t-test. No adjustment for multiple comparisons was applied. d Comparing length of DNA affected by gene conversion and recombination events. Data are presented as median values with IQR. Gene conversion: median = 3279 bases [IQR: 1816–5825], n = 76; recombination: median = 29,506 bases [IQR: 16,048–60,091], n = 55. Sample sizes (n) refer to independent biological events (distinct gene conversion or recombination events). Statistical significance was assessed using a two-sided t-test. No adjustment for multiple comparisons was applied. Source data are provided as a Source Data Fig. 5.

Splitstree analyses (Fig. 5b) of the concatenated breakpoint events predicted by RDP4 (Supplementary Data 5) identified no network reticulations between AI and AII, which is in marked contrast to those observed in AII alone (Supplementary Fig. 17). This strongly refutes true recombination between AI and AII and indicates non-reciprocal gene conversion⁶⁸, with AI donating to AII (Supplementary Fig. 18). Unlike true recombination, non-reciprocal gene conversion without genetic crossover⁶⁹ is consistent with an asexual G. duodenalis AI and might explain why AI is the minor parent (donor) in all AI-AII breakpoint events (Supplementary Data 5).

Gene conversion between AI and AII

Non-reciprocal gene conversion typically occurs within GC-rich regions⁷⁰ in centromeric chromatin, in which crossover events are suppressed⁷¹, and incorporates smaller DNA regions than true recombination⁶⁸. GENECONV⁷² identified 76 (79.2%) AI:AII breakpoint events as likely due to gene conversion (Supplementary Data 5). The 89 genes (Supplementary Data 7) within these events are GC-enriched compared to all other G. duodenalis genes (T-test: t = –6.548, DF = 89, p-value = 3.608e–09) or genes within true recombination breakpoints in AII (T-test: t = 6.7033, DF = 94.202, p-value = 1.486e–09) (Fig. 5c). In addition, although the G. duodenalis centromeres are unidentified, no proposed AI gene conversion events fell within telomeric or sub-telomeric (<60 kb from a telomere) regions (Chi-Square = 4.395, DF = 1, p-value = 0.036). In contrast, these regions were significantly enriched for true recombination events in AII (binomial test, p-value = 0.0001904). The regions encompassed by putative gene conversion events (n = 76) covered 390,975 bp at a median of 3280 bp (IQR-range: 1817–5826 bp) per event in AI. True recombination breakpoints (n = 55) covered 2,117,463 bp at a median of 29,506 bp (IQR-range: 16,048–60,091 bp) per event in AII. This difference was highly significant (T-test: t = -8.1516, DF = 58, p-value = 3.344e–11) (Fig. 5d).

Gene conversion allows some pathogens to evade the host immune response through antigenic variation⁷³. Notably, Variant-Specific Surface Proteins (VSPs) are significantly enriched in the 89-gene set (Chi-sq = 54.95, DF = 1, p-value < 2.2e–16). VSPs are involved in antigenic variation and host interactions in Giardia⁷⁴. Immune evasion may provide the selective pressure driving gene conversion in G. duodenalis. However, VSP genes are scattered throughout the G. duodenalis genome³⁹, and conserved regions within these genes might simply provide the homologous sequences necessary for non-reciprocal gene conversion. Thus, our analyses provide a plausible explanation for prior observations suggesting recombination within AI that does not involve sexual recombination^32,33.

Gene conversion in AI-introgression in the ‘cat-derived hybrid’ clade

RDP4 analysis of the ‘cat-derived hybrid’ clade identified 40 breakpoint events with AI and 8 with AII introgressing into an AIII-like genomic backbone (Supplementary Data 6). This predominance of AI introgression is supported by STRUCTURE (Fig. 1c). The concatenated AII-associated events produced a reticulated Splitstree network consistent with true recombination (Supplementary Fig. 19b). However, the AI-associated events did not (Supplementary Fig. 19a). GENECONV identified 30 AI-associated breakpoint events as consistent with gene conversion (Chi-Sq = 12.96, DF = 1, p-value = 0.0003182). These findings suggest non-reciprocal gene conversion could drive hybrid clade formation, which is common between sexual and asexual lineages^75,76, without true recombination.

A proposed model of how sexuality drives host-specificity

Lastly, we observed an apparent paradox. G. duodenalis AI is asexual and nearly clonal, whereas G. duodenalis AII is sexual and highly variable. However, AII is highly host-specific, and AI is not. Indeed, a relaxation of host specificity is prevalent in asexual parasites compared to their sexual counterparts¹³. Should the more diverse and adaptive lineage not be the more expansive? We explored a simple model that may explain how asexuality leads to a relaxation in host specificity before mutational meltdown leads to lineage extinction.

We hypothesised that in the absence of recombination, asexual lineages become prone to Hill-Robertson Interference and Muller’s Ratchet, leading to the accumulation of multiple deleterious SNPs in genes influencing host-specificity (i.e. ‘virulence’ or ‘effector’ genes) (Supplementary Fig. 20). This loss of specificity leads to a brief period when the asexual lineage can expand its host range. This expansion is caused by LoF mutations in ‘virulence’ genes recognised by other hosts. Although similar LoF mutations must also occur in sexual parasites, we propose that these mutants are rapidly purged so that genotypes with multiple LoF SNPs cannot evolve. In a sexually recombining population, ‘soft selection’ (i.e. density and frequency dependent selection) is more efficient due to the vast amount of variation present in many different recombinant genotypes. Consequently, sexual lineages cannot evolve multiple virulence genes with LoF mutations before they are outcompeted by other genotypes without these mutations. Assuming multiple virulence genes must be dysfunctional to infect a non-host, host range expansion cannot occur under sexual reproduction. Our simple evolutionary genetic computer model shows the purging of LoF mutations in ‘virulence’ genes is less efficient in asexuals (Supplementary Fig. 21), leading to their accumulation (Supplementary Fig. 22). Over time, this loss of host-specific adaptations favours host-expansion in asexuals (Supplementary Fig. 23) before they succumb to Muller’s Ratchet (Supplementary Fig. 24).

This model can explain why we see an accumulation of pseudogenised alleles in asexuals, including G. duodenalis AI, but retention of host-specificity in sexual counterparts. It could also explain how asexual parasites can expand into numerous hosts, and why LoF mutations in G. duodenalis AI are not purged, and increase in frequency. However, we do not know enough about the markers of host-specificity in Giardia to determine whether the observed mutations in these genes relax host specificity. Our model demonstrates that this is plausible and effective in computer simulations, but it requires further empirical validation with more suitable model species.

Discussion

Since early observations that Giardia lacked many seemingly essential eukaryotic organelles⁷⁷, it has attracted interest as an intriguing glimpse into the early origins of eukaryotes⁷⁸. Numerous theories on early eukaryotic evolution have been proposed, sometimes unseating, but most frequently and recently supporting the Excavata, including Giardia, as one of the earliest diverging eukaryotic lineages⁷⁹.

If Giardia is indeed ancient, is it one of last remaining ‘asexual scandals’⁸⁰? This question has persisted as an unresolved ‘thorn in the side’ of eukaryotic evolutionary biologists²⁴. Previous studies produced conflicting and puzzling results^{25,32–34,81}, seemingly showing that sex occurs within the genus, but, to paraphrase Birky et al.²⁴ ‘how much’? Our study concludes that Giardia are both sexual and asexual. However, it is not a question of ‘how much’ but ‘which ones’. G. duodenalis subassemblage AII shows all the genomic hallmarks of engaging in frequent and truly sexual (meiotic) recombination, driving Red Queen Dynamics in a complex suite of genes likely involved in host-parasite coevolution, immune-evasion and virulence.

In contrast, G. duodenalis subassemblage AI appears obligately asexual, displaying the first clear evidence of the Meselson effect within the order Diplomonadida²⁰. Although in vitro experiments suggest recombination within G. duodenalis AI³³, we find no genomic support for this and posit that this observation is driven by non-reciprocal gene conversion. Yet, we show that G. duodenalis AI is only recently diverged from AII. Although Giardia may sit within an ancient eukaryotic lineage, its asexual dalliance is recent. Furthermore, the inability for AI to sexually recombine likely results in genome-wide selective sweeps that wipe out diversity, resulting in a small effective population size (N_e). Despite accumulating mutations, which fuel adaptive evolution, G. duodenalis AI lacks the signatures of host-parasite Red Queen coevolution due to clonal interference and the inefficacy of selection. The lack of sexual recombination leads to an inevitable accumulation of deleterious mutations due to Muller’s Ratchet. Altogether, our findings consistently fit within evolutionary genetic theory, particularly the ancient origins and essentiality of eukaryotic sex. Giardia is basally sexual, and G. duodenalis AI, although asexual, is not ancient and it is likely caught up in a mutational meltdown.

G. duodenalis AI provokes a further intriguing question. How can an obligately asexual lineage undergoing mutational meltdown expand its host range? We think the loss of sexual recombination fundamentally alters Red Queen coevolution. We hypothesise that Muller’s Ratchet drives mutational meltdown in asexual pathogens, making genes that facilitate infection in a given host species increasingly dysfunctional. Although the loss of such virulence genes reduces the efficacy of infection, the asexual lineage can continue to persist as long as it can infect its hosts. In contrast, the loss of a single virulence gene would render a genotype uncompetitive in a sexually reproducing population with high diversity. Hence, the loss of multiple virulence genes is more likely to evolve in the asexual lineage. Furthermore, asexual lineages may initially expand their host range by losing effectors that trigger immune responses in non-host species⁸², thereby increasing their Malthusian fitness, and enabling rapid population growth under ideal conditions⁸³. This expansion allows asexual genotypes to exploit a broader range of hosts than their sexual counterparts. However, as these effector genes become increasingly dysfunctional due to the accumulation of deleterious mutations, asexual lineages suffer reduced fitness on specific hosts, such as humans, where immune evasion relies on functional effectors. In direct competition, sexual genotypes with more intact effector repertoires are favoured through soft selection, ultimately replacing asexual ones. Over time, the continued buildup of harmful mutations leads to mutation meltdown, driving asexual lineages to extinction not only on shared hosts but also on novel hosts through hard selection when they can no longer sustain positive population growth. In support of this hypothesis, we found evidence of LoF mutations at high prevalence and under positive selection in AI.

Moreover, our observation of a common truncation in one of AI’s four AGO alleles is intriguing. In Giardia, AGO-mediated RNA interference regulates switching of VSPs⁸⁴ as a form of antigenic variation and immune evasion⁷⁴. This truncated AGO allele occurs in >90% AI genotypes, but it was not found in any AII isolate. Furthermore, it has emerged at least twice within AI. Could these LoF mutations in AI AGO alleles be of a selective benefit by impacting VSP switching and altering host-parasite interactions? This may support our model but requires testing.

We encourage studies of asexual pathogens that have recently evolved from sexual ancestors to test our hypothesis. Notably, studies on non-neglected pathogens with well-characterised host-specificity markers could help assess whether the pseudogenization of virulence genes during Muller’s Ratchet drives host expansion. Such investigations would shed further light on the Red Queen processes during host-parasite coevolution, the preponderance of asexual pathogens and provide a better understanding of mechanisms driving expansion into new hosts.

Methods

Giardia duodenalis isolates

Isolates CAT-1, GP-1 and VAJ were kindly provided by Dr. G.S. Visvesvara (Centers for Disease Control and Prevention, Atlanta, Georgia, USA). Isolates CZ12, CZ13, CZ16, CZ18, CZ19, CZ2, CZ20, CZ23, CZ24, CZ25, CZ27, CZ28, CZ34, CZ36, CZ43, CZ44, CZ50, CZ51, CZ52, CZ53, CZ9, HJ, KC, LK16, NAX, and BER1 were obtained from Lecová et al.⁸⁵. All isolates were adapted to in vitro growth in standard TYI-S-33 medium, following the procedure described by Bingham and Meyer⁸⁶. Briefly, cysts were separated from stool debris by sucrose gradient centrifugation, suspended in cold water and stored for 2 days at 4 °C. Sterile solutions of 60 mM L-cysteine HCl (Sigma), 60 mM ascorbic acid (Sigma), both in 1× Hanks solution (Sigma H1641), 1 M NaHCO₃ in distilled water, and 0.01 N HCl, were freshly prepared immediately before use.

DNA extraction

Isolates were grown to obtain from 10⁷ to 10⁸ trophozoites, which were harvested by chilling on ice for 10 min and centrifugation at 900 × g for 10 min at +4 °C. Parasite pellets were washed with cold PBS (pH 7.2) and used for DNA extraction. The PureLink^TM genomic DNA Kit (Thermo Fisher Scientific) was used for isolates CZ2, CZ12, CZ13, CZ51, LK16 and HJ. A modified CTAB method⁸⁷ was used for isolates CZ18, CZ25, CZ16, CZ19, CZ23, CZ28, CZ34, CZ43, CZ44, CZ50, CZ52, CZ53 and Ber1. Finally, the Blood & Cell Culture DNA Mini Kit was used for isolates GP-1, CAT-1, VAJ, P2-MER and K188.

Whole genome amplification

Genomic DNA of isolates CZ2, CZ12, CZ13, CZ51, LK16, HJ, GP, CAT, VAJ, P2 and K188 was submitted to whole genome amplification using by Repli-G Midi kit (Qiagen), according to the manufacturer’s instrument, as described⁸⁷.

NGS sequencing

For Next Generation Sequencing experiments, about 1 μg of purified genomic or WGA product per sample was used to generate Illumina Nextera XT 2 ×150 bp paired-end libraries, which were sequenced on an Illumina NovaSeq 6000 SP platform. Library preparation and sequencing were performed by a commercial service (BioDiversa, Italy). Sequencing data generated for this study have been submitted to the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/bioproject/) under the BioProject PRJNA1110996.

Whole genome global data set

We supplemented our newly sequenced genome data set (n = 66) for a global comparative genomics analysis of G. duodenalis assemblage A by downloading all available published Giardia assemblage A genome sequences from the EMBL’s European Nucleotide Archive (see Supplementary Data 1). Collectively, these data represented 107 genome sequence isolates of humans and animals (including beaver, cat, dog, pig and sheep) from 20 countries across the world.

Whole genome dataset pre-processing and variant discovery

Raw reads from 107 genomes were pre-processed with Trimmomatic v.0.36 using the default settings to remove adapter sequences and low-quality nucleotides⁸⁸. Filtered reads were then mapped to the G. duodenalis reference genome (strain: WB C6, BioProject PRJNA1439) using BWA-MEM v.0.7⁸⁹ with default settings. MarkDuplicates was used to mask PCR duplicates (https://broadinstitute.github.io/picard/). Isolates with >80% reference genome coverage and at least fivefold depth were used for variant calling. The HaplotypeCaller module in the GATK v3.7.0⁹⁰ was used to identify SNPs with default hard filtering parameters⁹¹. Further, SNPs with QUAL < 30, allele depth (AD) < 5, MAF < 0.02 and missingness >0.5 were excluded. Additionally, we determined the Fws statistic, a fixation index used to evaluate within-host genetic differentiation, using MOIMIX R package⁹², which is intended to identify outlier isolates with potential for multiple infections. Normally, Fws statistics below 0.95 or 0.90 are indicators of mixed infections in haploid genomes. However, in this case of tetraploid genomes, we have relaxed the cutoff values Fws < 0.8, which is 1.5 standard deviation away from the mean (Supplementary Fig. 1). Finally, the phasing of heterozygous SNPs was done using WhatsHap v1.2.1⁹³ with ploidy=2 to generate haplotype assemblies for each isolate. To evaluate the Meselson effect, two datasets of phased haplotypes were generated based on whole genome SNPs and synonymous SNPs only (https://github.com/bahlolab/moimix), which is intended to identify ‘outlier’ isolates containing multiple assemblages or subassemblages. Finally, the phasing of heterozygous SNPs was done using WhatsHap v1.2.1⁹³ with ploidy set at 2 to generate haplotype assemblies for each isolate. To evaluate the Meselson effect, two datasets of phased haplotypes were generated based on whole genome SNPs and synonymous SNPs only.

Population genetic structure analyses

The filtered set of biallelic SNPs from single infection isolates of G. duodenalis assemblage A was used to construct a phylogenetic tree using FastTree⁹⁴ and IQ-TREE⁹⁵, PCA was performed using SNPRelate v.1.18R package⁹⁶, and STRUCTURE analysis using STRUCTURE v2.3⁹⁷. Best K value from STRUCTURE output was estimated by using evanno method in CLUMPAK⁹⁸. The phylogenetic tree was visualised in iTOL⁹⁹ and STRUCTURE results for the best K value were plotted using Genetic-Structure-Tools (https://github.com/DrewWham/Genetic-Structure-Tools). POPART program was used to construct a haplotype network¹⁰⁰. Genomic windows of positive selective sweeps were detected based on composite likelihood ratio test using SweeD¹⁰¹ with a grid size of 1000.

Demographic analysis

The divergence time between G. duodenalis subassemblages was estimated using a multi-species coalescent model based on SNP data implemented in the SNAPP package⁴⁶ from the BEAST2 program. Changes in the effective population size (Ne) of G. duodenalis assemblage A in recent years were inferred based on SNP frequency spectra using Stairway Plot 2¹⁰². TempEst⁴⁷ was used to estimate the Time to Most Recent Common Ancestor and substitutions per site per year by fitting the root to minimize the mean squared residuals. The divergence time between G. duodenalis subassemblages was estimated using a multi-species coalescent model based on SNP data implemented in the SNAPP package⁴⁶ from the BEAST2 program.

Recombination analyses

We used the DensiTree2 program to construct a consensus of 10⁷ trees and detect any inconsistencies in the topology of the phylogenetic tree of the Giardia assemblage A population that could inform on potential recombination events¹⁰³. We also used the NeighborNet method in SplitsTree6 to detect loops in a phylogenetic split network, which may be caused by recombination¹⁰⁴. We also performed bootstrapping and set a threshold for a minimum bootstrap support of 50% to detect a reliable split. We then used the RDP4 program¹⁰⁵ to identify breakpoints in DNA and recombinant and parental sequences. Breakpoints with p-values ≤10^–5 in at least three RDP4 programs among RDP, GENECONV, Bootscan, Maxchi, Chimaera, SIScan and 3Seq, were considered significant. Furthermore, breakpoints with unknown parental sequences, undetermined positions, and size >100,000 bp were excluded. HYBRIDCHECK¹⁰⁶ program was further used to detect genomic signatures of genetic exchange (introgression) between parental sequences by performing pairwise sequence similarity analyses. Linkage Disequilibrium (LD) analysis was performed using PopLDdecay to determine the rate of LD decay¹⁰⁷. In addition, we performed a sliding window topology weighting analysis to identify genomic mosaicism as evidence of potential recombination. To do this, we first generated trees with window-size 50 and GTR model by using the phyml_sliding_windows.py script from https://github.com/simonhmartin/genomics_general package, followed by computing the weighting of topologies using Twisst¹⁰⁸ and plotting it across chromosomes using R scripts provided in https://github.com/simonhmartin/twisst.

Genomic analyses of the coding region

Population genetic indices, Nucleotide diversity (π), Tajima’s D and Dxy were calculated using the PopGenome R-package¹⁰⁹. Ka/Ks ratio was calculated using Ka/Ks_Calculator¹¹⁰. For genes evolving through an Arms Race, genes having low population diversity (π) and evolving under positive selection (Ka/Ks > 1) were selected¹¹¹. In contrast, for genes evolving under Red Queen Dynamic, genes having high population diversity (π) that is maintained by balancing selection (Tajima’s D > 0) were selected. For added stringency, we only considered genes with Tajima’s D > 0 and being in the top decile ranking for Tajima’s D among all genes in AI or AII, respectively. Annotation of SNPs was performed using the SNPEff program⁵⁶. B-cell epitopes were predicted for each amino acid residue using the BepiPred-3.0 protein language model¹¹². Average predicted scores were calculated for each gene. All statistical tests and results were performed and plotted in R (v.3.6.1).

Modelling host-specificity in sexuals and host-range expansion in asexual lineages

A global macro in Minitab 12.1 was written to illustrate that virulence genes are more likely to become dysfunctional in asexual parasites due to Muller’s Ratchet, and how despite a local loss of fitness on a single host species, the asexual parasite is able to expand its host range. In total, 40 virulence genes were simulated that were under soft selection, such that the probability that a genotype survives and reproduce is dependent on the frequency and density of other genotypes in the population⁸³. Fitness is a function of the number of functional virulence genes (effectors) in the genome. There are 40 virulence genes in the modelled genome, which are initially all functional. If a virulence gene becomes dysfunctional due to a mutation (mutation rate μ = 0.05 per virulence gene), this incurs a 20% fitness loss (s = -0.2). Dysfunctional virulence genes could become functional with a backwards mutation rate μ_back = 0.005 per dysfunctional virulence gene. Fitness is multiplicative across virulence genes, such that, for example, an individual with five dysfunctional virulence genes possesses a genetic load that is equal to one lethal equivalent, with a the fitness equal w = (1–0.2)⁵ = 0.33¹¹³. The number of host species that can be infected is also a function of the number of dysfunctional virulence genes. With the loss of 5 functional virulence genes, the parasite is able to infect one additional host species, and with 10 dysfunctional virulence genes, the parasite is able to two additional host species, etc. All virulence genes recombined freely in the sexual parasites. In the asexual parasite, clonal reproduction created complete linkage between these genes. The population size of adult parasites was set at N = 100, reproducing 5000 offspring per generation.

As output, the mean number of dysfunctional genes was recorded every generation. In addition, the response to selection was recorded, calculated as the difference between the mean number of dysfunctional genes in the total population before selection minus the mean number of dysfunctional genes in the selected individuals. Furthermore, we recorded the mean number of host species that could potentially be infected. The simulations were run for 100 generations and repeated 10 times.

Reporting summary

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

Author contributions

ARJ, RBG, CVO and SMC conceived and guided the overall study. ST led the analysis with support from CVO, ARJ, FW, BB, MN, BREA, SJEC, ML and SS. ARJ, RBG, SMC and EN provided the isolates used in the study. BREA, SJEC, FW and LB supported aspects of the laboratory work underpinning the study. ST, CVO, ARJ and SMC led draft of the manuscript, with additional editorial comments provided by all co-authors.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data analysed in this study, including both newly generated and previously published datasets, are publicly available. Newly generated sequencing data have been deposited in the NCBI Sequence Read Archive (SRA) under the BioProject PRJNA1110996. Previously published datasets are available through the SRA under the SRA accession codes listed in Supplementary Data 1, which also provides direct hyperlinks to each dataset. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-65843-4.