Infant gut anellovirus composition associates with early life factors and childhood atopic disease Authors: Ulrika Boulund, Jonathan Thorsen, Frej Larsen, Casper S. Poulsen, Michael Widdowson, Jie Jiang, Kaare Tranæs, Ling Deng, Morten Arendt Rasmussen, Bo Chawes, Klaus Bønnelykke, Dennis S. Nielsen, Shiraz A. Shah, Jakob Stokholm Published: 2025 ABSTRACT Anelloviruses are single-stranded circular DNA viruses that infect eukaryotic cells. They are acquired in infancy and can be found in most body compartments. The Anelloviridae family is highly prevalent and genetically diverse, but factors shaping the composition and the relationship with disease remain elusive. Here, we characterise faecal viromes of 647 one-year-old children from the deeply phenotyped Copenhagen Prospective Studies on Asthma in Childhood2010 (COPSAC2010) mother-child cohort. In this study, 111 unique anellovirus species are identified and reveal immense prevalence and individual variability. The anellovirus and bacteriophage compositions are slightly correlated, while no correlation is observed with the bacterial composition. Early-life exposures such as attending day care and having older siblings are associated with a reduced anellovirus abundance and diversity, and influence the anellovirus composition. Additionally, genetic secretors of the FUT2 enzyme have a higher anellovirus abundance compared to non-secretors. Minor associations are observed between anelloviruses and later allergic rhinitis, but no associations are found for concurrent acute infections or the development of other atopic traits. Our findings indicate that anelloviruses are a commensal component of all infants’ faecal virome, and further investigations into how immune-training factors influence their composition longitudinally are pertinent. Introduction The gut microbiome is composed of millions of microorganisms, including bacteria and viruses. It is established in infancy and is influenced by both intrinsic and extrinsic factors, such as delivery mode1, nutrition2, host genetics3,4 and environmental exposures5. An increasing body of research is finding links between the gut microbiome and acute and chronic diseases6,7. Current research is largely focused on the bacterial component of the gut microbiome and, to some extent bacteriophages8. However, the members of the highly prevalent Anelloviridae family infect human cells and showcase a considerably high abundance in some individuals. Despite this, it is still largely unknown how these viruses colonise the host, which factors influence this colonisation8, and how they may influence health and disease in childhood. Anelloviruses are single-stranded circular DNA viruses9–11 that infect eukaryotic cells and exhibit an immense genetic diversity12–14. They are highly prevalent, believed to be present in nearly all humans across the earth14,15, and have been found in a multitude of sample types ranging from skin16, saliva17,18 and blood12,13,18, to faeces19–21. Coinfections with anelloviruses of multiple different genotypes in one individual also commonly appear, especially in early life14,22. Anelloviruses are acquired in infancy14,23, potentially already through transplacental transmission, considering that anelloviruses have been identified in umbilical cord blood24–26, amniotic fluid25 and newborns26, though the findings are not entirely concordant8,27. Mothers and their infants shared some, but not all, anelloviruses in the blood8,16. Similarly, adult twins share less faecal anelloviruses than child twins27,28. Several potential sources of transmission have been suggested, such as breast milk24,25 or contact with skin, hair or saliva from infected individuals8,16,29, though there is conflicting evidence8,27. Moreover, birth mode may influence the anellovirus composition, as one study found that children born by spontaneous vaginal delivery exhibited larger gut anelloviral diversity than children born by emergency Caesarean section27,30. Anelloviral richness in blood and faeces has been shown to be low during the first few weeks and months of life, possibly because unknown factors in mother’s milk protect against infection31,32. Subsequently, with weaning, both richness and abundance increase between 6 to 15 months of age23,27,33. Thereafter, it appears that the faecal anelloviral load decreases, demonstrated by a lower faecal anelloviral richness in mothers compared to infants33,34, which could reflect infant immunity development that is partly dependent on maternal antibodies and the host’s own immune maturation33. Due to their high prevalence, anelloviruses are not believed to be causal in acute or chronic disease development, but this remains an intense topic of investigation. Furthermore, it has been speculated that direct interactions with the developing immune system in childhood could potentially affect the risk of developing immune disorders such as asthma and allergy35. On one hand, anelloviruses from several sample types have reportedly been associated with a range of diseases such as IBD36(faeces), fever37(plasma), cancer38(cancer tissue) and asthma35,38,39(airways, plasma). On the other hand, another study reported no association of intestinal anelloviruses with type 1 diabetes development or seroconversion40. In summary, there is a lack of concrete evidence that anelloviruses are causal in disease, and it remains unknown whether certain anelloviral species have more pathogenic potential than others8. In this study, we characterise the infant gut anellovirus composition in 647 children from the deeply phenotyped Copenhagen Prospective Studies on Asthma in Childhood2010 (COPSAC2010) cohort at age 1 year. The faecal viromes were previously sequenced and meticulously catalogued and described, identifying hundreds of novel viral families41. The concurrent bacterial1 and bacteriophage6 compositions have previously been associated with asthma development. With this study, we aim to describe the composition and intrinsic and extrinsic shaping factors of the anelloviruses in early life and its association with infection susceptibility and development of the atopic diseases asthma, eczema and allergy. Results Characterisation of the early-life virome We analysed the sequencing data from DNA viruses extracted from faecal samples of 647 1-year-old children in the COPSAC2010 cohort (Table 1). The early-life faecal virome consisted mainly of bacteriophages, but also contained two families of eukaryotic DNA viruses, namely the adenoviruses and anelloviruses. We have previously presented a deep characterisation of the bacteriophageome41 and its relation to asthma6, as well as a report on the associations of environmental exposures on the overall gut virome42 in the same cohort. Even though the bacteriophages accounted for the vast majority of relative abundance in most samples (median = 97%, range = 7−100%), the anelloviruses were the most abundant eukaryotic viral family (median = 2%, range = 0−93%), while adenoviruses dominated in just a few individuals (median = 0%, range = 0−76%) (Fig. 1A). More than 9000 unique species of bacteriophages were identified, whereas the anelloviruses consisted of 111 unique species (970 viral OTUs or vOTUs) and the adenoviruses only comprised five unique species (Fig. 1B).Fig. 1Characterisation of the early-life gut virome in 647 samples.A Stacked barchart of the relative abundance of bacteriophages, adenoviruses and anelloviruses in each sample. This overall relative viral abundance, which includes all three viral compartments detected in our samples, was used for all further analyses. B Barchart with the number of unique observed taxa at the vOTU, species, subgenus, genus and family level for adenoviruses, anelloviruses and bacteriophages. C Heatmap of the log2 fold change of relative abundance of anelloviruses, adenoviruses and bacteriophages between cases and controls of the acute diseases measures within 7 days of faecal sampling. The asterisk denotes nominal p = 0.047 from a two-sided generalised linear model adjusted for covariates. D Percentage of vOTUs, species, subgenera, and genera of the anelloviruses that are classified as either very rare, rare, common or core.Table 1Cohort characteristicsChild age (mean (SD))372.76 (34.17)BMI mother (mean (SD))24.57 (4.33)Age mother (mean (SD))32.17 (4.30)Atopic mother (%)343 (53.1)Atopic father (%)275 (44.1)Preeclampsy (%)30 (4.6)Vitamin D intervention (%)276 (50.5)Fish oil intervention (%)321 (49.7)Smoking pregnancy (%)46 (7.1)Abx pregnancy (%)235 (36.4)Female (%)310 (47.9)FUT2 secretor = AG/GG (%)475 (82.5)Gestational age (days) (mean (SD))279.00 (11.74)Birth weight (mean (SD))3.54 (0.55)Abx birth (mother) (%)204 (31.7)C-section (%)137 (21.2)Emergency C-section (%)78 (16.8)Birth season spring (%)170 (26.3)Birth season summer (%)136 (21.0)Birth season winter (%)202 (31.2)Birth season autumn (%)139 (21.5)Induction vaginal birth (%)123 (24.2)Exclusive breastfeeding (mean (SD))104.16 (59.40)Prior cow milk exposure (%)615 (95.6)Prior egg exposure (%)532 (83.4)Prior fish exposure (%)617 (95.7)Hypoallergenic formula (%)109 (16.8)Abx ever at 1 year (%)297 (46.3)Any older siblings (%)360 (55.6)Number of older siblings (median [IQR])1.00 [0.00, 1.00]Urban living (%)342 (52.9)Cat or dog (%)222 (34.5)Sampling spring (%)176 (27.2)Sampling summer (%)141 (21.8)Sampling winter (%)182 (28.1)Sampling autumn (%)148 (22.9)Started day care (%)132 (20.5)Gastroenteritis (%)68 (11.0)Cold (%)268 (43.4)Fever (%)185 (29.9)Asthma ever by 6 years (%)140 (22.3)Atopic dermatitis ever by 6 years (%)201 (31.8)Rhinitis by 6 years (%)41 (6.6)BMI body mass index, Abx antibiotic use, FUT2 fucosyltransferase 2. Characterisation of the early-life gut virome in 647 samples. A Stacked barchart of the relative abundance of bacteriophages, adenoviruses and anelloviruses in each sample. This overall relative viral abundance, which includes all three viral compartments detected in our samples, was used for all further analyses. B Barchart with the number of unique observed taxa at the vOTU, species, subgenus, genus and family level for adenoviruses, anelloviruses and bacteriophages. C Heatmap of the log2 fold change of relative abundance of anelloviruses, adenoviruses and bacteriophages between cases and controls of the acute diseases measures within 7 days of faecal sampling. The asterisk denotes nominal p = 0.047 from a two-sided generalised linear model adjusted for covariates. D Percentage of vOTUs, species, subgenera, and genera of the anelloviruses that are classified as either very rare, rare, common or core. Table 1Cohort characteristicsChild age (mean (SD))372.76 (34.17)BMI mother (mean (SD))24.57 (4.33)Age mother (mean (SD))32.17 (4.30)Atopic mother (%)343 (53.1)Atopic father (%)275 (44.1)Preeclampsy (%)30 (4.6)Vitamin D intervention (%)276 (50.5)Fish oil intervention (%)321 (49.7)Smoking pregnancy (%)46 (7.1)Abx pregnancy (%)235 (36.4)Female (%)310 (47.9)FUT2 secretor = AG/GG (%)475 (82.5)Gestational age (days) (mean (SD))279.00 (11.74)Birth weight (mean (SD))3.54 (0.55)Abx birth (mother) (%)204 (31.7)C-section (%)137 (21.2)Emergency C-section (%)78 (16.8)Birth season spring (%)170 (26.3)Birth season summer (%)136 (21.0)Birth season winter (%)202 (31.2)Birth season autumn (%)139 (21.5)Induction vaginal birth (%)123 (24.2)Exclusive breastfeeding (mean (SD))104.16 (59.40)Prior cow milk exposure (%)615 (95.6)Prior egg exposure (%)532 (83.4)Prior fish exposure (%)617 (95.7)Hypoallergenic formula (%)109 (16.8)Abx ever at 1 year (%)297 (46.3)Any older siblings (%)360 (55.6)Number of older siblings (median [IQR])1.00 [0.00, 1.00]Urban living (%)342 (52.9)Cat or dog (%)222 (34.5)Sampling spring (%)176 (27.2)Sampling summer (%)141 (21.8)Sampling winter (%)182 (28.1)Sampling autumn (%)148 (22.9)Started day care (%)132 (20.5)Gastroenteritis (%)68 (11.0)Cold (%)268 (43.4)Fever (%)185 (29.9)Asthma ever by 6 years (%)140 (22.3)Atopic dermatitis ever by 6 years (%)201 (31.8)Rhinitis by 6 years (%)41 (6.6)BMI body mass index, Abx antibiotic use, FUT2 fucosyltransferase 2. Cohort characteristics BMI body mass index, Abx antibiotic use, FUT2 fucosyltransferase 2. Infant gut anelloviruses are taxonomically diverse vOTUs were delineated by virome assembly and deduplication at 95% nucleotide identity, whereafter anelloviral vOTUs were defined based on encoded protein similarity. After aligning the resulting 970 anelloviral vOTUs against anellovirus type species from the ICTV.global (International Committee for the Taxonomy of Viruses website) and constructing an average nucleotide identity (ANI)-based tree (Fig. 2), we found that most (711, 73%) of the vOTUs belonged to the genus betatorquevirus, while alpha- and gammatorquevirus comprised a smaller fraction of the vOTUs. Furthermore, we found that 22 vOTUs formed two genus-level clades of their own that were related to, but separate from alpha- and gammatorquevirus respectively. Here, we propose naming these novel anelloviral genera phitorquevirus and tautorquevirus. They were present in 39% and 49% of the children, respectively (Supplementary Fig. 1A). The 970 anelloviral vOTUs were further grouped into species, using the recent ICTV-ratified anellovirus species cutoff at >35% average nucleotide divergence (100-ANI = AND). This yielded 111 distinct species. In order to bridge the large gap between the more than 100 species and the few genera, we defined anellovirus subgenera by cutting the tree at a depth corresponding to 50%, resulting in 36 anelloviral subgenera, which were named according to the genus to which they belonged. In further analyses, we calculated alpha and beta diversity at the species level; however, focused taxa-specific analyses at the subgenus and genus level for improved interpretability and to reduce the multiple testing burden.Fig. 2Infant gut anelloviruses are taxonomically diverse.Neighbour-joining tree based on pairwise whole-genome alignments of vOTUs and ICTV reference anellovirus genomes. The tree was rooted based on a pig itotatorquevirus reference genome. Reference branches are coloured black. Concentric rings indicate the cut distances used to define anellovirus genera, subgenera and species, indicated by dots. Branches belonging to each of the human anelloviral genera are coloured purple for tau-, red for alpha-, yellow for phi-, green for gamma- and blue for betatorquevirus, respectively. Distances between the species and vOTU levels were truncated to save space, and the truncation is indicated by the outer grey band. Terminal tips are coloured in alternating shades to denote subgenus boundaries within each genus. Infant gut anelloviruses are taxonomically diverse. Neighbour-joining tree based on pairwise whole-genome alignments of vOTUs and ICTV reference anellovirus genomes. The tree was rooted based on a pig itotatorquevirus reference genome. Reference branches are coloured black. Concentric rings indicate the cut distances used to define anellovirus genera, subgenera and species, indicated by dots. Branches belonging to each of the human anelloviral genera are coloured purple for tau-, red for alpha-, yellow for phi-, green for gamma- and blue for betatorquevirus, respectively. Distances between the species and vOTU levels were truncated to save space, and the truncation is indicated by the outer grey band. Terminal tips are coloured in alternating shades to denote subgenus boundaries within each genus. Anelloviruses do not associate with common childhood infections We tested the associations of acute infections (gastroenteritis, cold, and fever) on the total relative abundance of the bacteriophages, adenoviruses and anelloviruses in the COPSAC2010 cohort. Participants with reported gastroenteritis symptoms within 7 days of faecal sampling had a significantly higher total adenovirus relative abundance (log2 fold change = 2.4, p < 0.047, Wilcoxon rank-sum test, Fig. 1C). The total relative abundance of anelloviruses and bacteriophages were not related to any acute symptoms (p > 0.05, Wilcoxon rank-sum test, Fig. 1C). Since the role of adenoviruses on infections is already well established we excluded them from further analysis, and focused on the unexplored anelloviruses. Anelloviruses are unique yet highly prevalent in the 1-year-old infant gut As we demonstrated previously for bacteriophages6,41,43, the anelloviruses were highly unique for each individual at the vOTU level, with 80% of the anellovirus vOTUs characterised as rare (present in 1−20% of participants), and only four vOTUs (0.4%) were part of the core anelloviruses (present in >50% of participants) (Fig. 1D). However, at the species, subgenus and genus level, the anelloviruses were increasingly shared among participants, and all samples contained at least one anellovirus. The anellovirus subgenera ranged in prevalence from 4% of participants (subgenus gamma 3 virus), to the subgenus betatorquevirus 1 (beta 1) being present in 100% of participants (Fig. 3A). The prevalence, median relative abundance and species richness of the anelloviral subgenera and genera were strongly associated (Fig. 3A and Supplementary Fig. 1A). However there were exceptions, such as the subgenus tautorquevirus 1 (tau 1) with a median relative abundance of 0, a prevalence of 49% and consisting of only 1 observed species (Fig. 3A). Interestingly, there did not appear to be a competitive (inverse) relationship among most anelloviruses, and there were some subgenera that were highly positively correlated (Fig. 3B and Supplementary Data 1). The alpha, beta, and gamma genera were all present in 100% of the samples, while the two previously undescribed genera we identified in this study, tautorquevirus and phitorquevirus, were present in 49 and 39% of the children, respectively (Supplementary Fig. 1A). Similar to the subgenera, the anelloviral genera showed largely positive correlations, in particular, the beta and alpha genera were highly correlated with each other, but also gamma and beta were significantly correlated (Supplementary Fig. 1B and Supplementary Data 1). Of the 111 anelloviral species detected in the cohort, each child carried on average 43 unique species (range = 9−86). The most prevalent anellovirus species was the alphatorquevirus 3, which was present in 631 (98%) of the children, whereas the least common species was the phitorquevirus 57, which was only present in 11 (2%) of the children in our cohort. Furthermore, we found that the anellovirus composition was significantly associated with the bacteriophageome (Procrustes, Supplementary Fig. 2A), but it was not associated with the bacterial composition as determined by 16S rRNA gene amplicon sequencing (Supplementary Fig. 2B).Fig. 3The early-life gut anellovirus subgenus composition.A Boxplot of the log10 transformed relative abundance of anelloviral subgenera (relative to the abundance of bacteriophages and adenoviruses as in Fig. 1A), ordered by median relative abundance, if the median relative abundance is below 0 they are ordered by prevalence. The first column to the right represents the prevalence of the subgenus in the samples, and the second column represents the number of unique species of the subgenus. The box represents the 25th and 75th percentiles, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. B Spearman two-tailed correlation plot of anelloviral subgenera. Blue colour represents a negative correlation and red colour represents a positive correlation. The asterisks represent a significant (nominal p value <0.05) correlation. The early-life gut anellovirus subgenus composition. A Boxplot of the log10 transformed relative abundance of anelloviral subgenera (relative to the abundance of bacteriophages and adenoviruses as in Fig. 1A), ordered by median relative abundance, if the median relative abundance is below 0 they are ordered by prevalence. The first column to the right represents the prevalence of the subgenus in the samples, and the second column represents the number of unique species of the subgenus. The box represents the 25th and 75th percentiles, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. B Spearman two-tailed correlation plot of anelloviral subgenera. Blue colour represents a negative correlation and red colour represents a positive correlation. The asterisks represent a significant (nominal p value <0.05) correlation. Extrinsic and intrinsic factors associated with the anellovirus alpha diversity and abundance How anelloviruses colonise the host in early life and what factors influence the anellovirus composition is still not fully known. To shed light on this, we analysed the total anellovirus relative abundance and Faith’s phylogenetic diversity (PD) in relation to several social, environmental, genetic, pre-, peri- and postnatal factors. Using a generalised linear model adjusted for technical covariates, we found several significant links (Fig. 4A and Supplementary Data 2). The pregnant mothers in the COPSAC2010 cohort took part in a randomised placebo-controlled fish oil intervention, which has been described previously44. In the current study, we found that children born to mothers who received the fish oil intervention during pregnancy had a higher total relative anelloviral abundance (model estimate = 1.18 [1.03, 1.35], p = 0.05) and an increased Faith’s PD at 1 year of age (1.16 [1.00, 1.35], p = 0.02). There were no significant associations between high-dose Vitamin D intervention, delivery mode, or breastfeeding duration and anellovirus diversity or abundance. However, samples collected in the summer had a lower total anellovirus abundance and Faith’s PD compared to samples collected in other seasons (abundance: model estimate = 0.78, CI = [0.64, 0.96], p = 0.02, Faith’s PD: 0.80 [0.67, 0.97], p = 0.02). Similarly, children who had started day care before faecal sampling had a lower anellovirus abundance and alpha diversity (abundance: 0.75 [0.62, 0.90], p = 0.002, Faith’s PD: 0.61 [0.52, 0.72], p = 1e-08), and the number of older siblings associated with a lower anellovirus abundance (abundance: 0.92 [0.86, 1.00], p = 0.04).Fig. 4Extrinsic and intrinsic factors associate with the anellovirus composition in 647 samples.A Barchart of the linear model estimate and standard error for each variable in relation to either overall anellovirus relative abundance or anellovirus Faith’s phylogenetic diversity. Asterisks denote nominal p < 0.05. Bars are filled if nominal p < 0.05, and bars are outlined if nominal p > 0.05. B Barchart of the F value from a PERMANOVA on either Bray-Curtis dissimilarity, or Jaccard or Aitchison distance for the anellovirome. Asterisks denote nominal p < 0.05. Bars are filled if nominal p < 0.05, and bars are outlined if nominal p > 0.05. Extrinsic and intrinsic factors associate with the anellovirus composition in 647 samples. A Barchart of the linear model estimate and standard error for each variable in relation to either overall anellovirus relative abundance or anellovirus Faith’s phylogenetic diversity. Asterisks denote nominal p < 0.05. Bars are filled if nominal p < 0.05, and bars are outlined if nominal p > 0.05. B Barchart of the F value from a PERMANOVA on either Bray-Curtis dissimilarity, or Jaccard or Aitchison distance for the anellovirome. Asterisks denote nominal p < 0.05. Bars are filled if nominal p < 0.05, and bars are outlined if nominal p > 0.05. The FUT2 gene encodes the fucosyltransferase 2 enzyme, which determines a person’s histo-blood group antigen secretor status. Noroviruses and rotaviruses have been shown to bind to secreted histo-blood group antigens in the gut mucosa45,46, and non-secretors demonstrate a decreased susceptibility to infection47, thus we hypothesised that there may be a similar relationship for anelloviruses. Analysis of secretor status revealed that children who were secretors (GA or GG allele, N = 475) based on the SNP rs601338 had a significantly higher total relative anelloviral abundance compared to non-secretors (AA allele, N = 101) (1.28 per SD increase in abundance, [1.04, 1.59], p = 0.02), though there was no effect on Faith’s PD. Extrinsic and intrinsic factors associated with the anellovirus beta diversity When comparing all the previously mentioned extrinsic and intrinsic parameters against the overall anellovirus composition using several beta diversity metrics, we again identified significant associations (Fig. 4B and Supplementary Data 3). Some perinatal factors, such as the mother’s antibiotic use at birth (Aitchison distance R2 = 0.2, p = 0.02) and whether the vaginal birth was induced (Aitchison distance R2 = 0.3, p = 0.002, Jaccard distance R2 = 0.4, p = 0.006) were significantly associated with the overall anellovirus composition. Concordant with the results on alpha diversity and overall relative anelloviral abundance, postnatal factors that may be involved in immune system development, such as whether children had started day care before the faecal sampling, is associated with a significantly different anellovirus composition compared to children who had not yet started day care (Bray-Curtis dissimilarity and Jaccard and Aitchison distance R2 = 0.3, 0.5 and 0.2, respectively, all p < 0.05). Extrinsic and intrinsic factors associated with the anellovirus taxa abundances Finally, we associated the relative abundance of anelloviral subgenera and genera with all previously mentioned extrinsic and intrinsic parameters to investigate taxa-specific associations. Here we again found several strong signals in the same direction as for the analyses on overall anelloviral abundance (Fig. 4A). Particularly, being in day care and the number of older siblings were negatively associated with anellovirus abundance across many subgenera, as were several birth-related factors such as sample date, season and induction (Fig. 5 and Supplementary Data 4, 5).Fig. 5Heatmap of the relationship between either binary (two-sided Wilcoxon rank-sum test, colour denotes log2 fold change) or continuous (Spearman correlation, colour denotes Spearman’s rho) variables in relation to the anelloviral subgenera relative abundance (N = 647). Blue colour denotes either a negative log2 fold change or a negative correlation, and red colour denotes either a positive log2 fold change or a positive correlation. Asterisks denote nominal p < 0.05. BMI body mass index, Abx antibiotics, FUT2 fucosyltransferase 2. Heatmap of the relationship between either binary (two-sided Wilcoxon rank-sum test, colour denotes log2 fold change) or continuous (Spearman correlation, colour denotes Spearman’s rho) variables in relation to the anelloviral subgenera relative abundance (N = 647). Blue colour denotes either a negative log2 fold change or a negative correlation, and red colour denotes either a positive log2 fold change or a positive correlation. Asterisks denote nominal p < 0.05. BMI body mass index, Abx antibiotics, FUT2 fucosyltransferase 2. The anellovirus composition is linked to the gut bacterial maturation Due to the numerous findings relating to the decreased anellovirus abundance and diversity with day care start, we employed additional analyses to elucidate this relationship. We found that children who had entered day care had a higher incidence of cold and fever (Chi-squared test, OR 1.63 [1.09−2.47], p = 0.02 and 2.02 [1.27−3.31], p = 0.003), though not gastroenteritis (1.25 [0.67−2.53], p = 0.48). Thus, we considered whether the lower abundance of anellovirus in children in day care was a compositional decrease driven by an increased abundance of other eukaryotic viruses, such as adenovirus. However, we did not find an association between day care attendance and the overall adenovirus abundance. We have previously calculated microbiota-by-age z-scores that capture the maturation of the bacterial component of the gut microbiome of the children in the COPSAC2010 cohort48. There was a slight negative correlation between the gut microbial maturation score and the overall relative anelloviral abundance (Spearman’s correlation rho = −0.08, p = 0.04). The gut anellovirus composition does not associate with the development of childhood atopic disease While the anellovirus composition was not associated with infections in the COPSAC2010 cohort, previous studies have suggested potential links to chronic diseases and immune system development. Thus, we investigated the relationship between the anelloviruses and the development of asthma, allergic rhinitis or atopic dermatitis by age 6 years. The total relative anellovirus abundance and individual anellovirus diversity (alpha diversity, as measured by multiple metrics) was not associated with atopic dermatitis or asthma development (Figs. 6A, B, 7A, B and Supplementary Data 6). However, the anellovirus Shannon index and evenness were lower in children who developed allergic rhinitis by 6 years of age (Shannon index: estimate 0.6 [0.41, 0.98], p = 0.04 and evenness: 0.1 [0.01, 0.67], p = 0.01, Fig. 8A, B and Supplementary Data 6). We delved further into allergic sensitisation to food and inhaled allergens at ages 6 months, 18 months or 6 years. Here we found suggestive results of aeroallergen sensitisation at 18 months associated with evenness (estimate 0.04 [0.00, 1.20], p = 0.067) and Shannon index (0.54 [0.25, 1.09], p = 0.089), and food sensitisation at 6 years associated with inverse Simpson index (1.04 [1.00, 1.08], p = 0.058) (Supplementary Data 6). All other tests on allergic sensitisations were not significant. Likewise, the overall anellovirus composition (beta diversity, as measured by multiple dissimilarity metrics) was not associated with atopic dermatitis or asthma development (PERMANOVA adjusted for technical covariates, Figs. 6C, 7C and Supplementary Fig. 3A, B, 3C, D and Supplementary Data 7). Nevertheless, Bray-Curtis dissimilarity was significantly associated with allergic rhinitis development (F = 1.62, R2(%) = 0.26, p = 0.01, PERMANOVA adjusted for covariates, Fig. 8C and Supplementary Fig. 3E, F, and Supplementary Data 7). When investigating the beta diversity in relation to allergic sensitisation, we found several metrics to be significantly different between cases and controls of any sensitisation at 6 years and aeroallergen sensitisation at 6 years and a suggestive association with aeroallergen sensitisation at 6 months (Supplementary Data 7). All other tests on sensitisation were not significant.Fig. 6The gut anelloviruses do not associate with the development of asthma (N = 647).A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any asthma diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by asthma status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates.Fig. 7The gut anelloviruses do not associate with the development of atopic dermatitis (N = 647).A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any atopic dermatitis diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by atopic dermatitis status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates.Fig. 8The gut anelloviruses do not associate with the development of rhinitis (N = 647).A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any allergic rhinitis diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by rhinitis status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates. The gut anelloviruses do not associate with the development of asthma (N = 647). A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any asthma diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by asthma status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates. The gut anelloviruses do not associate with the development of atopic dermatitis (N = 647). A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any atopic dermatitis diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by atopic dermatitis status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates. The gut anelloviruses do not associate with the development of rhinitis (N = 647). A Boxplots of the relative log10 transformed anellovirus abundance or B Faith’s phylogenetic diversity split by disease status (any allergic rhinitis diagnosis up until the age of 6 years). The box represents the 25th and 75th percentile, the middle line represents the median, and the upper and lower whiskers represent the 5th and 95th percentiles. The nominal p values are from a two-sided generalised linear model adjusted for covariates. C Principal coordinate analysis (PCoA) of the Bray-Curtis dissimilarity, participants are illustrated as points and coloured by rhinitis status. Ellipses illustrate the 90% confidence level. Percentages on the X and Y axis denotes the variance explained by the principal coordinate. R2 and nominal p values are from a PERMANOVA test adjusted for covariates. To investigate the relationship between the gut anellovirus composition and atopic disease at a more detailed level, we tested the differential relative abundance of anelloviral genera, subgenera, and species for all three atopic diseases. None were significantly different (Supplementary Figs. 4A, D, G, 5A, D, G for subgenera and genera, respectively, Supplementary Data 8). In addition to this, we employed an indicator species analysis, aiming to find genera, subgenera, and species that were indicative of cases or controls for all three atopic diseases. None of the results were significant after adjusting for multiple testing (Supplementary Figs. 4B, C, E, F, H, I, 5B, C, E, F, H, I for subgenera and genera, respectively, Supplementary Data 9). Moreover, we employed two separate multivariate methods, namely sparse partial least squares regression (sPLS) and random forest, for investigating the relationship between the faecal anellovirus relative abundance at the species or subgenus level at age 1 year and the development of these atopic diseases. These methods did not yield sufficiently discriminatory results, with a median area under the curve (AUC) for the test set below 0.562 (Supplementary Data 10). Discussion In this study, we demonstrate the high prevalence of anelloviruses in early life, based on 647 faecal samples from 1-year-old children in the COPSAC2010 cohort. Furthermore, we present an in-depth characterisation of anelloviruses, describe two novel anellovirus genera, and demonstrate that the infant gut anellovirus composition is diverse and highly individual at the species level. The total anellovirus relative abundance (relative to the entire virome) was generally lower than the bacteriophageome; however, in 20 (3%) subjects, the anelloviruses accounted for a majority of viral reads. There are no published reports suggesting interactions between anelloviruses and bacteriophages, but we did observe a small and significant correlation between their overall compositions. By contrast, the anellovirus and the bacteriome compositions were not associated. Thus, the anellovirus and bacteriophageome correlation is likely due to a shared environmental niche that influences the two viral compartments in similar directions, or an artefact stemming from the combined processing and sequencing of the bacteriophageome and anelloviruses. Despite the extremely high prevalence of anelloviruses, very little is known about how they are established in the host, as well as the extrinsic and intrinsic factors that are associated with their composition. Here, we link the anellovirus composition to extensive social, environmental, genetic, pre-, peri- and postnatal variables in addition to several concurrent or future disease outcomes. The most consistent and strong effect on the anellovirus composition was whether the child had started day care before the sample collection, which was significantly associated with a reduced anellovirus alpha diversity and abundance, both overall as well as on genus or subgenus level. The association of day care start was also clear on all metrics of anellovirus beta diversity. These findings are contrasting to what has been shown for bacteria and bacteriophages, where such environmental factors are considered seeding sources and, in general, are believed to increase diversity and bacterial abundances42,49. It has previously been shown that anellovirus closely interacts with the human immune system50,51, as exemplified by higher levels in immunocompromised patients8. Our results suggest that immune-training activities such as interaction with other children may reduce anelloviral relative abundance. These findings are in line with hypotheses from previous studies, suggesting that the immune development of infants after waning of maternal antibodies is associated with a decrease in the anelloviral load31. Similarly, several studies suggest that children with a paucity of social interactions have a higher risk of immune-related diseases52. Moreover, a previous study found that children who had entered day care around 1 year of age had an enrichment of bacterial taxa that were more common in older populations53, though this was not seen in children entering day care around the age of 3−4 months54. This indicates that starting day care may have a maturing effect on the gut microbiome, though timing and immune development may be of importance. A previous longitudinal study of eight infants showed a higher anellovirus richness after day care start as compared to before55, contradicting our results of a lower anellovirus richness in children who had started day care. However, these results could be affected by the children’s age, and the children started daycare on average several months earlier than the children in the COPSAC2010 cohort (median 137 and 314 days old, respectively). In our study, we adjusted the analyses for age, and did not see a correlation between anellovirus abundance or alpha diversity and the children’s age, though this could be hampered by the limited age span in our cohort. Appropriately powered longitudinal studies on the development of the anelloviruses are needed to elucidate whether there is a decrease in anelloviral abundance and alpha diversity after day care initiation. Furthermore, we identified a negative association of gestational age with the overall anellovirus abundance, which appeared to be mainly driven by the alpha genus. We hypothesised that starting day care reflects a maturation of the immune system and gut anellovirus composition, based on the above-mentioned findings and previous studies demonstrating that children in day care have lower anelloviral abundance, which is similar to adult levels27. In the present study, we did find that children with higher microbiota-by-age z-scores had lower overall anelloviral abundance, which supports our hypothesis. Overall, our results suggest that a competent immune system reduces the anelloviral load, whereas parameters associated with less immune activation (e.g. no siblings, not started day care, lower gestational age) are associated with a higher anelloviral load. Another possible explanation for the lower anelloviral abundance in children in day care is the competitive effect of other eukaryotic viruses. The term 'superinfection exclusion' refers to the process whereby a viral infection disrupts the potential of subsequent infections by the same or a closely related virus56,57. In this study, we only measured the abundance of the eukaryotic viruses adenovirus and anellovirus, not other common viruses (including RNA viruses), which may be responsible for colds and fever, which we found to be increased in children in day care in the COPSAC2010 cohort. We could not find evidence of superinfection exclusion or a compositional effect between adenoviruses and anelloviruses. However, viral exclusion does not appear to be a process internally among anelloviruses, as several different anelloviral genera were detected in the same samples. Despite this, it is unclear whether superinfection exclusion by other eukaryotic viruses, which were not detected in this study, could result in a lower overall anellovirus abundance or prevalence. A previous report demonstrated a higher anelloviral richness in children born via spontaneous vaginal delivery as compared to acute C-section8,30, which we could not replicate in our data, despite a more than 32 times larger sample size. Additionally, there was no association between C-section (emergency or planned) compared to vaginal delivery (induced or spontaneous) on the anellovirus alpha or beta diversity in the COPSAC2010 cohort. However, there was a significant association between induced vaginal delivery as opposed to spontaneous vaginal delivery and the overall anellovirus composition. Additionally, there were several associations between sample date season to the anellovirus alpha and beta diversity as well as the taxa abundances, where especially summer presented with a lower diversity and overall abundance of anellovirus. As the faecal samples were collected around the age of 1 year, the birth and sample seasons are inherently linked; thus, it is not possible to disentangle individual associations. One could speculate whether anelloviruses follow a seasonal variation, as many other eukaryotic viruses do; however, we did not see any relation between the anellovirus composition and acute infections, which typically follow a seasonal variation. Regarding nutrition, we did not find any substantial associations between the introduction of cow’s milk, eggs, fish or hypoallergenic formula and the overall anellovirus composition or abundance. Similarly, we only found the Jaccard distance and the subgenus betatorquevirus 3 (beta 3) to be positively correlated with breastfeeding duration, which has otherwise been suggested as a potential colonisation mechanism, though some other studies align with our results23,30, and one study demonstrated low abundance of anelloviruses in breastmilk58. Interestingly, several factors that are known to be strongly associated with the bacterial composition such as urbanicity5 and pets5,59,60 were not associated with the anelloviral genus or subgenus abundances. We demonstrated in a previous report42 that the overall viral richness was increased in the children born to mothers who received a fish oil intervention during pregnancy, which aligns with the results we demonstrate in the current study, specifically investigating anelloviruses. The FUT2 gene encodes the fucosyltransferase 2 enzyme, which is highly expressed in the gastrointestinal tract and influences the secretion of histo-blood group antigens. People with the AA allele encode a non-functional enzyme and are non-secretors, protecting them from norovirus44 and rotavirus infections due to the lack of a binding site to known spike proteins45,46, which led us to hypothesise that secretor status might also impact anelloviral load. We demonstrate a significantly lower total relative anelloviral abundance in non-secretors as compared to secretors, which, as far as we are aware, has not been shown before. No unique genus or subgenus of anellovirus was associated with secretor status, pointing to a general association. The exact mechanism is currently unknown, as the functional aspects of the anelloviral genome are still under intense research. There are, for example, several studies investigating anellovirus proteins that allow for immune system evasion61,62, which are important to elucidate the relationship between this commensal member of the virome and the human host. We did not find any substantial associations of anelloviruses with infections (fever, cold, gastroenteritis), nor consistent associations with the development of childhood atopic disease, despite the employment of a multitude of robust methodologies, including univariate and multivariate statistics and data-driven machine learning. We presented the association between the bacteriophageome and asthma development in a previous study6, demonstrating that the findings presented in the current study would be sufficiently powered to yield reliable negative results. There were a few signals from the overall analyses on alpha and beta diversity linking the anellovirus composition to the development of allergic rhinitis by 6 years of age, though nothing passed adjustment for multiple testing at individual species, subgenus or genus abundances. We further analysed allergic sensitisation, and found a few significant associations of the beta diversity to aeroallergen and any sensitisation at six years of age. Overall, our results suggest that the faecal anelloviruses are a largely commensal part of the microbiome and are not involved in infection susceptibility nor the development of immune-mediated disease in early life, such as asthma or atopic dermatitis, though a possible link to allergic rhinitis cannot be dismissed. In summary, we present an in-depth and comprehensive characterisation of the 1-year-old faecal anelloviruses in 647 children, possibly the largest sample size published to date. We analysed the relationship between the anellovirus composition and extensive and diverse social, environmental, genetic, pre-, peri- and postnatal variables. Of note, immune-training factors such as having older siblings and starting day care appeared to be exposures strongly influencing the faecal anellovirus composition at 1 year of age across several measures, such as overall abundance, alpha and beta diversity, and relative abundance of several genera and subgenera. However, we did not find robust associations between the anellovirus composition and infections, nor atopic disease, indicating that anelloviruses are a commensal component of all infants’ faecal virome. Future studies should focus on detailed genetic analysis of the mechanisms involved in the anellovirus-host relationship, and on longitudinal faecal sampling to further elucidate relevant extrinsic and intrinsic factors influencing the early-life development of the anellovirus composition. Methods Study population The study is composed of samples collected in the Danish population-based Copenhagen Prospective Studies on Asthma in Childhood2010 (COPSAC2010) mother-child cohort44. Samples were collected from 700 children and their mothers, starting at 24 weeks of pregnancy up to 6 years of age. Inclusion and ethics The study was conducted in accordance with the guiding principles of the Declaration of Helsinki and was approved by the National Committee on Health Research Ethics (H-B-2008-093) and the Danish Data Protection Agency (2015-41-3696). Both parents gave written informed consent before enrolment. Faecal sample collection, virome extraction and computational processing Faecal samples from 660 children at 1 year of age were collected either by the parents following comprehensive instructions, or at the clinic. Samples were sent to the laboratory by post and mixed with 1 ml glycerol broth (10% vol/vol glycerol) on arrival and stored at −80 °C48. Detailed protocols on extraction and sequencing have been described previously41,63. In short, to obtain viromes the faecal samples were filtered to enrich for viral particles and the DNA was extracted using the QIAmp viral RNA mini kit and underwent multiple displacement amplification for 30 min. The virome libraries were prepared following the manufacturer’s directions for the Illumina Nextera XT kit (FC-131-1096) and sequenced on the Illumina HiSeq X platform (2 ×150 bp paired-end reads), resulting in an average depth of 3 Gbp per sample. The sequencing was successful for 647 of the 660 samples. The reads were filtered, trimmed using Fastq Quality Trimmer/Filter v0.0.14 (options -Q 33 -t 13 -l 32 -p 90 -q 13), and adaptors were removed using cutadapt (v2.0). The processed reads were then de-replicated using VSEARCH71 (v2.4.3) “derep_prefix” and assembled with Spades72 (v3.10.1) using the meta flag with hamming disabled. Resulting contigs were deduplicated across all samples at the 95% average nucleotide identity (ANI)-level to yield vOTUs. Virus family-level clusters of vOTUs were first defined based on encoded protein similarity in order to identify all vOTUs belonging to the anellovirus family. This protein-level identification step was to ensure sensitivity towards novel anelloviruses distant from those already found in public databases. The workflow is described in detail on GitHub (https://github.com/shiraz-shah/VFCs). In short, pairwise alignments (from FASTA64 v. 36) between all vOTU-encoded open reading frames were used to (1) construct an all-against-all vOTU distance matrix for a neighbour-joining tree that was cut deeply, at the virus family-level, and (2) for Markov Clustering65 in order to define viral gene families, that had their functions resolved by profile-profile alignment against the Pfam66 database. Finally, the family-level vOTU cluster that carried the anelloviral ORF1 gene family was selected as the anellovirus family in our data. Next, whole-genome Smith−Waterman alignments64 were made at the DNA-level, between all identified anelloviral vOTUs in addition to public genome representatives for all ICTV-ratified genera, all the way from alpha- to zayintorquevirus. The whole-genome nucleotide alignments were used to define pairwise distances for a neighbour-joining tree. The tree was rooted using iotatorquevirus (known to infect non-human mammals such as pigs) as an outgroup. Subsequently, cut distances from the root were measured that would reproduce all the ICTV genera. This revealed two additional and undescribed genera that were named based on the vacant Greek letters tau and phi. A similar species-level root cut distance was derived based on the reproduction of select ICTV species belonging to alpha, beta and gammatorquevirus. The representative ICTV species were chosen because they had been revised recently, thus likely comprising borderline cases with a rigorous and up-to-date application of the most recent ICTV species-level cutoff. Cutting the tree at a distance reproducing those species enabled comprehensive species-level resolution for the entire data set. While inspecting the tree distances, an additional local minimum was found between the genus and species-level root distance, and this was used as the cut distance for defining anelloviral subgenera. Bacterial extraction and computational processing Details on the processing of faecal samples for 16S rRNA sequencing and processing have been published previously41,48. In short, the bacterial DNA was extracted from the faecal samples using MoBio PowerSoil kits targeting the V4 region of the 16S rRNA gene and sequenced using the v2 kit on the Illumina MiSeq platform to generate paired-end 250 bp reads. Samples with more than 2000 reads were kept before primers were removed, the reads were trimmed and ASVs were inferred. Clinical phenotypes Asthma The asthma diagnosis was assessed prospectively based on a quantitative symptom algorithm published previously67,68. A case is in this study defined as having ever had an asthma diagnosis by the age of 6 years. Atopic dermatitis The atopic dermatitis diagnosis was based on the criteria of Hanifin and Rajka69 and with clinical assessments throughout childhood70. To be labelled as a case, at least three of four major criteria and at least 3 of 23 minor criteria were needed, and a case was considered as having ever had an atopic dermatitis diagnosis by the age of 6 years. Allergic rhinitis Allergic rhinitis at age 6 years was diagnosed based on allergic sensitisation and clinical interviews of the parents on history of significant nasal congestion, sneezing and/or runny nose outside periods with common cold and in the relevant period of exposure to the sensitised aeroallergen71. If the participant did not self-report rhinitis, the participant was labelled as a control. If the participant self-reported rhinitis and at least one of the allergens was positive, the participant was labelled as a case. If the participant self-reported rhinitis and all allergens were negative, the participant was labelled as a control. If the participant self-reported rhinitis and the above-mentioned criteria for being a case was not fulfilled, and at least one of the allergens or antibodies were blank, the participant case status was set as missing. If the participant self-reported rhinitis but the IgE and skin prick test were inconclusive, the participant's case status was set as missing. Infections Symptoms of infections included colds, fever, and gastroenteritis within (+-) 7 days of the faecal sample were collected by daily diaries. Covariates Information on physician-diagnosed asthma in the mother and father, maternal age at birth, smoking during pregnancy, antibiotics in pregnancy, preeclampsia, gestational age, birth weight, C-section or vaginal birth, induction of birth, parents atopy, antibiotics administered to the child, pet ownership, older children in the home, introduction of foods (cow’s milk, eggs, fish and hypoallergenic formula), day care start and duration of exclusive and total breastfeeding period was obtained during the scheduled visits to the research clinic. C-section was either encoded as 'C-section': any C-section (acute or planned) compared to any vaginal delivery (induced or spontaneous, or as 'Emergency C-section': acute C-section compared to spontaneous vaginal delivery. Information on intrapartum antibiotics was validated against birth records from hospitals, and information on antibiotic use during pregnancy and childhood was validated against national registries72. The living environment was based on home address at birth using the CORINE satellite-based land cover database5. Fish oil and high-dose vitamin D intervention groups in pregnancy were included73,74. Statistical analyses All analyses were conducted in the programming language R version 4.2.2 (Vienna)75. Preparation and normalisation of relative abundance data We calculated the relative abundance per sample, including all viruses detected in our samples, because we wanted to investigate the role of anelloviruses relative to the entire virome community, partly due to the potential biological relationship between these viral compartments, as demonstrated by the procrustes analysis described in the results section 'Anelloviruses are unique yet highly prevalent in the 1-year-old infant gut'. Technical covariates Several analyses were adjusted for covariates as noted; the selected covariates were the proportion of viral reads, age at faecal sampling, sequencing lane and sequencing depth. These were chosen based on their relationship with the anellovirus composition as measured by several alpha and beta diversity measures. Alpha and beta diversity calculations Alpha diversity was calculated at the species level without filtering, with Faith’s phylogenetic diversity (PD) (package picante version 1.8.276), Shannon index, inverse Simpson’s index (phyloseq package version 1.44.077, estimate_richness function) and evenness (Shannon index divided by the number of observed species). Beta diversity was calculated at the vOTU level without filtering using Aitchison (pseudocount of half of the lowest non-zero value in the data), Bray-Curtis, Jaccard, Canberra (function vegdist from vegan package version 2.6-478) and Spearman distance (function separman.dist from bioDist package version 1.70.0), as well as weighted UniFrac and unweighted UniFrac distance (unifrac function from phyloseq package version 1.44.0). The difference in beta diversity between cases and controls was tested using a permutational multivariate analysis of variance using distance matrices (PERMANOVA, adonis2 function from the vegan package version 2.6-4), with 9999 permutations, sequencing lane set as strata, and covariates as described above. Indicator species analysis The relative abundance data was filtered to remove species with a prevalence <50%, all subgenera and genera were included for analyses. Indicator species were identified with the multipatt function (999 permutations) from the indicspecies package version 1.7.1479). This allows for the identification of particular taxa that are indicators of certain sample groups, such as cases or controls of clinical outcomes. Indicator species are defined as either being specific (positive predictive value, the probability that the sample belongs to a group due to the species being present) or being sensitive (the probability of finding the species in samples belonging to a group). Species displaying high specificity to a group occur only in samples from that group (though not necessarily present in all samples) and are missing from the samples of the other group. Species displaying high sensitivity to a group occur in all samples belonging to that group and are largely missing from samples of the other group. Differential abundance analysis Differential abundance of anellovirus taxa was tested with a generalised linear model (glm, family = 'binomial’) adjusted for the above-mentioned covariates, with the relative abundance zero mean unit variance scaled. Secretor status analysis Based on the genome-wide genotyping of the blood cell fraction of plasma using the Illumina OmniExpress imputed with the Hapmap and 1000 genomes reference populations44, the rs601338 SNP was encoded to either non-secretors (AA allele, N = 101) or secretors (GA or GG allele, N = 475). The statistical test was performed using a generalised linear model (glm, family = 'binomial’) adjusted for the above-mentioned covariates. Multivariate analyses using sPLS Sparse partial least squares (sPLS) from the mixOmics package80 was used to test the multivariate association between the anelloviruses and clinical outcomes. This methodology maximises the covariance between latent variables of the data, rather than the variance explained on the outcome side. This results in a robust model that is resistant to overfitting and can handle high-dimensional data as both dependent and independent variables, even with small sample sizes. This model identifies linear combinations of the variables from the datasets as a measure to reduce the dimensionality of the data. The microbial data agglomerated at the species level was filtered to remove species with a prevalence <30%, while at the subgenus level, all taxa were kept, before log transforming with a pseudocount of half the minimum non-zero value. The data was then split into a training (80%) and testing (20%) set ten times (termed stability runs) using the function createDataPartition from the caret package version 6.0-9481 to achieve stable estimates of the model accuracy, thus avoiding spurious results from certain sample combinations. The model was trained on the training set using the train function from the caret package with scale and centre as pre-processing, using fivefold repeated cross-validation with three repeats using the trainControl function. The hyperparameters used were keepX (number of taxa to keep) which started at 5 to the maximum number of taxa, with a step size of 4 for subgenera, and a step size of 8 for species, and ncomp (number of components) which was set to 1:4. The model hyperparameter set with the best median area under the curve (AUC) calculated across all repeats was selected and evaluated on the testing set. The median test AUC of the ten stability runs was used to evaluate the model performance. Multivariate analyses using random forest A random forest model was additionally used to test the multivariate association between the anelloviruses and clinical outcomes. This type of methodology can identify nonlinear and multivariate relationships between a high number of independent variables to a single dependent variable. Given a sufficient sample size, and with appropriate training and testing splits, these models are robust and can be operated to avoid overfitting. This model was applied to all taxa agglomerated at the subgenus level, or at the species level after removing species with a prevalence <30%. This was performed on the relative taxa abundance, and 80% of the data were used for training and 20% for testing, over ten separate stability runs to obtain stable estimates of the model accuracy and avoid spurious results from certain sample combinations. The hyperparameters were tested using the train function (method = 'ranger’, importance = 'impurity’) from the caret package version 6.0-94, where the control was set to fivefold repeated cross-validation with three repeats and random search using the trainControl function. The hyperparameter grid used was: ntrees = (500 and 2000), nodesizes = (20, 50 and 80), splitrule = (variance, extratrees and beta), and mtry started at 5 to the maximum number of taxa, with a step size of 4 for subgenera, and a step size of 8 for species. The model parameters with the best median AUC calculated across all repeats was used to evaluate the model on the test set, using the function ranger (importance = impurity) from the package ranger version 0.15.182. The median test AUC of the ten stability runs was used to evaluate the model performance. Procrustes analysis Procrustes is a statistical method that compares two ordinations or matrices from, e.g. environmental and microbial data, by scaling (stretching) and rotating them to minimise the sum of squared differences. The protest function thereafter performs permutation of the Procrustes analysis repeatedly to estimate the significance of the Procrustes statistic and returns a correlation-like statistic from the symmetric Procrustes sum of squares. This was used to test associations between the bacteriophage composition and the anellovirus composition (N = 647), and the association between the anellovirus composition and the 16S rRNA gene amplicon-sequencing-based bacterial composition (N = 611). The data were first re-normalised separately, e.g. the anelloviruses and bacteriophages were re-normalised to 100%. The ordination was created using the distance-based redundancy analysis (dbrda) function, conditioned on lane and sequencing depth, and the association between the two ordinations was assessed with the procrustes function (symmetric = TRUE) and tested with the protest function (9999 permutations), all from the vegan package version 2.6-4. SpiecEasi analysis SPIEC-EASI (SParse InversE Covariance Estimation for Ecological Association Inference) is a method for identifying ecological networks among microbes (Kurtz et al. 2015). The SpiecEasi package version 0.1.4 was applied to all taxa agglomerated at the subgenus level, with the following settings: method = ‘mb’, lambda.min.ratio = 1e-1, nlambda = 50. P value threshold for significance All p values were considered significant at the nominal p < 0.05 level, or at the multiple testing adjusted (p.adjust function with method fdr) pFDR < 0.05 level as noted. All statistical tests were two-tailed. No statistical method was used to predetermine sample size. No data were excluded from the analyses. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of additional supplementary files Supplementary data table 1 Supplementary data table 2 Supplementary data table 3 Supplementary data table 4 Supplementary data table 5 Supplementary data table 6 Supplementary data table 7 Supplementary data table 8 Supplementary data table 9 Supplementary data table 10 Reporting Summary Transparent Peer Review file Supplementary Information Description of additional supplementary files Supplementary data table 1 Supplementary data table 2 Supplementary data table 3 Supplementary data table 4 Supplementary data table 5 Supplementary data table 6 Supplementary data table 7 Supplementary data table 8 Supplementary data table 9 Supplementary data table 10 Reporting Summary Transparent Peer Review file