Exploring the ecological drivers of bacteriophage diversity and functional viral potential in the skin of the axolotl Ambystoma altamirani

Abstract

Bacteriophages play important roles in shaping microbial community dynamics across diverse environments. In the amphibian skin, most microbiome studies have focused on bacteria and their interactions with the fungus Batrachochytrium dendrobatidis (Bd), leaving other microbial components, including viruses, largely unexplored. Here, we present the first characterization of the viral community in the amphibian skin microbiome, focusing on ecological drivers of bacteriophage diversity and functional potential in the axolotl Ambystoma altamirani. Using public shotgun metagenomes, we found that the viral fraction was dominated by bacteriophages of the class Caudoviricetes. Bacteriophage diversity was significantly associated with local physicochemical parameters at the time of sampling, and showed a strong positive correlation with bacterial diversity, whereas no significant associations were detected with the presence of Bd. In addition, seasonality influenced the composition and properties of bacteria–bacteriophage co-abundance networks. Functional annotation of assembled bacteriophage sequences revealed a diverse functional potential, including putative auxiliary metabolic genes, superinfection exclusion, toxin–antitoxin, and virulence factors. Overall, these findings highlight the ecological relevance of bacteriophages in amphibian skin microbiomes and underscore the need for further studies on their role in the amphibian host’s health.

Analysis of shotgun metagenomes from the skin of the axolotl Ambystoma altamirani revealed a viral community dominated by bacteriophages, whose diversity is tightly associated with environmental factors and revealed a diverse functional repertoire potentially involved in microbial interactions.

Introduction

Bacteriophages—viruses that infect bacteria—are the most abundant biological entities in the world and are key players across diverse ecosystems where they can account for a large proportion of bacterial mortality and significantly contribute to global processes, including biogeochemical cycles (Suttle 2007, Breitbart et al. 2018, Zhang et al. 2022). Phages can selectively lyse dominant bacteria, thereby modulating bacterial population dynamics and exerting significant effects on bacterial diversity and composition (Winter et al. 2010, Chen et al. 2019). Additionally, temperate phages can adopt the lysogenic replication cycle where they integrate their genome into the bacterial host chromosome, resulting in increased host fitness through superinfection exclusion, the expression of prophage-encoded virulence factors, antibiotic resistance genes or auxiliary metabolic genes (Breitbart et al. 2018, Chevallereau et al. 2022), Kondo et al. 2021, Luo et al. 2022). Furthermore, both virulent—phages that only undergo the lytic replication cycle—and temperate bacteriophages can contribute to bacterial genetic diversity through horizontal gene transfer (HGT) (Chiang et al. 2019).

In host-associated systems phages have been shown to modulate the gut microbiome and metabolome with cascading effects on non-host bacteria (Hsu et al. 2019). Phages can also facilitate bacteria–eukaryote symbioses (Jahn et al. 2019, Braga et al. 2021) or even confer direct benefits to eukaryotic hosts (Metcalf and Bordenstein 2012, Yan et al. 2023). For instance, the bacteriophage adherence to mucus model (BAM) posits that phages can bind to mucosal surfaces of diverse metazoan hosts (including cnidarians, fish skin, human gum, and mouse intestine) and kill incoming pathogenic bacteria, thereby providing an additional non-host defense barrier (Barr et al. 2013), whose dynamics may be modulated by a mucin gradient (Silveira and Rohwer 2016). Moreover, the specific ecological impact of bacteriophages in both environmental and host-associated systems is likely determined by variations in different environmental factors across spatial and temporal gradients (Suttle 2007, Łusiak-Szelachowska et al. 2017, Breitbart et al. 2018, Lourenço et al. 2022, Zhang et al. 2022)

Amphibians, the most threatened vertebrate class in the world, are highly impacted by chytridiomycosis, a skin disease caused mainly by the globally distributed fungus Batrachochytrium dendrobatidis (Bd) (Berger et al. 1998). This disease—characterized by a range of skin disorders, including excessive shedding, erythema, and ulceration (Van Rooij et al. 2015)—has been identified as one of the primary factors leading to amphibian population declines since the 1980s (Scheele et al. 2019), Luedtke et al. 2023). Up to date, thousands of amphibian skin-associated bacteria have been isolated, cultured and tested for their antifungal properties, including the capacity to inhibit the growth of Bd (Woodhams et al. 2015, Wax et al. 2023, González-Serrano et al. 2025). However, other components of the amphibian skin microbiome, such as microeukaryotes and viruses, remain largely understudied (Kueneman et al. 2017, Belasen et al. 2021, Soto-Cortés et al. 2024, Bouchali et al. 2025).

The advent of -omics disciplines has facilitated the recognition of numerous amphibian viruses from at least 18 viral families (Shi et al. 2018, Harding et al. 2022). However, only a limited number of amphibian viruses have been thoroughly characterized, and fewer have been associated with the manifestation of disease signs (Essbauer and Ahne 2001, Johnson and Wellehan 2005, Pessier 2014). Notably, ranaviruses (family Iridoviridae), which have been linked to severe population declines in amphibians (Gray and Chinchar 2015), are the best characterized amphibian viruses to date, and are the only ones reported to co-infect with Bd, resulting in interactions ranging from neutral to synergistic or antagonistic effects (Herczeg et al. 2021). Concerning bacteriophages, the existing body of knowledge in relation to their ecological role in the amphibian skin microbiome remains limited to specific reports such as the characterization of two lytic phages infecting Aeromonas hydrophila (Rai et al. 2023)—a known amphibian pathogen (Herczeg et al. 2021)—, and the identification of prophages integrated in the genome of Janthinobacterium lividum (Friedrich et al. 2020)—a bacterium that can inhibit the growth of Bd in vitro and in vivo (Brucker et al. 2008, Harris et al. 2009).

Recently, a high prevalence of Bd has been detected in the skin of Ambystoma altamirani populations, a Mexican salamander (commonly known as axolotl) that is presumed to be tolerant to the fungal pathogen in the wild (Basanta et al. 2023). Subsequent 16S rRNA profiling showed that skin samples were enriched in the bacterial families Burkholderiaceae and Chitinophagaceae, which are families that include Bd-inhibiting bacteria reported in other amphibian hosts (Martínez-Ugalde et al. 2022). Further functional analyses using shotgun metagenomes revealed a high prevalence of potential antifungal genes (Martínez-Ugalde et al. 2024), and genome mining of A. altamirani bacterial isolates identified a wide repertoire of biosynthetic gene clusters with predicted antifungal activity and chitin-degrading families linked with their antifungal capacity (González-Serrano et al. 2025).

In A. altamirani populations, water physicochemical parameters, such as temperature and pH, were reported to vary significantly across seasons and sampling sites, having significant impacts on Bd prevalence and infection intensity (Basanta et al. 2023), as well as on the bacterial community diversity and composition (Martínez-Ugalde et al. 2022). The recognized ecological importance of bacteriophages in diverse environmental and host-associated systems and the detection of viral reads in A. altamirani skin metagenomes (Martínez-Ugalde et al. 2024) raises a general question of what is the ecological role of viruses in the amphibian skin microbiome, and whether they could contribute to shaping the microbiome structure and function in ways relevant to maintaining host health in the presence of Bd.

To gain deeper insights into the ecological role of viruses in amphibian skin we explored the diversity and functional potential of viruses—particularly dsDNA bacteriophages—associated with the skin of A. altamirani by analyzing published shotgun metagenomic data (Martínez-Ugalde et al. 2024) and genomes of bacterial isolates (González-Serrano et al. 2025). Specifically, we aimed to (i) describe the taxonomic composition of viruses associated with the amphibian skin, (ii) evaluate the extent to which phage community diversity and composition are structured by ecological factors, including seasonality, sampling site, and host Bd infection status, and (iii) explore the functional repertoire of skin-associated dsDNA phages.

Based on evidence from environmental and host-associated systems (Adriaenssens et al. 2017, Arkhipova et al. 2018), Łusiak-Szelachowska et al. 2017, Gregory et al. 2019, Hevroni et al. 2020, Alanazi et al. 2022, Lourenço et al. 2022), we hypothesize that phage communities will vary across seasons and sites in response to differences in physicochemical conditions, and that phage diversity and composition will be associated with host Bd infection status. Furthermore, we explored whether phage genomes encode functions with the potential to exert cascading effects on downstream microbial interactions (Hsu et al. 2019). To our knowledge, this study represents the first ecological analysis of bacteriophages associated with amphibian skin microbiomes.

Materials and methods

Metagenomic dataset selection and read filtering criteria

The viral community on the skin of A. altamirani was analyzed from public shotgun metagenomes (Martínez-Ugalde et al. 2024) derived from a sampling effort carried across four sites over a year (Basanta et al. 2023, Martínez-Ugalde et al. 2022). Briefly, skin swabs were collected from 194 non-metamorphic (with gills) and 84 metamorphic (without gills) A. altamirani axolotls at four different times, encompassing the four seasons of one full year (Summer—July 2019, Autumn—October 2019, Winter—January 2020, and Spring—April 2020), and at four different sites with elevations ranging from 3087 to 3462 MASL (Sites A –D) at the Sierra de las Cruces, Estado de México, México (Supplementary Figure S1).

From the previous sampling design, a more recent study (Martínez-Ugalde et al. 2024) selected 40 samples of exclusively non-metamorphic axolotls for shotgun metagenomic sequencing using the Illumina HiSeq 2×150 technology (no reverse-transcription step was used at any stage of library preparation) to describe the functional bacterial diversity in the amphibian skin. The selected samples encompassed ten samples per season, categorized into five Bd-infected and five non-infected groups, except for the Autumn season, which included eight infected and two non-infected groups (Supplementary Table S1).

For this study, we downloaded the 40 metagenomes from the BioProject PRJNA1010953. These samples were pre-processed by the authors through the elimination of adapters and poor-quality reads, as well as filtering out reads originating from the host and potential human contaminants. For this study, the quality of the sequenced reads was verified using FastQC v0.11.9 (Andrews 2010), and the sequences were deduplicated with fastp (Chen et al. 2018), leaving a total of 69 656 993 reads for further analysis (Supplementary Table S2).

Short read taxonomic classification

For taxonomic classification, paired-end reads were processed with Kaiju v1.8 .0 (Menzel et al. 2016) using the nr_euk database. Classification summary tables were generated with the program kaiju2table using the -e flag to enable viral classification across seven taxonomic ranks: superkingdom, phylum, class, order, family, genus, and species. We used a custom script named COMETS (Cisneros-Martínez et al. 2023) to create a count table and a taxonomy table, which were used to build a phyloseq object with the phyloseq v1.48.0 package (McMurdie and Holmes 2013). In the count table, taxonomic identifiers assigned by kaiju are treated as operational taxonomic units (OTUs), which can be filtered at different taxonomic ranks. Taxonomy was updated by matching taxonomic identifiers to the latest release of the NCBI Taxonomy (accessed on July 3rd, 2026) (Schoch et al. 2020). Site B (n = 2) was excluded from statistical analyses due to insufficient replication. Nonetheless, these two samples were included in the viral assembly pipeline to ensure comprehensive recovery of viral sequences. Finally, to account for differences in sequencing depth across samples, retaining proportional abundance and avoiding data loss, classified reads were normalized by applying a Total Sum Scaling (TSS) to the median read sample sum (192,534).

Bacterial and bacteriophage diversity analysis

All statistical and ecological analyses were conducted in R v4.4.1 (R Core Team 2024). Alpha and beta diversity of bacterial and bacteriophage communities were evaluated separately. Bacteriophage reads were separated from other viral reads by applying a taxonomy filter to retain OTUs belonging to the classes Caudoviricetes, Faserviricetes, and Tectiliviricetes, and excluding the family Adenoviridae. Bacterial reads were selected by applying a taxonomy filter to keep OTUs classified as Bacteria under the superkingdom rank.

Bacteriophage alpha diversity was assessed by estimating observed Richness (Hill q = 0), Shannon diversity (Hill q = 1), and inverse Simpson (Hill q = 2) with phyloseq (McMurdie and Holmes 2013). To evaluate the influence of sequencing depth on bacteriophage alpha diversity, rarefaction/extrapolation curves of the three diversity metrics were calculated as a function of sample coverage (SC) with iNEXT v3.0.1 (Chao et al. 2014). Pearson correlations between alpha diversity metrics and library size were also calculated to evaluate sampling bias before (raw) and after (median-scaled) normalization. Finally, Wilcoxon signed-rank tests and Pearson correlations were calculated between raw, median-scaled and iNEXT coverage-standardized (SC = 0.75) alpha diversity metrics to evaluate changes in estimate values and patterns among samples, respectively. Although estimated SC values where relatively high (Supplementary Table S2) Richness was strongly biased by library size (Supplementary Table S3), and rarefaction/extrapolation curves suggested that Richness was more sensitive to incomplete sampling than Shannon and Simpson diversity indices (Supplementary Figure S2). The consistency of Shannon and Simpson diversity estimates across standardization approaches suggested that diversity patterns driven by dominant bacteriophage taxa were substantially less affected by differences in sampling effort. Therefore, subsequent analyses were restricted to Shannon (Hill q = 1), and Simpson (Hill q = 2) diversity indices.

Resulting estimates were compared across sites (n=3) and seasons (n=4) using the Kruskal-Wallis H test, with post-hoc Dunn tests adjusted using the Benjamini-Hochberg method. Differences between infected and non-infected samples were evaluated using Wilcoxon rank-sum tests. To examine the factors potentially contributing to alpha diversity, Shannon diversity and inverse Simpson were modeled with Gaussian (continuous response) generalized linear models (GLMs). For each diversity metric, two models were built: one to isolate the independent contribution of categorical predictors (site, season, and Bd infection status) while controlling for potential covariate confounding effects, and another to assess whether the observed differences could be explained by water physicochemical variables [seasonal mean temperature, seasonal Δ temperature (difference between maximum and minimum seasonal temperature), pH, dissolved oxygen, and conductivity].

For bacteriophage beta diversity analyses, Aitchison distance matrices (robust.aitchison method) were calculated and subsequent multivariate analyses were conducted with vegan v2.6–6.1 (Oksanen et al. 2024). Specifically, statistical differences between sample groups were assessed via permutational multivariate analysis of variance (PERMANOVA) with 999 permutations. Pairwise PERMANOVA tests were performed with the pairwiseAdonis v0.4.1 package (Martinez Arbizu 2020) , and P-values were adjusted for multiple comparisons using the Benjamini–Hochberg method. To assess the effect of environmental variables on differences in community structure, a distance-based redundancy analysis (dbRDA) was performed, with significance of the constrained model and individual predictors tested via permutation-based ANOVA. To identify specific OTUs contributing to differences in bacteriophage diversity across sites, seasons, and between Bd infection categories, a Linear Discriminant Analysis Effect Size (LEfSe) (Segata et al. 2011) was performed as implemented in the microbiomeMarker v1.10.0 package (Cao et al. 2022). Hosts of enriched phage OTUs were identified with the latest release of the Virus–Host Database (accessed on July 3rd, 2026) (Mihara et al. 2016).

Bacterial alpha and beta diversity were estimated following the same procedures as described for bacteriophages, using Shannon diversity (Hill q = 1), inverse Simpson (Hill q = 2), and Aitchison distance as diversity metrics. Potential associations between bacteriophage and bacterial alpha diversity were assessed using Spearman rank correlations, comparing bacteriophage and bacterial Shannon diversity as well as inverse Simpson. For beta diversity, Mantel tests were performed to evaluate correlations between bacteriophage and bacterial Aitchison distance matrices.

To assess whether seasonality influenced putative bacteria–bacteriophage interactions, we performed co-abundance network analyses using the SparCC algorithm, which estimates linear Pearson correlations between log-transformed counts to control for compositional effects (Friedman and Alm 2012), as implemented in the SpiecEasi v1.1.3 package (Kurtz et al. 2025). Networks were constructed with data from Site A, since it was the only site with consistent temporal coverage (samples collected in at least three seasons: Summer, Autumn, and Winter) and sufficient replication (at least five samples per season) for reliable detection of co-abundance patterns (see Supplementary Table S1). Correlations with coefficients ≥ 0.9 and P-values ≤ 0.05 were retained for downstream network analysis. The dataset used for the networks was aggregated at the genus level, and mean prevalence and mean abundance were calculated independently for bacterial and bacteriophage genera. Genera with both prevalence and abundance below their respective means were removed, resulting in 479 bacterial and 53 bacteriophage genera retained for analysis. Network topology metrics were computed in Cytoscape v3.10.3 (Shannon et al. 2003) using the built-in Network Analyzer tool. The clustering coefficient and average shortest path length were compared across seasons using the Kruskal–Wallis H test followed by post hoc Dunn tests with Benjamini–Hochberg adjustment for multiple comparisons. The hosts of bacteriophage nodes were retrieved from the Virus–Host Database, and corresponding virus–host associations were plotted with the package circlize v0.4.16 in R (Gu 2014).

Identification and classification of assembled bacteriophage sequences

The PenguiN v5.cf8933 program (Jochheim et al. 2024) was employed to perform 13 co-assemblies, in which samples were grouped by location and season: Summer (Site A = 6, Site C = 2, Site D = 2), Autumn (Site A = 7, Site B = 1, Site D = 2), Winter (Site A = 5, Site C = 2, Site D = 3), Spring (Site A = 1, Site B = 1, Site C = 3, Site D = 5) (see Supplementary Table S12). In addition to these assemblies, 50 metagenome-assembled genomes (MAGs) (Martínez-Ugalde et al. 2024) and 24 bacterial genomes—derived from bacterial isolates previously tested for Bd inhibition and isolated from the skin of A. altamirani (González-Serrano et al. 2025) (BioProject PRJNA1200739)—were included for prophage detection. All assembled contigs, MAGs, and bacterial genomes were processed with VIBRANT v1.2.1 (Kieft et al. 2020) to identify bacteriophage sequences. The identified sequences were subsequently assessed for quality with CheckV v.0.9.0 (Nayfach et al. 2021) and dereplicated with an average nucleotide identity (ANI) of 95% and an aligned fraction (AF) of 85% using nucleotide–nucleotide BLAST v2.12.0+ (Camacho et al. 2009) and CheckV supporting scripts anicalc and aniclust.

Taxonomic classification of bacteriophage sequences was performed using vConTACT2 v0.11.3 (Jang et al. 2019), which is designed to cluster viral genomes based on shared gene content and similarity scores, and assigns a viral cluster (VC) status to each query genome reflecting its relationship to known reference taxa: clustered, overlap, outlier, or singleton. Following the developers’ recommendations, bacteriophage sequences classified as clustered or overlap—indicating reliable gene-sharing connections with reference genomes—were assigned genus-level taxonomy according to the closest reference virus within the same cluster. Those identified as outliers or singletons—which lack confident clustering with known taxa—were further analyzed using the custom script VirTaK (Virus Taxonomy from Kmers; available at: https://github.com/AleCisMar/VirTaK). VirTaK integrates multiple metrics for inferring viral taxonomy, including nucleotide and amino acid identity, protein domain content, k-mer profiles, coding sequence (CDS) counts, and genome length, against reference viral genomes from the Virus Metadata Resource (VMR_MSL38_v1) curated by the International Committee on Taxonomy of Viruses (ICTV). These metrics are integrated into a unified similarity score ranging from 0 (no similarity) to 1 (100% similarity). Bacteriophage sequences with a value greater than 0.285 were assigned to an order, those with scores above 0.476 were assigned to a family, and those exceeding 0.714 were assigned to a genus. Bacteriophage sequences associated with the class Caudoviricetes were uploaded to the ViPTree server v4.0 (Nishimura et al. 2017) to obtain genomic similarity scores (S_G) to reference dsDNA prokaryotic viruses. S_G ≥ 0.075 was used to visualize clusters of similar viruses with Cytoscape v3.10.3 (Shannon et al. 2003).

Bacteriophage sequences were further classified as temperate if they met one or more of the following criteria: (1) obtained from isolated bacterial genomes or MAGs (presumably prophages), (2) predicted as prophages by VIBRANT or CheckV, (3) contained integration/excision genes, as predicted by pharokka (Bouras et al. 2023). Those that failed to meet any of the aforementioned criteria were classified as having an undetermined replication strategy. Flanking host regions were trimmed from the sequences identified as prophages by either VIBRANT or CheckV before proceeding with further analyses. The prediction of hosts was conducted with iPHoP v1.1.0 (Roux et al. 2023) exclusively for phage sequences identified from assembled contigs. For predictions with a score greater than 0.8, host taxonomy was taken to the family level. Predicted host taxonomy was elevated to the genus level for predictions with a score greater than 0.9. De novo assembled viral sequences and prophages identified within MAGs were deposited to the NCBI under BioProject PRJNA1010953. Prophages identified within genomes of isolated bacteria were uploaded to a public figshare repository (https://doi.org/10.6084/m9.figshare.32194227).

Functional annotations of phage-derived sequences

Prediction of coding sequences (CDS) and functional annotation of bacteriophage sequences were performed with pharokka v1.3.0 (Bouras et al. 2023), which uses remote homology detection against the PHROGs database of bacterial and archaeal viruses (Terzian et al. 2021) and screens for antimicrobial resistance and virulence genes using the Comprehensive Antibiotic Resistance Database (CARD) (Alcock et al. 2020) and the Virulence Factor Database (VFDB) (Liu et al. 2019). Viral sequences were further annotated with lovis4u v1.0.3 (Egorov and Atkinson 2025), which detects defense and anti-defense proteins.

The predicted functions were grouped into the following categories: “Structural”, “Nucleotide metabolism”, “Transcription regulation”, “auxiliary metabolic genes (AMGs)”, “Integration and excision”, “Lysis”, “Toxin/antitoxin”, and “Other”. AMGs predictions were also obtained from VIBRANT. Putative AMGs predicted by pharokka and VIBRANT were validated with HHpred (Zimmerman et al. 2018) against PDB_mmCIF70_20_Feb, Pfam-A_v38.2, UniProt-SwissProt-viral70_3_Nov_2021, and PHROGs_v4 databases, by analyzing the annotations of hits with probabilities > 99. Candidates located at either end of the contig and probable peptidases, methyltransferases, structural, and ribosomal proteins were discarded, as recommended (ter Horst et al. 2021). Anti-CRISPR proteins (Acrs) were identified by protein-protein BLAST v2.12.0+ (Camacho et al. 2009) against the 122 known anti-CRISPRs from the AcrDB (Khatri et al. 2025). Sequences matching known Acrs with e value < 1e-03 and bidirectional coverage (calculated as 2 L/(Q+S), where L is alignment length, and Q and S are query and subject lengths) > 25% were subjected to structural similarity search against known Acrs with TM-Vec (Hamamsy et al. 2024). Matches with TM-score > 0.6 were designated as candidate Acrs. Bacteriophage sequences (contigs) with candidate Acrs were screened for anti-CRISPR associated (Aca) domains (AcaFinder; Yang et al. 2022) using hmmscan v3.3.2 (Eddy 2011) with an e-value cutoff of 1e-05. Identified Acrs and Acas were grouped into the “Immune evasion” category. The CDS under the “Other” category were screened with hmmscan for antibiotic resistance domains against the Resfams database (Gibson et al. 2015). Significant matches were searched with phmmer on the HMMER web server (Potter et al. 2018) against the UniProt (2025_01) and SwissProt (2025_01) databases for validation. CDS in the “Other” category were further annotated with PfamScan v1.6 (Finn et al. 2014) against the Pfam-A database (Mistry et al. 2021). Putative functions were assigned according to the categories defined by Jahn et al. (2019), which were manually curated based on the domain descriptions. This process helped in the identification of CDS potentially linked to “Superinfection exclusion” and “Virulence” categories.

Results

Bacteriophages dominate the viral fraction of the axolotl skin microbiome

In agreement with Martínez-Ugalde et al. (2024), the taxonomic classification of metagenomic reads from the skin of A. altamirani revealed a microbial community predominantly composed of bacteria, which accounted for an average relative abundance of 82.3% across the 7 320 155 classified reads. Eukaryotes followed at 16.6%, while archaea and viruses represented 0.59% and 0.58%, respectively (Supplementary Table S2).

Within the viral fraction (42 769 reads), bacteriophages of the class Caudoviricetes were the most abundant (22.6%), followed by viral classes that infect eukaryotic organisms, including protists and animals: Megaviricetes (20.3%), Revtraviricetes (8.32%), Pokkesviricetes (2.89%), Herviviricetes (1.73%), and Polintoviricetes (1.69%). Other less abundant groups included Repensiviricetes (0.65%), Naldaviricetes (0.54%), Papovaviricetes (0.53%), and Cardeaviricetes (0.51%) (Fig. 1A).

Figure 1.

Viral relative abundance at(A) class, and (B) family levels. The most abundant classes or families are shown in descending order of average abundance across all samples. Classes and families with average abundance lower than 0.5% are grouped into the “Other” category. Lower legend indicates the features associated to the samples and the genomic/biological features of viral taxa. Reads assigned to the class Pisoniviricetes and the family Coronaviridae are likely classification artifacts (see results).

At the family level, aside from the bacteriophage families Peduoviridae, Kyanoviridae, Straboviridae, and unclassified Caudoviricetes, which together represented 14.02% of the viral reads, we detected abundant viral families known to infect protists (Koonin and Yutin 2019), such as Mimiviridae (7.72%), Schizomimiviridae (5.38%), Pithoviridae (0.76%), and Marseilleviridae (0.51%). Viral families with members capable of infecting amphibians were also present (Davison et al. 2009, Krupovic et al. 2018, Harvey and Holmes 2022), including Retroviridae (6.23%), Iridoviridae (4.29%), and Orthoherpesviridae (1.42%) (Fig. 1B). Given that the data set did not include any metatranscriptome, and that the library preparation did not include any reverse-transcription step, reads assigned to the class Pisoniviricetes and the family Coronaviridae (ssRNA viruses without DNA intermediates) are likely classification artifacts. Knowing the amphibian health risk represented by ranaviruses (family Iridoviridae) and the reported Bd-Ranavirus co-infections (Gray and Chinchar 2015, Herczeg et al. ) we tested if the abundance of reads classified as Iridoviridae was associated with the presence of Bd or with the abundance of reads assigned to the genus Batrachochytrium. The Iridoviridae abundance did not differ significantly between Bd-infected and non-infected individuals, but a positive and significant association with Bd abundance was detected (Supplementary Figure S3).

Bacteriophage alpha and beta diversity exhibit differences across seasons and sites

To evaluate the factors associated with bacteriophage diversity and composition we compared alpha and beta diversity across seasons, sampling sites, and Bd infection status. We found significant seasonal variation for Shannon diversity (Kruskal-Wallis, H = 8.42, P = 0.038; Fig. 2A), particularly between Summer and Winter (Dunn’s test, Z = −2.86, P = 0.026). The difference in diversity between Summer and Winter was also found with the Simpson index (Dunn’s test, Z = −2.68, P = 0.045; Supplementary Figure S4A) and generalized linear models (GLMs), that showed higher Shannon (Model Shannon1; β = 5.65, P = 0.014) and Simpson diversity values (Model Simpson1; β = 1.8, P = 0.017) in Winter compared to Summer (Supplementary Table S7). Regarding beta diversity (Aitchison distance), season did not significantly explain variation in overall community composition (PERMANOVA, R² = 0.089, P = 0.263; Fig. 2D) (Supplementary Table S4).

Figure 2.

Differences in bacteriophage Shannon diversity across(A) seasons, (B) sites, and (C) A. altamirani Bd infection status. Asterisk indicates pairwise post-hoc Dunn test with P < 0.05. Bacteriophage beta diversity across (D) seasons, (E) sites and (F) A. altamirani Bd infection status represented as principal coordinate analysis (PCoA) based on Aitchison distance. Ellipses denote 95% confidence intervals. Test statistics are displayed only for significant comparisons (P-value < 0.05). Complete statistical test results are provided in Supplementary Table S4.

We did not find significant differences in Shannon (Fig. 2B) nor Simpson alpha diversity (Supplementary Figure S4B) across sites (Supplementary Table S4). However, beta diversity analyses showed that community composition differed significantly among sites (Fig. 2E), with Site D and C showing the greatest pairwise dissimilarity (R² = 0.17, P = 0.015), and Site A and C the lowest (R² = 0.07, P = 0.016) (Supplementary Table S4).

No significant differences in alpha or beta diversity were observed between Bd-infected and non-infected samples using Wilcoxon or PERMANOVA tests, respectively (Fig. 2C and F) (Supplementary Table S4).

To identify specific OTUs contributing to the observed differences in alpha and beta diversity, we performed Linear Discriminant Analysis Effect Size (LEfSe) across seasons, sites, and Bd infection status (Supplementary Figure S5). With respect to seasons, Spring had the highest amount of enriched OTUs (4), followed by Winter (3), Autumn (2), and Summer (1). In agreement with the PERMANOVA results, LEfSe detected more differentially abundant OTUs across sites than across seasons or Bd infection status. Site D showed the highest number of enriched OTUs (20), followed by Site C (11), and Site A (3). Non-infected samples had eight enriched OTUs, while only two were found in infected samples. All enriched viral OTUs were bacteriophages belonging to the class Caudoviricetes, most of which were associated with hosts from the bacterial classes Gammaproteobacteria, Actinomycetes, Betaproteobacteria, Alphaproteobacteria, Bacilli, and Flavobacteriia (Supplementary Table S5).

Differences in water physicochemical conditions explain differences in bacteriophage diversity

All measured water physicochemical variables (seasonal mean temperature, seasonal Δ temperature, pH, dissolved oxygen, and conductivity) showed significant differences across seasons and sites of samples analyzed in this study, except for dissolved oxygen which did not differ across sites (Supplementary Figure S6; Supplementary Table S6).

To evaluate whether physicochemical variables could contribute to observed differences in alpha diversity, we fitted additional GLMs that combined categorical (Season, Site, Bd Infection status) and continuous environmental predictors. Separate models were constructed for observed Shannon and Simpson diversity (Model Shannon 2 and Model Simpson 2, respectively). For both diversity metrics, these extended models showed improved fit (i.e. lower deviance and AIC values; see Table 1), suggesting that environmental conditions may mediate part of the variation previously attributed to categorical factors. Specifically, Shannon was significantly associated with mean temperature, Δ temperature, conductivity, and Site C (Fig. 3A; Supplementary Table S7). Similarly, Simpson diversity was significantly associated with mean temperature, and Site C (Fig. 3B; Supplementary Table S7).

Table 1.

Summary of generalized linear models (GLMs) evaluating the effects of categorical and environmental variables on phage Shannon and Simpson diversity values. The table shows the improved model fit after including water physicochemical variables (lower deviance and AIC in bold).

Model	Explanatory variables	DF	Deviance	AIC
Shannon1	Season + Site + Bd Infection status	31	722.77	235.77
Shannon2	Season + Site + Bd Infection status + Mean Temp. + Δ Temp. + pH + Diss. O2 + Conductivity	26	491.59	231.12
Simpson1	Season + Site + Bd Infection status	31	78.94	151.6
Simpson2	Season + Site + Bd Infection status + Mean Temp. + Δ Temp. + pH + Diss. O2 + Conductivity	26	58.37	150.1

Figure 3.

Effects of physicochemical variables on bacteriophage diversity. Significant model estimates (± SE) for (A) Shannon diversity and (B) Simpson diversity, respectively. Red points indicate negative estimates; blue points indicate positive estimates. (C) Distance-based redundancy analysis (dbRDA) ordination based on Aitchison distance. Vectors (uniformly scaled 2x for visualization) represent the direction and strength of correlation between the environmental variables and the displayed canonical axes.

The contribution of environmental variables on beta diversity was tested using distance-based redundancy analysis (dbRDA). The first and second canonical components explained 62.6% and 16.2% of the total variation, respectively. ANOVA on the dbRDA model revealed that mean temperature (F = 2.04, P = 0.032) and dissolved oxygen (F = 2.5, P = 0.017) significantly influenced bacteriophage community composition (Supplementary Table S8). Although the ordination plot revealed a clear separation of samples by site (Fig. 3C), comparisons of adjusted R² values from models including only categorical predictors showed that season explained a greater proportion of variance (R² = 0.084) than site (R² = 0.045). This suggests that seasonal variation contributes more strongly to the observed community structure and that the apparent site-specific differences are largely driven by underlying water physicochemical conditions rather than spatial location per se.

Seasonal variation shapes bacteriophage-bacteria co-abundance networks

To evaluate whether bacteriophage alpha and beta diversity were associated with bacterial diversity, we performed Spearman correlations between bacteriophage and bacterial Shannon and Simpson diversity (alpha diversity), and a Mantel test between bacteriophage and bacterial Aitchison distances (beta diversity). Spearman rank correlations revealed a positive association between bacteriophage and bacterial Shannon diversity (Rho = 0.457, P = 0.0043). A similar positive association was observed for the Simpson diversity (Rho = 0.336, P = 0.039). The Mantel test further indicated a significant and strong covariation between bacteriophage and bacterial community composition (r = 0.924, P = 0.001; Supplementary Figure S7).

Given the observed influence of seasonality on bacteriophage alpha and beta diversity and the strong correspondence between bacteriophage and bacterial diversity patterns, we constructed three (summer, autumn, and winter) co-abundance networks from samples within Site A, to explore how seasonal dynamics shape potential bacteriophage-bacteria associations. Network topology differed significantly among seasons (Fig. 4A and B; Supplementary Table S9), with the summer network exhibiting a higher clustering coefficient and a lower average shortest path length, suggesting a more connected and cohesive structure than those of Autumn and Winter. This pattern was reflected in the nodes directly involved in bacteriophage–bacteria connections, which showed seasonal shifts in the proportion of positive and negative correlations (Fig. 4C–F).

Figure 4.

Bacteriophage-bacteria co-abundance networks across season at Site A. Boxplots of network (A) clustering coefficients and (B) average shortest path lengths. Letters denote pairwise significant differences based on Dunn’s test (Supplementary Table S9). (C) Barplots showing the proportion of positive and negative bacteriophage-bacteria correlations across seasons. Co-abundance networks for (D) summer, (E) autumn, and (F) winter showing only bacteriophage-bacteria edges. Bacterial and bacteriophage nodes are included in large and small circles, respectively. Core microbial taxa (present in the three networks) are depicted as black nodes. (G–I) Circos plots showing the bacteriophage nodes (bottom-colored segments) present in every network linked by colored ribbons to their associated hosts (top-colored segments). Dark grey and light grey segments represent phages without family-level classification and unknown host, respectively. The names of phage and bacterial families represented only in one or two networks are shown in the figure. Phage families represented in the three networks and their associated hosts are indicated by shapes: ° = Vilmaviridae, □ = Vandenendeviridae, △ = Straboviridae, ▽ = Kyanoviridae, ◇ = Drexlerviridae, = Chimalliviridae, = Autonotataviridae, ● = Alcaligenaceae, ■ = Burkholderiaceae, ▲ = Comamonadaceae, ▼ = Flavobacteriaceae, ◆ = Lysobacteraceae, ★ = Procholorococcaceae, + = Pseudomonadaceae, X = Streptomycetaceae, = Synechococcaceae.

Despite these differences, 74 bacterial and 11 bacteriophage nodes were consistently part of a microbial core that was shared across the three seasons (Fig. 4D–F; Supplementary Figure S8A-B), representing on average 35.42% and 39.67% of bacterial and bacteriophage nodes, respectively. Core bacterial nodes were mostly represented by the families Chitinophagaceae, followed by Burkholderiaceae, Sphaerotilaceae, Oxalobacteraceae, Spirosomataceae, Methylophilaceae, Flavobacteriaceae, and others (Supplementary Tables S10-S11). Most of the bacteriophage nodes belonged to the family Chimalliviridae, followed by Autonotataviridae, Kyanoviridae, Vandenendeviridae, Straboviridae, Drexlerviridae, and Vilmaviridae. These phages were mainly associated with bacterial hosts (Virus–Host Database) from the family Pseudomonadaceae, followed by Lysobacteraceae, Synechococcaceae, Burkholderiaceae, Comamonadaceae, Flavobacteriaceae, Streptomycetaceae, Procholorococcaceae, and Alcaligenaceae (Fig. 4G–I). Summer and Autumn had phages associated with bacteria from the families Mycobacteriaceae, Hafniaceae, and Yersiniaceae (not present in Winter), whereas Autumn and Winter had phages associated with Gordoniaceae, Aeromonadaceae, and Pectobacteriaceae (not present in Summer), and one phage associated with a host from the family Thermaceae was only present in Autumn (Fig. 4G–I).

Genomic diversity and functional potential of amphibian skin bacteriophages

For each sampling site, metagenomes collected during the same season were co-assembled, producing 13 co-assemblies from which a total of 155 916 contigs were recovered (Supplementary Table S12). These contigs, along with 50 bacterial metagenome-assembled genomes (MAGs) (Martínez-Ugalde et al. 2024) and 24 genomes from bacteria isolated from the skin of A. altamirani (González-Serrano et al. 2025), were screened for viral signatures, resulting in 2 541 putative assembled viral sequences. Following quality assessment and dereplication, 63 viral sequences were recovered, ranging in quality from medium to complete (Supplementary Table S13).

Of the 63 assembled viral sequences, 61 were identified as bacteriophages of the class Caudoviricetes, one as a filamentous bacteriophage of the class Faserviricetes, and one as a Revtraviricetes-related sequence. The 62 assembled bacteriophage sequences were affiliated with nine clusters of reference dsDNA prokaryotic viruses based on a genomic similarity score (S_G) ≥ 0.075 (Fig. 5). The average S_G of 0.206 indicates substantial novelty among the bacteriophages recovered in this study. Some of the highest S_G values were observed for prophages integrated into the genomes of isolated bacteria. The bacterial host taxonomy (from both isolated bacteria and MAGs), together with in silico host predictions (iPHoP), suggests that most amphibian-associated bacteriophages are linked to Betaproteobacteria (30), followed by Gammaproteobacteria (12), and Bacilli (10), which was highly consistent with the host class of similar reference bacteriophages (ViPTree) (Fig. 5). Furthermore, four of the identified bacteriophages were predicted to infect Alphaproteobacteria, another four Flavobacteriia, and two Actinomycetes. The four sequences predicted to infect Flavobacteriia, one predicted to infect Alphaproteobacteria and one predicted to infect Betaproteobacteria could not be affiliated with any reference virus at the established S_G cutoff (Fig. 6, Supplementary Table S13).

Figure 5.

Taxonomic clusters based on genomic similarity scores (S_G). The 62 assembled bacteriophage sequences are shown as yellow circles. Reference dsDNA prokaryotic viruses are represented as circles with their corresponding host class. Edges are depicted according to the S_G values. Clusters were arranged using an edge-weighted spring-embedded layout; consequently, the edge length is also indicative of the S_G values.

Figure 6.

Stacked bar plots representing the abundance of functional categories (excluding “Other”) across the 62 assembled bacteriophage sequences. The predicted phage Class, phage type (Temperate/Undetermined), host Class, and corresponding taxonomic clusters are also indicated. The total CDS count (including those under the “Other” category) is indicated with numbers at the end of each bar.

Of the 62 identified bacteriophages, 58 were predicted to be temperate, whereas the replication strategy of four was undetermined (Fig. 6). Functional annotations of the assembled bacteriophage sequences (aside from the “Other” category, which accounted for 68.49% of the CDS) revealed that structural proteins were the most abundant (18.29%), followed by functions related to nucleotide metabolism (4.93%), transcription regulation (2.86%), integration and excision (2.01%), and lysis (1.98%) (Fig. 6) (Supplementary Table S14). Among the less frequent categories (< 1%) immune evasion was the best represented with 34 CDS (including methyltransferases, anti-CRISPR, anti-CBASS and anti-Pycsar), followed by superinfection exclusion (8) (e.g. PF14373, PF07751, and PF07553), toxin/antitoxin (5) [potentially involved in genetic element maintenance, virulence, stress resistance, phage inhibition (Qiu et al. 2022), or immune evasion (LeRoux and Laub 2022)], AMGs (4), and virulence (1).

A protein identified as 7,8-dihydroneopterin aldolase or FolB was found in four different assembled bacteriophage sequences. FolB is a lyase which takes part of the folate biosynthetic pathway by catalyzing the removal of a glycolaldehyde from 7,8-dihydroneopterin to form 6-(Hydroxymethyl)-7,8-dihydropterin. Different enzymes involved in the folate biosynthetic pathway have been identified as AMGs in other systems (Anderson et al. 2017, Graham et al. 2023). Folate and its active derivative tetrahydrofolate (THF) are cofactors involved in a variety of metabolisms (Ragsdale and Pierce 2008). However, given that THF is also involved in one-carbon reactions associated with nucleotide and amino acid biosynthesis it is likely that these phage-encoded enzymes play a role in the phage replication cycle rather than being true AMGs (Martin et al. 2025).

Finally, one CDS was annotated as a PGA cap domain (PF09587) which is found on proteins involved in poly-gamma-glutamate (PGA) capsule biosynthesis in bacilli and staphylococci bacteria. PGA is a polymer that forms a capsule that protects the bacteria from the immune system, therefore acting as a virulence factor (Candela and Fouet 2006).

Discussion

In this study, we characterized the viral community associated with the skin of the endangered axolotl A. altamirani, revealing a virome dominated by bacteriophages of the class Caudoviricetes. While their predominance is consistent with their global abundance (Ackermann and Prangishvili 2012, Gregory et al. 2019, Du et al. 2023, Boeckel et al. 2026), the key insight here is their tight ecological integration within the amphibian skin microbiome, which reinforces the idea that bacteriophages may play a central ecological role in the amphibian skin (Woodhams et al. 2023).

The strong positive association between bacterial and bacteriophage diversity suggests a close coupling between these communities. Such patterns can arise through multiple, non-mutually exclusive mechanisms. On one hand, bacteriophages may exert top-down control on bacterial populations, for instance through kill-the-winner dynamics, thereby promoting diversity (Winter et al. 2010, Chen et al. 2019, Liang et al. 2020). On the other hand, environmentally driven variation in bacterial communities may impose bottom-up constraints on bacteriophage diversity, particularly given the typically narrow host ranges of many phages (Ndiaye et al. 2025). Although our data do not allow to disentangle these mechanisms, the observed coupling is consistent with growing evidence that bacteriophage diversity often follows, rather than drives, bacterial community dynamics (Castledine and Buckling 2024, Ndiaye et al. 2025).

Environmental variation, linked to location or seasonality, emerged as a major factor structuring bacteriophage communities. Seasonal shifts in physicochemical parameters such as temperature, pH, and dissolved oxygen are well known to influence viral communities across aquatic and terrestrial systems (Adriaenssens et al. 2017, Arkhipova et al. 2018, Gregory et al. 2019, Hevroni et al. 2020, Alanazi et al. 2022) as well as bacterial communities in a range of ecosystems (Shigyo et al. 2019, Mestre et al. 2020, Palladino et al. 2021, Bell et al. 2024), including the amphibian skin (Longo et al. 2015, Estrada et al. 2019, Kueneman et al. 2017, Zhang et al. 2024) and the axolotl A. altamirani (Martínez-Ugalde et al. 2022). Similar drivers likely underpin the bacteriophage diversity patterns observed here, with the parallel seasonal structuring of bacterial and bacteriophage communities further supporting the idea that environmental conditions indirectly shape phage diversity through their effects on host availability, while also potentially exerting direct effects on phage physiology, including infectivity and replication strategies (Shan et al. 2014, Jurczak-Kurek et al. 2016, Howard-Varona et al. 2017).

Although integration and excision genes and integrated prophages are commonly overrepresented in whole shotgun metagenomes in comparison to viral-like particle enriched metagenomes (Kosmopoulos et al. 2024), the prevalence of positive associations in co-abundance networks (Shkoporov et al. 2022), together with the widespread detection of prophage-like sequences, points toward lysogeny as a common strategy in the amphibian skin microbiome. Temperate phages can influence bacterial communities not only through infection dynamics but also via HGT (Chaing et al. 2019, Kondo et al. 2021) and lysogenic conversion (Hannigan et al. 2015, Chevallereau et al. 2022, Luo et al. 2022, Graham et al. 2023). The functional repertoire identified here—including genes putatively associated with auxiliary metabolism and virulence—supports the idea that bacteriophages may contribute to bacterial adaptation and ecological interactions on the amphibian skin. Although these annotations require experimental validation, they highlight the potential for phages to modulate microbial function beyond simple predator–prey dynamics.

Despite the apparent ecological relevance of bacteriophages in this biological system, we did not detect a clear association between bacteriophage diversity and Bd infection status. This result is consistent with previous observations of limited bacterial community shifts in these populations despite high Bd prevalence and may reflect a degree of host tolerance or microbiome stability (Basanta et al. 2023, Martínez-Ugalde et al. 2022). While speculative, the positive association between bacterial and bacteriophage diversity could indicate a relatively stable and healthy microbial system (Wheatley et al. 2025). However, further work is needed to determine whether phages play any direct or indirect role in mediating host–pathogen interactions. For instance, two Pseudomonas aeruginosa phages have been reported to directly inhibit pathogenic biofilm-forming ascomycete fungi such as Aspergillus fumigatus and Candida albicans in vitro (Penner et al. 2016, Nazik et al. 2017).

Beyond bacteriophages, we identified several viral families with host ranges extending to amphibians, most notably the family Iridoviridae, which includes important amphibian pathogens such as ranaviruses (Gray and Chinchar 2015). This prompted us to explore potential association between Iridoviridae abundance and the presence and abundance of Bd. However, the absence of significant differences in Iridoviridae abundance between Bd-infected and non-infected samples, alongside a positive association with Batrachochytrium abundance, points to a weak and likely indirect relationship between these pathogens, potentially reflecting shared environmental drivers or host susceptibility gradients. Additionally, we detected signatures of viruses associated with microeukaryotes, suggesting that protists may represent an important, yet underexplored, component of the amphibian skin microbiome (Kueneman et al. 2017, Rebollar et al. 2020, Belasen et al. 2021, Martínez-Ugalde et al. 2024, Bouchali et al. 2025). However, the presence of certain eukaryotic viral groups should be interpreted with caution. Particularly coronavirus-like sequences which were most likely classification artifacts, and retrovirus- and adintovirus-like sequences, which may represent endogenous elements (Krupovic et al. 2018, Starrett et al. 2021) rather than active infections.

Although this study provides valuable insights into an underexplored component of amphibian skin microbiomes, several limitations should be considered. A major constraint was the underrepresentation of viral reads in the whole-shotgun metagenomic datasets (<1%), which hindered the recovery of high-quality phage genomes and restricted diversity analyses to taxonomically annotated unassembled reads. Although Kaiju is considered a high-performing taxonomic classifier and often performs better on short reads than on assembled contigs (Di Gloria et al. 2025), it may still generate a non-negligible rate of false-positive assignments (Ye et al. 2019, Di Gloria et al. 2025). Consequently, the detection of RNA viruses from reads generated through a DNA-based library preparation should be interpreted with caution.

Another limitation of viral analyses based on whole-shotgun metagenomes is the potential overrepresentation of integration and excision genes, which were used here as indicators of a temperate lifestyle. The persistence of these genes likely reflects the presence of integrated prophages within bacterial genomes rather than the recovery of genomes from free viral particles (Kosmopoulos et al. 2024). The substantial taxonomic novelty observed in this study underscores the largely unexplored viral diversity associated with the amphibian skin and, together with the technical limitations discussed above, highlights the need for more robust viromic analyses based on viral enrichment methods (Hall et al. 2014). Such approaches generally recover a broader representation of viral communities and may lead to different ecological interpretations (Kosmopoulos et al. 2024). Nevertheless, although the effect of methodological choice on viral community characterization appears to be sample-dependent, studies of human stool viruses suggest that both whole-shotgun metagenomics and viral-enrichment approaches can effectively recover the underlying biological patterns (e.g. alpha and beta diversity) despite their specific methodological biases (Zhang et al. 2023).

In this study, we show that bacterial and bacteriophage diversity in the skin of the axolotl A. altamirani are tightly coupled with environmental fluctuations, and that assembled bacteriophage sequences encode a functional repertoire including potential AMGs and virulence factors. These findings provide a foundation for future studies aimed at disentangling the ecological and functional roles of viruses in amphibian health and disease, including studies on the interactions between viruses and other components of the skin microbiome (Herczeg et al. 2021) and host-associated factors such as the immune system or host genetic diversity (Monroy-Vilchis et al. 2019, Rebollar et al. 2020, Cianci et al. 2025). These studies will also contribute to determining if there are viral diversity patterns that can be associated with the host health status (Woodhams et al. 2023, Wheatley et al. 2025) or if phages can participate in non-host-derived immunity, for example, through the BAM model (Barr et al. 2013) or by contributing to biofilm formation (Chen et al. 2022, Huang et al. 2024). In the light of the pivotal roles of bacteriophages in diverse environments (Zhang et al. 2022) and host-associated systems (Metcalf and Bordenstein 2012, Hsu et al. 2019, Jahn et al. 2019, Braga et al. 2021, Yan et al. 2023), it is imperative to persist in the exploration of the ecological roles of bacteriophages in the skin microbiomes and their potential impact on host health, particularly in amphibians, which are currently facing numerous threats including emerging diseases (Luedtke et al. 2023).

Author contributions

Alejandro M Cisneros-Martínez (Conceptualization, Data curation, Formal Analysis, Visualization, Writing – original draft, Writing – review & editing), Miguel A Flores Varela (Formal Analysis, Writing – review & editing), Francisco Maximiliano González-Serrano (Formal Analysis, Investigation, Writing – review & editing), and Eria A Rebollar (Conceptualization, Funding acquisition, Investigation, Project administration, Visualization, Writing – original draft, Writing – review & editing)

Conflicts of interest

The authors declare no conflicts of interests.

Funding

This work was supported by Secretaría de Ciencia, Humanidades, Tecnología e Innovación (SECIHTI) Ciencia de Frontera 2019-373914 granted to E. A. R.