Host-linked viruses are predicted to infect key C cycling microbes.

In order to examine these viruses’ impacts on the Mire’s resident microbial communities and processes, we sought to link them to their hosts via emerging standard in silico host prediction methods, significantly empowered by the recent recovery of 1,529 MAGs from the site (508 from palsa, 588 from bog, and 433 from fen [17]). Tentative bacterial hosts were identified for 17 of the 53 vOTUs (Fig. 3; Table S3): these hosts spanned four genera among three phyla (Verrucomicrobia: Pedosphaera, Acidobacteria: Acidobacterium and “Candidatus Solibacter,” and Deltaproteobacteria: Smithella). Eight viruses were linked to more than one host but always within the same species. The four predicted microbial hosts are among the most abundant in the microbial communities and have notable roles in C cycling (16, 17). Three are acidophilic, obligately aerobic chemoorganoheterotrophs and include the Mire’s dominant polysaccharide-degrading lineage (Acidobacteria), and the fourth is an obligate anaerobe shown to be syntrophic with methanogens (Smithella). Acidobacterium is a highly abundant, diverse, and ubiquitous soil microbe (88,–90) and a member of the most abundant phylum in Stordalen Mire. The relative abundance of this phylum peaked in the bog at 29% but still had a considerably high relative abundance in the other two habitats (5% in palsa and 3% in fen) (17). It is a versatile carbohydrate utilizer, has recently been identified as the primary degrader of large polysaccharides in the palsa and bog habitats in the Mire, and is also an acetogen (17). Seven vOTUs were inferred to infect Acidobacterium, implicating these viruses in indirectly modulating a key stage of soil organic matter decomposition. The second identified acidobacterial host was in the newly proposed species “Candidatus Solibacter usitatus,” another carbohydrate degrader (91). The third predicted host was Pedosphaera parvula, within the phylum Verrucomicrobia, which is ubiquitous in soil, is abundant across our soils (~3% in palsa and ~7% in bog and fen habitats, based on metagenomic relative abundance [17]), and utilizes cellulose and sugars (92,–96), and in this habitat, this organism could be acetogenic (17). Last, vOTU_28 was linked to the deltaproteobacterium Smithella sp. strain SDB, another acidophilic chemoorganoheterotroph but an obligate anaerobe, with a known syntrophic relationship with methanogens (97, 98). Collectively, these virus-host linkages provide evidence for the Mire’s viruses to be impacting the C cycle via population control of relevant C-cycling hosts, consistent with previous results in this system (46) and other wetlands (99).

Open in a separate window

FIG 3

Viral-host linkages between vOTUs and MAGs. Seventeen vOTUs were linked to 4 host lineages by multiple lines of evidence, with 15 linked by CRISPRs (solid line) and 2 linked by BLAST (dashed line). Node shape denotes organism (oval for microbe and triangle for virus), and number references vOTU or bin (see Table S2). Viral nodes are color coded by habitat of origin (green for bog and blue for fen).

We next sought to examine viral AMGs for connections to C cycling. To more robustly identify AMGs than by the standard protein family-based search approach, we used a custom-built in-house pipeline previously described in the work of Daly et al. (100) and further tailored to identify putative AMGs based on the metabolisms described in the 1,529 MAGs recently reported from these same soils (17). From this, we identified 30 AMGs from 13 vOTUs (Fig. 4; Tables S2 and S4), encompassing C acquisition and processing (three involved in polysaccharide binding, one involved in polysaccharide degradation, and 23 involved in central C metabolism) and sporulation. Glycoside hydrolases that help break down complex OM are abundant in resident microbiota (17) and may be especially useful in this high-OM environment; notably, they have been found in soil (at our site [46]), rumen (101, 158) and marine systems (albeit scarcely [73]). In addition, central C metabolism genes in viruses may increase nucleotide and energy production during infection and have been increasingly observed as AMGs (25,–35). Finally, two different AMGs were found in regulating endospore formation, spoVS and whiB, which aid in formation of the septum and coat assembly, respectively, improving spores’ heat resistance (102, 103). A WhiB-like protein has been previously identified in mycobacteriophage TM4 (WhiBTM4) and experimentally shown to not only transcriptionally regulate host septation but also cause superinfection exclusion (i.e., exclusion of secondary viral infections [104]). While these two sporulation genes have been found only in Firmicutes and Actinobacteria, the only vOTU to have whiB was linked to an acidobacterial host (vOTU_178 [Fig. 4]). A phylogenic analysis of the whiB AMG grouped it with actinobacterial versions and, more distantly, with another mycobacteriophage (Fig. 4), suggesting either (i) misidentification of host (unlikely, as it was linked to three different acidobacterial hosts, each with zero mismatches of the CRISPR spacer), (ii) that the virus could infect hosts spanning the two phyla (unlikely, as only ~1% of identified virus-host relationships span phyla [45]), or (iii) that the gene was horizontally transferred into the Acidobacteria. Identification of these 30 diverse AMGs (carried by 25% of the vOTUs) suggests a viral modulation of host metabolisms across these dynamic environments and supports the findings from bulk metagenome-derived viruses of Emerson et al. (46) at this site. That study’s AMGs spanned the same categories as those reported here, except for whiB, which was not found, but the study did not discuss them, other than the glycoside hydrolases, one of which was experimentally validated.

Open in a separate window

FIG 4

Characterization of select AMGs. FastTree phylogenies were constructed for select AMGs (one from each group), and their structures and those of their nearest neighbors were predicted using I-TASSER (detailed in Table S3). Tree lineages are shaded blue for bacteria and red for viruses. “vOTU” sequences are from the 53-vOTU virome-derived data set, while “SoilVir contig” represents homologs from the 378 probable viral contigs. The predicted protein structure for each AMG is labeled (A, B, C, or D) to match the location in the corresponding phylogenetic tree. The first predicted model for each soil virus is shown and was used for the TM-align comparison. The scale bar indicates the number of substitutions per site.

Thus far, the limited studies of soil viruses have identified few AMGs relative to studies of marine environments. This may be due to undersampling or to difficulties in identifying AMGs; since AMGs are homologs of host genes, they can be mistaken for microbial contamination (105) and thus are more difficult to discern in bulk-soil metagenomes (whereas marine virology has been dominated by viromes); also, microbial gene function is more poorly understood in soils (106). Alternately, soil viruses could indeed carry fewer AMGs. One could speculate on a link between host lifestyle and the usefulness of carrying AMGs; most known AMGs are for photo- and chemoautotrophs (73, 107, 108), although this may be due to more studies of these metabolisms or phage-host systems. Thus far, soils are described as dominated by heterotrophic bacteria (109,–113), and if AMGs were indeed less useful for viruses infecting heterotrophs, that could explain their limited detection in soil viruses. However, a deeper and broader survey of soil viruses will be required to explore this hypothesis.

Evaluating sample storage on vOTU recovery.

While our previous research demonstrated that differing storage conditions (frozen versus chilled) of these Arctic soils did not yield different viral abundances (by direct counts [41]), the impact of storage method on viral community structure was unknown. Here, we examined that in the palsa and bog habitats for which viromes were successfully recovered from both storage conditions. Storage impacted recovered community structure only in the bog habitat, with separation among the viromes’ reads (Fig. 5A and ​andB)B) and a broader recovery of vOTUs from the chilled sample (Fig. 5C and ​andD),D), leading to higher diversity metrics (Fig. S3) and appreciable separation of the recovered chilled-versus-frozen bog vOTU profiles in ordination (Fig. 5E). The greater recovery of vOTUs from the chilled sample was likely partly due to higher DNA input and sequencing depth, which was 107-fold more than bog frozen replicate A (BFA) and 350-fold more than bog frozen replicate B (BFB). This led to 1.6- to 9-fold more reads assembling into contigs (compared to viromes BFA and BFB, respectively [Table 1]) and 3.5- to 9-fold more distinct contigs; while one might expect that as the number of reads increased, a portion would assemble into already-established contigs, that was not observed. This higher proportional diversity in the chilled bog virome than in the two frozen ones could have several potential causes. Freezing might have decreased viral diversity by damaging viral particles, although these viruses regularly undergo freezing (albeit not with the rapidity of liquid nitrogen). Alternatively, there could be a persistent metabolically active microbial community under the chilled conditions with ongoing viral infections, distinct from those in the field community. Finally, there could have been bog-specific induction of temperate viruses under chilled conditions (since this difference was not seen in the palsa samples). The bog habit is very acidic (pH ~4 versus ~6 in palsa and fen [11, 46]), with a dynamic water table, and both of these have been hypothesized or demonstrated to increase selection for temperate viruses (80, 114,–118). In addition, galacturonic acid, a compound produced by the Sphagnum moss that dominates the bog habitats and is present at our site, is a known inhibitor of microbial activity and inferred to shape microbial communities (reviewed in references 119 and 120) and is an important bog chemical at this site (11, 121). In addition, of the 19 vOTUs shared between this study and the bulk-soil metagenome study of Emerson et al. (46) (in which the soil was likely to be enriched for temperate viruses based on its majority sampling of microbial DNA), 13 were unique to the bog, and of those, 10 were present in only the chilled rather than frozen viromes and the remaining 3 were enriched in the chilled viromes.

Open in a separate window

FIG 5

Viral community structure across the thaw gradient. (A and B) Social network analyses of the total reads from the seven viromes (A) and all the reads mapped to the 53 vOTUs (B). Dots in the social networks represent statistical samples taken from the marginal posterior distributions (Bayesian method), and each habitat is denoted by a black circle. (C) Euler diagram relating the seven viromes and their 53 vOTUs. (D) The relative abundance of vOTUs (columns) in the seven viromes (rows). Reads were mapped to this nonredundant set of contigs to estimate their relative abundance. “Chilled” and “Frozen” indicate sample storage at 4°C or flash-freezing in liquid nitrogen and storage at −80°C. “A” and “B” denote technical replicates. Dots after contig names indicate membership in a viral cluster, and fill color indicates habitat specificity. (E) Principal-coordinate analysis of the viromes by normalized relative abundance of the 53 vOTUs.

Finally, while the chilled bog sample was an outlier to all other viromes (Fig. 5E), a social network analysis (also known as k-mer analysis) of the total reads (Fig. 5A) and of the reads that mapped to the viromes (Fig. 5B) indicated that habitat remained the primary driver of recovered communities Because of this, the diversity analyses were redone with the chilled bog sample taken out (Fig. S3B) instead of subsampling the reads (in addition to the virome normalization that we had already done, called “total-sum scaling” [described in references 26, 56, and 74]), because this is a smaller data set (subsampling smaller data sets described further in reference 122) and the storage effect was observed only for the bog.

Habitat specificity of the 53 vOTUs along the thaw gradient.

We explored the ecology of the recovered vOTUs across the thaw gradients, by fragment recruitment mapping against (i) the viromes and (ii) bulk-soil metagenomes. Virome mapping revealed that the relative abundance of each habitat’s vOTUs increased along the thaw gradient; relative to the palsa vOTU abundances, bog vOTUs were 3-fold more abundant and fen vOTUs were 12-fold more abundant (Fig. 5D). This is consistent with overall increases in virus-like particles with thaw observed previously at the site via direct counts (41). Only a minority (11%) of the vOTUs occurred in more than one habitat, and none were shared between the palsa and fen (Fig. 5C). Consistent with this, principal-coordinate analyses (PCoA; using a Bray-Curtis dissimilarity metric) separated the vOTU-derived community profiles according to habitat type, which also explained ~75% of the variation in the data set (Fig. 5E). Mapping of the 214 bulk-soil metagenomes from the three habitats (17) revealed that a majority (41; 77%) of the vOTUs were present in the bulk-soil metagenomes (Fig. 6), collectively occurring in 62% (123) of them. Of the 41 vOTUs present, most derived from the bog, and their distribution among the 133 metagenomes reflected this, peaking quite dramatically in the bog (Fig. S4). This strong bog signal in the bulk-soil metagenomes—both in the proportion of bog-derived vOTUs present in the bulk metagenomes and in the abundance of all vOTUs in the bog samples—is consistent with the hypothesized higher abundance of temperate viruses in the bog, suggested by the chilled-versus-frozen storage results above. Overall, vOTU abundances in larger and longer-duration bulk-soil metagenomes indicated less vOTU habitat specificity than in the seven viromes: 10% were unique to one habitat, 22% of vOTUs were present in all habitats, 22% were shared between palsa and bog, 27% were shared between palsa and fen, and 68% were shared between bog and fen (Fig. 6). The difference in observations from vOTU read recruitment of viromes and bulk-soil metagenomes could be due to many actual and potential differences, arising from their different source material (but from the same sites) and different methodology, including vOTUs’ actual abundances (they derive from different samples), infection rates, temperate versus lytic states, burst size, and/or virion stability and extractability.

Open in a separate window

FIG 6

vOTU abundance in 133 bulk-soil metagenomes. The heat map represents abundance of vOTUs (rows) in the bulk-soil metagenomes (columns); metagenome reads were mapped to the nonredundant set of contigs to estimate their relative abundances. Only the 41 vOTUs present in the metagenomes (out of 53) and the 63 bulk-soil metagenomes (out of 214) that contained matches to the vOTUs are shown. Metagenome names denote source: habitat of origin (P, palsa; S, bog; E, fen), soil core replicate (1, 2, or 3), depth (3-cm intervals denoted with respect to geochemical transitions; generally, S=1 to 4cm, M=5 to 14cm, D=11 to 33cm, and X=30 to 50cm), month collected (5 to 10 for May to October, respectively), and year collected (2010, 2011, or 2012).

The vOTUs’ habitat preferences observed in both read data sets are consistent with our network analytics (Fig. 2), with the numerous documented physicochemical and biological shifts along the thaw gradient, and with observations of viral habitat specificity at other terrestrial sites. Changes in physicochemistry are known to impact viral morphology (reviewed in references 37, 124, and 125) and replication strategy (36, 37). In addition, at Stordalen Mire (and at other, similar sites [112]), microbiota are strongly differentiated by thaw-stage habitat, with some limited overlap among “dry” communities (i.e., those above the water table, the palsa and shallow bog) and among “wet” ones (those below the water table, the deeper bog and fen) (15,–17). These shifting microbial hosts likely impact viral community structure, creating distinctive viral communities across the palsa, bog, and fen habitats. Expanding from the 53 vOTUs examined here, the recent analysis by Emerson et al. (46) of nearly 2,000 vOTUs recovered from the bulk-soil metagenomes also showed strong habitat specificity among the recovered vOTUs (only 0.1% were shared among all habitats, with <4.5% shared between any two habitats). These findings are also consistent with observations of distinct viral communities from desert, prairie, and rainforests (126) and from grasslands and arctic soils (45). In contrast, an emerging paradigm in the marine field is “seascape ecology” (127), where the majority of taxa are detected across broad geographical areas, as are marine viruses (7, 26). This important difference in habitat specificity between soils and oceans may be due to the greater physical structuring of soil habitats.

Although vOTU richness and diversity appeared to increase along the thaw gradient (roughly equivalent in palsa and bog and ~2-fold higher in fen, omitting the chilled bog sample [Fig. S3B]), this data set captured only a small fraction of the viral diversity (based on a collector’s curve comparison [Fig. S5]), and therefore, the undersampling prevents diversity inferences. Intriguingly, while our virome-derived vOTU richness was lowest in the palsa, the much greater sampling by Emerson et al. (46) recovered the most vOTUs in the palsa, more than double that in the fen (42% versus 18.9% of total vOTUs). This major difference could potentially be due to the known increase in microbial alpha diversity along the thaw gradient (16, 17), causing increased difficulty of viral genome reconstruction in the bulk-soil metagenomes; specifically, this could be due to poorer assembly of temperate phages within an increasingly diverse microbiota or of lytic or free viruses due to concomitantly increasing viral diversity (which is consistent with the increased vOTU richness with thaw in our virome data set). Notably, these diversity inferences are limited to double-stranded DNA (dsDNA) viruses, because our methods did not capture single-stranded DNA (ssDNA) or RNA viruses, a known part of the soil virosphere (128,–130).

vOTU recovery.

Eight viromes were prepared, and seven samples were successfully sequenced (2 palsa, one chilled and one frozen; 3 bog, one chilled and two frozen; and 2 fen, both chilled). The sequences were quality controlled using Trimmomatic (139) (adaptors were removed, reads were trimmed as soon as the per-base quality dropped below 20 on average on 4-nucleotide[nt] sliding windows, and reads shorter than 50bp were discarded) and then assembled separately with IDBA-UD (140), and contigs were processed with VirSorter to distinguish viral from microbial contigs (virome decontamination mode [69]). The same contigs were also compared by BLAST to a pool of putative laboratory contaminants (i.e., phages cultivated in the lab: Enterobacteria phage PhiX17, Alpha3, M13, Cellulophaga baltica phages, and Pseudoalteromonas phages) that we cultured. All contigs matching these genomes at more than 95% ANI were removed. VirSorted contigs were manually inspected by observing the key features of the viral contigs that VirSorter evaluates (e.g., the presence of a viral hallmark gene places the contigs in VirSorter category 1 or 2, but further inspection is needed to confirm that it is a genuine viral contig and not a GTA or plasmid). To identify GTAs, we searched through all of our contigs assembled by IDBA-UD for (i) taxa related to the 5 types of GTAs (keyword searches were done on Rhodobacterales, Desulfovibrio, Brachyspira, Methanococcus, and Bartonella) and (ii) microbial DNA from the SILVA rRNA database (release 128 [136]), with all the assembled contigs with ≥95% ANI. The percentage of reads that mapped to these contigs is described in Text S1 in the supplemental material.

After verification that the VirSorted contigs were genuine viruses, quality-controlled reads from the seven viromes were pooled and assembled together with IDBA-UD to generate a nonredundant set of contigs. Resulting contigs were rescreened as described above, removing all identifiable contamination. The contigs then underwent further quality checks by (i) removing all contigs of <10kb and (ii) using only contigs from VirSorter categories 1 and 2.

To detect putative archaeal viruses, the VirSorter output was used as an input for MArVD (with default settings [141]). The output putative archaeal virus sequences were then filtered to include only those contigs of ≥10kb in size, resulting in the set of putative archaeal vOTUs described here.

Viral genes were annotated using a pipeline described in the work of Daly et al. (100). Briefly, for each contig, open reading frames (ORFs) were freshly predicted using MetaProdigal (142), and sequences were compared to KEGG (143) and UniRef and InterProScan (144) using USEARCH (145), with single and reverse best-hit matches with a bit score greater than 60. AMGs were identified by manual inspection of the protein annotations guided by known resident microbial metabolic functions (identified in reference 17). To determine confidence in functional assignment, representatives for each AMG underwent phylogenetic analyses. First, each sequence was used for a BLAST search and the top 100 hits were investigated to identify main taxon groups. An alignment with the hits and the matching viral sequence (MUSCLE with default parameters [146]) was done with manual curation to refine the alignment (e.g., regions of very low conservation from the beginning or end were removed). FastTree (default parameters with 1,000 bootstraps [147]) was used to make the phylogeny, and iTOL (148) was used to visualize and edit the tree (any distance sequences were removed). To see if this AMG was widespread across the putative soil viruses, a BLASTp search (default settings) of each AMG against all putative viral proteins from our viromes was done. The sequences from identified homologs (based on a bit score of >70 and an E value of 10⁻⁴) were used with the AMG of interest to construct a new phylogenic tree (same methods as used before). Finally, structures were predicted using I-TASSER (149) for our AMGs of interest and their neighbors. To assess correct structural predictions, AMGs of interest and their neighbors’ structures were compared with TM-align (TM-score normalized by the length of the reference protein [150]).

Gene-sharing network construction, analysis, and clustering of viral genomes (fragments).

We built a gene-sharing network where the viral genomes and contigs are represented by nodes and significant similarities as edges (74, 75). We downloaded 198,556 protein sequences representing the genomes of 1,999 bacterial and archaeal viruses from NCBI RefSeq (v 75 [151]). Including protein sequences from the 53 Stordalen Mire viral contigs, a total of 199,613 protein sequences were subjected to all-to-all BLASTp searches (default parameters, an E value threshold of 10⁻⁴, and a bit score of 50) and defined as protein clusters (PCs) in the same manner as previously described (70). The resulting output was parsed in the form of a matrix comprising the genomes (vOTUs) and PCs. We then determined the similarities between genomes by calculating the probability of finding the number of PCs shared between the genomes and/or vOTUs, based on the hypergeometric formula as previously described (74, 75). A similarity score was obtained by taking the negative logarithm (base 10) of the hypergeometric P value multiplied by the total number of pairwise genome (vOTUs) comparisons (i.e., 2,052×2,051). Genome (vOTU) pairs with a similarity score of ≥1 were previously shown to be significantly similar through permutation test of PCs and/or singletons (proteins that do not have close relatives) between genomes (vOTUs). The resulting network comprising 1,772 viral genomes, including 53 vOTUs and 58,201 edges, was visualized with Cytoscape (version 3.1.1; http://cytoscape.org/), using an edge-weighted spring-embedded model, which places the genomes or fragments sharing more PCs closer to each other. There were 398 RefSeq viruses not showing significant similarity to vOTUs that were excluded for clarity. To gain detailed insights into the genetic connections, the network was decomposed into a series of coherent groups of nodes (also known as VCs [70, 72, 73]), with an optimal inflation factor of 1.6. Thus, the discontinuous network structure of individual components, together with the vOTUs, indicates their distinct gene pools (71). To assign vOTUs into VCs, PCs needed to include ≥2 genomes and/or genome fragments, then the Markov clustering algorithm was used, and the optimal inflation factor was calculated by exploring values ranging from 1.0 to 5 by steps of 0.2. The taxonomic affiliation was taken from the NCBI taxonomy (https://www.ncbi.nlm.nih.gov/taxonomy) and the International Committee on Taxonomy of Viruses (ICTV) taxonomy (ICTV Master Species List v1.3, as of February 2018).

vOTU ecology.

Virome reads were mapped back to the nonredundant set of contigs to estimate their coverage, calculated as number of base pairs mapped to each read normalized by the length of the contig, and by the total number of base pairs sequenced in the metagenome in order to be comparable between samples (Bowtie 2, threshold of 90% ANI on the read mapping, and 75% of contig covered to be considered detected [58, 152]). The heat map of the vOTU’s relative abundances across the seven viromes, as inferred by read mapping, was constructed in R (CRAN 1.0.8 package pheatmap).

The 214 bulk-soil metagenomes and 1,529 associated recovered MAGs used here for analyses were described in the work of Woodcroft et al. (17). The paired MAG reads were mapped to the viral contigs with Bowtie 2 (as described above for the virome reads). The heat map of the vOTU’s relative abundances across the 214 bulk-soil metagenomes, as inferred by read mapping, was constructed in R (CRAN 1.0.8 package pheatmap); only microbial metagenomes with a viral signal were shown.

Viral-host methodologies.

We used two different approaches to predict putative hosts for the vOTUs: one relying on CRISPR spacer matches (45, 46, 100, 153) and one relying on direct sequence similarity between virus and host genomes (154). For CRISPR linkages, Crass (v0.3.6, default parameters), a program that searches through raw metagenomic reads for CRISPRs, was used (further information in Table S3) (155). For BLAST, the vOTU nucleotide sequences were compared to the MAGs (17) as described in the work of Emerson et al. (46). Any viral sequences with a bit score of 50, E value threshold of 10⁻³, and ≥70% ANI across ≥2,500bp were considered for host prediction (described in reference 154).

Phylogenetic analyses to resolve taxonomy.

Two phylogenies were constructed. The first had the alignment of the protein sequences that are common to all Felixounavirinae and Vequintavirinae as well as vOTU_4, and the second had an alignment of select sequences from PC_03881, including vOTU_165. These alignments were generated using the ClustalW implementation in MEGA5 (version 5.2.1; http://www.megasoftware.net/). We excluded noninformative positions with the BMGE software package (156). The alignments were then concatenated into a FASTA file, and the maximum likelihood tree was built with MEGA5 using a JTT (Jones-Taylor-Thornton) model for each tree. A bootstrap analysis with 1,000 replications was conducted with uniform rates and a partial depletion of gaps for a 95% site coverage cutoff score.

We thank Bonnie Poulos and Christine Schirmer for their assistance on different stages of this project. We also thank SWES-MEL, TMPL, The University of Arizona Genetics Core facility, the MAVERIC lab at the Ohio State University, the Abisko Naturvetenskapliga Station, and the Joint Genome Institute for support. We thank Moira Hough, Robert Jones, and Rachel Wilson for sample collection assistance.

Bioinformatics were supported by The Ohio Supercomputer Center and by the National Science Foundation under award numbers DBI-0735191 and DBI-1265383 (https://www.cyverse.org). This study was funded by the Genomic Science Program of the United States Department of Energy Office of Biological and Environmental Research (grants DE-SC0004632, DE-SC0010580, and DE-SC0016440) and by a Gordon and Betty Moore Foundation Investigator Award (GBMF#3790 to M.B.S.).