1. Microbiology and Infectious Disease

MetaHiC phage-bacteria infection network reveals active cycling phages of the healthy human gut

  1. Martial Marbouty  Is a corresponding author
  2. Agnès Thierry
  3. Gaël A Millot
  4. Romain Koszul  Is a corresponding author
  1. Institut Pasteur, Unité Régulation Spatiale des Génomes, CNRS, UMR 3525, France
  2. Institut Pasteur, Bioinformatics and Biostatistics Hub, CNRS, USR 3756, France
https://doi.org/10.7554/eLife.60608
Share this article
Cite this article
  1. Martial Marbouty
  2. Agnès Thierry
  3. Gaël A Millot
  4. Romain Koszul
(2021)
MetaHiC phage-bacteria infection network reveals active cycling phages of the healthy human gut
eLife 10:e60608.
https://doi.org/10.7554/eLife.60608
  2. Download BibTeX
  3. Download .RIS

Abstract

Bacteriophages play important roles in regulating the intestinal human microbiota composition, dynamics, and homeostasis, and characterizing their bacterial hosts is needed to understand their impact. We applied a metagenomic Hi-C approach on 10 healthy human gut samples to unveil a large infection network encompassing more than 6000 interactions bridging a metagenomic assembled genomes (MAGs) and a phage sequence, allowing to study in situ phage-host ratio. Whereas three-quarters of these sequences likely correspond to dormant prophages, 5% exhibit a much higher coverage than their associated MAG, representing potentially actively replicating phages. We detected 17 sequences of members of the crAss-like phage family, whose hosts diversity remained until recently relatively elusive. For each of them, a unique bacterial host was identified, all belonging to different genus of Bacteroidetes. Therefore, metaHiC deciphers infection network of microbial population with a high specificity paving the way to dynamic analysis of mobile genetic elements in complex ecosystems.

Introduction

The gut hosts a complex microbial ecosystem composed of bacteria, archaea, eukaryotic microorganisms, and viruses. High-throughput sequencing allows in-depth investigation of the gut microbiota (Arumugam et al., 2011; Li et al., 2014), unveiling links between the balance of this complex ecosystem with a wide range of human diseases (Cho and Blaser, 2012), including autoimmunity (Wu and Wu, 2012) or neurological disorders (Cryan and Dinan, 2012). The regulation of the human gut microbiome is therefore much investigated, notably the role played by mobile genetic elements, and especially the collection of viruses present in the gut (or virome), in influencing its homeostasis maintenance. Indeed, the virome can either directly influence the immune system or reshape the bacterial component of the gut (Chatterjee and Duerkop, 2018; Gogokhia et al., 2019; Keen and Dantas, 2018; Norman et al., 2015). The gut virome is dominated by temperate bacteriophages belonging to the Caudovirales family (Minot et al., 2011; Shkoporov et al., 2019), but how the microbiome and the virome influence each other over time remains largely unknown, with only a few studies characterizing concomitantly host-phage relationships in individual gut sample (Džunková et al., 2019; Shkoporov et al., 2018). Reaching at a better characterization of these relationships is nevertheless a key, primary step to fully understand the impact of phages on microbial population (Edwards et al., 2016), hence our ability to influence it for biomedical applications. The ratio between phages and bacteria in the gut is also a relevant metric to understand further the regulation of the microbiota equilibrium. This ratio is typically assessed from the quantification of virus-like particles (VLP), corresponding to free phages produced during lytic cycle (Kleiner et al., 2015), with respect to the amount of bacteria cells. First evidences indicated ratios between 1 and 0.001 (Hoyles et al., 2014; Kim et al., 2011; Shkoporov et al., 2019), suggesting that the intestinal tract is quite different from other ecosystems such as oceans where phages largely outnumber bacteria (Roux et al., 2016). There is therefore a general need for comprehensive studies aiming at characterizing (1) which phage infects which bacterial species and (2) the relative phage and bacterial concentrations for each phage-host pair in the human gut.

Several methods have been developed in recent years to capture the global host-phage network directly in their natural environment: proximity ligation (Marbouty et al., 2014; Marbouty et al., 2017; Stalder et al., 2019; Yaffe and Relman, 2020), viral tagging (Deng et al., 2014; Džunková et al., 2019), or single amplified genomes (Labonté et al., 2015; Munson-McGee et al., 2018). We notably showed that meta3C, a proximity ligation-based approach, allows the assembly and scaffolding of dozens of nearly complete or complete phages and bacteria genomes, as well as their relationships, when applied on a single mammalian gut sample (Marbouty et al., 2017). These genomes and the corresponding infection network are inferred from the quantification of collision frequencies between DNA segments in the population, captured using a derivative of chromosome conformation capture (Dekker et al., 2002).

In the present work, we developed and apply an improved meta3C protocol (metaHiC) to reconstruct the large landscape of phages and their interacting bacteria in 10 individual, healthy human gut microbiota. By capturing and quantifying the DNA-DNA collisions during phages replication inside their bacterial hosts, we characterized a total of 6,763 unique host-phage contigs pairing, which represent, to our knowledge, the largest infection network of the human gut microbiome to date. While three-quarters of the detected phages appeared to be dormant lysogenic phages, a significant proportion (~5%) of the phage’s population exhibits a sequencing read coverage higher than the one of their corresponding hosts, suggesting these phages are actively dividing. We also characterized different crAss-like phages candidates as well as their relatively elusive bacterial hosts, all belonging to Bacteroidales order. Through the use of such approach to determine phage's host target in single samples, this study can have practical consequences, notably for fecal microbiota transplantation (Chehoud et al., 2016) or phage therapy (Dufour et al., 2019; Kortright et al., 2019).

Results

High-quality proximity ligation meta3C / metaHiC libraries of human gut microbiome

Frozen stool samples from 10 healthy adults from the Institut Pasteur Biobanque (agreement N18) (Figure 1a) were processed with the established meta3C protocol using either HpaII or MluCI as a restriction enzyme (Marbouty et al., 2017). Each meta3C library was sequenced (between 43 and 206 million paired-end [PE] reads) and assembled separately using MEGAHIT (Li et al., 2015), resulting in 10 human gut assemblies ranging from 242 to 617 Mb (total size = 3.45 Gb for 1,485,156 contigs) ('Materials and methods') (Supplementary file 1). To assess the amount of useful, long-range 3D DNA contacts in the different meta3C libraries, each read of a pair was aligned against its corresponding assembly. The number of pairs for which both reads aligned unambiguously on two different contigs (mapping quality ≥20) was used to compute a ‘3D ratio’, by dividing it with the number of pairs that aligned unambiguously within individual contigs (Figure 1—figure supplement 1a and Supplementary file 2). 3D ratios using the standard meta3C protocol range from 1.92% to 14.58%, a variability that results most likely from samples heterogeneities and from the restriction enzymes used. The distribution of meta3C reads along the longest contigs (>100 kb) of the different assemblies confirms that most of the PE reads correspond to ‘shotgun’-like reads, with very few useful trans, or long-range cis contacts (Figure 1—figure supplement 1b). These 3D ratios compare favorably to those obtained in recently published works, including using commercial solutions that encompass a biotin enrichment step (compare with 0.36% and 2.38% from DeMaere et al., 2020; Press et al., 2017, respectively), but they remain weak overall.

Figure 1 with 3 supplements see all
Download asset Open asset
Binning of human gut microbiota using 3C and HiC protocols I and II.

(a) Sex, ID, and age of the 10 individuals investigated in this study. The colors indicate the proximity ligation protocol used to generate the libraries. (b) Computational pipeline of paired-end (PE) sequence analysis (MAG: metagenomic assembled genome). (c) Left diagram: proportion for each individual assembly of the different bins obtained. Colors represent groups of bins of similar quality (completion/contamination), according to the color code on the right. ND = not determined. Right table: for each individual number of MAGs corresponding to the left diagram.

To increase the 3D ratio, we tested derivatives of metagenomic Hi-C protocols (I and II), corresponding to the meta3C protocol but with the addition of a biotin enrichment step ('Materials and methods'; see also Cockram et al., 2021). We applied them on samples #16016 (protocols I and II), #8015 (protocol II), and #9010 (protocol II) (Figure 1a). Simply adding a biotin enrichment step to meta3C (protocol I) only marginally improved the ratio (14.9% and 5.9% compared to 4.4% and 5.1% for the HpaII and MluCI libraries, respectively; Figure 1—figure supplement 1b). However, protocol II applied on samples #16016, #8015, and #9010 (using a ligation at room temperature) yields an average 3D ratio of ~32% (with a min/max of 25% and 41%, respectively). This is a strong improvement compared to the meta3C ratio ranging from the same sample (mean = 8.4%; min/max of 4.4% and 12.2%, respectively).

Therefore, protocol II yielded approximately four times more informative trans-contacts between pairs of contigs compared to meta3C, for a same cost. These experiments show that the ligation step was one of the main limiting steps in the Hi-C protocol applied to human stool samples and should be considered carefully in future experiments. The resulting protocol was named metaHiC.

Binning of human gut microbiome using meta3C or metaHiC protocols

We applied metaTOR on the assemblies and the corresponding contact data generated by either meta3C or the metaHiC protocols I and II. MetaTOR is an iterative network segmentation pipeline based on the Louvain algorithm (Blondel et al., 2008) to identify communities of contigs (or bins) enriched in genuine trans-contacts and likely to correspond to individual genomes for each sample (Figure 1b and 'Materials and methods'; Baudry et al., 2019; Marbouty et al., 2014). Among the 1,485,156 contigs (3.45 Gb) present in the 10 assemblies, 300,677 (525 Mb) were grouped by metaTOR into 19,122 small bins ranging in size between 10 and 500 kb. An additional 855,633 (2.57 Gb) contigs were grouped into 1,100 larger bins containing more than 500 kb and further referred to as metagenomic assembled genome (MAG) (Supplementary files 4 and 5). For each sample, the corresponding MAGs represent between 72% and 91% of all mapped reads. The composition of MAGs was assessed using CheckM, which quantifies the contamination and completion of pools of contigs with respect to known bacterial genomes (Parks et al., 2015). Out of the 1,100 MAGs, 304 were identified by CheckM as 'high quality', that is, displaying a completion rate ≥90% and contamination rate ≤5%, and corresponding most likely to complete bacteria genomes, and 411 as “medium quality” (completion ≥50% and contamination ≤10%) (Figure 1c). Some of these 715 high- and medium-quality genomes belong to species with an abundance below ~0.1% (over the 10 individuals) (Figure 1—figure supplement 2a-b), paving the way to future in-depth individual sample analysis.

We focused on the contribution of the improved metaHiC protocol II to the generation of high-quality MAGs. Indeed, a high 3D ratio does not necessarily correspond to an enrichment in informative bridges between contigs, as high levels of noise (i.e. ligation events between random restriction fragments) could also increase the ratio. However, all metrics (number of high-quality MAGs, completion, etc.) recovered when comparing the binned metagenomic maps generated using metaHiC protocol II with their meta3C or protocol I counterparts point at a strong enrichment in informative contacts (Figure 1—figure supplement 1c; 'Materials and methods'). These results are further supported by the modularity for the corresponding networks (0.89, 0.79, and 0.93 for metaHiC protocol II compared to 0.73, 0.71, and 0.86 with meta3C) (Newman, 2006), pointing at a better-defined community structure when using the network generated by metaHiC protocol II. A comparison between the MAGs retrieved using either the meta3C, metaHiC protocol I or II datasets showed that all approaches are highly concordant (Figure 1—figure supplement 3). The meta3C protocol is therefore the least enriched in useful 3D signal, but this can be overcome either by a higher sequencing depth or by using the metaHiC protocol II.

Therefore, the metaHiC protocol II presented here is several times cheaper than conventional meta3C or current commercial solutions and should be chosen in future studies. The rest of the study was pursued by merging the libraries generated for each individual.

Diversity of the human gut microbiota assessed by metaHiC/meta3C

We generated a phylogenetic tree of the 715 high- and medium-quality MAGs (Figure 2), as well as for each individual sample (Figure 2—figure supplement 1). As expected, the trees cover the taxonomic groups found in a healthy human intestinal tract (Almeida et al., 2019). In addition, the abundance of the different MAGs is consistent with the outcome of taxonomic analysis of the raw reads by the classifier Kaiju (Menzel et al., 2016; Figure 2—figure supplement 2).

Figure 2 with 2 supplements see all
Download asset Open asset
Metagenomic assembled genomes (MAGs) recovered using meta3C/metaHiC applied on 10 healthy human gut samples.

Phylogenetic tree comprising the 715 reconstructed MAGs with a completion above 50% and a contamination below 10% (691 different species at a threshold of 95% identity). A very long branch was cut out (symbol / /) to resize the tree for clarity. Red dots indicate MAGs with no reference in the GTDB-tk database (threshold of homology = 95%). The colors of the first three layers indicate the taxonomy of the MAGs, as determined by CheckM (phylum, class, order from center to periphery). Green and red bars represent completion and contamination levels, respectively. Completion scale: max = 100%; min = 50%. Contamination scale: max = 10%; min = 0%. Gray bars: MAGs (bin) sizes (scale: max = 7.07 Mb, min = 766 kb). Black bars in the outmost layer indicate MAG abundance in the different samples (max = 4.73%; min = 0.0039%).

The 715 MAGs were then compared to the latest genome references of the GTDB-Tk databases (release 0.95; Chaumeil et al., 2020). Eighty-five (around 10%) of them did not present any close similarity to the GTDB-Tk references at a threshold identity of 95% (average nucleotide identity; Konstantinidis and Tiedje, 2005). New MAGs are proportionally distributed among the different clades, with a majority belonging to Clostridia, in agreement with Pasolli et al., 2019. This relatively low number of newly recovered species, although still significant, suggests that the databases about the human gut microorganisms (at least those describing western population) are converging toward completion.

Note that, starting from 200 million PE reads on a single sample (#9010), the metaHiC protocol II was able to catch 81 high-quality genomes including seven from unknown species. This result therefore demonstrates the interest of this approach when investigating complex microbial communities from single samples.

Phages-hosts network in the human intestinal tract

We used VIRSorter (Roux et al., 2015) and VIBRANT (Kieft et al., 2020) to annotate phage sequences among the 1,485,156 contigs generated from the 10 assemblies. VIRSorter and VIBRANT identified 4,649 and 7,918 putative phages, respectively (including all categories for VIRSorter and excluding identified prophages, see 'Materials and methods'), resulting in 9,488 unique annotated phages contigs, ranging in size between 1 and 10 kb. Taxonomic annotations of those sequences using DemoVir (https://github.com/feargalr/Demovir) confirmed the high prevalence of phages belonging to the Caudovirales order (families Siphoviridae 59%, Myoviridae 15%, Podoviridae 2%, Unassigned 22%). Trans-contacts between phage annotated contigs and the 1,100 MAGs unveiled the infection pattern of the different phages presents in the 10 samples (Marbouty et al., 2014; Marbouty et al., 2017). Three classes (A, B, C) of phages were defined (Figure 3—figure supplement 1; see also 'Materials and methods'). Class A corresponds to 6,503 phages contigs (out of 9,488) that can be unambiguously assigned to a single MAG based on their 3D contacts (Supplementary file 6). In total, 955 MAGs (out of 1,100) made specific contacts with one or more of these contigs. Class B corresponds to the 2,460 phage annotated contigs assigned to highly contaminated bins prior to recursive procedure, and class C to the 525 contigs that never cluster with a bin above 500 kb during the 100 iterations of the iterative partition procedure. This latter category corresponds either to free phages or to phage sequences associated with uncharacterized/poorly abundant MAGs. Among the 2,460 contigs from category B, 364 clustered with a MAG at least once over the 10 Louvain recursive partition procedure, with 260 of those displaying an association score of at least 5/10 (same MAG for 5 out of the 10 iterations of the recursive procedure). These 260 contigs were reintroduced into class A for a total number of phage annotated contig in this class of n = 6,763. The 2,200 remaining phages contigs of class B (i.e. potentially contacting multiple MAGs) and the 525 phage contigs from class C (making no contacts at all with large MAGs) may correspond to free phages. These results suggest that the majority (6,763 out of 9,488, i.e. ~70%) of the phages present within the human gut are specific, infecting only one bacterial species (Figure 3a).

Figure 3 with 2 supplements see all
Download asset Open asset
Phage-bacteria network of interactions in human gut.

(a) Pie chart of phages contigs distribution among the different classes (see 'Materials and methods' and Figure 3—figure supplement 1). (b) Gaussian smoothed heatmap of the sequence distance between all phages taken two by two (Y-axis) and the sequence distance between their associated metagenomic assembled genomes (MAGs) (X-axis). Distance varies between 0 (full sequence similarity) and 0.3 (no similarity). Color scale on the left indicates the counting. Dotted lines indicate mean value of distances between MAGs at different taxonomic levels (species = 0.0402, genus = 0.205, family = 0.263). (c) Phylogenetic tree of phage contigs (n = 2,726). Colored strips in the first layer indicate the taxonomy of the associated MAG at the order level (1,535 over the 2,726 contigs present in the tree are attributed to a MAG). Bar plot in the second layer indicates phages contigs size (min = 5 kb; max = 211 kb).

On average, a MAG was associated with 6.9 phages contigs, but this rate presented high fluctuations (SD = 7.4), with a maximum of 45 contigs (MAG #9010_74_0; Supplementary file 4 and 6). No correlation between this rate and taxonomic annotation was identified.

We tested the coherence of the host-phage pairing strategy by focusing on phylogenetically phages related (i.e. homologous) sequences to see whether phages-MAGs pairs were conserved across individual samples. We searched for pairs of class A phages contigs belonging to the same genus, but with each member of the pair belonging to a different sample out of the 10 individuals (i.e. sharing at least 70% identity on 70% of their length; Lavigne et al., 2008). Among 454 pairs identified, 394 of them corresponded to phages assigned to two different MAGs (from the corresponding individual) but presenting the same taxonomic assignment (Figure 3—figure supplement 2a). In addition, three phylogenetically related MAGs from three different individuals exhibited the same set of homologous phages assigned to them (Figure 3—figure supplement 2b). These results support a good level of confidence in the phage-host network.

The network was further corroborated by looking for CRISPR spacer matches between MAGs and phages contigs using PILER-CR and Blast over the entire network (all individuals data mixed; Edgar, 2007; Edwards et al., 2016). We found 1,252 robust spacer signatures shared by a MAG and one or more phages sequences ('Materials and methods'). Of those, 387 were associated to a host-phage pair from the same individual. Among the 865 remaining hits, 90% (781/865) corresponded to MAGs and phages contigs from different individuals; 84% of these hits (660/781), however, involved MAGs from two different individuals but with the same taxonomy at the genus level compared to the characterized hosts in the same subject. This suggests that these MAGs have already been in contact with these phages during their evolutionary history but are not in the present individual.

To further investigate the infection network, we computed the sequence distance between all MAGs and phages contigs using the program Mash (Ondov et al., 2016). We then generated a density plot of the distance between all phages sequences as a function of the distance between the sequences of their related MAGs (Figure 3b). The plot showed that homolog phage sequences are associated to homolog MAGs sequences (Figure 3b, bottom left corner). In addition, while distantly related phages are occasionally found infecting homologous MAGs (Figure 3b, top left corner), very few related phages infect non-homologous MAGs (Figure 3b, bottom right corner). This result supports the idea that phages are ‘locally adapted’ to their bacterial hosts and do not easily shift from one bacterial species to another (Koskella and Meaden, 2013; Vos et al., 2009). Plotting the mean distance score between MAGs sequence for different taxonomic levels (species, genus, families; Figure 3b – dashed vertical lines) highlights a disruption at the species level, further supporting the presence of interspecific boundaries.

Phage contigs were then positioned in a phylogenetic tree based on a set of 77 specific markers of double-stranded DNA phages (Low et al., 2019) ('Materials and methods'). The genus taxonomic level of the host, when identified, was then indicated on the tree (Figure 3c); 2,726 out of the 9,488 phages contigs, including 1,535 with an assigned host, contained at least three markers and were included in the tree. The tree unveils clusters of closely related phages infecting the same bacterial genus (Kruskal-Wallis test, p-value=0.0132) suggesting, again, that phages are specific to their hosts in human gut.

Phages-host ratio in the human intestinal tract

The phage-bacteria network has the potential to provide insights on the phages’ metabolism. For instance, the proportion of infected cells in the host population is unknown; but, comparing the read coverages of a phage contig and its associated MAG can provide indication about the former’s activity (Figure 4). Indeed, the genome of an actively replicating phage should, in theory, display a higher coverage than that of its bacterial host (although multiple factors will nevertheless influence the phage sequence coverage within a metagenome, including the number of copies or sequence composition bias). For the 6,763 class A phages contigs, their mean read coverage ratio compared to the MAGs ratio is 1.23 (SD = 1.70), with a maximum of 58. This distribution is similar to the one of all the contigs binned within our different MAGs (Figure 4A,a). There are nonetheless significant variations among the 10 individual microbiota, as the average ratio varies from 1 (sample #10015) to 2 (sample #17006) (Figure 4—figure supplement 1a). These variations could reflect differences between individual gut microbiomes or through time, which should be elucidated using larger cohort and time-series sampling. We further investigated the potential influence of GC% of the sequence on the ratio. Plotting the coverage ratio of the phages contigs as a function of the mean GC% of these contigs, or as a function of the GC ratio (phages contigs/MAGs mean GC%), did not unveil a correlation between the ratio and GC% (Figure 4—figure supplement 1b).

Figure 4 with 1 supplement see all
Download asset Open asset
Phages-host ratio in human intestinal tract.

(a) Boxplot of the coverage ratio between different classes of contigs and the mean coverage of their associated MAGs. From left to right, coverage ratio of: (1) all binned contigs into metagenomic assembled genomes (MAGs) (n = 861,082) vs. mean MAG, (2) class A phage contigs (n = 6,763) vs. mean MAG, (3) dnaA contigs (n = 856) vs. mean MAG coverage, (4) all binned contigs into MAGs (n = 672,156) vs. dnaA contig of their associated MAGs, (5) class A phage contigs (n = 6,239) vs. dnaA contig of their associated MAGs. Log2 ratio thresholds define colored areas reflecting the potential phage status: green, dormant (cat1: −1 < log2(ratio) <1); gray, undefined (cat2: log2(ratio) < −1); light red, potentially active (cat3: 1 < log2(ratio) <2); red, likely active (cat4: log2(ratio) >2). Inside bar of the box, median; box, quartiles; whiskers, 1.5× interquartile range. (b) Read coverage of phage contigs (n = 6,239) as a function of the read coverage of the dnaA contig of their associated MAG (n = 856). Dashed lines indicate the limit of the categories described in a. (c) Scatterplot of class A phage contigs coverage vs. dnaA contig coverage as a function of the growth rate index calculated using GRID for each MAG. (d) Scatterplot of class A phage contigs coverage vs. dnaA contig coverage as a function of the MAGs abundance. (e). Bar plot of MAGs taxonomic abundance depending on the categories specified in (a).

Then, we investigated the influence of replication activity in the host by searching in each associated MAGs contigs encompassing the dnaA gene, and therefore likely to carry the origin of replication of the bacterial genome (n = 856 MAGs implicating 6,239 phages contigs) (Emiola and Oh, 2018). As expected, ‘dnaA’ contig (dubbed oriMAG contig) exhibits a mean read coverage ratio higher than the other contigs (mean = 1.37; SD = 0.51; max = 5.15). The coverage ratio analysis was then refined considering the phage vs. oriMAG contigs coverage ratio instead of the mean coverage of the MAGs (Figure 4Aa). We classified the phages into four categories based on their read coverage ratio with oriMAGs contigs (Figure 4a and b).

Phages displaying a ~1:1 ratio with their oriMAG contig (category 1, 0.5 < ratio < 2.0) represented the vast majority of the 6,239 phages-host pairs (n = 4,477; ~75%). They correspond to likely dormant (pro)phages, able to reactivate or not. This population may also encompass active phages that incidentally exhibit the same coverage of their host. Then, for 1,395 phage-hosts pairs, the corresponding oriMAG contig was significantly more covered than the associated phage (category 2 – undefined). This profile is consistent with an abortive phage infection cycle, a pseudo-lysogenic cycle, or even with prophages not present in the whole host population. The remaining 367 pairs (~5%) corresponded to phages with a higher coverage than their assigned oriMAG contig: category 3 corresponded to a ratio between 2 and 4 (n = 291) and category 4 to a ratio ≥4 (n = 76). These phages may be (category 3) or are probably (category 4) actively replicating, potentially impacting the ecosystem through lytic activities. Overall, these results support the conclusions of former and recent work suggesting that the major part of the phages found in human gut are temperate phages with few lytic activity (Džunková et al., 2019; Minot et al., 2011; Reyes et al., 2010).

To further question the influence of active phages, we investigated MAGs abundance and growth rate using GRID. This program measures for each MAG the read coverage ratio between the contigs (if any) that carry either the dnaA gene (i.e. oriMAG contig) or the marker of replication termination site dif (Emiola and Oh, 2018). No significant correlation was detected between the presence of active phages and the abundance/growth rate of their corresponding hosts, as assessed by this analysis (Figure 4c and d). We then asked whether some bacterial genera are more or less prone to phages activity (Figure 4e). Bifidobacteriales and Selenomodales appear underrepresented in the targeted MAGs, which suggests that they could be less subject to phages lytic activity in the gut than other genera.

CrAss-like phages family in the human intestinal tract

CrAss-like phages represent a large family of phages widespread in the human gut. They are predicted to infect bacteria from the Bacteroides phylum (Dutilh et al., 2014; Guerin et al., 2018; Shkoporov et al., 2018; Yutin et al., 2018). So far, only two hosts have been described ‘in vitro’ for one crAss member: Bacteroides intestinalis (Shkoporov et al., 2018) and Bacteroides thetaiotaomicron (Hryckowian et al., 2020). We searched for the presence of crAss-like phages in the different individual assemblies. Using a set of representative crAss-like phages sequences (Guerin et al., 2018), we detected 17 contigs, ranging in size from 58 to 176 kb and presenting several proteins homologs to the crAss-like family ('Materials and methods'). The phylogenetic tree encompassing these 17 contigs and the set of representative sequences pointed at phages belonging to genera I, III, IV, VI, and V (Figure 5 and Supplementary file 3; Low et al., 2019). All these contigs were unambiguously assigned to MAGs belonging to the Bacteroides clade, as expected from previous predictions and co-culturing (Dutilh et al., 2014; Guerin et al., 2018; Hryckowian et al., 2020; Shkoporov et al., 2018). More specifically, Bacteroides vulgatus, Bacteroides uniformis, Bacteroides ovatus, and Bacteroides thetaiotamicron were identified as hosts of these crAss-like phages. These four species were already suspected to be, or characterized as, hosts for crAss-like phages (Guerin et al., 2018; Hryckowian et al., 2020; Shkoporov et al., 2018). Interestingly, we also detected unknown species from the Prevotellaceae and Porphyromonadaceae families as hosts of other crAss-like phages. Some of these phages come from the same individual and were assigned to the same MAG, suggesting that crAss-like phages are highly abundant and potentially competing for hosts in the human gut.

Figure 5 with 1 supplement see all
Download asset Open asset
CrAss-like phages and their associated hosts.

Phylogenetic tree of crAss-like phages contigs found in the 10 assemblies. Representatives of the 10 genera of crAss-like phages described by Guerin et al. are included (in italic blue). Names of the different contigs are indicated on the left of each branch. Vertical colored stripes indicate, from left to right: host metagenomic assembled genome (MAG) order, host MAG family, host MAG genus, host MAG species, ratio of the read coverage of phages vs. read coverage of MAG (red – ratio > 4; light red – 4 > ratio > 2; green – 2 > ratio > 0.5; gray – ratio < 0.5). Red dot: circular signal characterized by VirSorter. Blue and gray dot: strong and weak circular signal characterized by 3C, respectively. Black lines: size of the contigs (min = 58,125,125 pb; max = 171,503 bp). Genera of the crAss-like phages are indicated on the right of the tree.

Among these 17 contigs, nine were predicted to be circular by VirSorter. Indeed, a circular signal was visible in their 3D contact map (Figure 5—figure supplement 1). 3D contacts also pointed at four more circular crAss-like phage contigs. Finally, the phage-host ratio of these 17 contigs showed that crAss-like phages are an active family of phages, with 11 contigs exhibiting a ratio ≥2 and therefore likely multiplying. On the other hand, we could not determine whether these ratios were associated to a low growth rate, and further analysis will be needed to understand the impact of crAss-like phages on their hosts. Altogether, these results significantly extend our understanding of the spreading of the ubiquitous but elusive crAss-like family, while confirming its link with Bacteroides.

Discussion

Bacteriophages, also called bacteria viruses, are the most abundant biological entities on earth and have a major impact on microbial communities (Clokie et al., 2011). High-throughput DNA sequencing of environmental samples has allowed to bypass virus isolation and revealed the existence of a wide variety of phages in all ecosystems (Cornelissen et al., 2012; Devoto et al., 2019; Dutilh et al., 2014). Several studies and computational developments have shed light on the composition and dynamics of the viral part of the human gut (Manrique et al., 2016; Minot et al., 2013; Shkoporov et al., 2019), unveiling entirely new phages family such as the crAss-like family (Dutilh et al., 2014; Guerin et al., 2018; Yutin et al., 2018). However, the concomitant investigation of the relationships between phages and bacteria remains challenging (Džunková et al., 2019; Marbouty et al., 2017). Moreover, metagenomic approaches often rely on covariance analysis involving multiple sampling, which can be cumbersome. Proximity ligation assays applied to metagenomic samples (meta3C, metaHiC) allow in situ investigation of microbial communities starting from single samples (Marbouty et al., 2014; Marbouty et al., 2017). In the present study, we applied those approaches to 10 human gut microbiota samples, characterizing hundreds of high-quality bacterial genomes (MAGs) as well as thousands of phage sequences and revealing the large multiscale network of interactions between them.

The approach unveiled hundreds of MAGs, with ~10% of them identified as new, a relatively low number. However, given the in-depth investigation of the gut with metagenomic studies, this is not surprising. It is a dynamic field: in an earlier version of this work, nearly 30% of the MAGs identified in the present study were considered as ‘new’. The exhaustive analysis of Pasolli et al., 2019 reduced that number to 10%. That number remains also significant (85 MAGs), given only 10 individuals were screened and no covariance approach performed. Therefore, there is still in our opinion a strong potential in the approach to quickly call new, complete MAGs de novo from complex microbial ecosystems not as intensively sampled as the human gut. This prediction is likely to hold true until long-read sequencing, which provides complete or near complete genomes at a high frequency, becomes the prevalent choice of future metagenomic studies. The ability to assign episomes to their hosts will nevertheless remain an asset of metaHiC in the future.

The phage-host ratio computed in this study suggests that phages appear slightly more abundant than bacteria in the human intestinal tract but nothing comparable to other environments (Rodriguez-Valera et al., 2009). Therefore, this work confirms that the majority of phages present in the gut are temperate phages with little or no lytic activity (Reyes et al., 2010). This result is in adequation with previous studies that showed a limited phage-host dynamic and a relative homeostasis of this environment, with few examples of a phage overrepresented compared to its host. On the other hand, nearly 5% of the phage population exhibits a sequence coverage ratio higher than twice that of their hosts, representing actively replicating phages that could play an important role in the maintenance of the bacterial population. Notably, members from the Bifidobacteriales and the Selenomonadales bacteria clades appear to be less subject to phages lytic activity. The work also confirms that most of the phages are specific to their host and do not shift easily to another host, at least in the human intestinal tract (Koskella and Meaden, 2013; Vos et al., 2009).

As expected from their overrepresentation in the human gut population (Edwards et al., 2019), we found several potential members of the crAss-like phages family. Those phages and their hosts have been sought in vitro since their discovery in 2014, with two successful attempts (Hryckowian et al., 2020; Shkoporov et al., 2018). In a single analysis, meta3C/HiC unveiled the host, at the species level, of 17 crAss-like contigs, all belonging to different members of the Bacteroides clade. This result is in agreement with previous prediction at the class level (Dutilh et al., 2014), and suggests crAss-like phages are widely distributed in the Bacteroides clade.

It also underlines the interest of this affordable approach, which is easily applicable to human gut samples. Application of metaHiC on larger cohorts, including or not longitudinal tracking, will allow to better understand the relationships not only between crAss-like phages and this clade but also the specificity of each phages sub-group to their hosts in the Bacteroides clade. CrAss-like phages share features characteristics to both phages and plasmids and generally encode proteins allowing for their own replication. The crAss-like phages are therefore reminiscent of the concept of plasmid or phage plasmid (Lagos and Goldstein, 1987).

Finally, coupling metaHiC with other approaches on larger human cohort will shed new lights and findings about the relationships between phages, plasmids, and bacteria. For instance, coupling meta3C with VLP purification could offer an unprecedent view of phage-bacteria relationships in this important ecosystem.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 17006–19 yo – femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 16026–20 yo – maleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 16021–28 yo – femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 7016–64 yo – maleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 10015–26 yo – femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 8016–69 femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 7020–73 yo – maleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 16016–35 yo – femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 8015–64 yo – femaleFreshly frozen from healthy donor
Biological sample (human)Frozen human fecal samplesInstitut Pasteur Biobanque (ICAReB)Patient 9010–60 yo - femaleFreshly frozen from healthy donor
Chemical compound, drugFormaldehyde (35–37%)Sigma AldrichF8775(also contain methanol as stabilizer – 15%)
Chemical compound, drugMyOne streptavidin beadsLife Science
Chemical compound, drugdCTP-14 BiotinLife Science
Software, algorithmCutadaptMartin, 2011v.1.9.1
Software, algorithmFastQCAndrews, 2010v.0.10.1
Software, algorithmMEGAHITLi et al., 2015v.1.1.1.2
Software, algorithmQuastGurevich et al., 2013v.2.2
Software, algorithmbowtie2Langmead and Salzberg, 2012v.2.2.3
Software, algorithmMetaTOR PipelineBaudry et al., 2019
Software, algorithmMetaBatKang et al., 2015
Software, algorithmCheckMParks et al., 2015v1.1.2
Software, algorithmGTDB-TkChaumeil et al., 2020release 0.95
Software, algorithmseqtkseqt, 2020; https://github.com/lh3/seqtk
Software, algorithmhicstuffMatthey-Doret, 2020; https://github.com/koszullab/hicstuff
Software, algorithmLASThttp://last.cbrc.jp/
Software, algorithmitol https://itol.embl.de/
Software, algorithmVirSorterRoux et al., 2015v.1.0.3
Software, algorithmVIBRANTKieft et al., 2020v.1.0.1
Software, algorithmMashOndov et al., 2016v.2.0
Software, algorithmPILER-CREdgar, 2007
Software, algorithmProdigalHyatt et al., 2010v.2.6.3
Software, algorithmMUSCLEEdgar, 2004
Software, algorithmAMASBorowiec, 2016
Software, algorithmIQ-TREENguyen et al., 2015v.1.5.5
Software, algorithmR environmentR Development Core Team, 2020
OtherCovaris S220CovarisAFA tubes
OtherPrecellys TUBEBertin TechnologyVK05 + VK01 glass beads

Feces sampling and meta3C libraries generation

Request a detailed protocol

Frozen feces samples of 10 healthy human adults (19–73 years old) were collected from the Institut Pasteur Bobanque (ICAReB) (agreement N18). For each sample, 100 mg was thawed directly in 50 mL of crosslinking solution (1× PBS supplemented with 5% formaldehyde) and incubated for 1 hr at room temperature under strong agitation. Formaldehyde was quenched by adding 20 mL of 2.5 M glycine during 20 min at room temperature under gentle agitation. Samples were then recovered by centrifugation, washed with 10 mL 1× PBS, re-centrifuged, and aliquots of 20 mg were stored at −80°C until processing. One aliquot (20 mg) of fecal matter was resuspended in 4 mL TE 1× supplemented with antiprotease (mini tablets – Roche), transferred in two Precelys tubes (2 mL – VK05 supplemented with 100 µL of VK01 glass beads) and disrupted (6,700 rpm – 20 s ON/30 s OFF – six cycles). Lysates were recovered, pooled, and 10 µL of Ready-to-lyze lysozyme (Epitech) was added prior a 1 hr incubation at 37°C under gentle agitation. Sodium dodecyl sulfate (SDS) 10% was added to a final concentration of 0.5% and the lysate was incubated 10 min at room temperature. For each library, 1 mL of lysate was transferred to a tube containing the digestion reaction solution (500 µL NEB1 10× buffer, 500 µL Triton 10%, 1000 U HpaII or MluCI, H2O, final volume = 4 mL). Digestion was allowed to proceed for 3 hr at 37°C under gentle agitation. Tubes were then centrifuged for 20 min at 4°C and 16,000 × g and the supernatants discarded. Pellets were resuspended in 500 µL H2O and transferred to a tube containing the ligation reaction (160 µL NEB ligation buffer 10×, 16 µL ATP 100 mM, 16 µL BSA 10 mg/mL, 500 U T4 DNA ligase, final volume = 1.1 mL). Ligations were processed for 4 hr at 16°C; 20 µL EDTA 0.5M, 80 µL SDS 10%, and 2 mg proteinase K were added to each reaction and incubated overnight at 65°C to digest proteins. DNA was extracted using phenol-chloroform and precipitated with 2.5 vol ethanol 100%. Pellets were suspended in a final volume of 130 µL TE 1× supplemented with RNAse, incubated 1 hr at 37°C and stored at −20°C until use. Libraries were then processed for sequencing as described (Baudry et al., 2019; Marbouty et al., 2017).

MetaHiC libraries generation

Request a detailed protocol

MetaHiC libraries were generated similarly to 3°C with minor changes. After digestion, tubes were centrifuged for 20 min at 4°C and 16,000 × g, supernatants were discarded, and pellets were resuspended in 400 µL H2O. Biotinylation was done by adding 50 µL NEB ligation buffer 10× (without ATP), 4.5 µL of 10 mM dATP/dTTP/dGTP (HpaII), 37.5 µL biotin-dCTP 0.4 mM, and 8 µL Klenow (5 U/µL) (for MluCI: 10 mM dCTP/dTTP/dGTP and biotin-dATP). Reactions were incubated for 45 min at 37°C. Ligations were performed at 16°C for 4 hr in the case of the metaHiC protocol I and at room temperature for 3 hr in the case of the metaHiC protocol II. DNA was extracted, purified, and processed into sequencing library as described above (Moreau et al., 2018).

Sequencing, processing, and assembly

Request a detailed protocol

Proximity ligation libraries were sequenced using PE Illumina sequencing (2 × 75 bp, NextSeq500 apparatus) (Supplementary file 1; Baudry et al., 2019). Reads were filtered and trimmed using cutadapt (Martin, 2011) (parameters: -a file:adapters.fa -A file:adapters_illumina.fa --match-read-wildcards-q 20,20 m 45). Quality was controlled with FastQC (Andrews, 2010; http://www.bioinformatics.babraham.ac.uk). 1,478,113,712 PE reads were retained in total (from the 20 3C libraries) (Supplementary file 1). These reads were used to perform for each sample an assembly using MEGAHIT v.1.1.1.2 (Li et al., 2015) and default parameters. Contigs under 500 bp were discarded from further analyses. Assemblies were analyzed using Quast (Gurevich et al., 2013; Supplementary file 1).

Alignment step and 3D ratio calculation

Request a detailed protocol

PE reads were aligned separately in single-end mode using bowtie2 (option: --very-sensitive-local) (Langmead and Salzberg, 2012). For each sample, both alignment files (forward and reverse) were then sorted and merged using the samtools and pysam libraries. Alignments with mapping quality below 20 and ambiguous alignment were discarded. 3D ratio was then calculated by dividing the numbers of PE reads where each end mapped on different contigs by the total number of mapped PE reads (Supplementary file 2).

MetaTOR binning procedure

Request a detailed protocol

For the binning procedure, we applied the metaTOR pipeline (https://github.com/koszullab/metaTOR) (Baudry et al., 2019). Briefly, each PE reads that aligned unambiguously and with a mapping score >20 on two different contigs were kept to generate the network of contigs’ interactions. Interaction scores between contigs were normalized by dividing their interaction score by the geometric mean of the contigs’ respective coverage. Contig coverage was calculated using MetaBat script: jgi_summarize_bam_contig_depths with default parameters for every set of reads and a minimum contig size of 500 bp (Kang et al., 2015). We then applied 100 iterations of the Louvain procedure (Blondel et al., 2008) and recovered bins that clustered together at least 90 times over the 100 iterations. Bins above 500 kb were evaluated for completeness and contamination by CheckM (Parks et al., 2015). CheckM was also used to assign taxa to these sequences using the lineage workflow. A bin was validated if its contamination rate fell under 10%. Bins with contamination levels higher than 10% were selected for recursive binning as previously described (Baudry et al., 2019). Briefly, the partition step was re-run 10 times on the corresponding sub-network of each contaminated bin, yielding smaller bins which were then re-evaluated using CheckM. The resulting bins above 500 kb were then also retained as MAGs. We used the Genome Taxonomy Database Toolkit with default parameters (GTDB-Tk) (Chaumeil et al., 2020) to validate taxonomic classifications consistent with the GTDB taxonomy and to define the novelty of the characterized MAGs.

Protocols comparison

Request a detailed protocol

To compare the different protocols, we subsample the different libraries using (https://github.com/lh3/seqtk, seqt, 2020) and the same seed. We then fed the pipeline with the corresponding subsampled PE reads and their corresponding assemblies and process them as described previously in this section. Contact matrices and pie chart of the different ligation events were obtained using a home-made pipeline (https://github.com/koszullab/hicstuff, Matthey-Doret, 2020). Characterized MAGs for 3C, Hi-C, and eHi-C were also compared using the program LAST (http://last.cbrc.jp/).

MAGs phylogentic tree

Request a detailed protocol

MAGs phylogenetic tree was obtained using CheckM and the tree-qa workflow (out format 3). Tree was pruned, visualized, and annotated using itol (https://itol.embl.de/) (Letunic and Bork, 2016).

Phages contigs detection and classification

Request a detailed protocol

VirSorter v.1.0.3 (Roux et al., 2015) and VIBRANT v.1.0.1 (Kieft et al., 2020) were used to screen the 10 different assemblies. We removed the contigs annotated as prophages by the two programs and kept phages from all categories for VIRSorter. The remaining phages contigs were then classified as described in the Figure 3—figure supplement 1: (1) class A corresponding to contigs assigned to only one MAG after the iterative/recursive procedure (Baudry et al., 2019), (2) class B corresponding to contigs assigned to only one contaminated bin prior the recursive procedure and potentially corresponding to phages able to infect several species, (3) class C corresponding to contigs assigned to a bin smaller than 500 kb after the iterative procedure (but prior the recursive procedure).

Concerning contigs from category B, we calculated an association score of these contigs with the MAGs defined during the recursive procedure. This association score corresponds to the number of Louvain iterations where a contig was clustered with a MAG during the recursive process. Contigs exhibiting an association score ≥5 with only one preferential MAG were reintroduced in category A (Figure 3—figure supplement 1).

Phages-host network analysis

Request a detailed protocol

Phage contigs were compared using two strategies. First, we used the software Mash v.2.0 to compute the distances between all phages contigs as well as between all characterized MAGS (Ondov et al., 2016), using Mash score = 1 as absence of homology and redefine it as a value of 0.3 as the maximum value after 1 is 0.28. Heatmap of the distance between all phages taken two by two and the sequence distance between their associated MAGs was smoothed using the kde2d() function of the MASS package of R (Figure 3b). Second, we used Blast to compare the different phage contigs from the 10 assemblies. All contigs sharing at least 70% of identity on 70% of their length were considered as belonging to the same genus.

We also used Blast to compare the different phage contigs. All contigs sharing at least 70% of identity on 70% of their length were considered as belonging to the same genus. We then compared the taxonomy/homology of their attributed MAGs and represented it using a circos plot.

CRISPR array in the characterized MAGs was identified using PILER-CR (Edgar, 2007). We then used Blast to screen spacers among the identified phages; 1252 robust hits between MAGs and phages were found (BLAST e-value <10–5, bitscore ≥45), 387 of which in agreement with the host-phages assignments defined by 3D contacts. Most of the other hits involve MAGs and phages contigs from different samples (n = 781) but, in 84% of the cases (n = 660), these hits point to MAGs with same taxonomy.

Phages contigs phylogenetic tree

Request a detailed protocol

Putative open reading frames (ORFs) inside phage contigs were detected using prodigal and the option -meta (Hyatt et al., 2010). A set of 77 protein hidden Markov model shared by dsDNA phages and defined previously (Low et al., 2019) was then used to screen the different predicted proteins (e-value <10−3). For each contigs, the hit with the highest bit score was identified as the best hit and only the highest scoring sequence was selected. Only contigs with at least three markers among the 77 proteins were retained for further analysis. Markers were individually aligned using MUSCLE (Edgar, 2004), and individual multiple sequence alignment was trimmed using trimAl (Capella-Gutiérrez et al., 2009) at a threshold of 50%. Alignments were then concatenated using AMAS (Borowiec, 2016) by introducing gaps in positions where markers were absent from a contig. A phylogenetic tree was computed using IQ-TREE v.1.5.5 (Nguyen et al., 2015) and the following parameters: LG + GAMMA model and midpoint rooted, followed by 100 non-parametric bootstrap replicates. The tree was pruned, visualized, and annotated using itol (https://itol.embl.de/) (Letunic and Bork, 2016).

Statistical analysis

Request a detailed protocol

R environment was used for all analyses (R Development Core Team, 2020). Statistical significance was set to p≤0.05. In Figure 3b, a chi-squared goodness-of-fit test was applied using the raw (unsmoothed) contingency table of the heatmap and cell equiprobability (1/number of cells) to define the theoretical contingency table. In Figure 3c, a Kruskal-Wallis test with continuity correction was used, considering that the different MAGs genus are ranked all along the circos plot.

CrAss phages identification and phylogenetic tree

Request a detailed protocol

CrAss phages sequences from Guerin et al., 2018 were downloaded and used as a database to search for homologous proteins against our set of contigs using prodigal and Blast (e-value <10−5). Contigs with at least 10 significant hits were classified as potential crAss-like phages and were integrated in the phylogenetic tree. Phylogenetic tree was built as described in the upper section. We also integrated in the phylogenetic tree two crAss-like phages genomes of each genera from Guerin et al., 2018.

Data availability

Request a detailed protocol

Sequence data (raw reads, assemblies) have been deposited in the NCBI Sequence Read Archive under the BioProject number PRJNA627086.

Code and additional files containing details on all the MAGs, bins, contigs, and phages can be found at the following address https://github.com/mmarbout/HGP-Hi-C; (Marbouty, 2021; copy archived at swh:1:rev:f2a185ed6638d445884177e319e831f88d67dba7).

Data availability

Sequence data (raw reads, assemblies) have been deposited in the NCBI Sequence Read Archive under the BioProject number PRJNA627086. Code and additional data on MAGs, Bins, Contigs and Phages can be found at the following address https://github.com/mmarbout/HGP-Hi-C (copy archived at https://archive.softwareheritage.org/swh:1:rev:f2a185ed6638d445884177e319e831f88d67dba7/).

The following data sets were generated
    1. Koszul R
    2. Marbouty M
    (2020) NCBI Sequence Read Archive
    ID SRX8404055. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404055
    1. Koszul R
    2. Marbouty M
    (2020) NCBI Sequence Read Archive
    ID SRX8404054. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404054
    1. Marbouty M
    2. Koszul R
    (2020) NCBI Sequence Read Archive
    ID SRX8404053. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404053
    1. Marbouty M
    2. Koszul R
    (2020) NCBI Sequence Read Archive
    ID SRX8404052. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404052
    1. Marbouty M
    2. Koszul R
    (2020) NCBI Sequence Read Archive
    ID SRX8404051. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404051
    1. Marbouty M
    2. Koszul R
    (2020) NCBI Sequence Read Archive
    ID SRX8404050. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404050
    1. Koszul R
    2. Marbouty M
    (2020) NCBI Sequence Read Archive
    ID SRX8404049. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404049
    1. Koszul R
    2. Marbouty M
    (2020) NCBI Sequence Read Archive
    ID SRX8404048. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404048
    1. Marbouty M
    2. Koszul R
    (2020) NCBI Sequence Read Archive
    ID SRX8404047. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404047
    1. Koszul R
    2. Marbouty M
    (2020) NCBI Sequence Read Archive
    ID SRX8404046. proximity ligation of human gut samples.
    https://www.ncbi.nlm.nih.gov/sra/SRX8404046
    1. Marbouty M
    2. Koszul R
    (2021) NCBI Sequence Read Archive
    ID SRX9848788. proximity ligation of human gut samples 8015_metaHiC_hpaII.
    https://trace.ncbi.nlm.nih.gov/Traces/sra/?run=SRR13435229
    1. Marbouty M
    2. Koszul R
    (2021) NCBI Sequence Read Archive
    ID SRX9848789. proximity ligation of human gut samples 8015_metaHiC_MluCI.
    https://trace.ncbi.nlm.nih.gov/Traces/sra/?run=SRR13435228
    1. Marbouty M
    2. Koszul R
    (2021) NCBI Sequence Read Archive
    ID SRX9848786. proximity ligation of human gut samples 16016_metaHiC_hpaII.
    https://trace.ncbi.nlm.nih.gov/Traces/sra/?run=SRR13435231
    1. Marbouty M
    2. Koszul R
    (2021) NCBI Sequence Read Archive
    ID SRX9848787. proximity ligation of human gut samples 16016_metaHiC_MluCI.
    https://trace.ncbi.nlm.nih.gov/Traces/sra/?run=SRR13435230
    1. Koszul R
    2. Marbouty M
    (2020) NCBI BioProject
    ID PRJNA627086. human gut metagenome Metagenomic assembly.
    https://www.ncbi.nlm.nih.gov/bioproject/PRJNA627086

References

    1. Lagos R
    2. Goldstein R
    (1987)
    Evolutionary relationship between plasmids and phages: phasmid P4 as a model
    Archivos De Biologia Y Medicina Experimentales 20:325–332.
  69. Software
    1. R Development Core Team
    (2020) R: A language and environment for statistical computing
    R Foundation for Statistical Computing, Vienna, Austria.
    http://www.r-project.org
  74. Software
    1. seqt
    (2020) Toolkit for Processing Sequences in FASTA/Q Formats, version 1.3
    Toolkit for Processing Sequences in FASTA/Q Formats.
    https://github.com/lh3/seqtkVersion

Decision letter

  1. Breck A Duerkop
    Reviewing Editor; University of Colorado School of Medicine
  2. Wendy S Garrett
    Senior Editor; Harvard T.H. Chan School of Public Health, United States
  3. Breck A Duerkop
    Reviewer; University of Colorado School of Medicine
  4. Christopher Quince
    Reviewer; University of Warwick, United Kingdom

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This exciting and rigorous study utilizes a modified HiC method to increase metagenome assembled genome (MAG)-phage contig connections and the resulting data supports phage modulation of bacterial populations within the gut microbiota. This work lays the foundation for metagenomic applications that will advance our understanding of how phages impact their hosts and thus the microbiota both in the gut and in microbiotas other than the gut. This work is broadly applicable to many fields of research where microbial communities are studied.

Decision letter after peer review:

Thank you for submitting your article "Phages-bacteria infection network of the healthy human gut" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Breck A Duerkop as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Wendy Garrett as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Christopher Quince (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

In this study, the authors apply chromosome conformation capture (3C) library preparation methods to ten fecal samples from healthy individuals. 3C allows physically collocated DNA molecules to be identified. It is well established for resolving genomic structure but its application to metagenomes is more novel. Here, the authors focus on its application to link bacteriophages in the human microbiome to their hosts. They determine that about 25% of the phages are lytic, whereas ~50% of the phages are dormant lysogens. They use this methodology to identify human phages belonging to an emerging family of viruses, CrAss phages, greatly extending the predicted host range of these phages. Overall, this paper largely supports previously observed biology of intestinal phages. What makes this new and of broad interest is the ability to increase metagenome assembled genome (MAG)-phage contig connection using a modified eHiC method, coupled with some innovative computational analysis to correlate MAG replication to phage contig abundances that suggest phage modulation of bacterial populations within the microbiome. The techniques outlined here have the potential to provide profound understanding of how phages impact their hosts and thus the microbiota beyond just the human gut.

The revisions below come from three independent reviewers, each who deemed this work important and of high quality. However, there was consensus that revisions are needed to improve the work, including more rigorous statistical analyses and the inclusion of more scientific detail. This study came across like a proof of principle exploring the alternate 3C methods and their effectiveness in this context rather than dramatically novel science. Taking this into consideration, it was our collective opinion that the methods themselves are still sufficiently novel that even as a proof of principle this is an interesting study that could provide guidance to future larger scale studies utilizing these methods. Additionally, we commend the authors for establishing their tools on a GitHub repository. The readme document contained within was very helpful in describing the scripts as well as what was used in this manuscript. This resource will be valuable to the field as a whole.

Essential revisions:

1) A major concern is the number of replicates used to verify the superiority of the eHiC technique to that of the HiC or 3C method. Only one sample is used to make this conclusion. Ideally the author's would have performed this on a few more samples and then compared these to several replicates of just metaHiC and meta3C. The authors say they identified 81 high-quality genomes. Are these the same as what was identified via the metaHiC and meta3C? This is a critical result to support the utility of this developed assay. Since the authors have already published that 3C technology can make phage-bacteria connections, the real novelty here is that the eHiC method greatly increases the number of connections that can be made. However, this should be more rigorously proven.

2) From these data, a major observation is reinforced, that temperate phages seem to dominate the virome. If most phages are temperate then is Hi-C actually needed to make valid and in depth host connections?

3) The authors indicate that they verified their connections via CRISPR matching; however, they do not show the data. This is important to include for two reasons: 1) they indicate that it complements their current analysis and 2) CRISPR matching is often thought to be state of the field for making connections and it is implied that this method works better. Additionally, an explanation of why 15% validation with CRISPR is reasonable.

4) The association between MAG taxonomy and the phylogeny of their associated phages. This is presented graphically in Figure 3C by comparing the phage phylogeny with the order level assignment of a MAG. There does appear to be an association, but this needs to be quantified. This could be demonstrated with a permutation ANOVA of the phylogenetic distance between phages against their order assignment (using adonis in vegan for example). It could then be determined if this association breaks down at higher MAG taxonomic resolution. Similarly, the data in Figure 3B could be used in a Mantel test to determine strength of association between MAG distance and phage distance.

5) Why were the “category B” phages excluded from the analysis. The distribution of their contacts across MAGs would be a good sanity check for the methodology. Ignoring these results seems inappropriate.

6) Simply observing twofold higher coverage for a phage than the host is not definitive evidence that the phage is lytic since multiple factors influence sequence coverage, the phage could be multicopy, it could be near the origin of replication, or possess a sequence composition biased to higher coverage. The authors need to demonstrate that the phage sequences have significantly higher coverage than expected given these sources of variation. Since they are using growth estimation methods they could look at position in the genome and compare with non-viral contigs that are at a similar point. It would be helpful if the authors could pull out specific phage-host pairs as examples of what these patterns look like.

7) When discussing the novelty of MAGs, this should really be placed in the context of the recent large-scale binning surveys (e.g. Pasolli et al., 2019) which similarly observed most novel diversity in the Clostridia.

8) Was there any taxonomic signal in the number of phages associated with a MAG?

9) Without a cell harvesting step; intra cellular crosslinking, digestion and ligation; or a large ligation reaction volume, how is random proximity ligation and / or crosslinking of extracellular DNA avoided following a freeze thaw cycle of sample with no storage buffer? This may not be an issue as the data seems to suggest a small degree of contamination in the fused reads, yet this was not explicitly tested experimentally with mock communities. Having the molarity, copy number or even mass of DNA entering the ligation step would likely help as it is suspected it is quite dilute in the 1.1 ml volume.

10) Overall the manuscript indicates that two method derivatives are developed, but only meta-eHiC is discussed. The Materials and methods list the procedures for preparing the meta-Hi-C and meta-eHiC libraries so one can deduce the differences in the methods, but the Results should take a moment to explain these two derivatives. Is not developing them a result of this study? This would frame the results and interpretation of this work for the reader.

11) What is the correlation supposedly presented in Supplementary Figure 2, Results? A statistic is needed to support this claim.

12) For VIBRANT and VIrSorter results clarification is needed concerning uniqueness. Are those considered unique between the two tools or unique across samples (and possibly also between the two tools)? Also, were all category 1-3 VirSorter phages included or were just category 1 or category 1 and 2 (like VIBRANT does)? This detail should be included in the Materials and methods.

13) By looking just at the active phage population (Class A), one could speculate that they were lytic/infectious which may be reducing bacterial population numbers. But the ori/ter ratio is not really providing insight into their population spread rather just their replication rate. It is unclear how the authors draw their conclusion of an inverse correlation from Figure 4A, right panel. Is there a statistic to quantify this observation?

https://doi.org/10.7554/eLife.60608.sa1

Author response

Essential revisions:

1) A major concern is the number of replicates used to verify the superiority of the eHiC technique to that of the HiC or 3C method. Only one sample is used to make this conclusion. Ideally the author's would have performed this on a few more samples and then compared these to several replicates of just metaHiC and meta3C. The authors say they identified 81 high-quality genomes. Are these the same as what was identified via the metaHiC and meta3C? This is a critical result to support the utility of this developed assay. Since the authors have already published that 3C technology can make phage-bacteria connections, the real novelty here is that the eHiC method greatly increases the number of connections that can be made. However, this should be more rigorously proven.

The reviewers are correct to point that more “eHiC” experiments are needed to strengthen the robustness of the conclusions drawn from the improved protocol. In order to avoid multiple names of related methods, we first simplified the nomenclature and called the improved protocols: metaHiC protocol I and metaHiC protocol II. In the revised version, we performed new experiments using these metaHiC protocols and generated four new metaHiC libraries (corresponding to sample 8015 and 16016, each using HpaII and MluCI as restriction enzymes). These libraries confirmed that the metaHiC protocol II systematically recovers much more enriched 3D signal ratio and performs a better binning than the other protocols, with less raw reads. All new data were now included in the global analysis, which led us to update the whole Results section.

We now compare MAGs retrieved using the different protocols (Figure 1—figure supplement 2) and show that MAGs retrieved using either meta3C, metaHiC protocol I or metaHiC protocol II are highly concordant (at a threshold of 80% of identity).

Please note that because of restriction to lab access and the overall situation, generating more metaHiC libraries than the three we now present in the manuscript was not realistic. We hope the reviewers will be satisfied with the inclusion of these experiments, that in our opinion clearly demonstrate the improvement we brought to the approach. That the quality of the data is increased using this protocol is indirectly backed by a recent work on pure cultures of bacteria and archaea (Cockram et al., 2020), where we improved the yield of Hi-C experiments by bringing related modifications to the protocol.

“A comparison between the MAGs retrieved using either the meta3C, metaHiC protocol I or II datasets showed that all approaches are highly concordant (Figure 1—figure supplement 3). The meta3C protocol is therefore the least enriched in useful 3D signal, but this can be overcome either by a higher sequencing depth, or by using the metaHiC protocol II.”

2) From these data, a major observation is reinforced, that temperate phages seem to dominate the virome. If most phages are temperate then is Hi-C actually needed to make valid and in depth host connections?

Indeed, our data confirmed that temperate phages engaged in lysogenic cycle seem to dominate the virome. We agree that in this case Hi-C related approaches are not necessarily essential (or maybe one would say the most convenient way) to link them to their hosts. However, the interest of the approach relies primarily on three points. First, it is useful to assign virulent phages to their hosts, even though they don’t represent the majority of the phage population. Second, temperate phages can be in a lytic and replicative cycle, in which case Hi-C is valuable to characterize their host as well (or/and sort things out). Indeed, a significant fraction of the virome is actually cycling phages or multicopy phages (~5%) of the detected phage contigs exhibit a coverage ratio over 2, (n = 367). Finally, Hi-C related methods have proven highly efficient on single samples as they are not based on covariance analysis, and therefore their interests extend beyond gut microbiota investigation.

3) The authors indicate that they verified their connections via CRISPR matching; however, they do not show the data. This is important to include for two reasons: 1) they indicate that it complements their current analysis and 2) CRISPR matching is often thought to be state of the field for making connections and it is implied that this method works better. Additionally, an explanation of why 15% validation with CRISPR is reasonable.

CRISPR is indeed efficient at associating phages to host (Minot et al., 2013; Stern et al., 2012). However, CRISPR is not distributed in all bacteria species (it is estimated that ~40% of bacteria and ∼70% of archaea encode a CRISPR system; Edwards et al., 2015; Held et al., 2013). It is also well documented that the spacers in a CRISPR array are rapidly turned over in the environment (Minot et al., 2013; Pride et al., 2011). Consequently, CRISPR spacers don’t necessary match any known sequence. Therefore, although this approach is specific (few false positives) it is not very sensitive (many false negatives) (Edwards et al., 2015). It has been estimated that bacterium with the most similar CRISPR spacer is the correct host for ~15% of the phages and that bacterium with the highest number of CRISPR spacers is the correct host for ~21% of phages (Edwards et al., 2015). Although these numbers are likely to vary, this leaves room for complementary techniques aiming at the same objectives. To better contextualize our results, we have modified the text as follows:

“The network was further corroborated by looking for CRISPR spacer matches between MAGs and phages contigs using PILER-CR and Blast over the entire network (all individuals data mixed)(Edgar, 2007; Edwards et al., 2015). […] This suggests that these MAGs have already been in contact with these phages during their evolutionary history, but are not in the present individual.”

4) The association between MAG taxonomy and the phylogeny of their associated phages. This is presented graphically in Figure 3C by comparing the phage phylogeny with the order level assignment of a MAG. There does appear to be an association, but this needs to be quantified. This could be demonstrated with a permutation ANOVA of the phylogenetic distance between phages against their order assignment (using adonis in vegan for example). It could then be determined if this association breaks down at higher MAG taxonomic resolution. Similarly, the data in Figure 3B could be used in a Mantel test to determine strength of association between MAG distance and phage distance.

As suggested, we have performed statistical tests to support this association. Regarding more specifically the phage phylogenetic tree, we did a permutation test (Krukal-Wallis) by considering the order of the genus of the associated MAG as the observed order and obtained a p-value of 0.0132.

We added sentences in the Results section and the Materials and methods section to explain our tests.

“Phage contigs were then positioned in a phylogenetic tree based on a set of 77 specific markers of double stranded DNA phages (Low et al., 2019) (Materials and methods). […] The tree unveils clusters of closely related phages infecting the same bacterial genus (Kruskal-Wallis test, p-value = 0.0132) suggesting, again, that phages are specific to their hosts in human gut.”

“The R environment was used for all the analysis (R Core Team, 2020). […] In Figure 3C, a Kruskal-Wallis test with continuity correction was used, considering that the different MAG genus are ranked all along the circos plot.”

5) Why were the “category B” phages excluded from the analysis. The distribution of their contacts across MAGs would be a good sanity check for the methodology. Ignoring these results seems inappropriate.

Phage contigs encompassed in the category B are not unambiguously assigned to a MAG. Therefore, they could represent different types of phages like free phages or phages able to infect different but phylogenetically related MAGs.

We have now analyzed those sequences further. Most of them (2,096 out of 2,460) do not cluster at all with a MAG over the 10 iterations of the recursive partition procedure. Over the 364 remaining contigs, only 7 clustered with at least two different MAGs over the 10 iterations, while 260 exhibit an association score of ≥ 5 / 10 (i.e. they cluster at least 5 times with a MAG over the 10 iterations of the recursive partition procedure). Therefore, it appears that most phages contigs sequences from categories B may correspond to free phages, and that very few exhibit clustering with several MAGs, reinforcing the idea that phages are highly specific and infecting only one species. Consequently, we have integrated 260 contigs from category B (the ones that exhibit an association score ≥ 5 / 10 during the recursive procedure) into category A, and added a paragraph to explain these results and better explain this category. See also Figure 3—figure supplement 1 which has been modified.

“Three classes (A, B, C) of phages were defined (Figure 3—figure supplement 1; see also Materials and methods). […] These results suggest that the majority (6,763 out of 9,488, i.e. ~ 70%) of the phages present within the human gut are specific, infecting only one bacterial species (Figure 3A).”

6) Simply observing twofold higher coverage for a phage than the host is not definitive evidence that the phage is lytic since multiple factors influence sequence coverage, the phage could be multicopy, it could be near the origin of replication, or possess a sequence composition biased to higher coverage. The authors need to demonstrate that the phage sequences have significantly higher coverage than expected given these sources of variation. Since they are using growth estimation methods they could look at position in the genome and compare with non-viral contigs that are at a similar point. It would be helpful if the authors could pull out specific phage-host pairs as examples of what these patterns look like.

These are excellent suggestions to validate this point, as all the aforementioned factors would have an influence on the coverage (an inherent bias of metagenomic analysis). We thank the reviewers for raising this point which has allowed us to refine the analysis, and most likely removed some false positive events.

We further investigated the potential influence of GC content on the ratio. Plotting the coverage ratio of the phages contigs as a function of the mean GC content of these contigs, or as a function of the GC ratio (phages contigs / MAGs mean GC%), did not unveil a correlation between the ratio and GC content (Figure 4—figure supplement 1B).

Then, we investigated the influence of replication activity in the host by searching in each associated MAGs contigs encompassing the dnaA gene and likely to carry the origin of replication of the bacterial genome (n = 856 MAGs implicating 6239 phages contigs) (Emiola and Oh, 2018). As expected, “dnaA” contigs (dubbed oriMAG contigs) exhibit a mean read coverage ratio higher than the other contigs (mean = 1.37; SD = 0.51; max = 5.15). The coverage ratio analysis was then refined considering the phage vs. oriMAG contigs coverage ratio instead of the mean coverage of the MAGs (Figure 4Aa). We classified the phages into four categories based on their read coverage ratio with oriMAGs contigs (Figure 4A and B).

Phages displaying a ~1:1 ratio with their oriMAG contig (category 1, 0.5 < ratio < 2.0) represented the vast majority of the 6,239 phages-host pairs (n = 4,477; ~75%). They correspond likely to dormant (pro)phages, able to reactivate or not. This population may also encompass active phages that incidentally exhibit the same coverage of their host. For 1,395 phage-hosts pairs, the oriMAG contig was significantly more covered than the associated phage (category 2 – undefined). This profile is consistent with an abortive phage infection cycle, a pseudo – lysogenic cycle or even with prophages not present in the whole host population. The remaining 367 pairs (~5%) corresponded to phages with a higher coverage than their assigned oriMAG contig: category 3 corresponded to a ratio between 2 and 4 (n=291), and category 4 to a ratio ≥ 4 (n=76). These phages may be (category 3) or are probably (category 4) actively replicating, potentially impacting the ecosystem through lytic activities. Overall, these results support the conclusions of former and recent work suggesting that the major part of the phages found in human gut are temperate phages with few lytic activity (Džunková et al., 2019; Minot et al., 2011; Reyes et al., 2010).

Figure 4 and the corresponding Results section have been changed accordingly.

7) When discussing the novelty of MAGs, this should really be placed in the context of the recent large-scale binning surveys (e.g. Pasolli et al., 2019) which similarly observed most novel diversity in the Clostridia.

We have performed a new analysis regarding the novelty of our retrieved MAGs using the latest GTDB-Tk database (release 0.95, July 2020), and have added a sentence in light of the publication from Pasoli et al.

“The 715 MAGs were then compared to the latest genome references of the GTDB-Tk databases (release 0.95; Chaumeil et al., 2020). […] This relatively low number of newly recovered species, although still significant, suggests that the databases about the human gut microorganisms (at least those describing western population) are converging towards completion.”

8) Was there any taxonomic signal in the number of phages associated with a MAG?

We haven’t detected any specific taxonomic signal in the number of phages associated with specific clades, this is now indicated in the main text.

“On average, a MAG was associated with 6.9 phages contigs, but this rate presented high fluctuations (SD = 7.4), with a maximum of 45 contigs (MAG #9010_74_0; see Supplementary files 4 and 6). No correlation between this rate and taxonomic annotation was identified.”

9) Without a cell harvesting step; intra cellular crosslinking, digestion and ligation; or a large ligation reaction volume, how is random proximity ligation and / or crosslinking of extracellular DNA avoided following a freeze thaw cycle of sample with no storage buffer? This may not be an issue as the data seems to suggest a small degree of contamination in the fused reads, yet this was not explicitly tested experimentally with mock communities. Having the molarity, copy number or even mass of DNA entering the ligation step would likely help as it is suspected it is quite dilute in the 1.1 ml volume.

The influence of fixation on frozen samples has been tested in the lab using human gut samples (the same sample was fixed directly – fresh – or after a frozen period of 6 month). It appears that the variation in 3D signal resulting from the frozen step affects slightly the DNA topology (less self-interacting domains, less DNA loops detected, etc.) However, we never observed a significant increasing of the noise signal between species.

The molarity of DNA entering the ligation step for metagenomics samples is unfortunately not easy to define, even though we don’t think this will be an important limitation in applying the technique to other samples. This quantity is highly dependent of the response of the sample to the breakage step, the restriction enzymes, the richness of the raw sample… In extreme cases (stool samples from sick children) we noticed that the protocol fails at generating optimal results, but typically these cases are highly predictable from the nature of the sample.

Note that in the present protocol (similar to current Hi-C protocols) we isolate the insoluble fraction of the sample, containing mostly DNA-proteins complexes, prior to the ligation step (Gavrilov et al., 2013). If DNA escape from the thawing cells, it is likely to be discarded as well at this step, and it is unlikely that significant amounts of free DNA fragments are engaged in this step. Moreover, most of the protocols published lastly on pure culture have significantly reduced the volume of the ligation step without any loss of contact map resolution or increasing of noise signal (Helmsauer et al., 2020; Rao et al., 2014). Therefore, we think it is reasonable to assume this step is not a limitation.

We nevertheless added a warning in the Materials and methods to raise awareness about this potential limitation, in some samples.

10) Overall the manuscript indicates that two method derivatives are developed, but only meta-eHiC is discussed. The Materials and methods list the procedures for preparing the meta-Hi-C and meta-eHiC libraries so one can deduce the differences in the methods, but the Results should take a moment to explain these two derivatives. Is not developing them a result of this study? This would frame the results and interpretation of this work for the reader.

We actually tried to “lighten” the technical part of the manuscript, simplifying the nomenclature, etc. We have also brought more details in the Materials and methods section. We hope that the manuscript reads more easily now with respect to this question, and that the interest of the derivative described here is now obvious.

11) What is the correlation supposedly presented in Supplementary Figure 2, Results? A statistic is needed to support this claim.

We agree that the correlation is not obvious and after reflecting and thinking further about it, it appeared to us that it is difficult to compare the different libraries regarding their noise signal. We have, therefore, removed the corresponding sentence and figure.

12) For VIBRANT and VIrSorter results clarification is needed concerning uniqueness. Are those considered unique between the two tools or unique across samples (and possibly also between the two tools)? Also, were all category 1-3 VirSorter phages included or were just category 1 or category 1 and 2 (like VIBRANT does)? This detail should be included in the Materials and methods.

We include for each sample all the phages detected by VirSorter and VIBRANT (all contigs identified as phage by one of these programs is retained for downstream analysis). As we wanted to compare phages and their characterized hosts between samples, we have not removed identical phages between samples. Regarding VirSorter, we have included all the phages (not prophages) from category 1 to 3. We have added the details in the Materials and methods section.

13) By looking just at the active phage population (Class A), one could speculate that they were lytic/infectious which may be reducing bacterial population numbers. But the ori/ter ratio is not really providing insight into their population spread rather just their replication rate. It is unclear how the authors draw their conclusion of an inverse correlation from Figure 4A, right panel. Is there a statistic to quantify this observation?

Actually, class A phages were not characterized as lytic/infectious but as non-ambiguously assigned phages. We have tried to be clearer in the main text. Within the Class A, phages exhibiting a read coverage ratio over 2 with respect to their assigned MAGs were suspected to be actively replicating phages and potentially active actors of bacterial population behavior and growth.

We agree with reviewers that the conclusion of an inverse correlation from Figure 4A is wrong and we did not have a metric for that. We thank them to point at this problem, since we could not detect any correlation (or anticorrelation) between bacterial growth rate and phages activity. We have also explored a possible correlation between phages ratio and MAGs abondance without success.

“To further question the influence of active phages, we investigated MAGs abundance and growth rate using GRID. […] No significant correlation was detected between the presence of active phages and the abundance/growth rate of their corresponding hosts, as assessed by this analysis (Figure 4C and D).”

https://doi.org/10.7554/eLife.60608.sa2

Article and author information

Author details

  1. Martial Marbouty

    Institut Pasteur, Unité Régulation Spatiale des Génomes, CNRS, UMR 3525, Paris, France
    Contribution
    Conceptualization, Formal analysis, Investigation, Methodology, Writing - original draft, Writing - review and editing
    For correspondence
    martial.marbouty@pasteur.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1668-8423

  2. Agnès Thierry

    Institut Pasteur, Unité Régulation Spatiale des Génomes, CNRS, UMR 3525, Paris, France
    Contribution
    Methodology
    Competing interests
    No competing interests declared

  3. Gaël A Millot

    Institut Pasteur, Bioinformatics and Biostatistics Hub, CNRS, USR 3756, Paris, France
    Contribution
    Formal analysis
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0591-3509

  4. Romain Koszul

    Institut Pasteur, Unité Régulation Spatiale des Génomes, CNRS, UMR 3525, Paris, France
    Contribution
    Conceptualization, Supervision, Funding acquisition, Investigation, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    romain.koszul@pasteur.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3086-1173

Funding

European Research Council (771813)

  • Romain Koszul

Agence Nationale de la Recherche (ANR-16-JPEC-0003-05)

  • Romain Koszul

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This research was supported by funding to RK from the European Research Council under the Horizon 2020 Program (ERC grant agreement 771813) and JPI-EC-AMR STARCS ANR-16-JPEC-0003–05. We thank all our colleagues from the laboratory, especially Pierrick Moreau, Lyam Baudry, and Cyril Matthey-Doret for discussions, feedback, and comments. We also thank MA Petit and A Laffitte for constructive comments on the data analysis and the manuscript, as well as our colleagues from the JPI STARCS consortium.

Ethics

Human subjects: The work involved feces samples of healthy human individuals, stored in the Institut Pasteur biobanque (library). This research receives the ethical agreement n°N18 from Institut Pasteur (ICAReB), and through this process we dont need informed consent from the individual donors.

Senior Editor

  1. Wendy S Garrett, Harvard T.H. Chan School of Public Health, United States

Reviewing Editor

  1. Breck A Duerkop, University of Colorado School of Medicine

Reviewers

  1. Breck A Duerkop, University of Colorado School of Medicine
  2. Christopher Quince, University of Warwick, United Kingdom

Publication history

  1. Received: July 1, 2020
  2. Accepted: February 20, 2021
  3. Accepted Manuscript published: February 26, 2021 (version 1)
  4. Version of Record published: March 16, 2021 (version 2)

Copyright

© 2021, Marbouty et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 4,323
    Page views
  • 557
    Downloads
  • 28
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Martial Marbouty
  2. Agnès Thierry
  3. Gaël A Millot
  4. Romain Koszul
(2021)
MetaHiC phage-bacteria infection network reveals active cycling phages of the healthy human gut
eLife 10:e60608.
https://doi.org/10.7554/eLife.60608

Categories and tags

Research organisms

Further reading

    1. Microbiology and Infectious Disease

    The Plasmodium falciparum apicoplast cysteine desulfurase provides sulfur for both iron-sulfur cluster assembly and tRNA modification

    Russell P Swift, Rubayet Elahi ... Sean T Prigge
    Research Article Updated

    Iron-sulfur clusters (FeS) are ancient and ubiquitous protein cofactors that play fundamental roles in many aspects of cell biology. These cofactors cannot be scavenged or trafficked within a cell and thus must be synthesized in any subcellular compartment where they are required. We examined the FeS synthesis proteins found in the relict plastid organelle, called the apicoplast, of the human malaria parasite Plasmodium falciparum. Using a chemical bypass method, we deleted four of the FeS pathway proteins involved in sulfur acquisition and cluster assembly and demonstrated that they are all essential for parasite survival. However, the effect that these deletions had on the apicoplast organelle differed. Deletion of the cysteine desulfurase SufS led to disruption of the apicoplast organelle and loss of the organellar genome, whereas the other deletions did not affect organelle maintenance. Ultimately, we discovered that the requirement of SufS for organelle maintenance is not driven by its role in FeS biosynthesis, but rather, by its function in generating sulfur for use by MnmA, a tRNA modifying enzyme that we localized to the apicoplast. Complementation of MnmA and SufS activity with a bacterial MnmA and its cognate cysteine desulfurase strongly suggests that the parasite SufS provides sulfur for both FeS biosynthesis and tRNA modification in the apicoplast. The dual role of parasite SufS is likely to be found in other plastid-containing organisms and highlights the central role of this enzyme in plastid biology.

    1. Cell Biology
    2. Microbiology and Infectious Disease

    Profiling the bloodstream form and procyclic form Trypanosoma brucei cell cycle using single-cell transcriptomics

    Emma M Briggs, Catarina A Marques ... Keith R Matthews
    Research Article Updated

    African trypanosomes proliferate as bloodstream forms (BSFs) and procyclic forms in the mammal and tsetse fly midgut, respectively. This allows them to colonise the host environment upon infection and ensure life cycle progression. Yet, understanding of the mechanisms that regulate and drive the cell replication cycle of these forms is limited. Using single-cell transcriptomics on unsynchronised cell populations, we have obtained high resolution cell cycle regulated (CCR) transcriptomes of both procyclic and slender BSF Trypanosoma brucei without prior cell sorting or synchronisation. Additionally, we describe an efficient freeze–thawing protocol that allows single-cell transcriptomic analysis of cryopreserved T. brucei. Computational reconstruction of the cell cycle using periodic pseudotime inference allowed the dynamic expression patterns of cycling genes to be profiled for both life cycle forms. Comparative analyses identify a core cycling transcriptome highly conserved between forms, as well as several genes where transcript levels dynamics are form specific. Comparing transcript expression patterns with protein abundance revealed that the majority of genes with periodic cycling transcript and protein levels exhibit a relative delay between peak transcript and protein expression. This work reveals novel detail of the CCR transcriptomes of both forms, which are available for further interrogation via an interactive webtool.

    1. Epidemiology and Global Health
    2. Microbiology and Infectious Disease

    Integrating contact tracing and whole-genome sequencing to track the elimination of dog-mediated rabies: an observational and genomic study

    Kennedy Lushasi, Kirstyn Brunker ... Katie Hampson
    Research Article

    Background:

    Dog-mediated rabies is endemic across Africa causing thousands of human deaths annually. A One Health approach to rabies is advocated, comprising emergency post-exposure vaccination of bite victims and mass dog vaccination to break the transmission cycle. However, the impacts and cost-effectiveness of these components are difficult to disentangle.

    Methods:

    We combined contact tracing with whole-genome sequencing to track rabies transmission in the animal reservoir and spillover risk to humans from 2010-2020, investigating how the components of a One Health approach reduced the disease burden and eliminated rabies from Pemba Island, Tanzania. With the resulting high-resolution spatiotemporal and genomic data we inferred transmission chains and estimated case detection. Using a decision tree model we quantified the public health burden and evaluated the impact and cost-effectiveness of interventions over a ten-year time horizon.

    Results:

    We resolved five transmission chains co-circulating on Pemba from 2010 that were all eliminated by May 2014. During this period, rabid dogs, human rabies exposures and deaths all progressively declined following initiation and improved implementation of annual islandwide dog vaccination. We identified two introductions to Pemba in late 2016 that seeded re-emergence after dog vaccination had lapsed. The ensuing outbreak was eliminated in October 2018 through reinstated islandwide dog vaccination. While post-exposure vaccines were projected to be highly cost-effective ($256 per death averted), only dog vaccination interrupts transmission. A combined One Health approach of routine annual dog vaccination together with free post-exposure vaccines for bite victims, rapidly eliminates rabies, is highly cost-effective ($1657 per death averted) and by maintaining rabies freedom prevents over 30 families from suffering traumatic rabid dog bites annually on Pemba island.

    Conclusions:

    A One Health approach underpinned by dog vaccination is an efficient, cost-effective, equitable and feasible approach to rabies elimination, but needs scaling up across connected populations to sustain the benefits of elimination, as seen on Pemba, and for similar progress to be achieved elsewhere.

    Funding:

    Wellcome [207569/Z/17/Z, 095787/Z/11/Z, 103270/Z/13/Z], the UBS Optimus Foundation, the Department of Health and Human Services of the National Institutes of Health [R01AI141712] and the DELTAS Africa Initiative [Afrique One-ASPIRE/DEL-15-008] comprising a donor consortium of the African Academy of Sciences (AAS), Alliance for Accelerating Excellence in Science in Africa (AESA), the New Partnership for Africa's Development Planning and Coordinating (NEPAD) Agency, Wellcome [107753/A/15/Z], Royal Society of Tropical Medicine and Hygiene Small Grant 2017 [GR000892] and the UK government. The rabies elimination demonstration project from 2010-2015 was supported by the Bill & Melinda Gates Foundation [OPP49679]. Whole-genome sequencing was partially supported from APHA by funding from the UK Department for Environment, Food and Rural Affairs (Defra), Scottish government and Welsh government under projects SEV3500 & SE0421.