Statistical Genomics Foundry
Paper
domain/pangenomicstopic/pangenome-openness

Delmont & Eren 2018 — the Prochlorococcus metapangenome (anvi’o pangenomics)

1. Citation

Delmont TO, Eren AM. 2018. “Linking pangenomes and metagenomes: the Prochlorococcus metapangenome.” PeerJ 6:e4320. DOI 10.7717/peerj.4320. Open access: https://peerj.com/articles/4320/ (PMC: PMC5804319). License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/; ©2018 Delmont and Eren) — verbatim load-bearing quotes permitted with attribution.

2. Access note

Full text read via Europe PMC JATS XML (PMC5804319/fullTextXML); PeerJ HTML and one PMC route returned 403 / a mismatched cached article, so extraction is from the EBI full-text XML. All body methods/results sections and the abstract were readable. Figures/tables were not parsed as images; numeric results here come from the running text. This is the primary pangenomics-workflow paper (the anvi’o pangenomic workflow “developed for this study”), distinct from the anvi’o platform paper.

3. Thesis (one sentence)

The authors present an integrated, reproducible anvi’o workflow to build pangenomes and analyze them in conjunction with metagenomes — a metapangenome — and demonstrate it on a Prochlorococcus pangenome placed against a large marine (TARA Oceans) metagenomic survey.

4. Problem & context

Pangenomes characterize core/accessory genes across closely related genomes by clustering genes on sequence homology; metagenomes resolve relative distribution of genomes/genes across environments via read recruitment. Combining them can reveal the functional basis of niche partitioning and fitness, but “advanced software solutions are lacking.” The paper supplies that integrated analysis+visualization strategy and applies it to Prochlorococcus.

5. Method / approach (programs and exact parameters)

Genomes & metagenomes

  • 31 isolate genomes and 74 SAGs (single-amplified genomes, minimum length >1 Mbp) of Prochlorococcus from NCBI; 93 TARA Oceans metagenomes from EBI.

Read recruitment / metagenomic prep

  • Quality filtering: iu-filter-quality-minoche from illumina-utils v1.4.1.
  • FASTA header cleanup: anvi’o script reformat-fasta; concatenated 31 isolate FASTAs.
  • Mapping: Bowtie2, default parameters plus --no-unal; SAM→sorted/indexed BAM via samtools.

Pangenome workflow — three major steps (as stated)

  1. anvi-gen-genomes-storage — generate an anvi’o genomes storage database (DNA + amino acid sequences + functional annotations per gene). Here built from the 31 isolate FASTAs using the --internal-genomes flag.
  2. anvi-pan-genome — compute the pangenome by identifying gene clusters.
  3. anvi-display-pan — display the pangenome; interactively bin gene clusters; inspect per-cluster gene alignments.

anvi-pan-genome parameters used (verbatim flag values)

  • --use-ncbi-blast
  • --minbit 0.5
  • --mcl-inflation 10

What anvi-pan-genome does internally (6 sub-steps, as described)

  1. all-vs-all amino-acid similarity via blastp (every AA sequence vs every other);
  2. removes weak hits using the minbit heuristic (originally described in ITEP), filtering on the aligned fraction between the two sequences;
  3. MCL algorithm identifies gene clusters from remaining blastp results;
  4. computes gene-cluster occurrence across genomes and total gene counts;
  5. hierarchical clustering of gene clusters (by distribution across genomes) and of genomes (by shared gene clusters) — Euclidean distance and Ward clustering by default;
  6. generates an anvi’o pan database for downstream analysis / anvi-display-pan.

Definition of a gene cluster (in this study): sequences of one or more predicted ORFs grouped by homology at the translated DNA sequence level; a cluster with >1 sequence may contain orthologs and/or paralogs from one or more genomes.

Metapangenome workflow

  • anvi-gen-samples-database — with per-genome coverage estimates from the anvi’o profile-DB summary, to connect environmental distribution to the pangenome.
  • anvi-script-gen-environmental-core-summary (default parameters) — quantify the ratio of ECGs vs EAGs within each gene cluster.
  • anvi-display-pan visualizes the metapangenome; anvi-summarize summarizes gene clusters.

Environmental core/accessory gene (ECG/EAG) classification

  • anvi-script-gen-distribution-of-genes-in-a-bin with --fraction-of-median-coverage 0.25: sums per-gene coverage across all metagenomes where the population is “detected”; genes with < 25% of the median coverage of all genes in the genome are marked EAGs (else ECGs). Visualized with anvi-interactive.

Phylogenomics (note: NOT ANI)

  • Phylosift v1.0.1 (default parameters): 37 marker gene families → concatenated alignment → FastTree 2.1 tree; midpoint-rooted in FigTree v1.4.3.

SAG analysis

  • Combined 74 SAGs + 31 isolates through the same pangenome workflow; then 5 phylogenetically distant SAGs run through the metapangenome workflow (few-distant selection intended to minimize the dilution effect of competing read recruitment onto identical regions).

Operational definitions used

  • “Population”: an agglomerate of naturally occurring cells whose genomes align to the same reference at high identity under the read-recruitment stringency.
  • “Detected”: a genome is detected in a metagenome only if >50% of its nucleotide positions have at least 1X coverage (to avoid false detection from non-specific recruitment).

6. Key claims / findings (atomic)

  • Pangenome of the 31 isolates: 60,054 genes → 7,385 gene clusters.
  • Gene clusters binned into five groups: HL+LL core (n=766), HL core (n=492), LL core (n=144), singletons (single-genome clusters, n=2,215), other (n=3,768).
  • Singletons = 30% of all clusters; HL+LL core = 10.4%. Small core is consistent with a Prochlorococcus “open pangenome.”
  • 49.1% of all clusters had genes with COG functional annotation; annotation rate 90.5% for HL+LL core clusters vs 37.2% for singletons.
  • Gene-cluster–based organization of genomes matched the six Prochlorococcus clades but grouped genomes primarily by light-regime adaptation, whereas 37-core-gene phylogenomics (and ITS-based trees) did NOT recapitulate the environmental-distribution patterns. Suggests whole gene-content is better than marker genes for inferring ecological (vs evolutionary) relationships.
  • The metapangenome revealed very closely related isolates (same phylogenetic cluster, differing by few gene clusters) with markedly different environmental abundance / fitness.
  • A small set of core gene clusters occurred mostly in hypervariable genomic islands and systematically lacked read recruitment from surface-ocean metagenomes; these were all linked to sugar metabolism — suggesting benefit from high sequence diversity of sugar-metabolism genes.

7. Load-bearing statements — VERBATIM (CC BY 4.0; short quotes, with location)

  • Metapangenome definition (Methods, “Computing the metapangenome”): “Here we define ‘metapangenome’ as the outcome of the analysis of pangenomes in conjunction with the environment where the abundance and prevalence of gene clusters and genomes are recovered through shotgun metagenomes.”
  • Workflow (Methods, “Computing the pangenome…”): “The anvi’o pangenomic workflow developed for this study consists of three major steps: (1) generating an anvi’o genome database (‘anvi-gen-genomes- storage’)… (2) computing the pangenome (‘anvi-pan-genome’)… and (3) displaying the pangenome (‘anvi-display-pan’)…”
  • Parameters (Methods): “we then used the program ‘anvi-pan-genome’ with the genomes storage database, the flag ‘—use-ncbi-blast’, and parameters ‘—minbit 0.5’, and ‘—mcl-inflation 10’.”
  • minbit (Methods): “removes weak hits using the ‘minbit heuristic’, which was originally described in ITEP …, to filter weak hits based on the aligned fraction between the two reads”.
  • Detection criterion (Methods, “Criterion for ‘detection’”): “we assumed that a genome was ‘detected’ in a given metagenome only if more than 50% of its nucleotide positions had at least 1X coverage.”

8. Stated scope, assumptions, limitations (source’s own)

  • Read recruitment can non-specifically map reads to regions conserved across populations (basis for the >50%-at-1X detection criterion); e.g. genomes recruited up to 0.01% of reads from Southern Ocean metagenomes where Prochlorococcus is virtually absent.
  • The “population” concept is operational and reference-dependent: isolate genomes are assumed to give access to the environmental populations they belong to via read recruitment.
  • SAG metapangenome deliberately used only 5 distant SAGs to avoid a “dilution effect” from competing recruitment onto identical regions among many similar genomes.
  • Method demonstrated on one taxon/biome; generalizes only “for any taxon and biome for which genomic and sufficiently deep metagenomic data are available.”

9. Failure modes / invalidity patterns (referee-relevant)

  • False detection of low-abundance genomes from non-specific short-read recruitment to conserved regions → mitigated by the >50% of positions at ≥1X detection gate; without it, absent populations appear present (Southern Ocean example).
  • Dilution effect: including many near-identical genomes/SAGs splits read recruitment across identical regions, distorting per-genome coverage → they restricted to few phylogenetically distant SAGs.
  • ECG/EAG designations are metagenome-relative and can invert pangenome core/accessory labels: a gene that is pangenome-accessory can be environmental-core and vice versa — the paper flags this as a source of confusion and names the two class systems distinctly (ECG/EAG vs pangenome core/accessory).
  • Marker-gene / concatenated-single-copy-core-gene phylogenies did NOT recover the environmental-distribution structure that gene-cluster content did — using phylogeny as a proxy for ecological relationship is a stated pitfall here.

10. What the source does NOT address (confident silences)

  • No ANI / average nucleotide identity computation. The strings “average nucleotide identity”, “ANI”, anvi-compute-genome-similarity, and PyANI do not appear; genome-to-genome relatedness is assessed via Phylosift/FastTree phylogenomics and via shared gene-cluster content, not ANI.
  • No anvi-meta-pan-genome command (that string does not appear; the metapangenome is assembled from anvi-gen-samples-database + anvi-script-gen-environmental-core-summary + anvi-display-pan).
  • No general statement of a minimum number of genomes required for a “meaningful pangenome” — only the concrete counts used (31 isolates; 74 SAGs; +5-distant-SAG subset) are given.
  • No sensitivity analysis of the chosen --minbit 0.5 / --mcl-inflation 10 values (values are stated, not justified or swept).

11. Open questions / ambiguities

  • Rationale for --mcl-inflation 10 (high vs the anvi’o default) and --minbit 0.5 is not given in the readable text; treat as the values used, not recommended defaults, unless corroborated elsewhere.
  • “translated DNA sequence level” homology + blastp implies protein-level clustering, but the exact gene-calling step for isolates (Prodigal via anvi’o) is not spelled out in the extracted methods.

12. Guidance answers

  1. What the anvi’o pangenomics workflow computes / pipeline + program names. Three steps: anvi-gen-genomes-storage (genomes storage DB) → anvi-pan-genome (gene clusters) → anvi-display-pan (visualize/bin). Internally anvi-pan-genome does blastp all-vs-all → minbit filter → MCL clustering → occurrence/counts → hierarchical clustering (Euclidean + Ward) → pan DB. Related programs used: anvi-gen-samples-database, anvi-script-gen-environmental-core-summary, anvi-summarize, anvi-interactive, anvi-script-gen-distribution-of-genes-in-a-bin, and the reformat-fasta anvi’o script. Functional enrichment: the paper reports COG functional annotation of clusters (49.1% annotated) but does not name a dedicated enrichment command here. ANI: NOT computed — no anvi-compute-genome-similarity/ANI in this paper (see §10).
  2. Clustering algorithm + defaults. Algorithm = MCL on blastp hits after minbit filtering. Values used (verbatim flags): --minbit 0.5, --mcl-inflation 10, with --use-ncbi-blast (blastp). minbit = ITEP heuristic filtering weak hits by aligned fraction between the two sequences. Note these are the values used in the study; the paper does not label them as tool defaults, and the hierarchical-clustering defaults it does call “default” are Euclidean distance + Ward.
  3. “Metapangenome” definition — quoted (§7): “the outcome of the analysis of pangenomes in conjunction with the environment where the abundance and prevalence of gene clusters and genomes are recovered through shotgun metagenomes.”
  4. Number of genomes / minimum for a meaningful pangenome. 31 Prochlorococcus isolate genomes (plus 74 SAGs; a 5-distant-SAG subset for the SAG metapangenome), and 93 TARA Oceans metagenomes. No minimum genome count for a “meaningful pangenome” is stated.

Ingest note (per guidance): this PeerJ 2018 paper is the origin of the anvi’o pangenomic workflow (“developed for this study”); confirmed from the source’s own wording. The paper does not cite itself as “Eren 2021 Nat Microbiol”; no such claim appears here.