Statistical Genomics Foundry
Paper
domain/syntenydomain/whole-genome-duplication

Tang et al. 2008 — Unraveling ancient hexaploidy through multiply-aligned angiosperm gene maps (original MCscan)

1. Citation

Tang H, Wang X, Bowers JE, Ming R, Alam M, Paterson AH. 2008. “Unraveling ancient hexaploidy through multiply-aligned angiosperm gene maps.” Genome Research 18(12):1944–1954. doi:10.1101/gr.080978.108. Free full text: PMC2593578 (https://pmc.ncbi.nlm.nih.gov/articles/PMC2593578/).

This is the ORIGINAL MCscan paper (Genome Research 18:1944) — not the later Python JCVI port and not the Science 320:486 synteny paper.

2. Access note

Full text read via WebFetch against the PMC HTML (accessed 2026-07). The article body and Methods were reachable. Fidelity caveat: the text was relayed through an intermediary summarizing model, so single-sentence strings quoted below are high-confidence but not guaranteed character-exact; the numeric functional strings (E-values, kb distances, score cutoffs, scoring constants) were cross-checked across two independent fetches and returned identical values. Figure equations (the DP recurrence and the expected-occurrences formula) are rendered as images in the source and were NOT transcribable — captured here only structurally. Treat exact equation forms as [GAP: verify against PDF] if needed downstream.

3. Thesis

A γ (gamma) whole-genome triplication (paleo-hexaploidy) occurred in a common ancestor of the core eudicots, and it is recoverable by aligning multiple angiosperm gene maps and reading off the conserved syntenic multiplicity (depth) of aligned ancestral loci.

4. Problem & context

Ancient polyploidy events are obscured by gene loss and rearrangement, so pairwise synteny alone is ambiguous about how many rounds of duplication a lineage carries. The paper builds a multiple alignment of eudicot gene maps to bound the many-to-many syntenic depth of aligned loci and thereby distinguish a triplication (γ) from later independent duplications, across four sequenced eudicots (Arabidopsis, Carica papaya, Populus trichocarpa, Vitis vinifera) plus rice as an outgroup.

5. Method / approach (MCscan)

BLAST → pairwise collinear blocks (DP) → multiple alignment by progressive stacking → depth counting.

  • Homology input. Each genome compared against itself and others with BLASTP, retrieving the best five hits meeting an E-value threshold of 1 × 10⁻⁵.
  • Tandem collapse. Tandem matches <50 kb apart are collapsed, keeping a representative pair with the smallest BLASTP E-value. (Source notes this did not purge every tandem; a few long-distance tandems remained. It frames the 50-kb value as a trade-off — raising it would remove some genuine intra-chromosomal WGD duplicates.)
  • Chaining into collinear blocks (pairwise). Gene pairs are chained by dynamic programming. The default (configurable) scoring scheme is: match score = min(log10 E, 50) per gene pair, and a −1 gap penalty for each 10-kb distance between two consecutive gene pairs. (The explicit DP recurrence is a figure equation, not transcribed.)
  • Block reporting cutoff. All pairwise segments with scores above 300 are reported.
  • Statistical significance. An expected-number-of-occurrences model estimates how often a collinearity pattern arises by chance, as a function of the number of matching gene pairs (N), the number of collinear gene pairs (m), and the lengths (L1, L2) of the two regions; the expectation is multiplied by 2 for the two possible relative orientations of two collinear segments. The reported pairwise alignments are significant at E < 1 × 10⁻¹⁰. (Full formula is a figure equation; not transcribed — [GAP: exact formula].)
  • Multiple alignment (“multi-collinearity”). Pairwise blocks are combined by fixing one gene order as reference and heuristically stacking the pairwise synteny tracks one after another (threaded by the reference order, in the TBA / threaded-blockset-aligner framework). Because stacking order affects the result, a refinement procedure iteratively realigns each segment. Described as a top-down strategy needing only one deduction cycle (find pairwise synteny, then combine into a multi-way correspondence) rather than bottom-up sequential merging.
  • Downstream mapping. For placing Solanum and Musa BAC/unigene sequences, additional thresholds appear: BLASTN E < 1 × 10⁻⁶ and BLASTX E < 1 × 10⁻⁵.

6. Key claims / findings

  • Syntenic depth (multiplicity) of an aligned locus indexes duplication history. Expected multiplicities under the stated event histories: Carica = 3 (γ only), Vitis = 3 (γ only), Populus = 6 (γ × one additional duplication p), Arabidopsis = 12 (γ, β, α — three events).
  • Empirical support: 88 aligned loci in Carica have multiplicity 3 (only one aligned locus exceeded 3); 54 aligned loci in Populus have the expected multiplicity 6.
  • Cross-eudicot triplicated structure: aligned regions show a 1 Carica : 1 Vitis : 2 Populus : 4 Arabidopsis correspondence, consistent with γ predating eudicot divergence → γ occurred in the common ancestor of the four species.
  • Fraction of genes returned to ancestral locations by synteny: ~61% Arabidopsis, 44% Carica, 51% Populus, 46% Vitis; only ~14% of rice genes placed in cross-species clusters.
  • Estimated ancestral (angiosperm/eudicot) gene number varies by species from 10,149 (Carica) to 13,043 (Populus), consistent with a prior estimate of 12,000–14,000 ancestral genes.

7. Load-bearing statements — OWN-WORDS mode (restrictive: CSHL / Genome Research copyright)

License posture: Genome Research is freely readable but CSHL-copyrighted, not an open/CC license → own-words paraphrase; only functional strings (numbers, parameter names, thresholds) reproduced verbatim per the functional-strings exception.

  • Pairwise collinear blocks are found by DP over BLASTP gene pairs; default scoring is the match term min(log10 E, 50) minus a −1 per 10-kb gap term, and segments scoring >300 are reported.
  • Homology comes from BLASTP keeping the best five hits at E ≤ 1 × 10⁻⁵; tandems <50 kb are collapsed to one representative (smallest E).
  • Reported alignments are significant at E < 1 × 10⁻¹⁰; the chance model multiplies expected occurrences by 2 for the two segment orientations.
  • Multiple alignment = fix one genome’s gene order as reference, progressively stack pairwise tracks, then iteratively refine because stack order matters.

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

  • The 50-kb tandem-collapse threshold is acknowledged as imperfect (residual long-distance tandems) and chosen as a trade-off against removing real WGD duplicates.
  • Progressive-stacking result depends on the order of stacking; the refinement step is introduced specifically to mitigate this order dependence.
  • Depth/multiplicity interpretation assumes the stated event histories per lineage (γ; γ×p for Populus; γ,β,α for Arabidopsis).
  • Rice (monocot) recovers a much smaller synteny fraction (~14%) in the cross-species clusters, bounding cross-clade resolution.

9. Failure modes / invalidity patterns

  • Unpurged tandems inflate/distort multiplicity. The source explicitly reports a few long-distance tandems survived the 50-kb collapse — a symptom is aligned loci with multiplicity above the expected level (e.g., the single Carica locus exceeding 3).
  • Stacking-order artifacts. Multiple-alignment output can depend on progressive-stacking order; the refinement/iterative-realignment step is the named remedy — its absence is the failure mode.
  • Score/E-value cutoffs gate sensitivity. Blocks below score 300, or hits above E 1 × 10⁻⁵, are not reported; weak/eroded ancient synteny can drop out (relevant for deep events).
  • Depth misreading. Interpreting multiplicity without the correct per-lineage event history (γ vs γ×p vs γ,β,α) would misassign ploidy — the method reads depth relative to an assumed history.

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

  • No “cscore” / “C-score” concept and no reciprocal-best-hit stringency parameter. The strings “c-score”, “cscore”, “reciprocal”, and “quota” do NOT appear anywhere in the paper (confirmed by targeted string search over the full text). MCscan-2008 controls stringency via the BLASTP E-value (1 × 10⁻⁵), the best-five-hits cap, the DP scoring scheme, and the score-300 / E<10⁻¹⁰ block cutoffs — NOT via a C-score.
  • No --tandem_Nmax-style gene-count parameter. Tandem handling is a distance threshold (<50 kb), not a max-tandem-gene-count parameter. No --tandem_Nmax or equivalent named parameter appears.
  • No command-line flag catalog, software-version pin, or config-file schema is given (this is the algorithm/biology paper, not a software manual).

11. Open questions / ambiguities

  • Exact DP recurrence and exact expected-occurrences formula are figure equations not transcribable from the HTML relay — [GAP: transcribe both from the PDF].
  • Whether “best five hits” is itself configurable is not stated (scoring scheme is called “configurable”; the hit cap is stated as a fixed 5).

12. Guidance answers

  1. MCscan algorithm / chaining — yes. BLASTP gene pairs chained by dynamic programming; default configurable scoring = min(log10 E, 50) match score per pair and −1 gap penalty per 10-kb between consecutive pairs; segments scoring >300 reported; block significance via an expected-occurrences chance model (N matching pairs, m collinear pairs, region lengths L1/L2, ×2 for orientation), with reported alignments significant at E < 1 × 10⁻¹⁰. Homology from BLASTP best-5 hits at E ≤ 1 × 10⁻⁵. (Exact equation forms are figures — not transcribed.)
  2. cscore / C-score / reciprocal-best-hit stringency — ABSENT. The strings “c-score”, “cscore”, and “reciprocal” do NOT appear anywhere in the paper. This settles it: the skill’s --cscore 0.7/0.95/0.99 is JCVI-software-only, NOT Tang-2008-cited. In 2008 MCscan, stringency is controlled by the E-value (1 × 10⁻⁵), best-five-hit cap, DP scoring scheme, and score-300 / E<10⁻¹⁰ cutoffs.
  3. Multiply-aligned gene maps / depth bounding — yes (but the word “quota” is ABSENT — no “quota” algorithm here; that is later JCVI). Multiplicity/depth is read against per-lineage event histories: expected 3 (Carica, Vitis; γ), 6 (Populus; γ×p), 12 (Arabidopsis; γ,β,α); empirically 88 Carica loci at depth 3, 54 Populus loci at depth 6, and a 1:1:2:4 Carica:Vitis:Populus:Arabidopsis correspondence.
  4. Visualization — collinearity diagrams present (Fig. 2), plus phylogenetic trees (Fig. 5). The paper shows gene-order collinearity glyphs (genes as glyphs with transcriptional direction; shaded synteny matches). It does NOT describe the karyotype/dotplot rendering tooling associated with the later JCVI Python port; that is [summarizer-inferred] as out-of-scope for THIS paper.
  5. Anchor / E-value defaults stated: BLASTP best-5 hits, E ≤ 1 × 10⁻⁵; block score cutoff >300; block significance E < 1 × 10⁻¹⁰; tandem collapse <50 kb; downstream BLASTN E < 1 × 10⁻⁶, BLASTX E < 1 × 10⁻⁵. No minimum-anchor-COUNT (e.g. “≥5 gene pairs”) threshold is stated — the block gate is the score (>300), not an anchor count.

Functional-strings summary (verbatim facts): cscore = ABSENT; --tandem_Nmax-like gene-count param = ABSENT (tandem control is distance-based, <50 kb); scoring = min(log10 E, 50) match, −1 per 10-kb gap; block cutoff score >300; homology BLASTP best-5, E ≤ 1 × 10⁻⁵; significance E < 1 × 10⁻¹⁰.