Statistical Genomics Foundry
Paper
domain/ancestral-reconstruction

Some Limitations of Ancestral Character-State Reconstruction When Testing Evolutionary Hypotheses (Cunningham 1999)

1. Citation

  • Cunningham, Clifford W. 1999. “Some Limitations of Ancestral Character-State Reconstruction When Testing Evolutionary Hypotheses.” Systematic Biology 48(3):665–674.
  • DOI: 10.1080/106351599260238 (JSTOR stable/2585333).
  • Section: “Points of View.” Associate Editor: K. Omland. © 1999 Society of Systematic Biologists (all rights reserved).
  • Author affiliation (from source): Zoology Department, Duke University, Durham, North Carolina 27708-0325, USA; e-mail cliff@duke.edu.
  • Page range verified from the source header (“Syst. Biol. 48(3):665–674, 1999”; running head page number “665”).
  • Distinct from Cunningham, Omland & Oakley 1998, TREE 13:361 (“Reconstructing ancestral character states: a critical reappraisal”) — a different paper, cited within this one as “Cunningham et al., 1998.” NOT merged here.

2. Access note — PAYWALL BOUNDARY

  • The article is behind OUP / JSTOR paywall (all-rights-reserved). No open-access full text was obtainable (OUP, JSTOR, sci-hub, ResearchGate, Wayback, Internet Archive Scholar, shadow-library CDNs all blocked or gated; a stale Unpaywall “bronze” flag proved incorrect — OUP shows “You do not currently have access to this article”).
  • What was actually read: only page 665 (the article’s first page), captured legitimately from OUP’s free “Article PDF first-page preview” image. This covers the full title/byline, the introduction, and the opening of the first case study.
  • Pages 666–674 were NOT accessible. All content past the first page — including the detailed limitations, the parsimony-vs-likelihood analysis, the two echinoderm case-study results, uncertainty quantification, and any recommendations — is beyond the boundary and is NOT summarized below. Where a section can only be filled from the paywalled body, it says so explicitly rather than guessing.

3. Thesis (1 sentence)

Reconstructed ancestral character states are a weak basis for testing certain evolutionary hypotheses — especially hypotheses of irreversible evolution (Dollo’s law) — because the characters used are typically under selection and thus prone to homoplasy, making ancestral-state inference for individual characters harder and less reliable than phylogeny inference. (Stated in the p. 665 introduction; the paper says it illustrates this with two echinoderm life-history case studies.)

4. Problem & context (from p. 665)

  • The explosion of phylogenetic information has spread beyond systematics; workers across disciplines generate phylogenies and use them to reconstruct ancestral character states, “usually by parsimony.”
  • The appeal is simplicity: testing evolutionary hypotheses by mapping characters onto a phylogeny is attractive and widely adopted.
  • Counterpoint the paper presses: ancestral-state reconstruction is an “ambitious exercise.” Even though inferring phylogenies from thousands of characters is hard (Swofford et al. 1996), reconstructing the ancestral state of an individual character is correspondingly more difficult (cites Swofford & Maddison 1987, 1992; Maddison & Maddison 1992; Collins et al. 1994a; Frumhoff & Reeve 1994; Pagel 1994; Maddison 1995; Schluter 1995; Schultz et al. 1996; Omland 1997; Schluter et al. 1997; Cunningham et al. 1998).
  • Key framing argument (p. 665): data used for phylogeny are generally assumed selectively neutral, but many/most characters reconstructed to test evolutionary hypotheses are thought to be under selection (Brooks & McLennan 1991; Harvey & Pagel 1991) — indeed their presumed selective importance is why they are of interest. Therefore the convergence and parallel-evolution (homoplasy) problems that plague phylogenetic inference “should be that much more serious when we test evolutionary hypotheses with ancestral state reconstructions.”

5. Method / approach

  • Stated approach (p. 665): a “Points of View” argument piece. The author states he discusses the limitations of using reconstructed ancestral character states to test evolutionary hypotheses, arguing specifically that hypotheses of irreversible evolution (e.g., Dollo’s law; Dollo 1893) are particularly difficult to test this way, and that he illustrates the point “with two case studies of life history evolution in echinoderms.”
  • Case study 1 setup visible on p. 665 — section heading “DOLLO’S LAW AND THE BIASED LOSS OF LARVAL FEEDING IN ECHINODERMS”:
    • Strathmann (1978, 1985, 1993) argued obligate larval feeding has been repeatedly lost across echinoderm evolution; larval feeding appears homologous among echinoderm classes and dates back at least to the Ordovician; Strathmann attributes the absence of larval feeding in many echinoderms (including the entire Class Crinoidea) to a unidirectional bias of repeated losses.
    • Simpson’s (1953) reformulation of Dollo’s law: once lost, complex characters should be difficult to regain (a complex character’s expression can be interrupted by a few genes; unused developmental pathways become progressively harder to restore — Raff et al. 1990; Maddison 1994; Omland 1997).
    • Echinoderm larval feeding requires a ciliary system and a full digestive tract, so its loss “certainly qualifies as the loss of a complex character” (Strathmann 1978).
    • The visible text ends mid-sentence: Strathmann has also proposed functional reasons why regaining obligate feeding should be especially difficult in [continues on p. 666 — NOT accessible].
  • The actual analytical method (how parsimony vs. likelihood reconstructions were compared, model settings, rate parameters, the second case study, and results) is on pp. 666–674 and was NOT accessible.

6. Key claims / findings

From the accessible first page only:

  • Ancestral-state reconstruction is most commonly done by parsimony; its appeal is the simplicity of mapping characters onto a phylogeny. [p. 665]
  • Reconstructing the ancestral state of a single character is harder, not easier, than reconstructing a phylogeny from many characters. [p. 665]
  • Characters chosen to test evolutionary hypotheses are usually under selection (unlike the near-neutral characters assumed for phylogeny building), so homoplasy (convergence, parallelism) is a more severe problem for hypothesis tests based on ancestral states than for phylogeny inference itself. [p. 665]
  • The paper’s central thesis: irreversibility hypotheses (Dollo’s law) are especially difficult to test with ancestral-state reconstruction. [p. 665]
  • All quantitative findings, thresholds, and the parsimony/likelihood comparison are on pp. 666–674 and were NOT accessible. No numeric results are reported here because none appear on p. 665.

7. Load-bearing statements — MODE: own-words paraphrase (all-rights-reserved; no verbatim prose)

Under the ARR license, load-bearing statements are paraphrased, not quoted. Only p. 665 statements are available:

  • The paper’s stated purpose: to discuss the limitations of using reconstructed ancestral character states to test evolutionary hypotheses, arguing in particular that irreversible-evolution hypotheses (e.g., Dollo’s law) are especially hard to test by ancestral character-state reconstruction, illustrated with two echinoderm life-history case studies. [p. 665, intro]
  • The paper’s core caution: because the characters used in such tests are typically under selection rather than neutral, the convergence/parallel-evolution problems that already trouble phylogenetic inference are even more serious when ancestral-state reconstructions are used to test evolutionary hypotheses. [p. 665]
  • Functional strings (verbatim, license-exempt): DOI 10.1080/106351599260238; author e-mail cliff@duke.edu; case-study section title “DOLLO’S LAW AND THE BIASED LOSS OF LARVAL FEEDING IN ECHINODERMS.”
  • The specific load-bearing statements the guidance asks for (parsimony’s bias under rate asymmetry; the caution against treating a single reconstruction as certain) are in the paywalled body (pp. 666–674) and could NOT be captured.

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

Visible on p. 665: the paper positions ancestral-state reconstruction as an “ambitious exercise,” harder for individual characters than phylogeny inference, and worsened by selection-driven homoplasy; it scopes its argument to hypotheses of irreversible evolution and to two echinoderm life-history case studies. The author’s own enumerated caveats and limitations elsewhere in the paper are past the boundary.

9. Failure modes / invalidity patterns

From p. 665, the named invalidity driver is homoplasy in selected characters: convergence and parallel evolution corrupt ancestral-state inference more than they corrupt phylogeny inference, so hypothesis tests resting on reconstructed ancestral states are correspondingly unreliable; hypotheses of irreversibility (Dollo’s law) are the hardest case. Any explicit detector, diagnostic, rate-asymmetry symptom, or invalidity threshold the paper names is on pp. 666–674 and was NOT accessible; it is not reproduced or inferred here.

10. What the source does NOT address

Cannot be assessed — only p. 665 was read. The confident-silence judgement would require the full text.

11. Open questions / ambiguities

Not determinable from p. 665 alone. (The intended payload of this note — how parsimony and likelihood diverge under asymmetric rates, and how much uncertainty remains — lives in the inaccessible body.)

12. Guidance answers

Guidance file present; answered against the accessible text (p. 665) only. Guidance directs attention, not conclusions — where the source’s body is paywalled, that is stated rather than filled from memory or secondary sources.

  • Specific limitation(s) the paper enumerates about ASR when testing hypotheses (its own enumeration): Partially available. Page 665 states the thesis-level limitations: (a) reconstructing an individual character’s ancestral state is harder than phylogeny inference; (b) test characters are under selection, so homoplasy (convergence/parallelism) is more damaging than in phylogenetics; (c) irreversibility hypotheses (Dollo’s law) are especially difficult to test by ASR. The paper’s fuller, itemized enumeration is in the paywalled body (pp. 666–674) and was NOT captured.
  • Parsimony vs. ML under ASYMMETRIC forward/reverse rates — does parsimony (implicitly symmetric) bias toward the rarer state at ancestors; direction and cause? NOT ANSWERED by the accessible text. This analysis is in the paywalled body (pp. 666–674); page 665 does not state it. Not answered from memory or secondary sources per the faithfulness rule. (Page 665 only foreshadows the biological claim of a unidirectional loss bias in echinoderm larval feeding, i.e., Strathmann’s hypothesis under test — not the methodological parsimony-vs-ML bias statement.)
  • Statement on how much UNCERTAINTY remains in reconstructions and whether single-reconstruction hypothesis tests are unreliable (quote): NOT ANSWERED / not quotable — this material is past the paywall boundary; no such statement appears on p. 665. (Under ARR a must-quote would become own-words paraphrase, but the statement itself is inaccessible.)
  • Any recommendation (model comparison ER/SYM/ARD-style, reporting uncertainty, not over-interpreting a point reconstruction): NOT ANSWERED — no recommendations appear on p. 665; recommendations, if any, are in the inaccessible body.
  • Symposium context noted (from guidance, not from the source): part of the 1999 Syst Biol ASR symposium; companion pieces Pagel 48:612 and Schultz & Churchill 48:651 are not summarized here.