Marsupial herbivore evolution and the failure of morphological algorithmic phylogenetics

matt.phillips@anu.edu.auCentre for Macroevolution & Macroecology, Research School of Biology, Australian National University

New Guinea forest wallaby

“Phylogenetics is concerned with the problem of reconstructing the past evolutionary history of extant organisms from present day molecular data”

– Phylomania 2010 website

Darwin’s (the 1st ?) phylogenetic tree

Horse evolution & Macroevolutionary theory, e.g. Cope’s rule

Molecular data:

Invaluable for phylogenetic inference

Morphological studies had left us with 1.99 1021

possible relationships among the 29 orders

Molecular studies now leave us with ≈405 possible relationships

Phillips & Penny (2010)

Molecular data:

Is molecular phylogeny above the species level a pursuit of diminishing returns (for theoreticians)?

Remaining uncertainty involves lineage sorting: genomic retroposons better than species-tree methods for assigning ancestry

In either case, the interesting question of individual gene ancestry is defeated by stochastic error

99 mya(83-116HPD)

Cretaceous Period

21kb of nuclear genes for 57 marsupial&placental mammals

BEAST relaxed clock (lognormal dist. branch rates), 13 FR calibration priors – unconstrained, 20 lineages originate in the Cretaceous

Work with Kate Loynes

Loynes & Phillips (in prep)

Rates of DNA substitution (subs/site/Ma) on individual branches

Dark blue: unconstrainedLight blue: 4 placental lineages in CretaceousRed: no placentals cross into Cretaceous

Cretaceous Tertiary

82 mya(73-93HPD)

Cretaceous Period

21kb nuclear genes for 57 marsupial & placentals mammals

BEAST relaxed clock (as previous) – now constraining ≤4 placental lineages to originate in the Cretaceous

92

65

86

86

67

Cut-out section of the placental mammal tree, with putative relationships of fossils from close to or before the K/T boundary

More fossils confidently assigned to branches on the modern tree could immediately solve the K/T boundary problem

And for the overall evolutionary timescale , reduces reliance on assumptions for how rates vary among branches

Kulbeckia

All these fossils may be stem placentals

Meredith et al (MPE, 2009)Foraging height

Ancestral state reconstruction

Arboreality inferred at all deep nodes

Include 5 extinct sub-families

But megafaunal extinction was biased towards large/terrestrial

Palorchestes

Lineage through time analysis

Penny & Phillips (Nature, 2006)

Null hypothesis of constant net diversification (speciation-extinction) is linear

“Pull of the recent” peaksTurnover associated with recent biotic/aboitic events overwrites more ancient signals

Million years ago

Ln

(acc

umul

ated

bra

nchi

ng e

vent

s)

Marsupial divergence times

Hurdles for morphological phylogenetics: progress is being made in some areas

• Long branch attraction – A serious problem when MP is standard

ML models (e.g. Mk or Mkv of Lewis (Syst Biol, 2001) outperform MP

freq

uenc

y

Trait score

State 1 State 2

If no clear pattern or unimodal, exclude or score as constant

Other problems include:

• Developmental correlations (e.g. upper/lower molars)• Outgroup attraction of ecological long branches (e.g. turtles)• Objectivity in character state discrimination

Functional/ecological correlations

Emu

Pigeon

Chicken

Ducks

Not really three characters providing a strong phylogenetic signal Evolutionarily non-independent, associated with parenting strategy

• Babies cute/ugly• Wing development slow/rapid• Leg development rapid/slow

Galah

Marsupials arrived in Australasia 55-70 mya from S.America, via Antarctica

Diprotodontia “Polyprotodontia”

Microbiotheria

Diprotodontia: The most ecologically diverse mammal order

Diprotodon opatum ~2500kg

Thylacoleo carnifex 110kg

Terrestrial herbivores, arboreal insectivores and a multitude of niches in between

Diprotodontia: 10 extant families (≈ 120 species)

Phascolarctidae = koala (Arboreal folivores)

Vombatidae = wombats (Burrowing grazers)

Burramyidae = pygmy possums (Mostly-terrestrial to mostly arboreal gramnivores and generalized omnivores)

Macropodidae = kangaroos and potoroos (Bipedal hopping browsers/grazers and semi-fossorial root/fungi feeders)

Hypsiprymnodontidae = musky rat-kangaroo (Terrestrial, bounding frugivore-omnivore)

Tarsipedidae = honey possum (Arboreal nectivore)

Acrobatidae = feathertail possums (Gliding/arboreal omnivores)Pseudocheiridae = Ringtail possums (Arboreal folivores)

Phalangeridae = Brushtail possums and cuscuces (Scansorial to arboreal frugivores-folivores)

Petauridae = gliders and trioks (Gliding gumnivores and arboreal insectivores)

Diprotodontian consensus phylogeny: Cardillo et al. (J. Zool, 2004)

Vombatidae (wombats)

Phascolarctidae (koala)

Burramyidae (pygmy possums)

Pseudocheiridae (ringtail possums)

Phalangeridae (cuscuses and brushtail possums)

Tarsipedidae (honey possum)

Petauridae (gliders, stripped possums)

Macropodidae (kangaroos and potoroos)

Acrobatidae (feathertail possums)

Vombatiformes

“Cor

e”

Pet

auro

idea

Hypsiprymnodontidae (musky rat-kangaroo)Macropodoidea

Phillips and Pratt (MPE, 2008): mitochondrial (mt) genomes

Beck (J. Mammalogy, 2008): several mt & nuclear genes

Meredith et al. (MPE, 2009): 5-nuclear genes

Vombatidae

Phascolarctidae

Pseudocheiridae

Burramyidae

Phalangeridae

Macropodidae

Acrobatidae

Tarsipedidae

Petauridae

Hypsiprymnodontidae

Molecular “supermatrix”: 26 marsupials 20,654 nucleotides

Complete mt genome protein/RNA coding sequences & 5 nuclear genes (RAG1, BRCA1, IRBP, vWF, APOB)

• Analysed as 13 separately modelled process partitions

• Mitochondrial protein 3rd codons RY-coded to reduce saturation and compositional non-stationarity

wombatskoala

pygmy possums

ringtail possums

honey possumgliders

kangaroos

feathertail possums

musky rat-kangaroo

cuscuses

bandicootsmarsupial mole

dasyurids

All nodes MrBayes BPP = 1.00 and RAxML BP >95%, (except where noted)

0.97 / 72

Diproto

donti

a

“Poly

proto

donti

a”

Previous work on the family-level phylogeny of Diprotodontia

Algorithmic morphology (MP) MRP supertree summary

Informal-comparative morphology MRP supertree

Mt sequence analyses MRP supertree summary

Albumin M’CF Baverstock et al. 1990 (review)

Single nuclear genes MRP supertree summary

DNA hybridization Kirsch et al. 1997 (review)

Algorithmic morphology morphol352 (MP)

Algorithmic morphology morphol352 (ML, Bayesian)

Differences between informal-comparative and algorithmic morphology

Selection criterion

Character analysis

Algorithmic

MP, ML etc.

Homology, otherwise biology-free

Many and varied (inc. bootstrap)

Informal-comparative

vague

Homology, untangling funct/dev correlation form phylogenetic signal

Non-statisticalHypothesis testing

How do these data / methods perform?

One test would be whether or not they reject the molecular consensus - not helpful … Hypothesis testing is difficult with distance methods like DNA hybridization and impossible with informal-comparative morphology

Alternative: Likelihood disadvantage on the 20,654 nucleotide molecular matrix for a fairer comparison of data / methods

Example:

–lnL(consensus) = 121,316.3

–lnL(DNA hybridization tree) = 121,438.2

lnL disadvantage = 121.9

Algorithmic morphology (MP) 690.1

Informal-comparative morphology 71.7

Mt sequence analyses 84.4

Albumin M’CF 182.4

Single nuclear genes 96.0

DNA hybridization 121.9

Algorithmic morphology morphol352 (MP/ML) 594.6

Algorithmic morphology morphol352 (Bayes) 617.5

Likelihood disadvantages

Do the algorithmic analyses just suffer from stochastic blindness?

• Scaled the molecular-dated marsupial tree to the treelength of the morphol352 ML tree

• Simulated 60,000 character “pseudomorphological” dataset, Sim352 in Seq-gen (JC, equivalent to Mk4). 1000 boots, 352 chs

VombatidaePhascolarctidae

Pseudocheiridae

BurramyidaePhalangeridae

Macropodidae

AcrobatidaeTarsipedidaePetauridae

Hypsiprymnodontidae

95100

87

100

7361

5374

*

5 outgroup taxa

Vombatidae

Phascolarctidae

Pseudocheiridae

Burramyidae

Phalangeridae

Macropodidae

Acrobatidae

Tarsipedidae

Petauridae

Molecular consensus

Phascolarctidae

Pseudoch’idae

Phalangeridae

Acrobatidae

Tarsipedidae

Vombatidae

Burramyidae

Macropodidae

Petauridae

Algorithmic morphology

Can we mimic the real morphological data by combining molecular phylogenetic and ecological signals ?

VombatidaePhascolarctidae

Pseudocheiridae

BurramyidaePhalangeridaeMacropodidae

AcrobatidaeTarsipedidaePetauridae

5 outgroup taxa

60,000 characters

Sim352

phylogenetic signal as per the molecular dated tree, scaled to morphol352 treelength

Size Diet

0 = <50g1 = 50-200g2 = 200-800g3 = 800g-3kg4 = 3-12kg5 = >12kg

0 = herb1 = sub-herb2 = omniv3 = sub-carn4 = carn

ordered states

6

4

2

0

0

10

8

3224168

% M

P tr

ee le

ngth

dis

adva

ntag

e

% ecological contribution to MP tree length

Optimum fit to the molecular consensus tree (0% ecological contribution)

Optimum fit to the algorithmic morphology tree

(9% ecol. cont.)

Vombatidae

Phascolarctidae

Pseudocheiridae

Burramyidae

Phalangeridae

Macropodidae

Acrobatidae

Tarsipedidae

Petauridae

Molecular consensus

Phylogenetic randomization test P-value = 0.00016

Phylo-ecol sim

Phascolarctidae

Pseudoch’idae

Phalangeridae

Acrobatidae

Tarsipedidae

Vombatidae

Burramyidae

Macropodidae

Petauridae

Algorithmic morphology

Tempting next move: Reverse engineered phylogeny

If the algorithmic morphology (morphol352) data is effectively 91% phylogenetic signal, 9% ecological signal … what if we subtract the 9% ecological signal from the observed signal?

b. MP on diet+size: 14 steps 26 steps

a. MP on morphol352 741 steps 782 steps

Alg. Morphol. tree “True” tree

c. Ave. over 9% “true” TL 61.8 steps 114.7 steps

c. Rev Eng Phylogeny (a-c) 679.2 steps 667.3 steps

Improvements

Co-inferring the relative weightings of the ecological correlates simultaneously with the relative apparent contributions of phylogenetic and ecological signal

Searching tree space for the reverse engineered phylogeny - current phylogenetic programs are well set up for addition of log-likelihoods (e.g. for partitioned data), but not for subtraction

Molecular tree is employed in the discrimination of apparent phylogenetic and ecological signals - so has some influence on the reverse engineered phylogeny.

However, the ultimate aim here is the placement of fossils. The correction for ecological signal (inferred with extant taxa) can be employed for fossil taxa, independent of their DNA

Eomaia

Microraptor

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Their fossils provide the only direct evidence for answering many key questions in macroecology and macroevolution and for calibrating molecular timescales

>99% of all species are extinct

Acknowledgements

• Kate Loynes (ANU, PhD student)

• Emily Lake (ANU, Honours student)

• Australian Research Council