Does extinction wield an axe or pruning shears? How interactions
between phylogeny and ecology affect patterns of extinction
Walton A. Green, Gene Hunt, Scott L. Wing, and William A. DiMichele
Abstract.?Extinctions are caused by environmental and ecological change but are recognized and
measured in the fossil record by the disappearance of clades or lineages. If the ecological preferences
of lineages or taxa are weakly congruent with their phylogenetic relationships, even large ecological
perturbations are unlikely to drive major clades extinct because the factors that eliminate some species
are unlikely to affect close relatives with different ecological preferences. In contrast, if phylogenetic
relatedness and ecological preferences are congruent, then ecological perturbations can more easily
cause extinctions of large clades. In order to quantify this effect, we used a computer model to
simulate the diversification and extinction of clades based on ecological criteria. By varying the
parameters of the model, we explored (1) the relationship between the extinction probability for a
clade of a given size (number of terminals) and the overall intensity of extinction (the proportion of
the terminals that go extinct), and (2) the congruence between ecological traits of the terminals and
their phylogenetic relationships. Data from two extinctions (planktonic foraminifera at the Eocene/
Oligocene boundary and vascular land plants at the Middle/Late Pennsylvanian boundary) show
phylogenetic clustering of both ecological traits and extinction probability and demonstrate the
interaction of these factors. The disappearance of large clades is observed in the fossil record, but our
model suggests that it is very improbable without both high overall extinction intensities and high
congruence between ecology and phylogeny.
Walton A. Green,* Gene Hunt, Scott L. Wing, and William A. DiMichele. Department of Paleobiology,
National Museum of Natural History, Smithsonian Institution, Post Office Box 37012, MRC 121,
Washington, D.C. 20013-7012. E-mail: wagreen@bricol.net
*Present address: Department of Organismic and Evolutionary Biology, Harvard University, 26 Oxford
Street, Cambridge, Massachusetts 02138
Accepted: 4 June 2010
Comes the blind Fury with th?abhorre`d shears,
And slits the thin spun life.
?Milton Lycidas 75f.
Introduction
Despite extensive study and debate (e.g.,
Raup 1972, 1991, 1993; Raup and Sepkoski
1982; Knoll 1984; Sepkoski 1993; Seilacher
1998; Purvis et al. 2000; Wing 2004; Jablonski
2001, 2004, 2005; Rabosky 2009) there remain
questions about the importance of extinctions
in the history of life compared with other
factors like competition, integration, and
evolutionary innovation. Because historical
events are unique it is often difficult to
establish a causal relationship between a
particular extinction event and its ecological
trigger. Of the ??big five?? mass extinctions
identified by Raup and Sepkoski (1982) only
one, at the Cretaceous/Paleogene boundary,
has a generally accepted cause (Alvarez et al.
1980), and even its evolutionary consequences
remain incompletely understood. For in-
stance, Wilf and Johnson (2004) describe an
extinction of almost 60% of plant morpho-
types in the terrestrial flora of North Dakota
at the Cretaceous/Paleogene boundary. It is
not clear, however, that this local species
extinction is associated with loss of taxa above
the level of genus at a continental scale (Green
and Hickey 2005). Throughout this paper we
use the term ??extinction?? broadly, to include
local extirpations; the significance of any
extinction can be appreciated only at a
specific spatial, temporal, and taxonomic
scale.
In order to avoid the problems of scale
dependence and repeatability, in this paper
we offer an approach based on computer
modeling. Such models have the obvious
disadvantage of lacking realism, but the
compensating advantages of being scale-
independent and easy to generalize and
replicate. As is described below in depth,
our model assumes a strictly dichotomizing
tree of life whose twigs are organisms or
Paleobiology, 37(1), 2011, pp. 72?91
? 2011 The Paleontological Society. All rights reserved. 0094-8373/11/3701?0004/$1.00
clades that live in an environment represent-
ed as a vector space defined by ecological
variables (ecospace). We compare the output
of this model to data from the fossil record of
planktonic foraminifera at the Eocene/Oligo-
cene (E/O) boundary and vascular land
plants at the Middle/Late Pennsylvanian
boundary, which following DiMichele and
Phillips (1996) we will refer to as the
Westphalian/Stephanian (W/S) boundary,
despite difficulties precisely correlating the
W/S boundary with the global timescale
(DiMichele et al. 2009). Our model should be
generalizable to any taxonomic group and to
any scale of analysis, provided a strictly
branching tree does a good job of summariz-
ing historical relationships among taxa.
Typically, the intensity of an extinction is
measured as the proportion of described
species or higher taxa that disappear in a
particular time interval (or metrics derived
from similar quantities; see Foote 2000). There
are no standard criteria for determining when
a mass extinction rises above background
levels, but the ??big five?? mass extinctions
were originally identified exclusively by their
intensities at the family rank (the proportion
of marine families that went extinct in a
geological stage) and are generally accepted
as influential in the marine realm (Raup and
Sepkoski 1982). In contrast, global compila-
tions of the stratigraphic ranges of fossil
plants do not show sharp declines in diversity
like those defining the largest extinctions in
the animal fossil record (Niklas et al. 1980,
1985; Niklas and Tiffney 1994). The apparent
absence of mass extinctions in plants has been
attributed to the resistance of plant individ-
uals and populations to environmental
shocks?individuals can regrow from buried
storage organs after major tissue loss, and
populations can survive bad conditions as
spores or seeds in the soil (Knoll 1984). If the
absence of global mass extinctions in plants
reflects the resistance of individuals and
populations to perturbations, then low ex-
tinction levels should also be observed in
local, high-resolution studies. When the re-
cord is examined at the temporal scale of local
stratigraphic sections and the geographic
scale of a sedimentary basin, however, more
than half the described species or morpho-
types sometimes disappear at a boundary
(e.g., Looy et al. 2001; Wilf and Johnson 2004;
McElwain et al. 2007, 2009).
The apparent conflict between high extinc-
tion levels at local to regional geographic
scales and low extinction levels globally, or
extinctions in some groups but not in others,
can be resolved if the probability of a species
going extinct is uncorrelated with its mem-
bership in higher groups (Wing 2004). Hier-
archically higher groups can persist even
when most of the species in them go extinct.
Extinctions would then be seen only in
taxonomically detailed studies of local sec-
tions, because on larger temporal, spatial, and
taxonomic scales the extinction would be
masked by the rapid appearance during the
recovery of new species similar to or even
indistinguishable from the old ones (Wing
2004).
In this paper, we try to determine the effect
of an extinction on lineages that vary in the
degree to which their phylogeny is congruent
with their ecology. Many studies (e.g., McKin-
ney 1995; DiMichele et al. 2001; Prinzing et al.
2001; Crisp et al. 2009) have pointed out that
clades often show ecological preferences or
??centroids?? in ecospace that inhere through
time. To the best of our knowledge, however,
there have been no attempts to relate ecology-
phylogeny congruence to extinction dynamics
quantitatively.
Figure 1 illustrates the two extreme scenar-
ios, which we are calling the chop and the trim
situations. In chop extinctions (Fig. 1A) close-
ly related species are clustered in ecospace,
and ecologically selective extinctions remove
large branches from the phylogenetic tree. In
trim extinctions (Fig. 1B) closely related spe-
cies are ecologically dissimilar and extinctions
are therefore dispersed across the phyloge-
netic tree.
The models described in the following
section are intended to quantify the impor-
tance of the chop and trim scenarios in
determining the phylogenetic effect of an
extinction event that drives a particular
proportion of species extinct. (Note that the
terms ??terminal?? and ??species?? are used
interchangeably without implying that the
EXTINCTION?S AXE AND SHEARS 73
terminals of a phylogenetic tree have to
represent biological species.) In this paper,
we ask if extinctions have different effects
when individual clades are broadly distribut-
ed ecologically as opposed to being clustered
in ecospace. How important are phylogenetic
relationships in determining survival of line-
ages in ecologically targeted extinctions?
After describing the sensitivity of our
model to various parameters, we examine
two case studies: the extinction of planktonic
foraminifera at the Eocene/Oligocene bound-
ary and the extinction of vascular plants at
the Westphalian/Stephanian boundary. In
both cases extinction propensity appears to
be phylogenetically clustered, as are some
measured traits. This finding has wider
implications both for our evaluation of the
significance of extinction events in the past
and potentially for helping to minimize long-
term consequences of current anthropogenic
extinctions.
Methods
Modeling
In the basic model used in this paper, we
construct a random phylogenetic tree of a
given size using a pure birth model (function
birthdeath.tree in the R package geiger; Har-
mon et al. 2008). We then apply one or more
ecological variables (drawn from a centered
multivariate normal distribution) to the tips
of this tree to give a specified congruence
between the ecological measure and the
relationships shown by the tree. This congru-
ence of ecology with phylogeny is measured
by Pagel?s l (Pagel 1997, 1999; Freckleton et
al. 2002), a parameter that transforms a
phylogenetic tree as follows. The transforma-
tion is performed on a tree?s phylogenetic
variance-covariance matrix, which represents
the statistical dependence among tips of a
phylogenetic hypothesis (Fig. 2). Entries in
the phylogenetic variance-covariance matrix
on the main diagonal represent the distance
between each terminal and the root of the
tree, and off-diagonal elements represent the
shared path-length for each pair of tips (i.e,
the distance between their most recent com-
mon ancestor and the root of the tree). These
FIGURE 1. Schematic relationships between phylogeny
and ecology at an extinction. All scenarios have the same
distribution of terminal taxa in ecospace. A, High
congruence between ecology and phylogeny; extinctions
chop off major branches of the phylogeny because related
species tend to occupy similar niches. B, Low congruence
between ecology and phylogeny; extinctions trim only
small clades because close relatives occupy niches that are
unlikely to be affected by the same environmental
changes. C, The extinction viewed in two-dimensional
ecospace at the time of the extinction. D, The same
extinction event projected onto a single ecological
dimension.
74 WALTON A. GREEN ET AL.
entries are in the same units as the branch
lengths of the tree, usually time or a proxy
thereof. This matrix is called a variance-
covariance matrix because traits evolving
according to a random walk model are
multivariate normal, with variances and
covariances among tips given by this matrix
multiplied by the rate parameter (Pagel 1999;
Freckleton et al. 2002). For a particular
topology, the l transformation multiplies
off-diagonal elements of the phylogenetic
variance-covariance matrix by l, where l
ranges from zero to one. Low values of l
shorten internal tree branches, resulting in a
tree with proportionately more evolutionary
history in the terminal branches and less
history shared among taxa. When l equals
zero, internal branches disappear, leaving a
star phylogeny, in which case there is no
phylogenetic structure to traits and their
evolution (Fig. 2). A tree with l equal to 1 is
unaltered by this transformation.
If we evolve traits as random walks
(Brownian motion) on these transformed
trees, the resulting traits will have high
phylogenetic congruence when l is near 1,
and low congruence when l is close to zero. It
is mathematically possible for l to exceed
one, but such values are not readily interpret-
ed, and commonly only values between zero
and one are considered (Freckleton et al.
2002).
In addition to its phylogenetic location,
each terminal on the tree also has a location
in ecospace whose dimensions are biotic or
abiotic variables along which an environment
can vary (this is a fundamental niche in
ecospace, sensu Hutchinson 1978: p.159).
Note that the ecological variables represented
by the dimensions of our ecospace need not
represent single traits, but can be composite
axes of variation, like principal components,
which come from a rotation and rescaling of
an empirical ecospace of higher dimensional-
ity.
Modeled extinctions are deterministic (non-
probabilistic), are centered in ecospace on the
location of a randomly chosen terminal, and
affect all taxa within a given radius of the
extinction center. The intensity of extinction is
determined by the proportion of tree termi-
FIGURE 2. Schematic examples of a tree, transformed
with different values of l and the resulting variance-
covariance matrices. A, l 5 1.0. B, l 5 0.5. C, l 5 0.0.
EXTINCTION?S AXE AND SHEARS 75
nals that go extinct. So for a tree with number
of terminals nt and extinction intensity xti,
one terminal is chosen at random to go extinct
and then the nt 3 xti other terminals closest
in ecospace also go extinct. The measure of
ecological proximity is the euclidean distance,
which simplifies to the scalar distance (dif-
ference) in the one-dimensional case.
The model also allows an ecospace of any
number of dimensions. For example, if the
single ecological axis is soil wetness, and the
extinction is centered at the wet end of the
spectrum, this may represent climatic drying
in which species requiring wet substrates will
go extinct, whereas species preferring drier
soils go unaffected. A two-dimensional case
might add the effect of soil fertility, so that the
extinction is centered in habitats that have
high soil moisture and low fertility, preferen-
tially affecting plants of oligotrophic wet-
lands. As demonstrated in Figure 1C,D, the
same schematic extinction may be viewed in
two dimensions or projected onto a single axis
of variation.
The following are parameters of the model
that can be independently varied: the number
of tree terminals, nt; the intensity of extinction
(proportion of terminals that go extinct), xti;
the congruence of the ecological signal with
phylogeny, l; the ecological dimensionality,
dims; and the number of replicates or different
random tree topologies, n. We measure the
response to these simulated extinction events
by recording the mean size of extinct clades
and the proportion of surviving clades of a
given size, s, that go extinct. After extinction,
we identify the number and size of clades that
are entirely eliminated by this extinction of
terminals and are not subclades of larger
clades that also go extinct. All programming
is done in R (R Development Core Team 2008)
and the scripts and results of model runs are
available as supplementary data (http://dx.
doi.org/10.1666/09078.s1 and http://dx.doi.
org/10.1666/09078.s2).
The mean size of extinct clades is calculated
from the distribution of extinct clade sizes,
which is heavily right-skewed. In most cases,
more than half the extinct clades are single-
tons, so the median extinct clade size is
seldom larger than 1. The mean extinct clade
size is also a problematic response variable
because it is heavily dependent on the largest
clade to go extinct. Hence in addition to
showing the variation in mean clade size
(Fig. 3), we also illustrate the proportion of
clades of a given size that go extinct (Fig. 4).
This probability of extinction of a clade with a
particular number of terminals seems to be
more informative about extinction effects,
though when necessary we rely on the mean
extinct clade size as a point estimate for
extinct clade size.
Case Studies
In addition to exploring the sensitivity of
the model to variation in its parameters, we
compare its output with two real extinctions
for which the phylogenetic, stratigraphic, and
ecological data needed to test the model are
available. We found relatively few extinctions
for which all three types of data had been
recorded at the species level. Our conclusions
are naturally dependent on the data being
reanalyzed, but small errors (for instance of
taxonomic lumping or splitting) inherited
from the cited sources should affect our
results only incrementally without biasing
our overall conclusions.
Eocene/Oligocene Planktonic Foraminifera.?
The E/O boundary marks a significant
turnover in planktonic foraminifer assem-
blages, including the extinction of an entire
family, the Hantkeninidae (Pearson et al.
2008). Extinctions in these and other taxa are
linked to changes in ocean circulation and
climatic cooling (Wade and Pearson 2008). In
our analyses, we consider the extinction of
all species of planktonic foraminifera extant
during the last biozone of the Eocene (E16,
duration ,600 Kyr). We scored species as
victims if they did not survive into the
Oligocene according to range charts in Pear-
son et al. (2006), updated to reflect the
recently reported range extension of Subbotina
hagni into the Oligocene (Wade and Pearson
2008). Because the survivorship of Acarinina
echinata through this interval is questionable,
this species was omitted from the analysis.
A species-level phylogenetic tree was as-
sembled for species extant in the last interval
of the Eocene using the phylogenetic hypoth-
76 WALTON A. GREEN ET AL.
eses of Pearson et al. (2006), supplemented
where necessary by inferred relationships
among Paleocene taxa (Olsson et al. 1999).
Branch lengths were set according to the
divergence dates reported in these references.
In a few cases, there was reported uncertainty
in some of the inferred relationships. For
purposes of analysis, the following relation-
ships were assumed: Dentoglobigerina was
considered to descend from the acarininids;
the clade composed of Chilguembelina +
Streptochilus was considered to descend from
the Cretaceous/Paleogene survivor Guembeli-
tria via the intermediate Woodringina; and
Cassigerinella + Tenuitella was interpreted as
having descended from an independent evo-
lution of the planktonic habit from a benthic
ancestor. The divergence date of this last
group is not known, but was set to be mid-
Jurassic, which is the interval in which
planktonic foraminifera first appear in the
fossil record.
Body sizes were taken from Pearson et al.
(2006), who list maximum dimensions, usu-
ally for type specimens. When sizes were
reported as ranges, the range midpoint was
used. Any bias introduced by using the mid-
range as a point estimate of a distribution of
body sizes should be small relative to the six-
fold range of body sizes recorded (0.11?
0.69 mm), uncorrelated with phylogeny, and
have the effect of reducing somewhat the
concordance between body size similarity
and phylogenetic relatedness. Because body
size evolution is generally best considered on
a proportional rather than absolute scale
(Foote 1991), size measures were log-trans-
formed before analysis.
Westphalian/Stephanian Vascular Plants.?
The W/S boundary (Middle/Late Pennsylva-
nian transition) marks a change in the
composition and diversity of lowland wet-
land vegetation in North America and Eu-
rope, particularly in peat-forming environ-
ments (Phillips et al. 1974; Peppers 1996;
DiMichele and Phillips 1996). Middle Penn-
sylvanian peat swamps were dominated by
large lycopsids (Lepidophloios, Diaphoroden-
dron, Synchysidendron, Paralycopodites, Sigil-
laria) with subdominant tree ferns and pteri-
dosperms. These assemblages were even-
tually replaced by communities dominated
by tree ferns and pteridosperms, with a much
reduced array of locally abundant lycopsid
trees (mainly Sigillaria) and smaller lycopsids
(Chaloneria). Similar patterns are found in
clastic adpression floras (Pfefferkorn and
Thomson 1982; Blake et al. 1999; Dimitrova
et al. 2005; Dimitrova and Cleal 2007). In both
settings, tree ferns began to rise in importance
before the W/S boundary.
The data used in this paper come from coal
balls, masses of permineralized peat found in
coal beds that preserve the original vegetation
of the peat swamp (DeMaris 2000). Coal balls
were collected from coals of both late Middle
Pennsylvanian and early Late Pennsylvanian
age, an interval of several million years;
methods of collection and quantification are
described by Phillips et al. (1977).
Of the 85 taxa identified in Westphalian
and Stephanian coals, 54 were observed in the
latest zone of the Westphalian (Westphalian
D), of which 35 went extinct at the W/S
boundary. There are no published phyloge-
nies of Pennsylvanian plants resolved to the
level of detail needed for this study. There-
fore, a phylogeny was constructed by com-
bining published phylogenies or schemes of
evolutionary descent with various degrees of
resolution into a basic phylogenetic frame-
work for the major plant groups. The intent
was to create a phylogeny not for all known
Carboniferous wetland plants, but only for
those taxa present in the study interval. For
many of these groups, whole plants are not
known. In those instances, one organ was
used as proxy for the parent plant (e.g., seeds
for pteridosperms). The most reliable sources
are reconstructed whole-plant phylogenies
(e.g., representative tree lycopsids [Bateman
et al. 1992]; marattialean tree ferns [Lesni-
kowska 1989]; cordaitaleans [Trivett 1992;
Costanza 1985]). In groups where only a few
whole plants have been reconstructed, whole-
plant attributes were inferred from organ
phylogenies (e.g., medullosan pteridosperms
based on seeds [Taylor 1965]). In other
instances, a phylogeny was compiled from
one or more partial phylogenies based on
single organs used as proxies for the whole
plant (e.g., inclusion of Albertlongia incostata,
EXTINCTION?S AXE AND SHEARS 77
Taylor 1967, in the larger phylogeny of
Pachytesta, Taylor 1965). Relationships were
based on published interpretations of species
or genera within various clades and among
related clades (e.g., filicalean ferns [Phillips
1974; Rothwell 1991; Galtier and Phillips 1996;
Phillips and Galtier 2005]). Finally, relation-
ships were sometimes inferred from a combi-
nation of stratigraphic and morphological
data, in the absence of a relationship scheme
(e.g., Heterangium [Pigg et al. 1987]). Our
phylogeny includes a few taxa for which
placement was highly uncertain (e.g., Stellas-
tellara: DiMichele and Phillips, 1979), but in
no case was without some rationale. Because
of the large number of papers and variety of
data that had to be synthesized in this
process, full documentation is not possible,
but no taxon was included without consulta-
tion of a published phylogeny, a published
discussion of its possible evolutionary rela-
tionships, or a published description of its
morphology adequate to support reasonable
speculation about its relationships.
After arranging the 54 taxa present in the
latest Westphalian in a phylogenetic tree, we
constrained the branch lengths of the phylog-
eny by known stratigraphic ranges where
possible, and recorded two morphological
variables of ecological importance for each
terminal: disseminule size (megaspore or
seed volume in cubic millimeters, log-trans-
formed for analysis) and growth form (scored
semiquantitatively as herb, shrub, small tree,
large tree). Both stature and disseminule size
are considered to be related to plant ecolog-
ical strategies. Plant stature relates to overall
adult body size, which is correlated with
environmental features like site stability and
resource availability, as well as the ability to
compete with other plants for light (Grime
2002). The ecological significance of dissem-
inule size is strongly correlated with dispersal
mode and the light environment of seedling
establishment, among other factors (Moles et
al. 2005). Seed size and adult height are two of
the three dimensions proposed by Westoby
(1998) for a comprehensive description of
plant ecological strategy space. Homosporous
plants lacking macroscopic disseminules
were assigned a small arbitrary disseminule
size (equal to 90% of the size of the smallest
macrospore size), and stem or foliage taxa
whose associated disseminules are not known
were scored as missing data.
Analysis.?For each case study, we mea-
sured the phylogenetic signal in each ecolog-
ical variable using Pagel?s l (estimated with
the function fitContinuous in the R package
geiger [Harmon et al. 2008]). For comparison,
we also assessed phylogenetic signal using
Blomberg?s K statistic (Bloomberg and Gar-
land 2002; Blomberg et al. 2003), using the
functions phylosignal and Kcalc in the R
package picante (Kembel et al. 2008). l and
K as well as other descriptive statistics are
given for both groups in Table 1. The metrics
l and K are designed to measure phylogenetic
signal in continuously varying characters
like disseminule or body size. As such, they
cannot be used directly to measure the
phylogenetic signal in discrete traits like
extinction. Although lineage extinction is
clearly a binary outcome, it can reasonably
be related to an underlying probability or
susceptibility to extinction, which can be seen
as varying continuously, like the threshold
model for quantitative phenotypic traits (Fel-
senstein 2005).
Thus, if we treat ??extinction susceptibility??
as a continuous variable that is characteristic
of a clade, we can estimate its l using
simulations similar to those described earlier
but run with constraints, using the known
phylogenies of the groups and observed
extinction intensities as data (0.37 5 17
extinctions among 46 foraminifer species;
0.65 5 35 extinctions among 54 vascular plant
species). For a wide range of l transforma-
tions of the tree, we simulated the evolution
of extinction susceptibility according to a
random walk model. Assuming that the
species with the highest susceptibilities were
the ones that went extinct, we recorded the
simulated distribution of extinct clade sizes in
each simulated run and computed the pro-
portion of runs in which the simulated
distribution exactly matched the observed
distribution of extinct clade sizes (see Ta-
ble 1). For instance, in the E/O case study, we
run our model with a 46-terminal tree and
extinction intensity equal to 0.37 (the values
78 WALTON A. GREEN ET AL.
obtained from the observed foraminifer data).
By simulating many trees at different values
of l and recording how frequently the
observed pattern of extinctions occurs (one
clade with seven species, one clade with three
species, and seven clades composed of a
single species each), we can generate a
likelihood curve for l?a plot of the proba-
bility of observing particular data values,
given a model. Values of l are favored to
the extent that they commonly produce the
observed data in the form of the distribution
of extinct clade sizes. The maximum of this
plotted curve is then the maximum likelihood
estimate for l.
Our approach to estimating l for extinction
susceptibility uses simulation to approximate
the likelihood function because no direct
analytical solution is available. Logistic regres-
sion is often used to estimate the relationship
between continuous factors and the binary
state of survival/extinction (e.g., Payne and
Finnegan 2007; Finnegan et al. 2008; Wang and
Bush 2008), but this approach assumes that
taxon values are independent, and therefore it
cannot be used to assess the amount of
phylogenetic signal, which is a measure of
the statistical dependence among values as-
signed to the terminals of a tree. General
estimating equations have been applied to
account for these dependencies in the analysis
of discrete traits (Paradis and Claude 2002),
but this approach does not rely on a likelihood
foundation, and so l cannot be estimated
using a maximum likelihood approach. In the
future, suitable modification of the general
estimating equation approach may allow a
more direct analytical estimation of l.
Results
Modeling
The basic model response is shown in
Figure 3A,D as a landscape in which elevation
(shading) represents the mean size of extinct
clades. It can be seen that extinct clade size
increases both with species-level extinction
rate and with higher phylogenetic congruence
of the ecological traits. Note that large clades
seldom go extinct except when both species-
level extinction and phylogenetic congruence
of the ecological traits are high. With a 1000-
terminal tree, l 5 0.9 (phylogenetically clus-
tered ecological preferences), and xti5 0.9 (900
of the 1000 species go extinct), the mean extinct
clade size is still smaller than seven terminals
per extinct clade.
Because the mean extinct clade size is very
sensitive to the high outliers in a right-skewed
distribution of extinct clade sizes, we also
show probabilities of taxa of given sizes going
extinct (Fig. 4). The top panel of Figure 4
shows the probability of going extinct for
clades of size 1. Because the extinction of a
single species does not depend on the fates of
TABLE 1. Range estimates for values are given in parentheses: for xti, 61 binomial standard deviation, and for l, 61
log likelihood unit. Estimated mean extinct clade size is produced by runs of our model with parameters set to their
observed values.
E/O Forams E/O Forams W/S Plants W/S Plants W/S Plants
Number of terminals
(nt) 46 54
Extinction intensity
(xti) 0.37 (0.30,0.44) 0.65 (0.58,0.71)
Extinct clade sizes 7,3,1,1,1,1,1,1,1 10,8,3,2,2,1,1,1,1,1,1,1,1,1,1
Observed mean
extinct clade size 1.89 2.34
Mean extinct clade
size, l 5 0 1.21 (1.1,1.3) 1.47 (1.3,1.6)
Ecological character size* extinction
susceptibility
stature seed size{ extinction
susceptibility
Blomberg?s K 0.38 0.89 0.54
Pagel?s l 0.97 0.97 (0.78,1) 1 0.99 0.7 (0.3,1)
Estimated mean
extinct clade size 1.94 3.01 2.91
*Maximum dimension in mm.
{Volume in mm3.
EXTINCTION?S AXE AND SHEARS 79
other lineages, it is completely insensitive to
l. As the size of clades increases, moving
down through Figure 4, we see an increased
dependence on l. This substantiates the well-
known observation (Raup and Sepkoski 1982;
Sepkoski 1993; Janevski and Baumiller 2009)
that extinctions look less severe at higher
taxonomic levels. An extinction that elimi-
nates a given proportion of families will
eliminate a higher proportion of genera and
even higher proportion of species.
The probability of a clade of a given size
going extinct must of course also depend on
the total number of species available. If there
were only 1000 species in existence, then the
probability of 1001 going extinct would be
zero, independent of extinction intensity.
Empirically, however, this effect only seems
to affect the mean extinct clade size at high l
(Fig. 5). With realistic extinction intensities
and values of l below 0.9, the probability of a
large clade going extinct is virtually constant
above an absolute tree size of about 1000 taxa.
Even with l 5 1, the effect of overall tree size
on extinct clade size is weak (Fig. 5).
Another factor that might influence the
mean size of extinct clades is ecological
dimensionality. Ecospaces (or ecological strat-
egy spaces or morphospaces) are often de-
fined in many dimensions, with an axis for
each measurable ecological attribute, which
can include measures of biogeographic dis-
tribution or climatic tolerance. These are often
then reduced to a smaller number of axes that
are thought to capture the major features of
ecological diversity (Bambach 1983; Westoby
1998; Grime 2002). As can be seen from
Figure 6, extinct clade size is independent of
the number of ecological dimensions used in
our simulations over a wide range of l and
xti.
Because the ecospace offers only a simpli-
fied model of a biological community, it is
difficult to interpret this as support for the
idea that extinction risk is independent of
niche differentiation. In our model, however,
FIGURE 3. l-extinction intensity landscape. A and D show the same data (a 9 3 11 pixel landscape in which the
elevation or gray scale shows the mean extinct clade size for a given l and extinction intensity. B and C are higher
resolution transects showing three profiles of the same landscape from each margin.
80 WALTON A. GREEN ET AL.
it is clear that the dimensionality of the
modeled ecospace does not affect the way
that ecological congruence interacts with
extinction intensity to determine the sizes of
extinct clades.
Case Studies
Figure 7 gives curves showing the propor-
tions of model runs in which the observed
pattern of extinctions was observed for each
case study. Because these curves are analo-
gous to log-likelihood functions, the modes of
the curves can be interpreted as maximum
likelihood estimates of the parameter l for the
hypothetical continuous variable measuring
extinction susceptibility of a clade.
The statistics describing each case study are
given in Table 1. Blomberg?s K and Pagel?s l
both indicate high levels of phylogenetic
signal in foraminifer body size, plant stature,
and plant disseminule size.
The inferred phylogenetic relationships
among end-Eocene foraminifera lineages are
shown in Figure 8. Although it is not entirely
determined by phylogeny, body size is mostly
consistent among close relatives. There are
relatively small-bodied clades such as Tenui-
tella + Cassigerinella, and Pseudohastigerina,
and large-bodied clades such as the hantke-
ninids. The maximum likelihood estimate of l
for body size is quite high (l 5 0.97), very
close to the random walk prediction (l 5 1).
This pattern is consistent with previous
studies that document generally high phylo-
genetic signal for body size (Freckleton et al.
2002; Blomberg et al. 2003).
Figures 8 and 9 show the phylogeny and
survivorship at each extinction. The E/O
extinctions are not randomly distributed with
respect to phylogeny (Fig. 8). For example,
the hantkeninids, a clade of seven species,
goes entirely extinct. With independent bino-
mial extinction intensities of 0.37, a clade of
seven species is expected to go extinct only
about one time in ten thousand (p 5 0.00095),
and thus the extinction of a clade this large
FIGURE 4. Each panel represents a landscape of extinction
probabilities like the landscape of mean extinct clade
sizes in Figure 3A. The clade size considered increases
down through the four panels, showing how the
sensitivity to l increases as the clade size considered
r
increases. All results are for simulations with one
ecological dimension. Shading as in Figure 3A.
EXTINCTION?S AXE AND SHEARS 81
would not be likely unless extinction were
phylogenetically clumped. This reasoning is
confirmed by the simulation-based estimate
of l, which indicates very high phylogenetic
signal in extinction susceptibility (l 5 0.97).
The 95% confidence limits of this estimate
(taken as two units of log-likelihood) are quite
broad, encompassing l values as low as 0.62.
FIGURE 5. Sensitivity of mean extinct clade size to number of terminals. Each plotted point represents a synthesized
phylogeny. At small tree sizes there is a great deal of scatter, so points are only shown for the solid line (l 5 0.5). The
points from other values of l are summarized by dotted locally weighted regression (lowess) curves (Cleveland 1979).
Top: change in mean extinct clade size for 30 trees as total tree size increases for different levels of l. Bottom: same
representation, but varying extinction intensity (xti) instead of l. The logistic shape of the curves shows how the model
stabilizes at relatively small tree sizes except when l is very high.
82 WALTON A. GREEN ET AL.
Nevertheless, even this lower limit represents
substantial phylogenetic signal. Results from
the W/S extinction of vascular land plants
(Fig. 9) are similar: an estimated l of 0.7
suggests lower l for extinction susceptibility
than in the foraminifer case, but because of
the large error in the estimate (not statistically
different from 1; with lower 68% confidence
limit of 0.3) the two cases are not statistically
distinguishable.
The phylogenetic signal in the plant stature
and disseminule size measurements is visu-
ally apparent when plotted on the phylogeny
(Fig. 9). Because there are two measurable
FIGURE 6. Sensitivity to dimensionality, shown as in Figure 5. The change in mean extinct clade size among 30 trees as
ecospace dimensionality increases for different levels of l and xti. Each plotted point represents a synthesized
phylogeny and points are shown only for the solid l 5 0.5 (top) and xti 5 0.5 (bottom) lowess curves. The flat lines
indicate no response in the model to changes in dimensionality.
EXTINCTION?S AXE AND SHEARS 83
FIGURE 7. Maximum likelihood estimation of l for extinction susceptibility; solid lines are model output and dotted
lines are lowess smoothed curves of data from simulations. Interpreting each curve as a log-likelihood function, the
labeled modes are interpreted as maximum likelihood estimates for l.
FIGURE 8. Phylogeny of Eocene planktonic foraminifera. Symbol size is proportional to body size and a filled black
circle indicates a victim of the Eocene/Oligocene extinction.
84 WALTON A. GREEN ET AL.
ecological variables, the extinctions can also
be plotted in a two-dimensional ecospace
(Fig. 10). The W/S extinction is clearly not
spherical and determinate in ecospace like the
hypothetical extinction shown in Figure 1.
The propensity for plants of larger stature to
go extinct is also obscured by the granularity
of the stature data: because the variable takes
only four values (herb, shrub, small tree, large
tree), it is not immediately obvious that
survival at the W/S boundary is selective
with respect to stature (though it is; see
Fig. 11). When considering all 54 taxa, there
is no significant correlation between stature
and disseminule size, but note that there is a
weakly significant correlation between the
two variables in the sub-population of Ste-
phanian survivors (Pearson?s two-sided cor-
relation test; n 5 16, p 5 0.13, r2 5 0.16). This
appears to be because all of the large (.shrub
size) free-sporing plants went extinct at the
W/S boundary, which could suggest a tran-
sition to more shaded habitats in the Stepha-
nian (leading to direct selection against taxa
with small disseminules) or could be an
indirect result of the progressive increase in
tree-fern dominance through the Middle
Pennsylvanian.
This was a time of glacial-interglacial
oscillations during which the Tropics re-
sponded by drying out between periods of
peat formation. The wet-dry oscillations
became progressively, if irregularly, more
severe during the later Middle Pennsylva-
nian. During the drier periods, the wet flora
presumably withdrew to refugia including
channel bottoms and wet floodplains. As
drought severity increased, culminating in a
particularly severe wetland constriction at the
W/S boundary, tree ferns appear to have
FIGURE 9. Phylogeny of vascular land plants from the Westphalian D. Symbol sizes are proportional to disseminule
size and stature, and filled black circles indicate victims of the W/S extinction.
EXTINCTION?S AXE AND SHEARS 85
been favored increasingly with the return of
widespread wet conditions. The selectivity of
the extinction with respect to size is probably
an indirect effect due to localized environ-
mental change and taxonomic displacement
(DiMichele and Phillips 1996; DiMichele et al.
2009).
A non-phylogenetic analysis shows signif-
icant differences between survivors and vic-
tims in the cases of foraminifer size and plant
stature (Fig. 11). Surviving plants did not,
however, have significantly different dissem-
inule sizes from those that went extinct. So of
the two ecological variables measured for
W/S plants, both have very high congruence
with phylogeny, but only one has signifi-
cantly different values among survivors and
victims. If this pattern were found more gene-
rally, it would suggest that ecological extinc-
tions primarily work on lineages through
the congruence of ecological characters with
phylogeny.
Discussion
There is a long history in paleontology of
investigating the selectivity of extinctions
with respect to traits like body size. Many of
the traits shown to be selective for some
extinctions have also been demonstrated to
have phylogenetic signal or congruence: for
example, geographic range size (Jablonski
1987; Jablonski and Hunt 2006), ecology or
life history (Owens and Bennett 2000), and
body size or other morphological features
(Freckleton et al. 2002; Blomberg et al. 2003).
This suggests that extinction susceptibility
itself is likely to be phylogenetically inherited,
a phenomenon that would manifest on trees
as phylogenetically clumped extinctions, and
in traditional taxonomies as systematic vari-
ation in extinction rates among higher taxa.
Prior examination of the phylogenetic
structure of extinctions has often relied on
taxonomy to test for clumping of extinction
FIGURE 10. The W/S extinction in ecospace; analogous to the schematic diagrams in Figure 1C,D. Each point
represents a terminal taxon; solid diamonds are taxa that went extinct at the boundary, unfilled survived. Top: two-
dimensional plot with dimensions of plant stature (four-state), and disseminule volume. Bottom: points projected onto
the single dimension of disseminule volume. All points are moved by a small random amount to reduce over-plotting
of symbols.
86 WALTON A. GREEN ET AL.
risk, asking, for instance, if endangered or
extinct species are clumped within genera
(e.g., Russell et al. 1998; Lockwood et al. 2002;
Bielby et al. 2006; Janevski and Baumiller
2009). In this paper our approach is indepen-
dent of rank-based Linnean taxonomy, and
estimates congruence of ecology and phylog-
eny directly from phylogenies. Research on
living vertebrates has found good evidence
that extinction risk can be clustered taxonom-
ically and phylogenetically (Schwartz and
Simberloff 2001; Purvis 2008), and the same
has been suspected for paleontological extinc-
tions (e.g., Raup 1991; McKinney 1995).
Recently, Roy et al. (2009) demonstrated
taxonomic and phylogenetic clustering in
geological extinctions of bivalve genera. The
case studies described here, E/O planktonic
foraminifera and W/S land plants, are con-
sistent with this previous work. Extinctions in
both systems are phylogenetically clumped,
and in each case, at least one readily
measured morphological variable also shows
significant phylogenetic signal. Given these
results, and the frequency with which biolog-
ically important morphological, ecological,
and life-history traits show phylogenetic
structure (Freckelton et al. 2002; Blomberg et
al. 2003), this pattern of phylogenetically
clustered extinctions may be a very general
property of life.
Roy et al. (2009) report very high phyloge-
netic selectivity of extinction during the mass
extinction at the end of the Cretaceous, even
though this interval has consistently shown
less selectivity with respect to features (e.g.,
geographic range, larval mode) than earlier
background intervals (Jablonski 1986). In
general, selectivity can go undetected because
of low statistical power, or it may be masked
by differences among clades (Smith and Roy
2006; Purvis 2008). Alternatively, phylogenet-
ic selectivity without trait selectivity can also
result from extinctions that are selective with
respect to unmeasured (latent) variables that
are themselves phylogenetically congruent.
Selectivity in extinctions may therefore be
even more common than would be surmised
FIGURE 11. Boxplots of survivors versus victims for each ecological measurement?extinction pair. Medians marked by
solid points superimposed on boxplots; gray points are actual measurements. A, Eocene/Oligocene foraminifera by
test size. B, Westphalian/Stephanian plants by disseminule size; C. Westphalian/Stephanian plants by stature.
EXTINCTION?S AXE AND SHEARS 87
from studies that assess the relatively few
traits that paleontologists can measure across
many taxa.
The degree of phylogenetic selectivity can
have important effects on the macroevolu-
tionary impact of extinctions. Truly random
extinctions, as in Raup?s (1991) ??field of
bullets,?? remove a surprisingly small fraction
of total evolutionary history, even when
extinction is quite severe (Nee and May
1997), and are not expected to drive large
clades (or higher taxa) extinct. Our simulation
results support the intuition that clustered
ecological traits exacerbate the effects of
extinctions. When l is low, even a numeri-
cally intense extinction tends only to trim
dispersed tips from trees; when high, whole
limbs can sometimes be chopped off. Though
we approached the problem from a method-
ologically different perspective from Janevski
and Baumiller (2009), whose examination of
fossil occurrence data is taxonomically based
and reliant on data analysis instead of
computer modeling, we have come to a
similar conclusion. Neither intense extinction
nor strong congruence alone is sufficient for
removing large, diverse clades; both are
necessary.
Although our model does not explicitly
examine the aftermath of an extinction, it is
reasonable to infer that trim-type extinctions
will leave smaller gaps in ecospace, and that
recovery (diversification into emptied niches)
will be more rapid and complete than when
large gaps are left in ecospace by chop-type
removal of large clades. The distribution of
gaps and survivors also may have important
effects on the nature of morphological chang-
es that accompany the recovery of pre-
extinction diversity (Erwin et al. 1987).
Most observed extinctions have intensities
such that the effect of the events will depend
on l, so if plant groups have (on average)
lower l than animal groups, this would
provide a plausible explanation for the fact
that mass extinctions are less frequently
observed in plant groups. Although sufficient
data are not yet available to test this directly,
as noted by Wing (2004), it may be possible in
the future to compare this theory with the
alternative idea that properties of individual
plants account for the lack of major extinc-
tions.
Like all models, ours has necessitated some
simplifications of reality: for instance, we use
a birth-only model, so extinctions come in a
single event rather than continuously through
time. (This is an extreme pulsed model sensu
Foote 2005.) Extinction in our model is
deterministic rather than probabilistic and
has only one ecological focus, whereas real
extinctions may have multiple locations in
ecospace (or a single non-spherical shape)
and are likely to be probabilistic. Our reliance
on simulation cannot give the same kind of
mathematical insight that might be provided
by an analytical model but is a pragmatic
choice in a situation where no analytical
solution is available.
We also rely heavily on Pagel?s l as a
measure of phylogenetic signal. Calculating l
requires measuring an ecological or morpho-
logical variable for all terminals, so l cannot
be calculated for groups so divergent that
they lack comparable features. The choice of
characters may also be biased by practical
constraints and the presence of homologies.
Therefore, because comparable, ecologically
significant features cannot be measured for all
organisms, the true distribution in nature of
congruence between phylogeny and ecology
may be underestimated by l values calculat-
ed for particular groups of organisms. The
above difficulties aside, correspondence be-
tween modeled and measured phylogenetic
signal in susceptibility to extinction suggests
that our model is capturing some essentials of
the dynamics that are observed in the fossil
record. The results of our modeling substan-
tiate our intuition that mass extinctions
require both high species-level extinctions
and high congruence between phylogenetic
position and location in ecospace.
As the scale of current anthropogenic
extinctions becomes apparent (Butchart et al.
2005; Willis et al. 2008), it may be useful to
apply this result to the anthropogenic extinc-
tions we are now observing. Using the
extinction landscape shown in Figure 3, we
could predict surviving clade size in any
extant clade with known l and extinction
rate. Although the scope of this paper does
88 WALTON A. GREEN ET AL.
not permit an application of our model to
extant groups or full consideration of modern
conservation questions, it is easy to imagine
how such a consideration might help guide
conservation efforts toward groups whose
susceptibility to extinction is particularly
phylogenetically dependent (high l). Like
taxa with restricted geographic ranges or
low population densities, clades with greater
congruence between phylogeny and ecology
may be particularly sensitive to extinction.
Conclusions
1. It is very difficult to drive large clades
extinct. The two most important variables
controlling extinct clade size are the extinc-
tion intensity (the probability of a given
terminal going extinct), and the phyloge-
netic dependence of extinction susceptibil-
ity or ecological-phylogenetic congruence
(measured with Pagel?s l ). The main result
of our analyses is a documentation of the
mutual reinforcement of these two factors
and a quantification of their interaction and
importance. Large clades (e.g.,.20 species)
are likely to be lost at an extinction only
when the overall extinction rate is greater
than 0.8 and the congruence of ecology
with phylogeny yields a l of greater than
0.6. This puts observed species extinction
rates in the fossil record into perspective. A
50% species extinction may sound dramat-
ic, but if ecology-phylogeny congruence is
low, then it may be necessary to count even
extinctions of this magnitude as merely
background?the cost of evolving in a
cruel world. Note that this is consistent
with the generally accepted consensus that
the vast majority of species that ever
evolved have gone extinct (Raup 1986).
When the sum of extinction intensity and l
is less than 1 (e.g., extinction intensity 0.6,
l 5 0.4), only single terminals are likely to
go extinct.
2. Ecospace dimensionality appears to have
no effect on extinct clade size in our
simulations. Above a tree size of about
1000 terminals, the effect of tree size on
extinct clade size is weak and detectable
only when l is greater than about 0.9.
3. The E/O extinction of planktonic forami-
nifera was selective with respect to body
size; the North American W/S vascular
plant extinction was selective with respect
to plant stature, but not to disseminule
size. All traits also showed substantial
congruence with phylogenetic relation-
ships. In both case studies, the estimated
congruence between phylogeny and ex-
tinction susceptibility was significantly
greater than would be expected if the
terminals were located in ecospace at
random with respect to their evolutionary
relationships.
Acknowledgment
We would like to thank B. Huber for
assistance with compiling the phylogeny of
foraminifera; N. Atkins, K. Black, A. Janevski,
and M. McKinney for helpful comments on
the manuscript; and the Smithsonian Institu-
tion for financial support. This is publication
number 228 of the Smithsonian Evolution of
Terrestrial Ecosystems Program.
Literature Cited
Alvarez, L. W., W. Alvarez, F. Asaro, and H. V. Michel. 1980.
Extraterrestrial cause for the Cretaceous-Tertiary extinction.
Science 208:1095?1108.
Bambach, R. K. 1983. Ecospace utilization and guilds in marine
communities through the Phanerozoic. Pp. 719?746 in M.
Tevesz and P. McCall, eds. Biotic interactions in recent and
fossil benthic communities. Plenum, New York.
Bateman, R. M., W. A. DiMichele, and D. A. Willard. 1992.
Experimental cladistic analysis of anatomically preserved
arborescent lycopsids from the Carboniferous of Euramerica:
an essay on paleobotanical phylogenetics. Annals of the
Missouri Botanical Garden 79:500?559.
Bielby, J., A. A. Cunningham, and A. Purvis. 2006. Taxonomic
selectivity in amphibians: ignorance, geography or biology?
Animal Conservation 9:135?143.
Blake B. M. Jr, A. T. Cross, C. F. Eble, W. H. Gillespie, and H. W.
Pfefferkorn. 1999. Selected plant megafossils from the Carbon-
iferous of the Appalachian Region, Eastern United States:
geographic and stratigraphic distribution. Pp. 259?335 in L. V.
Hillis et al., eds. Carboniferous and Permian of the World.
Canadian Society of Petroleum Geologists Memoir 19.
Blomberg, S. P., and T. Garland Jr. 2002. Tempo and mode in
evolution: phylogenetic inertia, adaptation and comparative
methods. Journal of Evolutionary Biology 15:899?910.
Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for
phylogenetic signal in comparative data: behavioral traits are
more labile. Evolution 57:717?745.
Butchart, S. H. M, A. J. Stattersfield, J. Baillie, L. A. Bennun, S. N.
Stuart, H. R. Akc?akaya, C. Hilton-Taylor and G. M. Mace. 2005.
Using Red List Indices to measure progress towards the 2010
target and beyond. Philosophical Transactions of the Royal
Society of London B 360:255?268.
EXTINCTION?S AXE AND SHEARS 89
Cleveland, W. S. 1979. Robust locally weighted regression and
smoothing scatterplots. Journal of the American Statistical
Association 74:829?836.
Costanza, S. H. 1985. Pennsylvanioxylon of Middle and Upper
Pennsylvanian coals from the Illinois Basin and its comparison
with Mesoxylon. Palaeontographica, Abteilung B 197: 81?121.
Crisp, M. D., M. T. Arroyo, L. G. Cook, M. A. Gandolfo, G. J.
Jordan, M. S. McGlone, P. H. Weston, M. Westoby, P. Wilf, and
H. P. Linder. 2009. Phylogenetic biome conservatism on a
global scale. Nature 458:754?756.
DeMaris, P. J. 2000. Formation and distribution of coal balls in the
Herrin coal (Pennsylvanian), Franklin County, Illinois Basin,
USA. Journal of the Geological Society, London 157:221?228.
DiMichele, W. A., and T. L. Phillips. 1979. Stelastellara Baxter, axes
of questionable gymnosperm affinity with unusual habit?
Middle Pennsylvanian. Review of Palaeobotany and Palynol-
ogy 27:103?117.
???. 1996. Climate change, plant extinctions and vegetational
recovery during the Middle-Late Pennsylvanian transition: the
case of tropical peat-forming environments in North America.
Pp. 201?221 in M. B. Hart, ed. Biotic recovery from mass
extinction events. Geological Society of London Special
Publication 102.
DiMichele, W. A., W. E. Stein, and R. M. Bateman. 2001.
Ecological sorting of vascular plant classes during the Paleozoic
evolutionary radiation. Pp. 285?335 in W. Allmon and D. J.
Bottjer, eds. Evolutionary paleoecology. Columbia University
Press, New York.
DiMichele, W. A., I. P. Montan?ez, C. J. Poulsen, and N. J. Tabor.
2009. Climate and vegetational regime shifts in the late
Paleozoic ice age earth. Geobiology 7:200?226
Dimitrova, T. Kh., and C. J. Cleal. 2007. Palynological evidence for
late Westphalian-early Stephanian vegetation change in the
Dobrudzha Coalfield, NE Bulgaria. Geological Magazine
144:513?524.
Dimitrova, T. Kh., C. J. Cleal, and B. A. Thomas. 2005. Palynology
of late Westphalian-early Stephanian coal-bearing deposits in
the eastern South Wales Coalfield. Geological Magazine
142:809?821.
Erwin, D. H., J. W. Valentine, and J. J. Sepkoski Jr. 1987. A
comparative study of diversification events: the early Paleozoic
vs. the Mesozoic. Evolution 37:1177?1186.
Felsenstein, J. 2005. Using the quantitative genetic threshold
model for inferences between and within species. Philosophical
Transactions of the Royal Society of London B 360:1427?1434.
Finnegan, S., J. L. Payne, and S. C. Wang. 2008. The Red Queen
revisited: reevaluating the age selectivity of Phanerozoic
marine genus extinctions. Paleobiology 34:318?341.
Foote, M. 1991. Analysis of morphological data. Pp. 59?86 in N. L.
Gilinsky and P. W. Signor, eds. Analytical paleobiology.
Paleontological Society, Lawrence, Kans.
???. 2000. Origination and extinction components of taxo-
nomic diversity: general problems. In D. H. Erwin and S. L.
Wing, eds. Deep time: Paleobiology?s perspective. Paleobiology
26(Suppl. to No. 4):74?102.
???. 2005. Pulsed origination and extinction in the marine
realm. Paleobiology 31:6?20.
Freckleton, R. P., P. H. Harvey, and M. Pagel. 2002. Phylogenetic
analysis and comparative data: a test and review of evidence.
American Naturalist 160:712?726.
Galtier, J., and T. L. Phillips. 1996. Structure and evolutionary
significance of Palaeozoic ferns. Pp. 417?483 in J. M. Camus, M.
Gibby, and R. J. Johns, eds. Pteridology in perspective. Royal
Botanical Gardens, Kew, London.
Green, W. A., and L. J. Hickey. 2005. Leaf architectural profiles of
angiosperm floras across the Cretaceous/Tertiary boundary.
American Journal of Science 305:983?1013.
Grime, J. P. 2002. Plant strategies, vegetation processes, and
ecosystem properties, 2d ed. Wiley, New York.
Harmon, L., J. Weir, C. Brock, R. Glor, W. Challenger and G.
Hunt. 2008. geiger: analysis of evolutionary diversification. R
package version 1.2-13. http://www.webpages.uidaho.edu/,lukeh/
index.html
Hutchinson, G. E. 1978. An introduction to population ecology.
Yale University Press, New Haven, Conn.
Jablonski, D. 1986. Background and mass extinctions?the
alternation of macroevolutionary regimes. Science 231:129?
133.
??? . 1987. Heritability at the species level: analysis of
geographic ranges of Cretaceous mollusks. Science 238:360?363.
???. 2001. Lessons from the past: evolutionary impacts of mass
extinctions. Proceedings of the National Academy of Sciences
USA 98:5393?5398.
???. 2004. The evolutionary role of mass extinctions: disaster,
recovery and something in-between. Pp. 151?177 in P. D.
Taylor, ed. Extinctions in the history of life. Cambridge
University Press, Cambridge.
???. 2005. Mass extinctions and macroevolution. Paleobiology
31:192?210.
Jablonski, D., and G. Hunt. 2006. Larval ecology, geographic
range, and species survivorship in Cretaceous mollusks:
organismic versus species-level explanations. American Natu-
ralist 168:556?564.
Janevski, G. A., and T. K. Baumiller. 2009. Evidence for extinction
selectivity throughout the marine invertebrate fossil record.
Paleobiology 35:553?564.
Kembel, S., D. Ackerly, S. Blomberg, P. Cowan, M. Helmus and
C. Webb. 2008. picante: tools for integrating phylogenies and
ecology. R package version 0.4-0. http://cran.r-project.org/web/
packages/picante/index.html
Knoll, A. H. 1984. Patterns of extinction in the fossil record of
vascular plants. Pp. 21?68 in M. H. Nitecki, ed. Extinctions.
University of Chicago Press, Chicago.
Lesnikowska, A. D. 1989. Anatomically preserved Marattiales
from coal swamps of the Desmoinesian and Missourian of the
midcontinent United States: systematics, ecology, and evolu-
tion. Ph.D. dissertation. University of Illinois, Urbana-Cham-
paign.
Lockwood, J. L., G. J. Russell, J. L. Gittleman, C. C. Daehler, M. L.
McKinney, and A. Purvis. 2002. A metric for analyzing
taxonomic patterns of extinction risk. Conservation Biology
16:1137?1142.
Looy, C., R. J. Twitchett, D. L. Dilcher, J. H. A. Van Konijnenburg-
Van Cittert, and H. Visscher. 2001. Life in the end-Permian
dead zone. Proceedings of the National Academy of Sciences
USA 98:7879?7883.
McElwain, J. C., M. E. Popa, S. P. Hesselbo, M. Haworth, and F.
Surlyk. 2007. Macroecological responses of terrestrial vegeta-
tion to climate and atmospheric change across the Triassic/
Jurassic boundary in East Greenland. Paleobiology 33:547?573.
McElwain, J. C., P. J. Wagner, and S. P. Hesselbo. 2009. Fossil
plant relative abundances indicate sudden loss of Late Triassic
biodiversity in East Greenland. Science 324:1554?1556.
McKinney, M. L. 1995. Extinction selectivity among lower taxa:
gradational patterns and rarefaction error in extinction esti-
mates. Paleobiology 21:300?313.
Moles, A. T., D. D. Ackerly, C. O. Webb, J. C. Tweddle, J. B.
Dickie, and M. Westoby. 2005. A brief history of seed size.
Science 307:576?580
Nee, S., and R. M. May. 1997. Extinction and the loss of
evolutionary history. Science 278:692?694.
Niklas, K. J., and B. H. Tiffney. 1994. The quantification of plant
biodiversity through time. Philosophical Transactions of the
Royal Society of London B 345: 35?44.
90 WALTON A. GREEN ET AL.
Niklas, K. J., B. H. Tiffney, and A. H. Knoll. 1980. Apparent
changes in the diversity of fossil plants. Evolutionary Biology
12:1?89.
???. 1985. Patterns in vascular land plant diversification: an
analysis at the species level. Pp. 97?128 in J. W. Valentine, ed.
Phanerozoic diversity patterns: profiles in macroevolution.
Princeton University Press, Princeton, N.J.
Olsson, R. K., C. Hemleben, W. A. Berggren, and B. T. Huber.
1999. Atlas of Paleocene planktonic foraminifera. Smithsonian
Institution Press, Washington, D.C.
Owens, I. P. F., and P. M. Bennett. 2000. Ecological basis of
extinction risk in birds: habitat loss versus human persecution
and introduced predators. Proceedings of the National Acad-
emy of Sciences USA 97:12144?12148.
Pagel, M. 1997. Inferring evolutionary processes from phyloge-
nies. Zoologica Scripta 26:331?348.
???. 1999. Inferring the historical patterns of biological
evolution. Nature 401:877?884.
Paradis, E., and J. Claude. 2002. Analysis of comparative data
using generalized estimating equations. Journal of Theoretical
Biology 218:175?185.
Payne, J. L., and S. Finnegan. 2007. The effect of geographic range
on extinction risk during background and mass extinction.
Proceedings of the National Academy of Sciences USA
104:10506?10511.
Pearson, P. N., R. K. Olsson, B. T. Huber, C. Hemleben, and W. A.
Berggren. 2006. Atlas of Eocene planktonic foraminifera.
Cushman Foundation for Foraminiferal Research Special
Publication No. 41.
Pearson, P. N., I. K. McMillan, B. S. Wade, T. Dunkley Jones, H. K.
Coxall, P. R. Bown, and C. H. Lear. 2008. Extinction and
environmental change across the Eocene-Oligocene boundary
in Tanzania. Geology 36:179?182.
Peppers, R. A. 1996. Palynological correlation of major Pennsyl-
vanian (Middle and Upper Carboniferous) chronostratigraphic
boundaries in the Illinois and other coal basins. Geological
Society of America Memoirs 188.
Pfefferkorn, H. W., and M. Thomson. 1982. Changes in
dominance patterns in Upper Carboniferous plant-fossil as-
semblages. Geology 10:641?644.
Phillips, T. L. 1974. Evolution of vegetative morphology in
coenopterid ferns. Annals of the Missouri Botanical Garden
61:427?461.
Phillips, T. L., and J. Galtier. 2005. Evolutionary and ecological
perspectives of Late Paleozoic ferns, Part I. Zygopteridales.
Review of Palaeobotany and Palynology 135:165?203.
Phillips, T. L., R. A. Peppers, M. J. Avcin, and P. F. Laughnan.
1974. Fossil plants and coal: patterns of change in Pennsylva-
nian coal swamps of the Illinois Basin. Science 184:1367?1369.
Phillips, T. L., A. B. Kunz, and D. J. Mickish. 1977. Paleobotany of
permineralized peat (coal balls) from the Herrin (no. 6) coal
member of the Illinois basin.Geological Society of America
Microform Publication 7:18?49.
Pigg, K. B., T. N. Taylor, and R. A. Stockey. 1987. Paleozoic seed
ferns: Heterangium kentuckyensis sp. nov., from the Upper
Carboniferous of North America. American Journal of Botany
74:1184?1204.
Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2001. The niche of
higher plants: evidence for phylogenetic conservatism. Pro-
ceedings of the Royal Society of London B 268:2383?2389.
Purvis, A. 2008. Phylogenetic approaches to the study of
extinction. Annual Reviews of Ecology and Systematics
39:301?319.
Purvis, A., K. E. Jones, and G. M. Mace. 2000. Extinction.
BioEssays 22:1123?1133.
R Development Core Team. 2008. R: a language and environment
for statistical computing, Version 2.7.0. R Foundation for
Statistical Computing, Vienna. http://www.r-project.org/
Rabosky, D. L. 2009. Heritability of extinction rates links
diversification patterns in molecular phylogenies and fossils.
Systematic Biology 58:629?640.
Raup, D. M. 1972. Taxonomic diversity during the Phanerozoic.
Science 177:1065?1071.
???. 1986. Biological extinction in earth history. Science
231:1528?1533.
???. 1991. Extinction: bad genes or bad luck? Norton, New
York.
???. 1993. Extinction. Oxford University Press, Oxford.
Raup, D. M., and J. J. Sepkoski Jr. 1982. Mass extinctions in the
marine fossil record. Science 215:1501?1503.
Rothwell, G.W. 1991. Botryopteris forensis, a trunk epiphyte of the
tree fern Psaronius. American Journal of Botany 78:782?788.
Roy, K., G. Hunt, and D. Jablonski. 2009. Phylogenetic conserva-
tism of extinctions in marine bivalves. Science 325:733?737.
Russell, G. J., T. M. Brooks, M. M. McKinney, and C. G. Anderson.
1998. Present and future taxonomic selectivity in bird and
mammal extinctions. Conservation Biology 12:1365?1376.
Schwartz, M. W., and D. Simberloff. 2001. Taxon size predicts
rates of rarity in vascular plants. Ecology Letters 4:464?469.
Seilacher, A. 1998. Patterns of macroevolution: how to be
prepared for extinction. Comptes Rendus de l?Acade?mie des
Sciences, series II A, Sciences de la Terre et des Plane`tes
327:431?440.
Sepkoski, J. J., Jr. 1993. Ten years in the library: new paleonto-
logical data confirm evolutionary patterns. Paleobiology 19:43?
51.
Smith, J. T., and K. Roy. 2006. Selectivity during background
extinction: Plio-Pleistocene scallops in California. Paleobiology
32:408?416.
Taylor, T. N. 1965. Paleozoic seed studies: a monograph of the
American species of Pachytesta. Palaeontographica, Abteilung B
117:1?46.
??? . 1967. Paleozoic seed studies: on the structure of
Conostoma leptospermum n. sp., and Albertlongia incostata n.
gen. and sp. Palaeontographica, Abteilung B 121:23?29.
Trivett, M. L. 1992. Growth architecture, structure, and relation-
ships of Cordaixylon iowensis nov. comb. (Cordaitales). Interna-
tional Journal of Plant Sciences 153:273?287.
Wade, B. S., and P. N. Pearson. 2008. Planktonic foraminiferal
turnover, diversity fluctuations and geochemical signals across
the Eocene/Oligocene boundary in Tanzania. Marine Micro-
paleontology 68:244?255.
Wang, S. C., and A. M. Bush. 2008. Adjusting global extinction
rates to account for taxonomic susceptibility. Paleobiology
34:434?455.
Westoby, M. 1998. A leaf-height-seed (LHS) plant ecology
strategy scheme. Plant and Soil 199:213?227.
Wilf, P., and K. R. Johnson. 2004. Land plant extinction at the end
of the Cretaceous: a quantitative analysis of the North Dakota
megafloral record. Paleobiology 30:347?368.
Willis, C. G., B. Ruhfel, R. B. Primack, A. J. Miller-Rushing, and
C. C. Davis. 2008. Phylogenetic patterns of species loss in
Thoreau?s woods are driven by climate change. Proceedings of
the National Academy of Sciences USA 105:17029?17033.
Wing, S. L. 2004. Do plants suffer mass extinctions. Pp. 60?97 in
P. D. Taylor, ed. Extinctions in the history of life. Cambridge
University Press, Cambridge.
EXTINCTION?S AXE AND SHEARS 91