Geometric and Developmental Perspectives on the Evolution of the Skull and Internal Carotid Circulation in Turtles

Abstract

Internal carotid circulation is arguably one of the most intensively analyzed morphological characters in turtle systematics and, thus, it is critical for understanding turtle phylogeny. I used landmark-based geometric mor-phometrics to facilitate a quantitative analysis of variation in turtle skull shape and osteological correlates of the internal carotid circulation. The analysis indicates that the position of the foramen posterior canalis carotici interni differs among eucryptodires, paracryptodires, and pleurod-ires, but remains relatively stable within these lineages. This supports the hypothesis that the position of the foramen posterior canalis carotici interni is a reliable character for identifying higher turtle taxa. Results from the geometric morphometric analysis are consistent with two of the three traits recently proposed to affect patterns of internal carotid circulation. However, these three traits do not fully explain six different patterns of internal carotid circulation in turtles. I identify the association between the internal carotid artery and the palatine branch of the facial nerve (CN VII) as a developmental constraint in turtles. Using this association as a reference point, embryological observations reported in the literature suggest morphogenetic processes that may govern different patterns of internal carotid circulation. When mapped on a phylogenetic tree, some characters transform more than once independently. The evolution of internal carotid circulation in turtles may have been even more complex than reconstructed in this paper, because of uncertain relationships among basal eucryptodires and because of a mosaic of character states in that critical part of the tree.

Full text

Chapter 7
Geometric and Developmental Perspectives on the Evolution
of the Skull and Internal Carotid Circulation in Turtles
Tetsuto Miyashita
Abstract Internal carotid circulation is arguably one of the
most intensively analyzed morphological characters in
turtle systematics and, thus, it is critical for understanding
turtle phylogeny. I used landmark-based geometric mor-
phometrics to facilitate a quantitative analysis of variation
in turtle skull shape and osteological correlates of the
internal carotid circulation. The analysis indicates that the
position of the foramen posterior canalis carotici interni
differs among eucryptodires, paracryptodires, and pleurod-
ires, but remains relatively stable within these lineages. This
supports the hypothesis that the position of the foramen
posterior canalis carotici interni is a reliable character for
identifying higher turtle taxa. Results from the geometric
morphometric analysis are consistent with two of the three
traits recently proposed to affect patterns of internal carotid
circulation. However, these three traits do not fully explain
six different patterns of internal carotid circulation in
turtles. I identify the association between the internal
carotid artery and the palatine branch of the facial nerve
(CN VII) as a developmental constraint in turtles. Using this
association as a reference point, embryological observations
reported in the literature suggest morphogenetic processes
that may govern different patterns of internal carotid
circulation. When mapped on a phylogenetic tree,
some characters transform more than once independently.
The evolution of internal carotid circulation in turtles may
have been even more complex than reconstructed in this
paper, because of uncertain relationships among basal
eucryptodires and because of a mosaic of character states
in that critical part of the tree.
Keywords Cranial arteries Cranial foramina Cranial
nerves Cryptodira Eucryptodira Pleurodira
Introduction
Internal carotid circulation has been an important source of
phylogenetic characters in turtles, partly because the vessels
irrigating the skull are associated with osteological corre-
lates, such as foramina and bony canals, and partly because
the pattern of the circulation is relatively conservative
within each purported clade. This is a long-standing view
(McDowell 1961; Albrecht 1967,1976; Gaffney 1975a) that
has been supported by subsequent morphological-based
phylogenetic studies. In a pair of recent studies, Jamniczky
(2008) and Jamniczky et al. (2006) provided an assessment
of homologies and a comprehensive review of the phylo-
genetic implications of turtle internal carotid circulation.
Given the additional literature entirely devoted to this
character (e.g., Jamniczky and Russell 2004,2007; Sterli
and de la Fuente 2010; Sterli et al. 2010), internal carotid
circulation arguably is one of the most intensively analyzed
morphological traits in turtle systematics.
The advent of molecular phylogenetics has added a new
facet to discussions regarding the phylogenetic utility of
patterns of internal carotid circulation in turtle systematics.
In a classic example of a conflict between molecular and
morphological data, the phylogenetic signal of these mor-
phological characters is inconsistent with phylogenetic trees
generated from molecular data. The Trionychoidea are a
clade of cryptodires supported by morphological characters,
encompassing carettochelyids (pig-nosed turtles), dermate-
mydids (Mesoamerican river turtles), kinosternids (mud
turtles), and trionychids (soft-shelled turtles) (Gaffney
1975a; Meylan and Gaffney 1989). Support for this clade
partly comes from earlier suggestions (McDowell 1961;
Albrecht 1967) that trionychids and kinosternoids share
similar relative diameters of their cranial arterial foramina
and canals. The relative dimensions of the foramina indicate
a decrease in stapedial circulation and an increase in
internal carotid circulation. In concert with other cranial and
postcranial characters, the Trionychoidea are repeatedly
T. Miyashita (&)
Department of Biological Sciences, University of Alberta,
CW 405 Biological Sciences Centre, Edmonton,
AB T6G 2E9, Canada
e-mail: tetsuto@ualberta.ca
D. B. Brinkman et al. (eds.), Morphology and Evolution of Turtles, Vertebrate Paleobiology and Paleoanthropology,
DOI: 10.1007/978-94-007-4309-0_7, ÓSpringer Science+Business Media Dordrecht 2013
71

recovered as a clade in morphological phylogenetic analy-
ses (Gaffney et al. 1991; Gaffney 1996; Shaffer et al. 1997;
Brinkman and Wu 1999). However, molecular phylogenetic
analyses using mitochondrial, ribosomal, and/or nuclear
DNA have consistently rejected a monophyletic Triony-
choidea (Shaffer et al. 1997; Fujita et al. 2004; Krenz et al.
2005; Near et al. 2005). This molecular versus morpho-
logical conflict does not entirely hinge on the reliability of
the phylogenetic signal furnished by the internal carotid
circulatory characters—Joyce’s (2007) morphological phy-
logenetic analysis still supported, albeit weakly, a mono-
phyletic Trionychoidea, even in the absence of any
synapomorphies based on internal carotid circulatory mor-
phology. Instead, this conflict in tree topologies highlights
the need to reassess evidence supporting either of the
hypotheses. On the morphological side of the conflict, a
reassessment presents a methodological challenge of how to
represent phylogenetic signals expressed in morphological
characters.
Out of concern with the non-independence of internal
carotid circulatory characters, Jamniczky (2008) relied on
overall internal carotid circulatory pattern as the locus of
homology, in contrast to Gaffney (1975a) who compared
sizes of the foramina for the cranial arteries one to one. This
led to an unexpected interpretation that the internal carotid
circulatory pattern of kinosternoids is more similar to that of
testudinoids than to that of trionychians (Jamniczky 2008;
‘‘Trionychia’’ sensu Gaffney 1975a). Jamniczky’s (2008)
analysis concluded that the conditions in kinosternoids and
trionychians were autapomorphic for each clade, not syna-
pomorphic between the two clades. Therefore, the phylo-
genetic signal expressed by the internal carotid circulatory
morphology does not contribute to resolution of the
molecular versus morphological conflict, because the
character is uninformative for defining cryptodire suprafa-
milial relationships. The lesson from Jamniczky’s (2008)
reappraisal, aside from the molecular versus morphological
conflict, is that rigorously tested definition and delimitation
are crucial to the ability of a morphological character to
reflect phylogenetic signal.
Here I look at another aspect of phylogenetic signal in
turtle internal carotid circulation, namely the position of
entry for the internal carotid artery into the skull. Using
geometric morphometrics, I quantified variation in position
of the entry of the internal carotid artery with respect to
variation in positions of other cranial landmarks. Simulta-
neously, the morphometric analysis identified cranial land-
marks that correlate with patterns of internal carotid
circulation. Some of the correlated landmarks were con-
sistent with factors recently proposed to affect patterns of
internal carotid circulation. This information helps support
and revises certain characters that have been widely used in
turtle systematics. I also used embryological evidence to
propose potential factors responsible for the anteroposterior
variation in position of entry of the internal carotid artery. In
this analysis, an association between the internal carotid artery
and the palatine branch of the facial nerve (CN VII) is posited
as a developmental constraint. Lastly, I mapped characters of
the internal carotid circulation onto a turtle phylogeny.
Review of Turtle Internal Carotid Circulation
and Previous Studies
General Morphology
The basic eucryptodire condition for the internal carotid
artery and its main banches are shown in Fig. 7.1. Names for
foramina and canals in the turtle skull follow Gaffney (1972a,
1979b). Before the cranial entry of the internal carotid artery,
the stapedial artery (sa in Fig. 7.1) branches off through the
foramen stapediotemporale; this foramen is vestigial in
kinosternoids and absent in Baptemys and Dermatemys
(McDowell 1961). After entering the skull through the fora-
men posterior canalis carotici interni (fpcci in Fig. 7.1) and
extending along the canalis caroticus internus, the internal
carotid artery (ica in Fig. 7.1) anteriorly bifurcates into the
cerebral carotid artery (cca in Fig. 7.1), which enters into the
foramen anterior canalis carotici interni and the palatine
artery (pa in Fig. 7.1), which passes through the canalis ca-
roticus lateralis entering into the foramen caroticus lateralis.
The mandibular and orbital arteries (ma and oa, respectively,
in Fig. 7.1) represent secondary major branches that originate
either from the stapedial artery (general pattern; Fig. 7.1)or
the palatine artery (kinosternoids; not shown). In triony-
chians, the terms mandibular artery and pseudopalatine artery
apply to the vessels that have the same topographical origins
as the cerebral carotid and palatine arteries, respectively, in
other turtles (Albrecht 1967; Jamniczky 2008). In the tri-
onychian scheme, the mandibular artery (entering into the
canalis caroticus lateralis) and pseudopalatine artery (enter-
ing into the foramen anterior canalis carotici interni) branch
off the internal carotid artery instead, and the orbital arteries
irrigate downstream of the pseudopalatine artery (Jamniczky
and Russell 2007). The trionychian scheme is independently
repeated in chelonioids (Gaffney 1984; Jamniczky 2008).
Variation in Internal Carotid Circulation
Branching patterns and relative sizes of vessels constitute
the primary phylogenetic signal in the internal carotid cir-
culation of turtles. Both branching patterns and arterial
72 T. Miyashita

diameters set different groups of turtles apart from each
other (McDowell 1961; Albrecht 1967,1976). Gaffney
(1975a) was the first to translate this information for sys-
tematic use in turtles. In addition to differences in the rel-
ative position of the foramen posterior canalis carotici
interni (discussed later in this section), he identified
reduction of the stapedial artery as diagnostic for the
Trionychoidea and enlargement of the stapedial artery as
diagnostic for the Testudinoidea. That condition is reliably
measured by the sizes of the foramina and canals through
which the vessels pass (Jamniczky and Russell 2004). Thus,
the foramen stapediotemporale (the dorsal opening of the
stapedial canal) is smaller relative to the foramen posterior
canalis carotici interni in trionychoids than in other turtles,
and the reverse is true in testudinoids. Furthermore, the
canalis caroticus lateralis increases in size in dermatemyd-
ids and kinosternids, which has been interpreted as a syn-
apomorphy linking the two families within the more
inclusive Kinosternoidea. However, Jamniczky (2008)
showed that sizes of the foramina are correlated with each
other because the cranial arteries diverge from the same
source artery. Therefore, those characters are not indepen-
dent but, instead, represent two conditions in the intercon-
nected internal carotid circulatory pattern. Jamniczky
(2008) argued that both conditions should be treated as
states under a single character. An extensive review by
Jamniczky and Russell (2007) brought the entire internal
carotid circulation within the scope of analysis, and the
internal carotid circulatory character in support of the Tri-
onychoidea was revised into separate states, one autapo-
morphic for trionychians and the other autapomorphic for
kinosternoids. Their revised character assessment neither
supported nor rejected monophyly of the Trionychoidea.
Evolution of the Foramen Posterior
Canalis Carotici Interni
Internal carotid circulation yields another heavily discussed,
clade-specific character in turtle systematics—the location
of the entrance of the internal carotid artery into the skull
via the foramen posterior canalis carotici interni (Fig. 7.2).
This is unexplored territory for quantitative character
analysis. So far, the following four interpretations have
been offered for evolution of this foramen.
(1) Gaffney (1975a) distinguished paracryptodires from
eucryptodires, partly based on the more anterior location of
the foramen posterior canalis carotici interni along the
basisphenoid-pterygoid suture in paracryptodires. In eu-
cryptodires, the foramen opens at or near the posterior end of
the pterygoid (Gaffney 1975a; Gaffney and Meylan 1988).
(2) Revising Gaffney’s (1975a) hypothesis, Evans and
Kemp (1976) and Rieppel (1980) offered an alternative
interpretation for the paracryptodire condition by suggesting
that the more anterior position of the foramen represents
a gradient, possibly a plesiomorphy with respect to the
eucryptodire condition. These authors viewed paracrypto-
dire condition as representing a gradual transition toward
Fig. 7.1 Schematic drawing showing generalized path of the internal
carotid artery and its major branches in a representative eucryptodire
turtle skull (Adocus sp.; modified after Meylan and Gaffney 1989), in
ventral view. The internal carotid artery enters the skull through the
foramen posterior canalis carotici interni and gives rise to major
cranial arteries. The parabasisphenoid complex encloses the bifurca-
tion point between the cerebral and palatine arteries. Homology and
terminology of vessels are based on Jamniczky (2008); terminology
for foramina and canals follows Gaffney (1972a,1979b). Shading of
arteries in black and in grey indicates, respectively, portions of arteries
outside of and within skull. Abbreviations: bcp bifurcation between the
cerebral carotid and palatine arteries; cca cerebral carotid artery; fpcci
foramen posterior canalis carotici interni; ica internal carotid artery;
ma mandibular artery; oa orbital arteries; pa palatine artery; sa
stapedial artery
7 Internal Carotid Circulation in Turtles 73

the eucryptodire condition and, hence, not a synapomorphy
that sets paracryptodires apart as a monophyletic group.
(3) Brinkman and Nicholls (1993) postulated independent
derivations for the foramen posterior canalis carotici interni
among clades. In their model, the internal carotid artery
initially was exposed ventrally, then later was covered by
extensions of palatal elements beneath the artery. These
extensions progressed from anterior to posterior in para-
cryptodires (thus the foramen formed anteriorly), but in the
opposite direction, from posterior to anterior, in eucryptod-
ires (thus the foramen formed posteriorly). This model
predicts three transitional stages in cryptodire evolution, all
Fig. 7.2 Variations in skull morphology and in position of the
foramen posterior canalis carotici interni across representative turtles.
Skulls are illustrated in ventral view and are shaded black;white dots
indicate the foramina posterior canalis carotici interni. Skulls and
foramina are not to scale. a,bBasal turtles: aProganochelys
quenstedti (after Gaffney and Meeker 1983), illustrating the primitive
condition in which the foramen posterior canalis carotici interni is
absent (= pattern VI of Sterli and de la Fuente 2010); bMeiolania
platiceps (Meiolaniidae; after Gaffney 1983). c–f Eucryptodires:
cAdocus sp. (Adocidae; after Meylan and Gaffney 1989); dChelydra
serpentina (Chelydridae; after Gaffney 1979b); eSolnhofia parsonsi
(Eurysternidae; after Gaffney 1979b); fLissemys punctata (Triony-
chidae; after Gaffney 1979b). g,hParacryptodires: gBoremys pulchra
(Baenidae; after Brinkman and Nicholls 1993); hPlesiobaena antiqua
(Baenidae; after Gaffney 1979b). i–lPleurodires: iBothremys mag-
hrebiana (Bothremydidae; after Gaffney et al. 2006); jChelodina
expansa (Chelidae; after Gaffney 1979a); kPelomedusa subrufa
(Pelomedusidae; after Gaffney 1979b); lBairdemys venezuelensis
(Podocnemididae; after Gaffney and Wood 2002)
74 T. Miyashita

corroborated by fossil examples: (1) the initial stage in
which the internal carotid artery was exposed ventrally (e.g.,
Kallokibotion bajazidi; Gaffney and Meylan 1992); (2) the
paracryptodire condition in which the artery is underlain by
palatal elements anterior to the foramen posterior canalis
carotici interni, and the foramen shifts posteriorward during
the course of paracryptodire evolution (Evans and Kemp
1976; Rieppel 1980; Brinkman and Nicholls 1993); and (3)
the transitional state into the eucryptodire condition in which
the artery is underlain by bone posteriorly, but the vessels
downstream of the internal carotid artery remain exposed
ventrally. For the last state, basal eucryptodires such as
Sinemys lens (Brinkman and Nicholls 1993) and Xinjiang-
chelys latimarginalis (Brinkman and Wu 1999) have a large
area in which the origin of the palatine artery from the
carotid artery is exposed. The foramen posterior canalis
carotici interni decreases in size in more derived eucryp-
todires (Meylan and Gaffney 1989) and it has disappeared in
many extant forms (Brinkman and Nicholls 1993). This
transition implies that enclosure of the carotid artery in eu-
cryptodires progressed anteriorly, thus supporting indepen-
dent originations for the different positions of the foramen
posterior canalis carotici interni in eucryptodires and para-
cryptodires. The implication of the multiple transitional
states in stem eucryptodires is that the position of this
foramen was acquired independently in pleurodires.
(4) Most recently, Sterli and de la Fuente (2010) rec-
ognized six distinct patterns of the internal carotid circu-
lation based on osteological correlates (patterns I–VI; see
Table 7.1). In a companion paper that provided detailed
studies of the basicrania of Plesiochelys and Pleurosternon,
Sterli et al. (2010) suggested three main factors influenced
variation in internal carotid circulation: (1) closure of the
interpterygoid vacuity, leading to formation of the canalis
caroticus lateralis (patterns I–V); (2) ventral expansion of
the parasphenoid, leading to development of a bony flooring
below the the split between the cerebral carotid and palatine
arteries (patterns I–III); and (3) posterior extension of the
pterygoid, leading to the foramen posterior canalis carotici
interni lying within the pterygoid (patterns I and IV). Sterli
and de la Fuente’s (2010) patterns I–VI largely adopted the
stages previously identified by Brinkman and Nicholls
(1993). Therefore, Sterli and de la Fuente’s (2010) scheme
can be viewed as a phylogenetic reappraisal of Brinkman
and Nicholls’s (1993) model. Sterli and colleagues’ major
conclusions were: (1) basal turtles exhibiting patterns V and
VI lack a foramen posterior canalis carotici interni; (2)
ventral expansion of the parasphenoid trapped the internal
carotid artery within the parabasisphenoid complex and
resulted in the ventral covering of the cerebral carotid-
palatine bifurcation (patterns I–III); and (3) posterior
expansion of the pterygoid characterizes pattern I and dif-
ferentiates pattern IV from pattern V, a reverse scenario
from Brinkman and Nicholls’s (1993) model.
Sterli et al.’s (2010) second conclusion raises an interest-
ing possibility that the postulated expansion of the parasph-
enoid and ventral exposure of the parabasisphenoid complex
may be positively correlated. If such a correlation can be
demonstrated, it would provide a morphometric surrogate to
correlate patterns I-III of the internal carotid circulation with
relative degrees of the ventral parasphenoid expansion.
Sterli et al. (2010) reinforced the hypothesis by Brinkman
and Nicholls (1993) that the foramen posterior canalis car-
otici interni might have been acquired multiple times in the
evolution of turtles, because the presence of the foramen is
variable among basal paracryptodires, basal eucryptodires,
and stem taxa of the crown-group Testudines (Sterli and de
la Fuente 2010; Sterli et al. 2010).
Table 7.1 Six patterns of the internal carotid circulation in turtles, based on Sterli and de la Fuente (2010)
Pattern Systematic distribution Split between verebral and
palatine arteries
Location of foramen posterior canalis
caroticus internus
Path of palatine
artery
I Eucryptodires Covered Pterygoid (posterior end) Canalis caroticus
lateralis
II Paracryptodires Covered Basisphenoid-pterygoid suture
(midway)
Canalis caroticus
lateralis
III Pleurodires Covered Prootic or elements underlying
prootic
Canalis caroticus
lateralis
IV Macrobaenids; Synemyids;
Meiolaniids
Exposed Pterygoid (posterior end; posteriorly
expanded)
Canalis caroticus
lateralis
VKallokibotion;Mongolochelys;
Pleurosternon;Glyptodus
Exposed Absent Canalis caroticus
lateralis
VI Proganochelys;Kayentachelys;
Condorchelys; and others
Exposed Absent Interpterygoid
vacuity
7 Internal Carotid Circulation in Turtles 75

The Foramen Posterior Canalis Carotici Interni
in Recent Cladistic Analyses of Turtles
Hirayama et al.’s (2000) cladistic analysis included the
following character (their character 30): foramen posterior
canalis carotici interni formed by: 0, basisphenoid only; 1,
both basisphenoid and pterygoid halfway along the
basisphenoid-pterygoid suture; 2, prootic only; 3, formed
mostly or fully by pterygoid, foramen positioned near the
posterior end of the basisphenoid). Joyce (2007) used the
same character in his comprehensive morphological phy-
logenetics of turtles (his character 56). This character takes
five independent steps from the basal condition to the three
derived conditions (twice for states 2 and 3; Joyce 2007).
Although the evolution hypothesized for the posterior
internal carotid foramen foramen posterior canalis carotici
interni in eucryptodires and paracryptodires is complex, its
position is even more variable within pleurodires. The
carotid artery enters the prootic in chelids, pelomedusids,
and Araripemys, whereas it enters the basisphenoid and
pterygoid in bothremydines, cearachelyinians, euraxemyd-
ids, podocnemidids, Arenila, and Sankuchemys (Gaffney
et al. 2006). Several bothremydids provide exceptions
(Gaffney et al. 2006)—the foramen opens in the medial wall
of the basisphenoid in Kurmademys, in the quadrate in
Labrostochelys, and at the junction of the basisphenoid,
pterygoid, and quadrate in Taphrosphys spp., Bothremys
kellyi, and Zolhafa. As a result, Gaffney et al. (Gaffney et al.
2006, character 74) recognized seven states for this char-
acter. Gaffney et al. (2006) used another character to dis-
tinguish araripemydids, chelids, and pelomedusids from
bothremydids, euraxemydids, and podocnemydids (charac-
ter 75: posterior margin of pterygoid: 0, does not form
anterior margin of foramen posterior canalis carotici interni;
1, forms anterior margin of the foramen). Importantly, the
internal carotid artery passes through the prootic in all
pleurodires, with the possible exception of Kurmademys,
and the variation comes from extension of the underlying
elements that cover this passage (Gaffney et al. 2006).
Phylogenetic Hypotheses on the Position
of the Foramen Posterior Canalis Carotici
Interni
In summary, two levels of phylogenetic hypotheses exist
regarding the position of the foramen posterior canalis
carotici interni. At the higher level, differences in the
position of the foramen have been used to define major
clades of turtles. Gaffney (1975a) differentiated paracryp-
todires and eucryptodires based on position of the foramen.
Brinkman and Nicholls (1993), Sterli and de la Fuente
(2010), and Sterli et al. (2010) implied independent origins
of the foramen among pleurodires, paracryptodires, and
eucryptodires. At the lower level, Gaffney et al. (2006)
posited that differences in the position of the foramen also
comprise a phylogenetic signal among pleurodires.
To define these phylogenetic signals at higher and lower
levels, variation in the position of the foramen should be
quantitatively described and the source of that variation
should be identified. This information is crucial to character
definition and coding strategy in a phylogenetic analysis. A
single character with numerous states could confuse inde-
pendent signals from multiple sources of variation. On the
other hand, multiple characters based on a single source of
the variation could bias the analysis toward one character
state that is prerequisite to code for other characters.
Here I explore the issue of character definition and
coding strategy based on the foramen posterior canalis
carotici interni in turtles. The following four hypotheses are
tested: (1) the foramen posterior canalis carotici interni
varies in position between eucryptodires, paracryptodires
and pleurodires, and therefore provides potential synapo-
morphies that define each clade (Gaffney 1975a; Brinkman
and Nicholls 1993; Sterli and de la Fuente 2010); (2) ventral
exposure of the parabasisphenoid complex is greater in
patterns I-III than in patterns IV-VI (Sterli et al. 2010); (3)
posterior expansion of the pterygoid is associated with the
relatively posterolateral position of the foramen; and (4)
variation in the position of the foramen with respect to other
cranial bones within pleurodires (Gaffney et al. 2006) is due
to shifts in the positions of cranial bones relative to the
foramen.
Materials and Methods
Taxon Sampling
Turtle skulls often are illustrated in ventral view in
descriptions, and the ventral portion of the braincase pro-
vides suitable landmarks that are relatively resistant to
distortion and crushing. Significantly distorted or crushed
skulls of fossil turtles were excluded from the study.
Photographs were chosen over drawings. Reconstructions of
fossil skulls were included if they satisfied any of the fol-
lowing criteria: (1) the skull only lacked fragments along
the edges of bones or sutures, meaning that the original
outlines of elements could be extrapolated with confidence;
(2) missing or distorted parts were reconstructed from
corresponding elements on the other side of the skull, where
those parts were present and undamaged; or (3) the skull
76 T. Miyashita

could be retro-deformed along the direction of crushing. To
cover the diversity of turtles adequately, I included recon-
structed skulls of fossil taxa in the data set, instead of
building a data set solely on osteological specimens of
extant turtles. After a thorough literature search, published
figures of skulls (n=145) from 132 turtle taxa were col-
lected and scanned (Appendix), the majority of which come
from E. S. Gaffney’s consistent and precise descriptions.
Among the taxa included, intraspecific variation in
skull morphology has been documented for the emydid
Pseudemys texana, the kinosternid Sternotherus odoratus,
and the baenid Palatobaena cohen (Bever 2008,2009a,b;
Lyson and Joyce 2009a). For those taxa, all illustrated
specimens that document the variants were included. More
than one reconstruction exists for some fossil turtles
(e.g., Mongolochelys). If those reconstructions were based
on different specimens and done by different authors, all of
them were included.
The sample of turtle taxa represents a wide spectrum of
morphological variation and taxonomic diversity. This
paper follows the turtle phylogeny presented by Joyce
(2007) and includes representatives from the three major
radiations of turtles: pleurodires (n=39), paracryptodires
(n=21), and eucryptodires (n=79). Four basal turtle taxa
(Kayentachelys,Meiolania,Mongolochelys, and Progan-
ochelys) also were included in the data set (n=5); these do
not form a clade. Kayentachelys and Meiolania initially
were placed amongst basal eucryptodires (Gaffney 1983;
Gaffney et al. 1987,2007b; Gaffney and Jenkins 2010)oras
basal cryptodires (Hirayama et al. 2000), but Joyce’s
analysis (2007) suggests that they represent basal lineages
outside crown clades of turtles. I follow this position for the
purpose of this paper, because Joyce (2007) presented
the most inclusive and up-to-date phylogenetic analysis of
turtles.
Landmark Descriptions
In total, 31 landmarks were assigned to skulls taken from
published figures (Fig. 7.3, Table 7.2) using TpsDig ver.
2.16 (Rohlf 2010). Three landmarks (2, 8, and 17) were
along the sagittal plane, 14 landmarks (1, 3–7, 9–16)
were on the right side, and the other 14 landmarks (18–31)
were on the left side and are counterparts of those on the
right. The landmarks on the left side (18–31) minimized
artificial variation in position of the midline landmarks
created by rotation of skull configurations during Procrustes
superimposition (Zelditch et al. 2004). Beyond Procrustes
superimposition and the following multivariate analysis of
variance (MANOVA) and canonical variate analysis
(CVA), however, the counterpart landmarks from the left
side were not considered. According to Bookstein’s (1991)
classification, seven landmarks (6, 7, 10–14) from the right
side and two landmarks from the sagittal plane (8 and 17)
are Type 2 (local minima and maxima of curvature),
whereas the rest are Type 1 (a juxtaposition of tissues).
These landmarks delineate the right half of an entire skull
from ventral view and also capture relative positions of the
ventral cranial elements. When assigned to a foramen, the
Fig. 7.3 A skull of Adocus sp. in ventral view (modified after
Meylan and Gaffney 1989), showing configurations of the 31
landmarks used in this study. Numbers along the midline and on
the anatomical right side of the skull correspond to landmarks listed
in Table 7.2. Three landmarks (2, 8, and 17) are along the sagittal
plane, 14 landmarks (1, 3–7, 9–16) are on the right side, and the
other 14 landmarks (18–31; not numbered in this figure) are on the
left side and are counterparts of those on the right. Positive and
negative directions of the anteroposterior (x) and lateromedial (y)
axes provide interpretation of the coordinates generated by Procrustes
superimposition (see Figs. 7.6,7.7)
7 Internal Carotid Circulation in Turtles 77

landmark was placed at the intersection of the long axis and
the axis of greatest width. Most of the landmarks were
taken from the palatal or braincase floor of the skulls and,
thus, satisfied coplanarity (Zelditch et al. 2004). Only
landmarks 14 and 15 (and their counterpart landmarks from
the left side) and landmark 17 violated the coplanarity
assumption, because they were taken from horizontal
planes intersecting the skull more dorsally. However, these
landmarks were included because they were necessary to
capture the overall shape of the skull from a ventral view.
Because this study was designed to measure horizontal
displacement of landmarks with respect to the foramen
posterior canalis carotici interni, other sources of potential
information along the dorsoventral axis intentionally were
excluded.
Analysis 1: Position of the Foramen Posterior
Canalis Carotici Interni
First, I tested for clade-specific positions of the foramen
posterior canalis carotici interni and variation in the position
of the foramen within clades (hypotheses 1 and 4) using
geometric morphometrics. Therefore, taxa that lack the
foramen posterior canalis carotici interni (i.e., most basal
turtles, Glyptops, and Pleurosternon) were excluded from
this part of the study. The reduced data set (n=139)
consists of one basal turtle (Meiolania), 80 eucryptodires,
19 paracryptodires, and 39 pleurodires. All sets of digitized
landmarks, or configurations, were scaled and rotated
to minimize distance between configurations using the
Procrustes method of generalized least square superimpo-
sition by CoordGen ver. 6 h (Sheets 2001). The Procrustes
method is explained in more detail in Rohlf (1990),
Bookstein (1991), and Zelditch et al. (2004).
The Procrustes coordinates of landmarks are Euclidian
projections from Kendall’s shape space to the tangent space
and, therefore, are not strictly interchangeable with the
Procrustes distance. However, the correlation between these
two was nearly perfect for this data set (a=0.994;
r[0.999) (tpsSmall ver. 1.20; Rohlf 2003), which indi-
cated that distance in the Kendall tangent space closely
approximated the Procrustes distance. Therefore, the strong
correlation allows the topology of Procrustes coordinates
projected onto the Kendall tangent space to be treated as
variables of the superimposed landmark data.
Using Procrustes coordinates, I performed the following
four analyses to test whether position of the foramen pos-
terior canalis carotici interni varies among eucryptodires,
paracryptodires, and pleurodires: (1) Goodall’s Ftest for
significant difference in mean skull shapes between the
three major clades (i.e., eucryptodires, paracryptodires, and
pleurodires); (2) MANOVA (multivariate analysis of vari-
ance) of partial warp scores calculated from the Procrustes
coordinates, which was followed by CVA (canonical variate
analysis); (3) CVA of the Procrustes coordinates and
Table 7.2 List of landmarks
used in this study Landmark number and description Type
1 Foramen posterior canalis carotici interni 1
2 Interpremaxillary suture at ventral margin of skull 1
3 Premaxilla-maxillary contact at ventral margin of skull 1
4 Internal naris (anterior extreme) 1
5 Maxilla-jugal contact at ventral margin of skull 1
6 Pterygoid at subtemporal fenestra (anterolateral extreme) 2
7 Lateral extreme of medial margin of subtemporal fenestra 2
8 Parabasisphenoid (anterior extreme) 2
9 Parabasisphenoid-basioccipital contact (lateral extreme) 1
10 Lateral extreme of quadrate condyle 2
11 Medial extreme of quadrate condyle 2
12 Quadrate (posterolateral corner on ventral side) 2
13 Pterygoid (posterolateral extreme) 2
14 Squamosal (posterior extreme) 2
15 Opisthotic-squamosal contact at posterior margin of skull 1
16 Basioccipital-exoccipital contact (lateral extreme) 1
17 Posterior extreme of occipital condyle 2
Numbers in left column correspond to landmarks labeled on Fig. 7.3. ‘‘Type’’ refers to the classification by
Bookstein (1991)
78 T. Miyashita

analysis of the loadings on canonical variate axes; and (4)
comparison of sample variances of the Procrustes coordi-
nates. Meiolania was used to calculate Procrustes distance
so as not to bias the superimposition, but it was excluded
from all subsequent analyses because it does not belong to
any of the subsets (i.e., eucryptodires, paracryptodires, and
pleurodires) compared here.
Goodall’s Fwas used to test for significant difference
between mean skull shapes of the eucryptodire, paracryptodire,
and pleurodire clades in a pair-wise comparison performed
by TwoGroup ver. 6 h (Sheets 2005). Fscore was boot-
strapped (n=1000, p=0.001). Similarly, a bootstrap
analysis (n=4900) provided 95% confidence intervals for
Procrustes distance between mean skull shapes of the
clades. Thin-plate spline deformation also was performed to
visualize divergence of a mean skull shape for each clade
from a mean skull shape of the data set. The area of
deformation suggests landmarks displaced from the global
mean of the data set. This test was followed by MANOVA
of partial warp scores using CVAGen ver. 6 l (Sheets 2004),
the purpose of which was to evaluate whether or not dif-
ference in mean skull shapes of the clades represents mor-
phological divergence. CVA is a graphical representation of
MANOVA of partial warp scores. Turtles show a wide
spectrum of morphology from an anteroposteriorly long and
lateromedially narrow skull to an anteroposteriorly short
and lateromedially wide skull. This variation occurs within
each major clade and results in superficial resemblance in
skull shapes among distantly related turtles. CVA is the
ordination method that maximizes distance between group
means and, therefore, is capable of detecting underlying
morphological differences that separate a clade containing a
diverse array of skull shapes from other clades. Well-sep-
arated clusters in CVA would indicate that at least some
landmarks have distinct topologies that characterize skull
morphology of the clade.
Next, I used CVA of Procrustes coordinates to explore
variances of the landmarks that resulted in distribution of
the specimens in CVA (PAST ver. 3.2; Hammer et al.
2010). Partial warp scores provide an accurate representa-
tion of shapes defined by the landmarks, but CVA based on
partial warp scores cannot answer the question as to which
landmark contributed differences between groups. This is
because partial warp scores describe shape space and,
therefore, only have 2N-4 variables (N=number of
landmarks) (Zelditch et al. 2004). To circumvent this
problem, I compared CVA of Procrustes coordinates (which
are translated into Euclidian space) with CVA of partial
warp scores for the three major clades. The use of Pro-
crustes coordinates is also justified by the nearly perfect
correlation between the coordinates and Procrustes distance
(a=0.994; r[0.999) (tpsSmall ver. 1.20; Rohlf 2003).
The expectation was that the CVAs would produce plots
with nearly identical data distributions. With the two results
deemed to be similar enough based on the ratios of eigen-
values and the distributions of specimens, loadings of each
landmark coordinate on canonical variate axes in CVA of
Procrustes coordinates were interpreted as components of
the difference explained by the canonical variate axis. In
other words, the loadings are the contribution of landmark
coordinates to the difference between mean skull shapes of
the clades.
By this rationale, CVA loadings would show which
landmark coordinates loaded heavily on the axes that best
discriminate skull shapes of the major clades. Instead of
using raw magnitudes of the loadings, contributions of the
landmark coordinates to CVA of partial warp scores were
inferred from relative magnitudes of the loadings. For
example, the hypothesis that position of the foramen pos-
terior canalis carotici interni can be used as a synapo-
morphy for each of the three major clades predicts
relatively higher loading of the coordinates of the foramen
on the first canonical variate axis than coordinates of other
landmarks.
Finally, I divided the data set into eucryptodires, para-
cryptodires, and pleurodires and performed Procrustes
superimposition separately for each clade. The resulting
Procrustes coordinates were generated without the influence
of skull configurations from other clades and, therefore,
were desirable for examining variation in the skull mor-
phology within each clade. Sample variance was calculated
for each Procrustes coordinate. This information is critical
for two reasons. First, landmarks that differ in position
significantly between the mean skull configurations (thin-
plate spline deformation and CVA) do not necessarily lead
to morphological characters that distinguish one clade from
another. This is because positional variation of the landmark
within clades may swamp the difference between the
means. An overlap between the clade-specific variations
would present difficulty when evaluating a character state.
Second, landmarks that are not helpful in distinguishing
clades may either be highly variable or highly conserved in
position within and across the subsets. Therefore, if a
landmark loaded heavily on any of the canonical variate
axes and varied greatly within the clades, it would suggest
that the variation within the clades may be too great to
detect the difference between the clades even though the
mean positions differ significantly. If a landmark that
loaded heavily in CVA varied little within the clades, it
would suggest that the position of the landmark differs
between the clades and remains stable within the clades. In
this case, the difference between the clades may be trans-
lated into discreet character states. A high variation of a
landmark within the clades may also be used to infer
potentially useful characters below the taxonomic level
examined in this study.
7 Internal Carotid Circulation in Turtles 79

Analysis 2: Cranial Landmarks Correlated
with Internal Carotid Circulation
After testing for the clade-specific positions of the foramen
posterior canalis carotici interni [thereby supporting Sterli
and de la Fuente’s (2010) patterns I–IV], I compared the
overall skull shapes of turtles. I evaluated if any of the other
landmarks vary in correlation with Sterli and de la Fuente’s
patterns I-VI and tested for the morphological characters
correlated with each pattern. This second set of analyses
included all of the taxa from the data set (n=144), but
excluded the landmark representing the foramen posterior
canalis carotici interni (1 and 18) for two reasons: first, to
include the taxa representing patterns V and VI, which lack
the foramen, and, second, to eliminate false variance
transferred from the foramen to the neighboring landmarks
(see discussion, ‘‘Potential Sources of Error’’).
I next divided the data set into two comparative groups in
the following four combinations: (1) turtles with the foramen
posterior canalis carotici interni (patterns I–IV) and turtles
without the foramen (patterns V and VI); (2) turtles with
the bifurcation of the internal carotid artery enclosed within the
parabasisphenoid complex (patterns I-III) and turtles with the
ventrally exposed bifurcation (patterns IV–VI); (3) turtles in
which the canalis caroticus internus is restricted the pterygoid
and parabasisphenoid (patterns I, II, and IV) and turtles in
whichthe canalis caroticusinternusisabsent orassociatedwith
the prootic (patterns III, V, and VI); and (4) turtles with the
foramen posterior canalis carotici interni associated with the
basicranium (patterns II and III) and turtles with the foramen
not associated with the basicranium (patterns I and IV). In each
of the combinations, the morphology characterized by the
alternative patterns represents a plesiomorphic condition with
respect to the morphology that characterizes the initial pat-
terns. Therefore, I performed thin-plate spline deformation of
the latter patterns to the former patterns, not between the mean
skull shape of each of the patterns and the global mean skull
shape of the entire data set as in Analysis 1. For the last com-
bination, turtles with patterns V and VI were excluded because
the foramen is absent in theoe taxa. The analytical procedure
for each combination followed Analysis 1 (steps 1–4).
Patterns V and VI were treated together throughout this
section, because those patterns are correlated with a discrete
morphological character, namely the presence or absence of
the interpterygoid vacuity (Sterli and de la Fuente 2010).
Potential Sources of Error
The data set used in this paper includes only 21 paracryptod-
ires, a number substantially lower than either eucryptodires or
pleurodires. The Paracryptodira (pattern II) consist entirely of
fossil taxa, many of which are known from incomplete skulls
unsuitable for digitization. Patterns IV–VI also are greatly
underrepresented in the data set, because suitable fossil skulls
of macrobaenids, synemyids, and basal turtles are rare. It is
unlikely that the sample size of paracryptodires, macrobae-
nids, synemyids, and basal turtles will dramatically increase in
the near future. The only strategy to deal with unequal sample
sizes is to exercise caution when interpreting variances within
these groups. The most extreme example of this situation in
this paper is that the sample size of turtles with the foramen
posterior canalis carotici interni(n=139) overwhelms that of
the turtles without the foramen (n=6). Goodall’s Ftest is
sensitive to an overlap of the ranges between two groups, if not
too stringent a method for some geometic morphometric
studies (Zelditch et al. 2004). Therefore, a P-value substan-
tially smaller than 0.001 in this test implies that the observed
difference is unlikely to be an artifact of sample size.
The large sample size of eucryptodires with pattern I
(n=74) is problematic, because a global mean skull shape
of the data set was biased toward pattern I. Therefore, the
mean skull configuration of pattern I would appear less
divergent from the global mean than the means of patterns
II and III, and vice versa. As a result, a thin-plate spline
deformation of the true global mean to the mean of pattern I
would be underrepresented. In Analysis 1, the results of the
thin-plate spline deformation were interpreted with this bias
in mind, and I did not invoke an argument as to which
pattern diverged more from the global mean than others. In
Analysis 2, the large sample size of eucryptodires precluded
comparisons between pattern II and all others and between
pattern III and all others, because the sample sizes of pattern
IV–VI were overwhelmed by that of pattern I. As such,
these comparisons largely would repeat the initial compar-
ison among patterns I ?IV, II, and III in Analysis 1.
To test for error by the observer, twenty specimens were
randomly selected from the data set. I digitized the land-
marks in these specimens twice with a four-month interval
intervening. Using Procrustes coordinates, I compared the
two sets of the digitized specimens in Goodall’s Ftest
and MANOVA of partial warp scores followed by CVA.
Goodall’s Ftest cannot reject the null hypothesis that these
two samples were identical with P[0.9999 (F =0.27; df =
58.00, 2204.00; distance between means =0.0258; SE =
0.0132). CVA of partial warp scores showed that these two
samples overlapped (data not shown). Consequently, obser-
ver error in this study is considered negligible.
Another potential source of error in this study is that I
compared three major clades, but did not take into account
phylogenetic distance among the taxa within each clade.
Phylogenetic distance cannot be calculated consistently,
because no single cladistic data set includes all the turtle taxa
used in this study. Eucryptodires, paracryptodires, and
pleurodires each have been consistently recovered as a
80 T. Miyashita

monophyletic clade (e.g., Gaffney 1975a; Gaffney et al.
2006; Joyce 2007). The purposes of this study are to evaluate
the previously proposed, qualitatively accepted hypothesis
that the position of the foramen posterior canalis carotici
interni differs markedly among these major clades (Gaffney
1975a; Brinkman and Nicholls 1993) and to find osteologi-
cal correlates with the major arterial patterns (Sterli and de la
Fuente 2010; Sterli et al. 2010). Both these clades and the
arterial patterns have been dealt with as discrete categories
in the literature and were incorporated in the design of this
study as such. Instead of estimating phylogenetic distance, I
used comparisons of sample variances within the clades or
the arterial patterns to rule out variation within the clades or
patterns (due to phylogenetic distance) as a possible source
of significant difference between the means.
Finally, superimposition of landmark data may introduce a
‘‘Pinocchio effect’’ in which one highly variable landmark
transfers its variance to other landmarks when fitting the con-
figuration to the reference (Walker 2000). The least squares
principle of the Procrustes method in scaling and rotating is
vulnerable to such distortion when incorporating a landmark
with exceedingly large relative displacement (Chapman 1990).
Resistant-fit superimposition of repeated medians is robust to
this problem, but could not be used in this study because the
methoddeparts from the Procrustes distance metric. Although it
is difficult to quantify landmark variance allocated by the Pi-
nocchio effect in a given data set, often the effect is assumed to
be negligible as long as no extreme local change is observed in
the sample (Zelditch et al. 2004); that is the position followed in
this paper. None of the landmarks used herein marks a mini-
mum or maximum of an extremely variable structure that could
extend longer than the skull itself, which would have warranted
invoking the Pinocchio effect. Nevertheless, deformation was
carefully evaluated if any neighbor landmark markedly shifted.
Results
Analysis 1: Position of the Foramen Posterior
Canalis Carotici Interni
A pair-wise comparison using Goodall’s Ftest finds support
for significant differences in mean skull shapes among
eucryptodires, paracryptodires, and pleurodires (Table 7.3).
Mean skull configurations and their transformation by the
thin-plate spline method (Fig. 7.4) indicate the following
three major differences in position of the foramen posterior
canalis carotici interni: (1) the foramen posterior canalis
carotici interni is more posterior in eucryptodires than in
paracryptodires and pleurodires; (2) the foramen is more
anterior and more medial in paracryptodires than in
eucryptodires and pleurodires; and (3) the foramen is more
lateral in pleurodires than in eucryptodires and paracryp-
todires. In association with these differences, the landmarks
for the parabasisphenoid complex show anterior displace-
ment in paracryptodires (landmarks 8 and 9) and lateral
expansion in pleurodires (9). However, the anteromedial
displacement of landmark 9 in paracryptodires may be due
to variances allocated by the foramen posterior canalis
carotici interni, which also is displaced anteromedially. The
pterygoid (6, 7, 13) shows anteroposterior expansion in
eucryptodires, posterolateral expansion in paracryptodires,
and anteroposterior shortening in pleurodires. In pleurod-
ires, the posterolateral extreme (13) is displaced markedly
anteriorly, and the lateral extreme of the subtemporal
margin of the pterygoid (7) is laterally expanded and more
posterior in position. Overall skull shape (2, 3, 17) is
anteroposteriorly shorter in paracryptodires and pleurodires.
The portion of the skull behind the palate is relatively short
anteroposteriorly and wide lateromedially in paracryptod-
ires compared to the other two clades. These topological
differences between the mean skull configurations are can-
didates for morphological characters to distinguish these
clades. However, these differences may be swamped by
variation within the clades (tested in CVA and plots of
variances).
CVA of partial warp scores results in two significant
canonical variate axes (Fig. 7.5), as follows: Axis 1:
eigenvalue =11.792; Wilk’s k=0.0073; df =116;
P\0.001; 55% of total variance; and Axis 2: eigen-
value =9.7225; Wilk’s k=0.0932; df =57; P\0.001;
45% of total variance. All of the taxa included in CVA
segregate into well-defined clusters comprised of eucryp-
todires, paracryptodires, and pleurodires. The first canonical
variate axis separates all three clusters. Amongst these, the
most distant are eucryptodires and pleurodires. Paracryp-
todires are closer to eucryptodires than to pleurodires along
the first axis. The second canonical variate axis separates
Table 7.3 Results from Goodall’s Ftest. Eucryptodires, paracryptodires, and pleurodires were compared pair-wise
Pair-wise Comparisons Fdf P D Low D High D SE
Eucryptodires Paracryptodires 19.37 58.00, 5568.00 \0.001 0.1572 0.1441 0.179 \0.01
Eucryptodires Pleurodires 21.64 58.00, 6728.00 \0.001 0.132 0.1219 0.1485 \0.01
Paracryptodires Pleurodires 10.4 58.00, 3248.00 \0.001 0.1374 0.1289 0.16 \0.01
Procrustes distance between means of the clades was bootstrapped to determine a 95% confidence interval (Low D and High D). Other
abbreviations for column heaadings: DProcrustes distance between means; df Degree of freedom; FF score; PProbability; SE Standard error
7 Internal Carotid Circulation in Turtles 81

paracryptodires from the rest of the data set. The test of
grouping based on Mahalanobis distance between the
specimens and the group means indicates that all of the
specimens are grouped correctly within the clades to which
they belong at the highly significant level P\0.01 (data
not shown). Despite superficial resemblance in overall
skull shape between turtles from distantly related clades,
Goodall’s Fand CVA do suggest divergence in skull
morphology between the major clades. In other words, one
or more landmarks consistently vary between the clades.
To correlate the divergence in skull morphology (CVA
of partial warp scores) and the landmarks that differ in
positions between mean skull configurations (Goodall’s F),
CVA of Procrustes coordinates was undertaken. This results
in two significant canonical variate axes (Axis 1: eigen-
value =13.01; Wilk’s k=0.0061; df =124; P\0.001;
Axis 2: eigenvalue =10.8; Wilk’s k=0.0052; df =146;
P\0.001) and well-separated clusters in similar topo-
graphical relationships. Furthermore, the ratios of eigen-
values of the first two significant canonical variate axes
Fig. 7.4 Landmark-vector plots
of mean skull configurations,
from Analysis 1, based on
Procrustes coordinates and their
thin-plate spline deformation
from the global mean skull
configuration: aeucryptodires;
bparacryptodires; cpleurodires.
Grey silhouettes represent the
global mean skull configuration.
Small arrows in the landmark-
vector plots of mean skull shapes
(left plot) indicate displacement
of landmarks as inferred by the
corresponding thin-plate spline
deformation (right plot)
82 T. Miyashita

almost coincide with each other between the two CVAs (the
ratio for partial warp scores =1.21; the ratio for Procrustes
coordinates =1.2). Thus, in both CVAs, the first canonical
variate axis explains approximately 20% more variance
than explained by the second axis. The almost identical
feature of the two CVAs allows treating loadings on CVA
of Procrustes coordinates as approximated contributions of
these landmark coordinates to CVA of partial warp scores.
A bivariate plot of loadings on the first two canonical
variate axes (Fig. 7.6) indicates outliers that load on one or
both of the axes more heavily than other landmark coordi-
nates. Variances of these outlier coordinates contribute to the
separation of the major clades in CVA (Fig. 7.5). Along the
first canonical variate axis that largely discriminates pleu-
rodires from eucryptodires and paracryptodires, six landmark
coordinates score an absolute value of loading greater than
0.002: anteroposterior coordinates of the foramen posterior
canalis carotici interni (1x), the snout (2x, 3x), and the pos-
terolateral extreme of the pterygoid (13x), and both antero-
posterior and lateromedial coordinates of the lateral extreme
of the subtemporal margin of the pterygoid (7x, y). Along the
second canonical variate axis that discriminates paracryp-
todires from eucryptodires and pleurodires, eight landmark
coordinates score an absolute loading greater than 0.002: both
anteroposterior and lateromedial coordinates of the foramen
posterior canalis carotici interni (1xy), anteroposterior coor-
dinates of the snout tip (2x, 3x), the quadrate (10x, 11x, 12x),
and the occipital condyle (17x). Amongst these coordinates,
three (1x, 7x, 13x) are outside a 95% ellipse. The antero-
posterior coordinates of the foramen posterior canalis carotici
Fig. 7.5 Bivariate plot of CVA
of partial warp scores, from
Analysis 1, showing the two
significant canonical variate axes.
Eucryptodires, paracryptodires,
and pleurodires segregate into
their respective, well-separated
clusters. Mean skull
configurations (from Fig. 7.4) are
depicted beside each cluster.
Symbols: solid circle
eucryptodires; 9paracryptodires;
star pleurodires
Fig. 7.6 Bivariate plot of loadings of Procrustes coordinates on the
two significant canonical variate axes, from Analysis 1. The farther
away from the origin (0, 0) along either or both of the axes, the higher
the loading on the axis and the more important the landmark
coordinate is for setting apart eucryptodires, paracryptodires, and
pleurodires from each other. Shaded area represents the zone of
relatively small variance (i.e., date points in that zone are less useful
for differentiating clades). Data points outside the shaded area
represent landmarks with relatively high loading on either of the
canonical variate axes (i.e., data points that are more useful for
differentiating clades). Numbers correspond to landmarks (see
Fig. 7.3, Table 7.2); ‘‘x’’ is an anteroposterior coordinate; and ‘‘y’’ is
a lateromedial coordinate. Only outlier landmarks are labeled
7 Internal Carotid Circulation in Turtles 83

interni (1x) and the occipital condyle (17x) load more heavily
on the second canonical variate axis than on the first axis,
whereas the subtemporal extreme (7xy) and the posterome-
dial extreme (13x) of the pterygoid load more on the first axis
than on the second axis.
Bivariate plots of sample variances for anteroposterior
(x) and lateromedial (y) Procrustes coordinates identify
landmarks that vary little and others that vary greatly within
the clades (Fig. 7.7). In all clades, variances along the
anteroposterior axis (x) tend to be greater than those along
the lateromedial axis (y). In eucryptodires (Fig. 7.7a), the
snout (2, 3), the quadrate (10, 12), and the occipital condyle
(17) show large anteroposterior variances among other
landmarks. The internal naris (4) and the occipital region
(14, 15, 17) are the areas of relatively high anteroposterior
variance, whereas the lateromedial width across the antor-
bital region (5) is variable. In contrast, the foramen pos-
terior canalis carotici interni (1) is the least variable of all
the landmarks. The parabasisphenoid complex (8, 9) shows
minimal lateromedial variances, which is expected because
the landmark 8 is along the midline of the skull. The pter-
ygoid (7, 13) also shows small variances.
In paracryptodires (Fig. 7.7b), the parabasisphenoid
complex (8, 9) shows large anteroposterior variances along
with the internal naris (4). The foramen posterior canalis
carotici interni (1) is one of the least variable landmarks.
The pterygoid (7, 13), the quadrate (10–12), and the occiput
(17) have relatively small variances. Anteroposterior vari-
ances of the snout (2, 3) are intermediate between the less
variable and the more variable.
In pleurodires (Fig. 7.7c), almost all the land-
marks shows both anteroposterior and lateromedial vari-
ances smaller than 0.001, except for anteroposterior
coordinates of the anteror extreme of the parabasisphenoid
complex (8) and the occipital condyle (17). The foramen
posterior canalis carotici interni (1) has relatively small
variances, although it is not amongst the least variable
landmarks. The snout (2, 3) is the area of the least
variances.
Analysis 2: Cranial Landmarks Correlated
with Internal Carotid Circulation
In the second set of analyses, I identified landmarks
contributing to difference in mean skull shapes between the
modes of the internal carotid circulation (Sterli and de la
Fuente 2010) and tested if the landmarks corroborate mor-
phological transitions proposed to correlate with the modes
of the internal carotid circulation (Sterli et al. 2010).
Goodall’s F test supports significant difference in skull
shapes between turtles with the foramen posterior canalis
carotici interni (patterns I–IV) and those without the fora-
men (patterns V and VI), as follows (abbreviations as for
Table 7.3): F=2.81; df =54.00, 7614.00; P\0.001; D
(95% confidence) =0.1181 (Low D =0.1038, High
D=0.1660); SE =0.0196. Deformation of the mean skull
shape of patterns V and VI (Fig. 7.8a) shows that the mean
Fig. 7.7 Bivariate plots of sample variances of the landmarks
calculated from Procrustes coordinates, from Analysis 1: aeucryptod-
ires; bparacryptodires; cpleurodires. Only landmarks either discussed
in text or associated with unusually high variances are labeled. Each
number represents a landmark from the right side of a skull
(Table 7.2). As in Fig. 7.3, ‘‘x’’ represents the anteroposterior
component (sample variance of the anteroposterior coordinates); ‘‘y’’
represents the lateromedial axis (sample variance of the lateromedial
coordinates); and shaded area represents the zone of relatively small
variances to exclude outliers. Small variance means that the landmarks
vary little within a clade, whereas large variances suggest significant
variation of the landmarks within the clade
84 T. Miyashita

skull shape of patterns I–VI has the following characteris-
tics: the basicranium is expanded strongly posterolaterally
(landmarks 9 and 16–18) and slightly anteriorly (8); the
internal naris (4) is more posterior in position; and the
antorbital region (5) and the occiput (15) are shortened
anteromedially. In CVA of the Procrustes coordinates that
followed after comparison with CVA of partial warp scores,
eleven landmark coordinates score high loadings (upper and
lower 25% of the range of variation) on the single
significant canonical variate axis: 4x, 5xy, 9y, 14xy, 15xy,
16xy, and 17x. Amongst these coordinates, bivariate plots
of the variances for patterns I–IV and for patterns V and VI
show that 4x, 5y, 15xy, 16x, and 17x have relatively small
variances (smaller than the mean) in both of the compara-
tive groups. The coordinates 5x and 14xy are highly vari-
able and are outliers in both of the plots.
Fig. 7.8 Pair-wise comparison of mean skull configurations between
combinations that represent character states of the internal carotid
circulation, from Analysis 2. For each pair-wise comparison, left
image is superposition of mean skull shapes and right image is
corresponding thin-spline deformation. A mean skull configuration of
the combination that represents the plesiomorphic state is outlined by
grey lines and shaded in light grey. A mean skull configuration of the
combination that represents the apomorphic state is drawn with dark
lines and is not shaded. Thin-plate spline deformation shows the
deformation of the mean skull configuration of the plesiomorphic state
into the mean of the apomorphic state. Roman numerals represent
patterns of the internal carotid circulation (see Table 7.1). Pair-wise
comparisons are: aturtles with the foramen posterior canalis carotici
interni (patterns I–IV) and turtles without the foramen (patterns V and
VI); bturtles with the bifurcation of the internal carotid artery
enclosed within the parabasisphenoid complex (patterns I–III) and
turtles with the ventrally exposed bifurcation (patterns IV–VI); c
turtles with the canalis caroticus internus passing through the
pterygoid and parabasisphenoid only (patterns I, II, and IV) and
turtles with the canalis either absent or associated with the prootic
(patterns III, V, and VI); dturtles with the foramen posterior canalis
carotici interni associated with the basicranium (patterns II and III) and
turtles with the foramen not associated with the basicranium (patterns I
and IV). Not to scale among a, b, c, and d
7 Internal Carotid Circulation in Turtles 85

Between turtles with the bifurcation of the cerebral
carotid and palatine arteries enclosed within the parab-
asisphenoid complex (patterns I–III) and turtles with the
bifurcation exposed ventrally (patterns IV–VI), Goodall’s
Ftest returns significant difference in mean skull shapes, as
follows: F=2.95; df =54.00, 7614.00; P\0.001; D
(95% confidence) =0.0805 (Low D =0.0638, High
D=0.1218); SE =0.0148. The deformation of mean skull
shape representing the plesiomorphic state of the exposed
bifurcation (Fig. 7.8b) reveals similar trends recovered in
the comparison between patterns I–IV and patterns V and
VI (4, 5, 8, 9, 15–17) with three main differences. First, the
magnitude of deformation is smaller in this comparison than
in the comparison between the taxa with the foramen pos-
terior canalis carotici interni and those without. Second, the
snout (2, 3) is elongate anteriorly in patterns I–III. Third,
the palate (6, 10–13) is lateromedially narrow in these
patterns. In CVA of Procrustes coordinates after compari-
son with CVA of partial warp scores, fourteen landmark
coordinates score high loadings (upper and lower 25% of
the range of variation) on the canonical variate axis (n=1):
3xy, 4x, 5xy, 6y, 9xy, 10y, 11y, 14y, 15x, and 16xy.
Amongst these coordinates, bivariate plots of the variances
for patterns I–III and for patterns IV–VI shows that the
coordinates 2x, 3x, 9y and 16xy have variances smaller than
the means in both comparative groups. The coordinates 5y,
7y, 8x, 14x, and 17x are outliers in patterns I–III, and the
landmarks 5, 15, and 17 are similarly highly variable both
anteroposteriorly and lateromedially in patterns IV–VI.
In comparison between patterns I, II, and IV (in which
the internal carotid artery passes through the pterygoid
and parabasisphenoid only) and patterns III, V, and VI,
Goodall’s Ftest supports significant difference between the
mean skull shapes, as follows: F=15.06; df =54.00,
7614.00; P\0.001; D (95% confidence) =0.1053 (Low
D=0.0941, High D =0.1211); SE =0.0071. Deforma-
tion of the mean skull shape of patterns III, V, and VI to the
mean skull shape of patterns I, II, and IV (Fig. 7.8c) reveals
the following differences: the snout (2, 3) is elongated
anteriorly; the internal naris (4) is more posterior in posi-
tion; the pterygoid (6, 7) and the antorbital region (5) are
displaced anteromedially; the parabasisphenoid complex
(8, 9) is smaller; the pterygoid (13) is expanded posteriorly;
and the occiput (14–16) is narrower lateromedially with the
occipital condyle (17) displaced more posteriorly. However,
in CVA of Procrustes coordinates following comparison
with CVA of partial warp scores, only two landmark
coordinates load heavily (upper or lower 25% of the range
of variation) on the single significant canonical variate axis:
7y and 13xy. In bivariate plots of the variances, a single
landmark coordinates (13x) have variances smaller than
the means in both of the comparative groups. In contrast,
five landmark coordinates show high variances in both
plots: 5y, 8x, 14xy, and 17x.
Finally, turtles with the foramen posterior canalis caro-
tici interni associated with the basicranium (patterns II and
III) are compared with turtles with the foramen not asso-
ciated with the basicranium (patterns I and IV). Goodall’s
Ftest returns significant support for difference between the
mean skull shapes, as follows: F=16.01; df =54.00,
7344.00; P\0.001; D (95% confidence) =0.1020 (Low
D=0.0865, High D =0.1194); SE =0.0081. The thin-
plate spline deformation from the mean shape of patterns I
and IV to that of patterns II and III (Fig. 7.8d) suggests that
the snout is shorter posteriorly (2, 3), the pterygoid (6, 7)
and the antorbital region (5) are expanded posterolaterally,
the parabasisphenoid complex (8, 9) is expanded anteriorly
and laterally, the quadrate (10–12) is expanded laterally,
and the occiput (14–17) is shortened anteriorly. Amongst
these changes in positions of the landmarks, loadings on the
single significant canonical variate axis are high for 7y, 8x,
9x, 13xy, 14x, and 17x in CVA of Procrustes coordinates
after comparison with CVA of partial warp scores. Despite
the high loadings of several landmark coordinates, only two
of the landmark coordinates (7x and 16y) show variances
smaller than the means in bivariate plots of both compara-
tive groups. The coordinates 5y, 14xy, and 17x are outliers
in both of the plots.
Discussion
Clade-Specific Position of the Foramen
Posterior Canalis Carotici Interni
Analysis 1 quantitatively demonstrated that, despite
superficial resemblances, three major radiations of turtles
(eucryptodires: patterns I ?IV; paracryptodires: pattern II;
pleurodires: pattern III) represent divergence in skull mor-
phology, and that the position of the foramen posterior
canalis carotici interni is one of the important factors in
each shape divergence. The divergence was supported by
two observations. First, mean skull configurations for
eucryptodires, paracryptodires, and pleurodires significantly
differed from each other (Goodall’s F; Table 7.3). Second,
the three clades formed well-separated clusters on the CVA
plot (Fig. 7.5). The position of the foramen posterior canalis
carotici interni as a factor in each of the divergences was
supported by three lines of evidence (Table 7.4): (1) the
thin-plate spline deformations (Fig. 7.4) showed marked
differences in the position of the foramen between the mean
skull configurations; (2) anteroposterior and lateromedial
86 T. Miyashita

coordinates of the foramen loaded heavily on the significant
canonical variate axes (Fig. 7.6); and (3) despite its large
variation between the clades, the foramen was amongst the
least variable of the landmarks within each clade (Fig. 7.7).
Taken together, the position of the foramen posterior canalis
carotici interni is conserved within each clade, but differs
greatly between clades when compared within the set of
landmarks examined in this study. Its high loadings on the
canonical variate axes and small variances within each of
the three clades reject the possibility that difference in its
position between the mean skull configurations were due to
large variation within each clade. Topographically, the
foramen posterior canalis carotici interni is more postero-
lateral in position in eucryptodires than in other clades. In
pleurodires, the foramen is anteromedial relative to that in
eucryptodires and posterior relative to that in paracryptod-
ires. The foramen is anteromedial in position in paracryp-
todiers with respect to eucryptodires and pleurodires.
Osteologically, these different positions can be distin-
guished by the elements that the foramen perforates: the
pterygoid in eucryptodires, the pterygoid and basisphenoid
in paracryptodires, and the prootic in most pleurodires
(Gaffney 1975a; Brinkman and Nicholls 1993; Hirayama
et al. 2000). This topographical variation is consistent with
the results from Analysis 1.
These interpretations, together with the new classifica-
tion of arterial patterns by Sterli and de la Fuente (2010),
warrant reassessment of characters of the foramen posterior
canalis carotici interni used in previous cladistic analyses of
turtle interrelationships. The plesiomorphic state of the
internal carotid circulation in turtles unambiguously is the
absence of the foramen posterior canalis carotici interni in
basal turtles (Sterli and de la Fuente 2010). The clade-
specific positions of the foramen supported by the present
analysis each represents an apomorphy. The identification
of these apomorphies is consistent with Gaffney’s (1975a)
initial observation on difference in the position of the
foramen between eucryptodires and paracryptodires, but not
with Evans and Kemp’s (1976) and Rieppel’s (1980)
hypothesized evolutionary transition from the paracrypto-
dire condition to the eucryptodire condition. It is also
consistent with independent evolutionary origins of the
Table 7.4 Summary of the landmarks correlated with the patterns of the internal carotid circulation
Characters and landmark
numbers
Patterns Analysis 1 Analysis 2
Deformation CVA Test of
Variance
Deformation CVA Test of
variance
Foramen posterior canalis
carotici interni (1)
I + IV Strong High Small – – –
II Strong High Small – – –
III Strong High Small – – –
Snout (2, 3) I ?IV Strong High Equivoval Strong Equivocal Equivocal
Internal naris (4) I + II + III + IV – – – Strong High Small
Maxilla-jugal contact (5) I ?II ?III ?IV – – – Strong High Equivocal
Anterolateral extreme of
pterygoid (6)
I Weak Low Equivocal – – –
II Strong Equivocal Equivocal – – –
III Weak Low Equivocal – – –
II ?III – – – Strong Equivocal Equivocal
Lateral extreme of pterygoid (7) III Strong High Equivoval Strong High Equivocal
Parabasisphenoid complex
(8, 9)
I + II + III – – – Strong High Small
II Strong Low Large – – –
III Weak Low Small – – –
Quadrate (10–12) I ?IV Weak High Equivoval Strong Equivocal Large
Posterolateral extreme of
pterygoid (13)
I + II + IV Strong High Small Strong High Small
Occiput (14–17) I ?II ?III ?IV – – – Strong High Equivocal
Groups of the patterns represent the least inclusive combinations for the landmark based on Analysis 1 and multiple analyses from Analysis 2.
Landmarks were all evaluated relative to each other. In the first column, numbers in parentheses correspond to landmarks in Fig. 7.3 and
Table 7.2. Pattern numbers in the second column follow (Sterli and de la Fuente 2010; see Table 7.1). In the deformation columns, ‘‘strong’’
means deformation greater than the ones exhibited by the lower two-thirds of the landmarks and ‘‘weak’’ means deformation within lower
two-thirds of landmarks. In the CVA columns, ‘‘high’’ refers to loading values that fall into upper and lower 25% of the range of variation and
‘‘low’’ refers to loading values that fall within the middle 50% of the range of variation. In the test of variance column, ‘‘large’’ refers to the
variance larger than the mean of the variances and ‘‘small’’ refers to the variance smaller than the mean of the variances. Bolded entries highlight
characters positively supported by all tests
7 Internal Carotid Circulation in Turtles 87

foramen in eucryptodires (pattern I ?IV) and
paracryptodires (pattern II) proposed by Brinkman and
Nicholls (1993). Origin of the foramen in paracryptodires
independent of eucryptodires and pleurodires is supported
by the absence of the foramen in the stem paracryptodires
Glyptops and Pleurosternon (Sterli et al. 2010). Because
pleurodires do not form a sister clade to either eucryptodires
or paracryptodires in any contemporary turtle phylogeny
(Gaffney 1975a; Gaffney et al. 2006; Joyce 2007), the
pleurodire condition (pattern III) also may have arisen
independently. The basal position (Hirayama et al. 2000;
Joyce 2007)ofMeiolania (pattern IV), posits either an
independent derivation of the foramen for this taxon or a
reversal in its putative sister-taxon Mongolochelys (pattern V).
Independent origins of the foramen posterior canalis
carotici interni challenge the homology of the foramen and
may eventually complicate its terminology. However, the
more important implication is that the position of cranial
entrance of the internal carotid artery provides a quantita-
tively supported morphological character diagnostic for
each major clade. Based on new information provided by
Sterli and de la Fuente (2010) and Sterli et al. (2010),
combined with quantitative support for each state from the
present analysis, I propose the following amendment to
Joyce’s (2007) character 56:
The internal carotid artery passes: 0, ventral to the bas-
icranium and palate before bifurcating into the cerebral
carotid and palatine arteries (no foramen posterior canalis
carotici interni); 1, between the basisphenoid and the pter-
ygoid halfway along the basisphenoid-pterygoid suture; 2,
through the prootic; 3, through the pterygoid near the pos-
terolateral end of the parabasisphenoid. Unordered.
(Remarks: In state 2, passage of the internal carotid artery
may be obscured ventrally by the underlying bones. Mod-
ified from Joyce 2007, character 56.)
Variation of the Foramen Posterior Canalis
Carotici Interni Within Clades
In paracryptodires, anteroposterior variance of the foramen
is small relative to that of other landmarks (Fig. 7.7b). This
small anteroposterior variance rejects the hypothesis (Evans
and Kemp 1976; Rieppel 1980) of an anteroposterior gra-
dient in the position of the foramen within the clade. The
foramen identified as the foramen posterior canalis carotici
interni by Evans and Kemp (1976) actually represents the
foramen for the cerebral carotid artery (Sterli et al. 2010).
Thalassemys, used by Rieppel (1980) in support of the
gradient, is now considered a basal eucryptodire (Joyce
2007), not a paracryptodire. The foramen previously
identified as the foramen posterior canalis carotici interni in
paracryptodires needs further examination.
Pleurodires show a variety of conditions in the position
of the foramen posterior canalis carotici interni (Gaffney
et al. 2006). The anteroposterior and lateromedial variances
of the foramen being smaller than those of almost all other
landmarks within pleurodires (Fig. 7.7c) suggest that the
position of the foramen is stable within the clade, at least
relative to the other landmarks used in this study. The
conservative position of the foramen within the skull is
inconsistent with the osteological observation that the
foramen opens into different elements (Gaffney et al. 2006).
Therefore, it is hypothesized here that this variation depends
on the relative dimensions and position of the elements
within the skull, not a shift in the position of the internal
carotid artery. The present analysis provides no evidence
that the internal carotid artery significantly shifts its position
of the cranial entrance relative to the skull configuration
within pleurodires. For this reason, variation of the elements
surrounding the foramen should not be treated at the same
level as differences in the position of the foramen between
eucryptodires, paracryptodires, and pleurodires.
By setting up two characters describing the position of
the foramen posterior canalis carotici interni, Gaffney et al.
(2006) identified both anteroposterior and lateromedial
changes in the relative dimensions of the elements that
surround the foramen posterior canalis carotici interni in
pleurodires. Their character 74 essentially specifies that the
prootic is covered ventrally by either the lateral expansion
of the parabasisphenoid or the medial expansion of the
quadrate in derived pleurodires. The internal carotid artery
pierces whichever element expanded and underlays the
prootic. This sometimes is coupled with posterior expansion
of the pterygoid, which complicates Gaffney et al.’s (2006)
character definition. Gaffney et al. (2006) defined another
character (75) that describes participation of the pterygoid
in the margin of the foramen posterior canalis carotici
interni due to the posterior expansion of the element. Per-
haps this complex interaction of several elements conspired
against the geometric morphometric methodology used here
from detecting evidence of the interactive changes in rela-
tive dimensions of the elements within pleurodires.
One strategy to code for this complex variation in a
phylogenetic analysis is to formulate two characters that
code for the pleurodire variation, in addition to the character
(modified version of character 56 of Joyce 2007) that dis-
tinguishes position of the foramen between the major
clades. The first two characters can be modified from two
characters originally proposed by Gaffney et al. (2006)—
one of those codes for lateromedial change in the dimen-
sions of the parabasisphenoid complex and the quadrate,
whereas the other codes for anteroposterior change in the
dimensions of the parabasisphenoid complex and the
88 T. Miyashita

pterygoid. These two characters separate the states among
pleurodires and are scored as ‘‘inapplicable’’ for all other
taxa. This approach avoids weighting the analysis in favour
of the plesiomorphy and the eucryptodire and paracrypto-
dire conditions. Seven states in Gaffney et al.’s (2006)
character 74 present difficulties in recovering a strong
phylogenetic signal. That character treats variation within
pleurodires at the same level as it does for the independently
derived states at the level of eucryptodires, paracryptodires,
and pleurodires. For this reason also, multiple characters are
preferred over formulating a single phylogenetic character
that encompasses all the morphological variation in the
position of the foramen posterior canalis carotici interni.
Gaffney et al.’s (2006) characters 74 and 75 may be mod-
ified as in the following:
Foramen posterior canalis carotici interni in pleurodires:
0, surrounded by the prootic; 1, by the parabasisphenoid
complex and/or the pterygoid; 2, entirely or partly by the
quadrate. Unordered. (Remarks: Scored as ‘‘inapplicable’’
in non-pleurodires. Modified from Gaffney et al. 2006,
character 74.)
Posterior margin of the pterygoid in pleurodires: 0, does
not form the anterior margin of the foramen posterior
canalis carotici interni; 1, forms the anterior margin of the
foramen posterior canalis carotici interni. (Remarks: Scored
as ‘‘inapplicable’’ in non-pleurodires. Modified from Gaff-
ney et al. 2006, character 75.)
Correlations with Other Cranial Landmarks
The complex distributions of character states associated
with the internal carotid artery (Table 7.1) strongly suggest
multiple factors independently or interactively affect the
morphology of the internal carotid circulation in turtles. The
results from analyses 1 and 2 (summarized in Table 7.4)
help tease apart correlations between patterns of the internal
carotid circulation and other cranial landmarks and test two
of the three factors proposed by Sterli et al. (2010) that
facilitate patterns.
Sterli et al. (2010) proposed that the expansion of the
parasphenoid ventral to the basisphenoid captured the
internal carotid artery within the parabasisphenoid complex
and the ventral covering of the bifurcation between
the cerebral carotid and palatine arteries (patterns I–III).
Evidence for the expansion of the parasphenoid comes from
the basicrania of the basal eucryptodire Plesiochelys (pat-
tern I) and the basal paracryptodire Pleurosternon (pattern
V) in which the parasphenoid can be distinguished from
the basisphenoid (Sterli et al. 2010). Expansion of the
parasphenoid would lead to increased ventral exposure of
the parabasisphenoid complex. That hypothesis is supported
by Analysis 2. Larger ventral exposure of the parabasisph-
enoid complex was identified in the deformation from pat-
terns IV–VI to patterns I–III. That expansion was an
important factor in distinguishing patterns I–III from pat-
terns IV–VI in CVA, and the variances were small relative
to other landmarks. A similar trend of the expanded pa-
rabasisphenoid complex was detected in the comparison
between patterns V and VI and patterns I–IV. However,
deformation was smaller than in the comparison between
patterns IV–VI and patterns I–III (Fig. 7.8). Posterolateral
expansion of the parabasisphenoid complex was not
explicitly supported by both CVA and the plots of variances
in the comparison between patterns V and VI and patterns
I–IV. All of these observations suggest that increased ven-
tral exposure of the parabasisphenoid complex correlates
with patterns I–III, which are characterized by the ventral
flooring of the cerebral carotid-palatine bifurcation by the
parabasisphenoid complex. The present analysis upholds a
modified version of Hirayama et al.’s (2000) character 31,
which was not adopted by either Gaffney et al. (2006)or
Joyce (2007). The modified character can be described as
follows:
Bifurcation of the internal carotid artery into the palatine
artery and the cerebral carotid artery: 0, not covered ven-
trally; 1, covered ventrally by the parabasisphenoid com-
plex. (Remarks: modified from Hirayama et al. 2000,
character 31.)
Additionally, Sterli et al. (2010) identified posterior
expansion of the pterygoid as a factor facilitating patterns I,
II, and IV. That hypothesis is consistent with the results
from my analyses 1 and 2. However, posterolateral expan-
sion of the pterygoid is not necessarily correlated with the
foramen forming within the pterygoid (patterns I and IV),
because the foramen forms more anteriorly in paracryp-
todires than in any other clades (pattern II). Perhaps the
posterolaterally expanded pterygoid explains the postero-
lateral position of the foramen posterior canalis carotici
interni in eucryptodires, but this is not the case