ArticlePDF Available

A new abelisaurid dinosaur from the end Cretaceous of Patagonia and evolutionary rates among the Ceratosauria

May 2024
Cladistics 40(2)

DOI:10.1111/cla.12583

License
CC BY-NC-ND 4.0

Authors:

Diego Pol

Museo Argentino de Ciencias Naturales "Bernardino Rivadavia"

Mattia Antonio Baiano

Chinese University of Hong Kong

David Černý

University of Chicago

Fernando E Novas

Museo Argentino de Ciencias Naturales "Bernardino Rivadavia"

Show all 6 authorsHide

Gondwanan dinosaur faunae during the 20 Myr preceding the Cretaceous-Palaeogene (K/Pg) extinction included several line-ages that were absent or poorly represented in Laurasian landmasses. Among these, the South American fossil record contains diverse abelisaurids, arguably the most successful groups of carnivorous dinosaurs from Gondwana in the Cretaceous, reaching their highest diversity towards the end of this period. Here we describe Koleken inakayali gen. et sp. n., a new abelisaurid from the La Colonia Formation (Maastrichtian, Upper Cretaceous) of Patagonia. Koleken inakayali is known from several skull bones, an almost complete dorsal series, complete sacrum, several caudal vertebrae, pelvic girdle and almost complete hind limbs. The new abelisaurid shows a unique set of features in the skull and several anatomical differences from Carnotaurus sas-trei (the only other abelisaurid known from the La Colonia Formation). Koleken inakayali is retrieved as a brachyrostran abeli-saurid, clustered with other South American abelisaurids from the latest Cretaceous (Campanian-Maastrichtian), such as Aucasaurus, Niebla and Carnotaurus. Leveraging our phylogeny estimates, we explore rates of morphological evolution across ceratosaurian lineages, finding them to be particularly high for elaphrosaurine noasaurids and around the base of Abelisauridae, before the Early Cretaceous radiation of the latter clade. The Noasauridae and their sister clade show contrasting patterns of morphological evolution, with noasaurids undergoing an early phase of accelerated evolution of the axial and hind limb skeleton in the Jurassic, and the abelisaurids exhibiting sustained high rates of cranial evolution during the Early Cretaceous. These results provide much needed context for the evolutionary dynamics of ceratosaurian theropods, contributing to broader understanding of macroevolutionary patterns across dinosaurs.

Content uploaded by Mattia Antonio Baiano

Content may be subject to copyright.

Content uploaded by Mattia Antonio Baiano

Content may be subject to copyright.

A new abelisaurid dinosaur from the end Cretaceous of Patagonia

and evolutionary rates among the Ceratosauria

Diego Pol*

a,b

, Mattia Antonio Baiano

b,c,d,e

, David

Cern´

, Fernando E. Novas

a,b

Ignacio A. Cerda

b,e,g,h

and Michael Pittman*

Museo Argentino de Ciencias Naturales Bernardino Rivadavia, Buenos Aires, Argentina;

Consejo Nacional de Investigaciones Cientı´ﬁcas y

T´

ecnicas (CONICET), Buenos Aires, Argentina;

School of Life Sciences, The Chinese University of Hong Kong, Shatin, Hong Kong SAR, China;

Area Laboratorio e Investigaci´

on, Museo Municipal Ernesto Bachmann, Villa El Choc´

on, Neuqu´

en, Argentina;

Universidad Nacional de Rı´o Negro

(UNRN), General Roca, Rı´o Negro, Argentina;

Department of the Geophysical Sciences, University of Chicago, Chicago, IL, USA;

Instituto de

Investigaci´

on en Paleobiologı´a y Geologı´a (IIPG), General Roca, Rı´o Negro, Argentina;

Museo Provincial Carlos Ameghino, Cipolletti, Rı´o

Negro, Argentina

Received 5 January 2024; Revised 24 April 2024; Accepted 24 April 2024

Abstract

Gondwanan dinosaur faunae during the 20 Myr preceding the Cretaceous–Palaeogene (K/Pg) extinction included several line-

ages that were absent or poorly represented in Laurasian landmasses. Among these, the South American fossil record contains

diverse abelisaurids, arguably the most successful groups of carnivorous dinosaurs from Gondwana in the Cretaceous, reaching

their highest diversity towards the end of this period. Here we describe Koleken inakayali gen. et sp. n., a new abelisaurid from

the La Colonia Formation (Maastrichtian, Upper Cretaceous) of Patagonia. Koleken inakayali is known from several skull

bones, an almost complete dorsal series, complete sacrum, several caudal vertebrae, pelvic girdle and almost complete hind

limbs. The new abelisaurid shows a unique set of features in the skull and several anatomical differences from Carnotaurus sas-

trei (the only other abelisaurid known from the La Colonia Formation). Koleken inakayali is retrieved as a brachyrostran abeli-

saurid, clustered with other South American abelisaurids from the latest Cretaceous (Campanian–Maastrichtian), such as

Aucasaurus,Niebla and Carnotaurus. Leveraging our phylogeny estimates, we explore rates of morphological evolution across

ceratosaurian lineages, ﬁnding them to be particularly high for elaphrosaurine noasaurids and around the base of Abelisauridae,

before the Early Cretaceous radiation of the latter clade. The Noasauridae and their sister clade show contrasting patterns of

morphological evolution, with noasaurids undergoing an early phase of accelerated evolution of the axial and hind limb skeleton

in the Jurassic, and the abelisaurids exhibiting sustained high rates of cranial evolution during the Early Cretaceous. These

results provide much needed context for the evolutionary dynamics of ceratosaurian theropods, contributing to broader under-

standing of macroevolutionary patterns across dinosaurs.

Introduction

During the Late Cretaceous, Gondwanan dinosaur

faunae comprised a number of different lineages. Espe-

cially in South America, the ﬁrst stage of the Late Cre-

taceous saw the origin of several groups, some of

which reached the Cretaceous–Palaeogene (K/Pg)

boundary, while others went extinct during the

Cenomanian–Turonian interval (Apesteguı´a, 2002;

Coria and Salgado, 2005; Bellardini et al., 2021;

Baiano et al., 2022). The herbivorous South American

faunae in the early stages of Late Cretaceous

(Cenomanian–Turonian) were dominated by rebbachi-

saurid (e.g. Rayososaurus,Cathartesaura,Limaysaurus,

Katepensaurus; Bonaparte, 1996; Salgado et al., 2004;

Gallina and Apesteguı´a, 2005; Ibiricu et al., 2013)

and titanosaurian sauropods (e.g. Argentinosaurus,

*Corresponding author: E-mail address:dpol@mef.org.ar (D.P.);

mpittman@cuhk.edu.hk (M.P.)

Cladistics

Cladistics (2024) 1–50

doi: 10.1111/cla.12583

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and

distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modiﬁcations or adaptations are made.

Andesaurus,Epachthosaurus; Powell, 1990; Calvo and

Bonaparte, 1991; Bonaparte and Coria, 1993), whereas

ornithischians seem to be uncommon (e.g. Jakapil;

Riguetti et al., 2022). The theropod fossil record is

characterized mostly by the huge carcharodontosaurids

(e.g. Giganotosaurus,Mapusaurus,Meraxes; Coria

and Salgado, 1995; Coria and Currie, 2006; Canale

et al., 2022), spinosaurids (e.g. Oxalaia; Kellner

et al., 2011), abelisaurids (e.g. Ekrixinatosaurus,Iloke-

lesia,Skorpiovenator, and Xenotarsosaurus; Martı´nez

et al., 1986; Coria and Salgado, 2000; Canale

et al., 2009; Ibiricu et al., 2021), possible noasaurids

(Huinculsaurus; Baiano et al., 2020), and small-sized

maniraptoriforms such as Buitreraptor and Alnashetri

(Makovicky et al., 2005,2012).

After the Cenomanian–Turonian interval, there was

faunal turnover with rebbachisaurids disappearing and

titanosaurs replacing them in the principal role in eco-

systems, and surviving until the end of the Maastrich-

tian. In fact, multiple titanosaur taxa are known from

this age in South America (e.g. Yamanasaurus,Neu-

quensaurus,Uberabatitan,Notocolossus,Baurutitan,

Rocasaurus,Pelligrinisaurus,Saltasaurus,Arackar;

Bonaparte and Powell, 1980; Powell, 1992; Sal-

gado, 1996; Salgado and Azpilicueta, 2000; Kellner

et al., 2005; Salgado and de Souza Carvalho, 2008;

Gonz´

alez Riga et al., 2016; Apesteguı´a et al., 2020;

Rubilar-Rogers et al., 2021). Ornithischian dinosaurs

were more diverse compared to those from the ﬁrst

stages of the Late Cretaceous, encompassing ankylo-

saurs (e.g. Stegouros,Antarctopelta; Salgado and Gas-

parini, 2006; Soto-Acu ˜

na et al., 2021), early

ornithopods (e.g. Mahuidacursor,Talenkauen,Gaspari-

nisaura; Coria and Salgado, 1996; Novas et al., 2004;

Cruzado-Caballero et al., 2019) and hadrosaurs (e.g.

Huallasaurus,Bonapartesaurus,Kelumapusaura; Bona-

parte et al., 1984; Cruzado-Caballero and Powell, 2017;

Rozadilla et al., 2021). Theropod faunae also experi-

enced a turnover during the Cenomanian–Turonian

interval, as carcharodontosaurids and spinosaurids dis-

appeared, and abelisaurids and megaraptorans became

the predominant lineages (e.g. Coria and Sal-

gado, 2005; Novas et al., 2005; Novas et al., 2013;

Baiano et al., 2022; Baiano and Filippi, 2022). Late

Turonian–Maastrichtian non-avian theropods include

a high diversity of furileusaurian abelisaurids (e.g.

Elemgasem,Carnotaurus,Pycnonemosaurus,Quilme-

saurus,Aucasaurus,Niebla,Guemesia,Llukalkan,Via-

venator,Kurupi; Bonaparte, 1985; Coria, 2001; Coria

et al., 2002; Kellner and Campos, 2002; Filippi

et al., 2016; Gianechini et al., 2020; Agnolı´n

et al., 2021; Aranciaga Rolando et al., 2021; Iori

et al., 2021; Baiano et al., 2022), noasaurids (e.g. Velo-

cisaurus,Noasaurus,Vespersaurus; Bonaparte and

Powell, 1980; Bonaparte, 1991; Langer et al., 2019),

megaraptorids (e.g. Aerosteon,Tratayenia,Maip,

Megaraptor,Murusraptor; Novas, 1998; Sereno

et al., 2008; Coria and Currie, 2016; Porﬁri

et al., 2018; Aranciaga Rolando et al., 2022), alvarez-

saurids (e.g. Patagonykus,Bonapartenykus;

Novas, 1996; Agnolı´n et al., 2012) and dromaeosaurids

(e.g. Unenlagia,Neuquenraptor,Pamparaptor,Austror-

aptor,Ypupiara; Novas and Puerta, 1997; Calvo

et al., 2004; Novas and Pol, 2005; Novas et al., 2009;

Porﬁri et al., 2011; Brum et al., 2021).

Abelisaurids were the most abundant theropods dur-

ing the latest Cretaceous, occurring in all Gondwanan

regions, except Antarctica and Australia where they

remain unknown. They are known principally from

South America (e.g. Novas et al., 2013), which has the

best fossil record for this group, as well as from India

(e.g. Wilson et al., 2003), northern Africa (e.g. Long-

rich et al., 2023), and Madagascar (e.g. Krause

et al., 2007; Farke and Sertich, 2013). Incomplete abe-

lisaurid specimens were also discovered in Europe (e.g.

Accarie et al., 1995; Tortosa et al., 2014), adding evi-

dence to the biogeographical link between southern

Europe and Gondwana (Ezcurra and Agnolı´n, 2012).

Despite the instability of some taxa (e.g. Abelisaurus,

Dahalokely,Genusaurus), recent phylogenetic analyses

typically agree that there are two principal clades

within Abelisauridae, ‘majungasaurines’ and brachyr-

ostrans (Canale et al., 2009; Tortosa et al., 2014). The

clade recognized as ‘Majungasaurinae’ (sensu Baiano

et al., 2022) mainly includes Indian and Malagasy

forms, such as Majungasaurus,Rahiolisaurus and Raja-

saurus (Sampson et al., 1998; Wilson et al., 2003;

Novas et al., 2010), whereas Brachyrostra includes

mainly, if not only, South American abelisaurids, such

as Carnotaurus,Aucasaurus,Elemgasem and Skorpiove-

nator (Bonaparte, 1985; Coria et al., 2002; Canale

et al., 2009; Baiano et al., 2022,2023). Within Bra-

chyrostra, an early diverging lineage is retrieved, which

includes Ekrixinatosaurus,Ilokelesia and sometimes

Skorpiovenator, and a more diverse clade, furileusaur-

ian abelisaurids, that encompasses most South Ameri-

can abelisaurids known from the latest Cretaceous

(e.g. Santonian–Maastrichtian; Filippi et al., 2016).

Here we present a new furileusaurian abelisaurid

Koleken inakayali gen. et sp. n., from the Maastrich-

tian of Patagonia (La Colonia Formation) in Chubut

Province, Argentina. The holotype is composed of a

partial skeleton, including cranial, axial and appendic-

ular elements, with a unique set of characters that dis-

tinguish it from all known theropods. In particular,

Koleken inakayali shows several anatomical traits,

especially in the skull bones, that differentiate it from

Carnotaurus sastrei, the only other abelisaurid known

from the La Colonia Formation. In order to determine

the phylogenetic relationships between this new speci-

men and other abelisaurids, we carried out multiple

phylogenetic analyses using a revised version of a

2D. Pol et al. / Cladistics 0 (2024) 1–50

10960031, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/cla.12583 by CochraneArgentina, Wiley Online Library on [21/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

previously published dataset (Baiano et al., 2022). In

addition to discussing the signiﬁcance of the specimen,

we used the newly inferred phylogenies to explore rates

of morphological change among ceratosaurs to better

understand their evolutionary dynamics.

Institutional Abbreviations: CCG, Chengdu College

of Geology, Chengdu, China; MNN, Mus´

ee National

Boubou Hama, Niamey, Niger; MPCN, Museo

Patag´

onico de Ciencias Naturales, General Roca, Rı´o

Negro, Argentina; MPEF, Museo Paleontol ´

ogico Egi-

dio Feruglio, Trelew, Argentina; MPM, Museo Padre

Molina, Rı´o Gallegos, Santa Cruz, Argentina;

UNPSJB, Universidad Nacional de Patagonia “San

Juan Bosco,” Comodoro Rivadavia, Argentina.

Materials and methods

The specimen MPEF-PV 10826 is composed of cranial, axial and

appendicular elements, including almost complete right and fragmen-

ted left maxillae, fragments of the left and right nasals, a partially pre-

served right postorbital, left and right squamosals, fragments of the left

and right frontals, posterior portion of the left and right parietals

(fused), an incomplete atlas, the ﬁfth to twelfth dorsal vertebrae, a

complete sacrum, the ﬁrst to ﬁfth caudal vertebrae and two middle and

one posterior caudal vertebrae, both ilia, partially preserved right and

left pubes and ischia, complete right and fragmented left femora,

almost complete right tibia and fragments of the left one, complete

right ﬁbula, right astragalus, right distal tarsal IV, ﬁrst to fourth right

metatarsals and second and third left metatarsals (incomplete), left and

right phalanges I-1, right phalanx II-1, left and right phalanges II-2,

right ungual phalanx II-3, left and right phalanges III-1, right phalanx

III-3, left ungual phalanx III-4 and right phalanx IV-5.

The specimen MPEF-PV 10826 was found in Maastrichtian layers

of the Upper Cretaceous La Colonia Formation of Chubut Province,

Argentina (Fig. 1). Chronostratigraphic data for the La Colonia

Formation were provided by Clyde et al. (2021), who dated the

underlying Puntudo Chico Formation. Through radioisotopic

uranium–lead (U–Pb) dating, they found an age of 71.7 Ma for this

unit, suggesting a younger age for the La Colonia Formation.

This result is also conﬁrmed by palynological data owing to the pres-

ence of Quadraplanus brossus, which seems to be restricted to the

Maastrichtian in several Gondwanan localities (e.g. Australia, New

Zealand, Antarctica) (Clyde et al., 2021). Finally, magnetostrati-

graphic data conﬁrmed that part of the La Colonia Formation unit

was deposited during Chron C29r, an interval of reversed polarity,

containing the K/Pg boundary (Clyde et al., 2021).

There are different hypotheses about the depositional environ-

ments for this lithostratigraphic unit (Clyde et al., 2021). It has been

interpreted as a succession of facies from base to top, from ﬂuvial,

estuarine tidal ﬂat or coastal plain to upper tidal ﬂats (Pascual

et al., 2000). C´

uneo et al. (2013,2014) suggested an origin related to

clastic coastal plains bathed by shallow seas and development of bar-

rier island/lagoon complexes. Gasparini et al. (2015) interpreted the

area near where MPEF-PV 10826 was found as being part of a

brackish to freshwater estuary.

The palaeontological record of the La Colonia Formation is rich in

vertebrates and plants, and known since the 1980s. Indeed, the ﬁrst ver-

tebrate described from the La Colonia Formation was the iconic abeli-

saurid theropod Carnotaurus sastrei (Bonaparte, 1985). Yet, it was

only in the last 20 years that periodic palaeontological expeditions

yielded several discoveries that enriched the fauna and ﬂora of this

lithostratigraphic unit (Sterli et al., 2021). So far, the fauna of the La

Colonia Formation is composed of ﬁshes (Apesteguı´a et al., 2007); tur-

tles (e.g. Sterli and de la Fuente, 2013; Gasparini et al., 2015; Orioza-

bala et al., 2020); snakes (Albino, 2000;G

omez et al., 2019);

plesiosaurs (e.g. O’Gorman et al., 2013a,2013b; Gasparini

et al., 2015); ornithischian (Gasparini et al., 2015), sauropod (Gaspar-

ini et al., 2015;P

erez-Moreno et al., 2024) and theropod (Bona-

parte, 1985; Lawver et al., 2011; Gasparini et al., 2015) dinosaurs; and

mammals (e.g. Rougier et al., 2009,2021; Harper et al., 2019). The

palaeontological record of the macro- and microﬂora includes espe-

cially aquatic ferns and angiosperms, but other plants as well as micro-

phytes and dinoﬂagellate cysts have been reported also (e.g. C ´

uneo

et al., 2014; Gallego et al., 2014; Gandolfo et al., 2014; Guler

et al., 2014; Borel et al., 2016; De Benedetti et al., 2018).

Comparative anatomy

Vertebral laminae and fossae were described using the nomencla-

ture proposed by Wilson (1999,2012) and Wilson et al. (2011). We

used terminology proposed by Hendrickx and Mateus (2014a)to

describe the anatomical traits of the maxilla.

Bone histological evaluation

A histological analysis was conducted to determine the ontoge-

netic stage and minimum age of MPEF-PV 10826. A transverse sec-

tion was obtained from the midshaft of the right tibia. The thin

section was prepared at the Paleohistological Laboratory of the

Museo Provincial Carlos Ameghino (Cipolletti, Argentina), using

standard methods outlined by Cerda et al. (2020), and analysed

using a petrographic polarizing microscope (Leica DM750P). The

nomenclature and deﬁnitions of structures used in this study are

derived from Francillon-Vieillot et al. (1990) and de Buffr´

enil and

Quilhac (2021). For minimum age estimation, we counted the num-

ber of lines of arrested growth (LAGs). Double LAGs are here con-

sidered to represent a single annual growth cycle.

Phylogenetic dataset

The data matrix used for the analyses is composed of 245 charac-

ters and 47 taxa (after Carrano and Sampson, 2008; Pol and Rau-

hut, 2012; Rauhut and Carrano, 2016; Langer et al., 2019; Baiano

et al., 2020,2021,2022; Ibiricu et al., 2021; Cerroni et al., 2022).

Characters 29, 30, 34, 116, 118, 141, 159, 170, 174, 185, 191, 204,

207 and 222 were treated as ordered. We deleted the character

describing the orientation of ischial peduncle of ilium (ch. 172 in Pol

and Rauhut, 2012) from the dataset, as its state 1 (vertically

directed) is only present in Carnotaurus and Genusaurus; however,

because the ischial peduncle was completely restored with plaster in

the former taxon, the vertical condition could be an autapomorphy

of the French abelisaurid (Supporting Information and Koleken_

data_matrix.nex in Supplementary Data).

Parsimony analyses

In order to determine the phylogenetic relationships between Kole-

ken inakayali and other abelisaurids, we carried out parsimony ana-

lyses using the program TNT v.1.6 (Goloboff et al., 2008; Goloboff

and Morales, 2023). The heuristic search was performed under

equally weighted parsimony, using new technology searches of up to

100 independent hits, followed by TBR branch swapping until the

tree buffer was ﬁlled (at 100 000 trees). We carried out the IterPCR

procedure (Pol and Escapa, 2009; see iterpcr.out in Supplementary

Data) as implemented by Goloboff and Szumik (2015) in TNT in

order to detect unstable taxa. Support values were evaluated in TNT

using Bremer support, jackknife and nonparametric bootstrap (BS).

D. Pol et al. / Cladistics 0 (2024) 1–50 3

We also explored the use of the pcrjak script (Pol and Golob-

off, 2020; see pcrjak.run and pcrjak.out in Supplementary Data)to

detect taxa that decrease support values as a consequence of their

instability in jackknife pseudoreplicates.

Parsimony analysis of evolutionary rates

We implemented a new custom TNT script (see evolrates.run

in Supplementary Data) to assess rates of morphological change based

on the parsimony optimization of characters. The TNT script uses all

trees in memory and takes input from a simple text ﬁle named fadlad

(see Supplementary Data) that contains the minimum age and maxi-

mum age for each terminal taxon (Table S2). Rates are calculated by

evaluating optimization on trees that are calibrated against geological

time by setting the minimum age of divergence for each node equal to

the oldest maximum age of the terminal taxa descended from that

node. All branches that are zero-length in temporal duration are

assigned the minimum branch duration (1 Myr). Trees were evaluated

individually but the output was also produced based on the strict con-

sensus of these topologies. A random selection of 1000 MPTs was

selected to calculate the evolutionary rates for this analysis. The parsi-

mony optimization of characters is used to determine the number of

changes occurring along each of the branches of the trees. The rate of

morphological change was obtained by dividing the number of changes

by the temporal duration of each branch. The script produces a text

table output that includes the average rate of character change for each

branch as well as a graphical output with a colour-coded scheme of

rates scaled between the maximum and the minimum rate obtained for

the branches of the tree (blue =low rate; red =high rate).

Ambiguous optimizations frequently occur in parsimony analyses,

particularly when fossils contain large amounts of missing data and

create uncertainty in the placement of character transformations on

the phylogenetic tree (Goloboff, 2022). As a result of the prevalence

of these ambiguous optimizations, the script evolrates.run conducts

iterative evaluations of alternative most parsimonious reconstructions

for each character and calculates the average number of reconstruc-

tions that optimize a transformation along each branch for each char-

acter. For example, a typical instance of ambiguous optimization is

the ACCTRAN/DELTRAN scenario (Swofford and Maddi-

son, 1987), which presents just two alternative parsimony reconstruc-

tions. In such scenarios, each branch affected by an ambiguous

transformation is assigned 0.5 steps for the rate analysis. The script

evolrates.run also computes two stratigraphic ﬁt measures, as outlined

by Pol and Norell (2001; MSM*) and Wills (1999; GER).

Bayesian analyses used to estimate the relaxed clock

priors

Using the program RevBayes v.1.2 (H ¨

ohna et al., 2016), we fur-

ther conducted a Bayesian tip-dating analysis under the fossilized

birth–death process (Stadler, 2010; Gavryushkina et al., 2014; Heath

et al., 2014). Bayesian tip-dating combines both character data and

stratigraphic age information to jointly infer tree topology, diver-

gence times and evolutionary rates (Wright et al., 2022). Here, we

used partitioned relaxed clocks to estimate the rate of morphological

evolution separately for four different anatomical regions (the skull,

axial skeleton, forelimb and the hind limb).

In order to derive informative hyperpriors for the relaxed clock

model, we followed the protocol of Ronquist et al. (2012a) and per-

formed a series of preparatory analyses, starting with the inference of

a topologically unconstrained nonclock tree. To this end, we employed

a hierarchical gamma-Dirichlet prior (Rannala et al., 2012; Zhang

et al., 2012) on branch lengths, which ﬁrst speciﬁes a gamma hyper-

prior on overall tree length and then apportions this total length

Provincial Road

REFERENCES

La Colonia Formation

(a)

67º45′

Co. El Buitre

Arroyo Mirasol Chico

67°30′

BAJADA DEL

DIABLO

BAJADA

MORENO

Sierra de La Colonia

(b)

5 km

100 km

43°00′

68º00′

Co. Bayo

Sierra de los Tehuelches

Co. El

Buitre Chico

CHUBUT

PROVINCE

SOUTH

AMERICA

ARGENTINA

Somuncurá

Co. Huichilan

AJA

Fig. 1. Locality map showing where the type specimen of Koleken inakayali MPEF-PV 10826 was found (black star) within the La Colonia For-

mation of Chubut Province, Argentina. Figure modiﬁed from Sterli et al. (2021).

4D. Pol et al. / Cladistics 0 (2024) 1–50

among individual branches according to a Dirichlet prior. Following

the recommendations of Zhang et al. (2012), we set the shape (α)

parameter of the gamma distribution to 1, and scaled the rate

parameter (β) to obtain an expected prior mean length of 2.5 sub-

stitutions per character, close to the value suggested by the parsi-

mony analysis (see Results and discussion). The concentration

parameter of the Dirichlet distribution was set to 1 (ﬂat prior), and

no distinction was made between internal and terminal branches.

The dataset was partitioned ﬁrst into ordered and unordered char-

acters as described above, and further by the number of observed

states to ensure that each partition received an instantaneous rate

matrix (Q) of appropriate dimensions. In total, four partitions were

speciﬁed; for the three partitions comprised of unordered charac-

ters, the substitution process was modelled using the Mkmodel of

Lewis (2001). Because all 245 characters included in the matrix

were parsimony-informative, we applied the appropriate ascertain-

ment bias correction (Allman et al., 2010); however, we note that

this correction cannot be easily extended to multistate characters

(Matzke and Irmis, 2018) and its implementation in RevBayes

(coding =“informative”) should be regarded as experimental (Sebas-

tian H¨

ohna, pers. comm.). A single discrete gamma model with

four rate categories (Yang, 1994) was linked across partitions to

account for among-character rate variation (ACRV), with the

shape parameter drawn from a diffuse Unif(0, 10

) prior. Two

Markov chain Monte Carlo (MCMC) runs were performed, with a

burnin period of 80 000 iterations and a sampling period of 240 000

iterations. Under the default settings in RevBayes, each MCMC

iteration consists of a number of moves (equal to the sum of move

weights), as opposed to other Bayesian phylogenetic software pack-

ages such as MrBayes (Ronquist et al., 2012b) or BEAST 2

(Bouckaert et al., 2019), where each move is counted as a separate

iteration (Turelli et al., 2018). Accordingly, our sampling period

corresponded to 39 768 000 individual moves (=MrBayes or

BEAST iterations).

Convergence of scalar parameters was assessed by visual inspec-

tion in the program Tracer v.1.7 (Rambaut et al., 2018). To assess

topological convergence, we used the package rwty (Warren

et al., 2017) for the R statistical computing environment (R Core

Team, 2022) to calculate the average standard deviation of split fre-

quencies (ASDSF) and approximate topological effective sample size

(Lanfear et al., 2016) based on path distances (Steel and

Penny, 1993), ensuring that the former was <1.01 and the latter was

>625 (Fabreti and H¨

ohna, 2021). The pooled posterior sample from

both runs was summarized using the maximum clade credibility

(MCC) tree, which was then compared to the reduced consensus tree

from the parsimony analysis using Robinson–Foulds (RF) distances

(Robinson and Foulds, 1981) with a polytomy correction following

Huerta-Cepas et al. (2016), as implemented in a custom R function

employing the TreeDist package (Smith, 2020).

In order to obtain informative hyperpriors on the mean and vari-

ance of branch rates, we followed Ronquist et al. (2012a)

and sampled nonclock as well as strict-clock branch lengths in units

of expected substitutions per character (s/c) on a ﬁxed topology cor-

responding to the MCC tree. In contrast to the scenario explored by

Ronquist et al. (2012a), in which these analyses were restricted to

the extant portion of the tree, our taxon sample consists exclusively

of fossils, and the strict-clock analysis therefore had to account for

the different ages of individual tips in order to yield interpretable

results. Accordingly, instead of using arbitrary time units, we ﬁrst

estimated branch durations in units of Myr and subsequently multi-

plied them by the mean posterior clock rate (in units of expected

substitutions per character per Ma, s/c/Ma) to recover branch

lengths in units of s/c, directly comparable to the nonclock estimates.

This approach required conducting a full tip-dating analysis under

the fossilized birth–death (FBD) tree prior (Gavryushkina

et al., 2014), which only differed from the ﬁnal analysis by employ-

ing a ﬁxed topology and a strict rather than relaxed clock.

The rates of speciation (λ), extinction (μ) and fossil sampling (ψ)

were assumed to be constant through time and reparameterized in

terms of the net diversiﬁcation rate d=λμ, extinction fraction r

=μ/λand fossil recovery probability s=ψ/(μ+ψ). We placed an

Exp(10) prior on d, a Beta(2, 1) prior on r, which favoured high

values in accordance with Marshall’s (2017) “third law of paleobiol-

ogy” (λtends to equal μwhen averaged over time), and a Unif(0, 1)

prior on s. The extant sampling fraction (ρ) was ﬁxed to 0. The root

age was drawn from an exponential prior with an offset of 197.5 Ma

(the minimum plausible age of Saltriovenator, which represents the

oldest currently known ceratosaur; Dal Sasso et al., 2018) and a rate

of 0.06623. This ensured that 95% of the total probability mass was

placed on ages younger than 212.6 Ma, the maximum age of the

Hayden Quarry (Petriﬁed Forest Member, Chinle Formation) at

Ghost Ranch according to Irmis et al. (2011). The Hayden Quarry

represents the youngest assemblage containing a diverse theropod

fauna but no ceratosaurs, suggesting that the clade’s origin postdates

its deposition. Following Barido-Sottani et al. (2019), tip dates were

assigned uniform priors spanning their respective uncertainty ranges.

The topology was ﬁxed to that of the 47-tip nonclock MCC tree

after stripping it of the four outgroups (Herrerasaurus,“Syntarsus”,

Dilophosaurus,Allosaurus). Outgroups tend to be sampled less

densely than the ingroup, and as such may result in bias in the esti-

mates of d,rand s, which could propagate downstream to the

parameters of interest (topology, divergence times, evolutionary

rates). The removal of outgroups rendered 33 of the original 245

characters constant; we chose to remove these from the dataset.

Another 42 characters were rendered parsimony-uninformative,

motivating a change in the ascertainment bias correction (from

coding =“informative” to coding =“variable”). The global clock rate

was drawn from a lognormal prior with a mean of 0.005 s/c/Ma and

σset to 1.17481 so as to ensure that the 95% highest prior density

interval would span exactly two orders of magnitude. Two runs were

performed, each with a burnin period of 90 000 iterations (12 510 000

moves) and a sampling period of 287 500 iterations (39 962 500

moves). The pooled posterior sample was summarized as an MCC

tree with mean node heights (Fig. S6).

The ﬁxed-topology nonclock analysis was performed on the full

MCC tree (including the outgroups) to ensure that the ingroup

topology was rooted, and thus directly comparable to the rooted

strict-clock tree. We used the same gamma-Dirichlet branch length

prior as in the unconstrained analysis. Two runs were carried out,

each with a burnin period of 112 500 iterations (12 498 750 moves)

and a sampling period of 360 000 iterations (39 996 000 moves).

After verifying that the effective sample sizes (ESS) of all scalar

parameters exceeded 625, we summarized the posteriors yielded by

the strict-clock and nonclock ﬁxed-topology analyses using MCC

trees. After pruning the outgroups from the nonclock tree, we

regressed the nonclock branch lengths against the strict-clock branch

durations, obtaining a slope of 7.5 ×10

4

s/c/Ma. This value was

used as the mean of a diffuse lognormal hyperprior on the mean

clock rate, again parameterized so that the 95% highest prior density

interval spanned exactly two orders of magnitude. Next, we multi-

plied the strict-clock branch durations by the estimated clock rate to

convert them into branch lengths, calculated the squared differences

between the strict-clock and nonclock branch lengths, and regressed

these against the strict-clock branch lengths. Following Ronquist

et al. (2012a), we used the slope of the resulting regression line as

the median of an exponential hyperprior on the variance-scaling

hyperparameter of the relaxed clock model (Fig. S7).

Tip-dating analyses

The main tip-dating analysis was conducted on the same dataset

as the strict-clock analysis (43 taxa and 212 characters; i.e. without

outgroups or constant characters) and under the same FBD prior.

D. Pol et al. / Cladistics 0 (2024) 1–50 5

We chose to run the main analysis without sampled ancestors, which

are unlikely to occur in datasets such as ours with a sparse sampling

of temporally and geographically disparate taxa (Sim ˜

oes et al., 2020;

Cern´

y et al., 2022). However, we also performed a sensitivity analy-

sis that allowed for sampled ancestors (Table S5 and Fig. S9). The

analysis employed the white noise clock model of Lepage

et al. (2007), as parameterized by Zhang and Wang (2019) and previ-

ously implemented in RevBayes by Sz ¨

oll˝

osi et al. (2022). For the

purposes of the clock model, we deﬁned four anatomy-based parti-

tions: the skull (ch. 1–95), axial skeleton (ch. 96–150, 235–245), fore-

limb (ch. 151–178) and the hind limb (ch. 179–234). All four

partitions drew their respective branch rates from gamma distribu-

tions with a shared mean cand (potentially distinct) variances σ2

...,σ2

4, and individual branch rates were reparameterized as products

of the mean clock rate cand relative rates (multipliers) r

,...,r

2n2

The relative rate r

for the i-th branch and the j-th partition then fol-

lows a new Γ(α,β) distribution with α=β=cti

σ2

, where t

is the dura-

tion of the i-th branch in calendar time, so that the mean α

β¼1 and

variance α

β2¼

σ2

cti.

We explored two different variations on this basic model: one in

which the variance scalars were linked across partitions (σ2

1=σ2

σ2

3=σ2

4) and a single set of branch rates was applied to all four ana-

tomical regions, and another one in which the variance scalars were

unlinked (that is, treated as i.i.d. draws from a common hyperprior)

and each partition received a separate set of branch rates. In both

cases, the clock model hyperparameters were drawn from the follow-

ing hyperpriors, derived using the approach described above: c

LogNorm(μ=7.88553, σ=1.17481); σ2

1,σ2

2,σ2

3,σ2

4Exp(3.24339).

We compared the linked and unlinked clock models using Bayes fac-

tors (BF; Kass and Raftery, 1995), computed from marginal likeli-

hoods estimated using the stepping-stone approach of Xie

et al. (2011). This method samples from a series of unnormalized

“power posterior” distributions of the form f

(prior) ×(likelihood)

, constructed along the path between the prior

(β=0) and the posterior (β=1). After a pre-burnin period of

200 000 (linked model; 50 400 000 moves) or 100 000 (unlinked

model; 51 600 000 moves) iterations, we initialized 128 βvalues, tak-

ing advantage of a recently introduced RevBayes routine for fully

parallelized power posterior sampling (H ¨

ohna et al., 2021). Each

power posterior was run for 6200 (linked model; 1 562 400 moves) or

3050 (unlinked model; 1 573 800 moves) iterations, with a burnin

fraction of 50%. The results were summarized using the 2 log BF

statistic (twice the difference of the log marginal likelihoods) and

interpreted following the guidelines of Kass and Raftery (1995).

The ﬁnal tip-dating analysis was conducted under the preferred

clock model. Because the clock-model partitions based on anatomy

intersected with the substitution-model partitions based on character

type and state number, the overall model consisted of 12 distinct phylo-

genetic continuous-time Markov chains. We performed four MCMC

runs, each with a burnin period of 100 000 iterations (51 600 000

moves) and a sampling period of 400 000 iterations (206 400 000

moves). Throughout the analysis, the partition-speciﬁc branch rates

were logged on each sampled tree. After checking for convergence (sca-

lar ESS >350, approximate topological ESS >625, ASDSF <0.01),

we processed the pooled posterior sample using TreeAnnotator v.2.7.6

(Bouckaert et al., 2019) to generate an MCC tree with median node

heights and annotations for branch rate summary statistics.

Systematic palaeontology

DINOSAURIA Owen, 1842.

THEROPODA Marsh, 1881.

CERATOSAURIA Marsh, 1884.

ABELISAURIDAE Bonaparte and Novas, 1985

BRACHYROSTRA Canale et al., 2009

FURILEUSAURIA Filippi et al., 2016

Koleken gen. n.

(LSID urn:lsid:zoobank.org:act:8E384B7B-6731-

40DF-B65B-035AD7DB753B)

Type species. Koleken inakayali, by monotypy.

Derivation of name. Adapted from K ´

oleken, a name

in Teushen language spoken by the native population

of central Patagonia that means “coming from clay

and water”, given the specimen was found in a

sedimentary section dominated by claystone

representing an estuarine environment.

Diagnosis. As for the type and only species.

Koleken inakayali sp. n.

(LSID urn:lsid:zoobank.org:act:A147BAD3-87F0-

403F-A1C0-2067D0B11169)

Derivation of name. Honouring Inakayal, one of the

last chiefs of Tehuelches, native people from central

Patagonia. He is known for his resistance against

Argentina’s Conquest of the Desert military campaign,

which resulted in the decimation and displacement of

native communities from Patagonia. After his capture

and eventual death in 1888, Inakayal’s skeleton was

stored at the La Plata Museum Anthropology collection

but in 1994 his skeleton was respectfully restituted in its

native place and buried by his people near the town of

Tecka, in central Patagonia (Chubut Province,

Argentina).

Holotype. MPEF-PV 10826. The specimen includes

closely associated (but disarticulated) remains of the

skull and atlas, as well as the articulated postcranial

skeleton composed of the posteriormost eight dorsal

vertebrae, a complete sacrum, eight caudal vertebrae,

an almost complete pelvis and hind limbs. Select

measurements are provided in Table S1.

Locality and horizon. The specimen was found at

the Cerro Bayo Norte area, east of the Sierra de La

Colonia, centre north of Chubut Province, Argentina

(Fig. 1). Precise geographical provenance is deposited

at the MPEF collection. The specimen was found close

to the base of the stratigraphic section of the La

Colonia Formation that crops out in this region (see

Gasparini et al., 2015: Fig. 2). The age of the base of

this unit has been recently restricted to the early

Maastrichtian (Clyde et al., 2021), constraining the age

of the new taxon to the Maastrichtian.

6D. Pol et al. / Cladistics 0 (2024) 1–50

Diagnosis. Koleken inakayali is a brachyrostran

abelisaurid different from other theropods in having

the following set of autapomorphies (marked with

a*): (i) medially smooth paradental plates; (ii) dorsal

surface of the nasal with a row of foramina orientated

obliquely with respect to the longitudinal skull axis;

(iii) *anterior ramus of the postorbital lacking the

lateral wall reﬂected by a dorsoventral height less than

half of its anteroposterior length, which makes the

orbital surface face ventrally instead ventromedially;

(iv) *dorsal surface of the postorbital with a shallow

depression facing dorsolaterally, located at the

junction between anterior and posterior rami, and

bounded anteriorly by a well-deﬁned ridge; (v) parietal

with a mediolaterally concave dorsal surface and

lacking a knob-like dorsal projection and; (vi) ventral

process of the squamosal slender and thorn-like,

lacking a kink on the anterior margin; (vii) *dorsal

neural spines in lateral view bearing an elevated rim

along the anterior, dorsal and posterior margins of the

neural spine; (viii) margin of the proximal articular

surface of tibia connecting lateral and medial condyles

obliquely orientated, facing posterolaterally.

Differential diagnosis. MPEF-PV 10826 shows

further anatomical differences from several Cretaceous

abelisauroids, especially the sympatric Carnotaurus

sastrei Bonaparte, 1985. The angle between the

premaxilla and maxilla rami of the nasal is lower than

in Carnotaurus,Majungasaurus and Aucasaurus. The

Fig. 2. Right and left maxillae of Koleken inakayali MPEF-PV 10826. Right (a, b) and left (c, d) elements in (a, c) lateral and (b, d) medial

views. aof, antorbital fossa; csj, contact surface for the jugal; csp, contact surface for the palatine; ip, interdental plate; nvf, neurovascular fora-

men; p, protuberance; pp, paradental plate; rt, replacement tooth. Scale bar: 5 cm.

D. Pol et al. / Cladistics 0 (2024) 1–50 7

proximal tip of the lesser trochanter ends proximally

to the ventral margin of the femoral head (ends below

in Carnotaurus,Ekrixinatosaurus and Aucasaurus), the

distal femur lacks a well-deﬁned transverse ridge

across the infrapopliteal fossa (present in Carnotaurus,

Aucasaurus,Eoabelisaurus,Xenotarsosaurus and

Quilmesaurus), the distal end of metatarsal II has a

faintly marked extensor fossa in which the distal

margin is poor developed (more conspicuous in

Eoabelisaurus,Aucasaurus,Majungasaurus,

Masiakasaurus and Skorpiovenator), metatarsal III has

a poorly developed ventrolateral corner of the

proximal articular surface (prominent in Aucasaurus or

Elaphrosaurus), metatarsal IV is slightly shorter than

metatarsal II, curving laterally (almost straight or

barely curved in noasaurids and in Aucasaurus and

Skorpiovenator).

Description

Skull

General information. The skull is represented by six

bones: partially preserved right maxilla, fragmentary

left maxilla, fragmentary left and right nasals, partial

right postorbital, partial right frontal, partially

preserved left and right squamosals, and fragmentary

fused parietals (Figs 2–4).

Maxilla. Koleken includes the partially preserved

right and left maxillae that represent the entire

morphology of the maxilla, except for the ascending

ramus and the posteriormost tip of the bone (Fig. 2a–d).

A bone fragment with an L-shaped cross-section, rugose

lateral surface and a broken ventral surface is probably

from the left maxilla. The left maxilla is preserved as the

ventral rim of the antorbital fossa, which is missing in

the right maxilla. The ventral rim of the body is slightly

convex, as in all abelisaurids and in most theropods

(Fig. 2a,b). The lateral surface is ﬂat and ornamented

with a dendritic pattern of vascular grooves orientated

dorsoventrally, with some containing ventrally directed

foramina (Fig. 2a,c). This pattern of ornamentation

is observed in all abelisaurids where the maxilla is

known (e.g. Aucasaurus,Carnotaurus,Majungasaurus,

Skorpiovenator and Spectrovenator). However, Koleken

lacks the curved grooves observed in the middle and

posterior portion of the maxillary body present in

Rugops,Tralkasaurus and the indeterminate abelisaurid

UNPSJB-PV247. Two small fragments with a similar

lateral surface that medially preserve the broken rim of

alveoli are probably from the left maxilla. The anterior

rim of the right maxilla preserves a notch for the

articulation of the maxillary ramus of the nasal, just

below the base of the anterior ramus which is distally

broken as in Carnotaurus,Kryptops,Majungasaurus,

Rugops and the indeterminate abelisaurid UNPSJB-

PV247. The left maxilla is represented by a portion of

the ventral margin of the antorbital fossa (Fig. 2c).

Laterally, this margin is smooth but punctuated with a

small pointed protuberance that marks the anterior end

of the contact surface for the jugal. The contact surface

for the jugal forms a shelf, which is smooth and inclined

dorsally at almost 45°,asinCarnotaurus and

Skorpiovenator (Fig. 2a). Medially, the ventral margin

of the antorbital fenestra is smooth, rounded and

stepped above the ventral surface of the antorbital

fossa.

The medial surface of the maxilla is relatively smooth

with only a faint texture and a very small number of

posteroventrally directed foramina (Fig. 2b,d). By con-

trast, most abelisaurids have a rugose texture in this

region comprising dorsoventrally-orientated ridges,

many of which contain ventrally directed foramina (e.g.

Carnotaurus,Llukalkan,Majungasaurus,Rugops,

Tralkasaurus and the indeterminate abelisaurid

UNPSJB-PV247). The absence of these ridges and of a

rugose surface is here considered as an autapomorphic

trait of Koleken within the context of Abelisauridae.

Posteriorly, the contact surface for the palatine is poorly

preserved. There are at least 12 maxillary teeth based on

preserved alveoli. The distalmost preserved alveolus is

partially broken, but given its small size it probably rep-

resents the last maxillary tooth position equating to a

maxillary tooth count of 12. This number of teeth is

shared with Carnotaurus, while Aucasaurus (14 or 15),

Ekrixinatosaurus (16+), Majungasaurus (17) and Skor-

piovenator (16+) have a higher number of teeth. The

maxillary teeth of Koleken are replacement teeth and

have the general form typical of abelisaurids, showing

moderately laterally compressed crowns with curved

serrated mesial carina and a tooth apex close to the level

of the straight serrated distal carina (Hendrickx

and Mateus, 2014b; Hendrickx et al., 2020; Meso

et al., 2021). There is a longitudinal row of pits on the

medial surface of the maxilla. These pits are associated

with tooth development (Carrano and Sampson, 2008),

and replacement teeth are visible within them. The par-

adental and the interdental plates are all fused, forming

a medial wall and the separation among the alveolus.

Nasal. The nasals are preserved as fragments of the

left and right anterior rami, a large fragment of

the posterior portion of the right nasal and a smaller

fragment of the posterior portion of