Relationships of † Codoichthys carnavalii Santos , 1994 ( Teleostei , Clupeomorpha , † Ellimmichthyiformes ) from the Late Aptian of São Luís-Grajaú Basin , NE Brazil

†Codoichthys carnavalii is a clupeomorph fish only found in calcareous concretions of Codó Formation, State of Maranhão. It is known based on three specimens housed in the paleontological collection of the Museu de Ciências da Terra of Departamento Nacional da Produção Mineral, Rio de Janeiro. It was omitted in most of recent cladistic analyses about clupeomorphs. We revisited its anatomy furnishing new data and additional restorations. Furthermore we explored the relationships of †Codoichthys with the computer program TNT based on a matrix with 30 taxa and 60 unordered and unweight characters. Elops was used to root the tree. The strict consensus was obtained from three shortest trees (L=181; CI=0.387; RI=0.632). The monophyly of †Ellimmichthyiformes is supported by a sigmoid cleithrum and an uniquely derived predorsal scute series. Most of subgroups showed low support indices. †Sorbinichthyidae and †Horseshoeichthys appear in the most basal position, and not closely related to †Armigatus or †Diplomystus. A †Diplomystus clade is more advanced than †Armigatus and sister-group of remaining †ellimmichthyiforms. Within †Paraclupeidae, †Codoichthys is sister-group of remaining †paraclupeids (including †thorectichthyines and †paraclupeines). Within †Paraclupeinae, †Triplomystini includes a †Triplomystus clade, a sister group of †Rhombichthys plus †Tycheroichthys, and †Paraclupeini with †S. itapagipensis and all other †ellimmichthyiform taxa.


INTRODUCTION
Clupeomorpha is a great radiation of teleostean fishes that produced two major lineages: Clupeiformes and †Ellimmichthyiformes (Grande 1985).The former consists of 84 genera (Nelson 2006) and 397 living species (Lavoué et al. 2014) of fishes popularly known as herrings, sardines and anchovies together with many fossil taxa known since the Barremian (Figueiredo 2009).The latter is an archaic group of double-armored herringlike fishes including about 13 genera and at least 35 species found in many marine, estuarine, and freshwater deposits around the world ranging from the Hauterivian-Barremian (Santos and Corrêa 1985) to Middle Eocene (Patterson 1993).Grande (1982Grande ( , 1985) ) was the first to perceive the existence of †Ellimmichthyiformes.He intuitively indicated monophyly of the group on the basis of derived presence of subrectangular predorsal scutes.Unlike ordinary scales, he noted that these unpaired scutes are positioned at the dorsal midline and they are heavily ossified, often covered with cycloid scales.He also pointed out that many taxa of distant lineages of teleosts bear some type of dorsal scutes (e.g., many aulopiforms) and these scutes are commonly preserved in fossil fishes unlike true scales.
At first only †Diplomystus and †Ellimmichthys were placed within †Ellimmichthyiformes (Grande 1982) that also originally contained a single family (i.e., †Ellimmichthyidae).Afterwards, due to nomenclatural priority, †Paraclupeidae was indicated by Chang and Grande (1997) as replacement name for this family.Since then new diagnoses, definitions and compositions of putative monophyletic subunits were proposed, and new non-clupeiform taxa coming from many fossil localities throughout the world were assigned to †Ellimmichthyiformes.Many phylogenetic analyses were carried out (e.g., Chang and Maisey 2003, Zaragueta-Bagils 2004, Alvarado-Ortega et al. 2008, Murray and Wilson 2013) but until now the monophyly of the group remains uncertain as well as its content.
The fish fauna from Reconcavo Basin is mainly distributed in outcrops of Santo Amaro Group, particularly in the Itaparica, Candeias and Maracangalha formations (Carvalho and Figueiredo Souto 2005).†E.longicostatus was studied by Cope (1886) after material collected by the geologist Joseph Mawson in localities near Salvador and Simões Filho (Carvalho and Figueiredo Souto 2005).Santos (1949) stated that †Ellimmichthys longicostatus was collected together with remains of †Lepidotes, †Cladocyclus, †Calamopleurus and †Mawsonia by the paleontologist Lewellyn Ivor Price and Abel Oliveira, in Ilha de Itaparica.More recently, †E.longicostatus was collected in shales of Maracangalha Formation (Carvalho and Figueiredo Souto 2005).
Chang and Maisey (2003) while revisiting †Ellimma branneri considered only †E.branneri and †E.cruzae (=E.cruzi) as valid species for this genus.They indicated that †Ellimma branneri only partially agree with the diagnosis of †Paraclupeidae because anterior dorsal scutes are longer than broad unlike the posterior ones, broader than long.Thus they suggested a reassessment of characters supporting the monophyly of this family.
At first †E.cruzae (=E.cruzi) was identified as †Ellimmichthys longicostatus (Costa et al. 1979) but Santos (1990) assigned it to †Ellimma and named †Ellimma cruzi in honor of the paleontologist Norma Maria da Costa Cruz, from the staff of DNPM, Rio de Janeiro.We note the ending of the specific epithet is an incorrect original spelling based on an inadvertent mistake (ICZN 1999: art. 32.5.1), and should be corrected to the feminine †E.cruzae.Murray and Wilson (2013) based on 24 taxa and 62 characters furnished the most recent phylogenetic review on †Ellimmichthyiformes.They divided the group in two major clades: †Armigatoidei (for †Armigatus and †Diplomystus) and †Ellimmichthyoidei (for remaining taxa).They tentatively named many subunits excluding or adding taxa.Thus †Paraclupeidae was divided in five subfamilies: †Scutatuspinosinae, †Thorectichthyinae, †Ellimminae, †Ellimmichthyinae, and †Paraclupeinae.Notwithstanding the relevance to the knowledge of the relationships of †ellimmichthyiform fishes, many problems still remain particularly regarding the phylogenetic status of the Brazilian taxa †Ellimma cruzae, †Scutatuspinosus itapagipensis and †Codoichthys carnavalii.
Certainly †Codoichthys carnavalli is one of the most puzzling clupeomorph taxa described from Brazil.At first Santos, in 1945, considered it a species of †Knightia.Afterwards the same author (Santos 1994), based on comprehensive revisions by Grande (1982Grande ( , 1985)), classified it as Clupeomorpha incertae sedis although recognizing putative affinities with non-clupeiform clupeomorphs as †Armigatus, †Ellimmichthys, and †Diplomystus.Since then no rigorous attempt to decipher its relationships was undertaken.
Our goal in this paper is to describe in detail the morphology of †Codoichthys carnavalii as soon as the material permits and exploring its relationships within Clupeomorpha using cladistic methodology.

Material
The specimens of †Codoichthys carnavalli herein studied belong to the paleontological collection of the Museu de Ciências da Terra of the Departamento Nacional de Produção Mineral, Rio de Janeiro, and consists of the type-series (see details in Santos 1994).They are referred with the abbreviation DGM followed by the institutional register number.
According to Santos (1994) the fishes were collected by the late geologists Odorico Albuquerque and Victor Dequech, even belonging to the Divisão de Geologia e Mineralogia of DNPM, in 1945, in the locality of Morro de Umburanas, State of Maranhão, northeastern Brazil (Fig. 1).All specimens examined are poorly preserved and laterally compressed.They are preserved in yellowish calcareous concretions and were previously mechanically prepared with steel needles of different sizes under dissecting microscope by Rubens da Silva Santos.We produced painted silicone peels from the typeseries specimens to enhance anatomical features.Camera lucida drawings were made using a Motic-Quimis stereomicroscope with a drawing-tube attachment.Digital photographs of high resolution were obtained with a USB Camera under MIAS® Software and photograph camera Nikon D7100.For enhancing anatomical details and reducing bright during photograph sessions fossils were coated with a sublimate of magnesium oxide.Anteriormost vertebrae hidden by opercle were counted as three according to Grande (1985).Measurements and meristic counts follow Forey et al. (2003).Proportions are presented as a percentage of the standard length (SL).A "dagger" ( †) preciding taxon indicates that it is known only by fossils.

GeoloGical SettinG
The São Luís -Grajaú Basin is included in the Gurupe Graben System and encompasses an extensive area of about 250,000 km 2 in the centernorth part of the State of Maranhão.It is a typical marginal rift basin, with predominantly Cretaceous filling.Codó Formation is in the lowermost sequence of Cretaceous deposits reaching of about 4,000m thick in the depocenters (Rossetti et al. 2004).This formation was deposited in a pre-rift phase and forms a bedding sequence of about 180 m through an area extending from Araguatins, Tocantins State, to Brejo Municipality, northeastern Maranhão.It is interspersed by Grajaú (below) and Itapecurú (above) formations and constituted mainly of lacustrine dark shales, evaporites and calciferous shales which include calcareous concretions and gypsum lenticular bedding (Santos 1994).There are two facies (Santos and Carvalho 2009): one lacustrine (lower unit), with †Dastilbe elongatus and other fossils also found in the Crato Formation of the Araripe Basin, and other estuarine (upper unit), represented by calcareous concretions from which Santos (1994) pointed out the occurrence of †Codoichthys carnavalii in association to fishes commonly found in Romualdo Member of Araripe Basin, for instance, †Tharrias araripis, †Calamopleurus cylindricus, †Brannerion latum, †Vinctifer comptoni, †Santanichthys diasii, †Rhacolepis buccalis, †Cladocyclus gardneri, and †Araripelepidotes temnurus.
The localities in the State of Maranhão from which Cretaceous fossil fishes were found are plotted in map (Fig. 1).According to Santos (1994) these localities are Morro de Umburanas in Brejo, Codó and Barra do Corda.

cladiStic analySiS
We used cladistic principles (e.g., Wiley and Lieberman 2011, Wheeler 2012) and underlying computational technology to explore the phylogenetic relationships of †Codoichthys carnavalii together with other clupeomorphs.The analysis was performed with TNT computer program version 1.1 (Goloboff et al. 2003) to generate the shortest trees based on a polarized data matrix (Table I) composed of 30 taxa and 60 unordered and unweight selected characters taken from foregoing cladistics analyses of the †ellimmichthyiforms (i.e., Chang and Maisey 2003, Forey 2004, Zaragueta-Bagils 2004, Alvarado-Ortega et al. 2008, Murray and Wilson 2013).Only characters suggestive of immediate common ancestry (synapomorphic) are accepted as criteria for the recognition of monophyletic groups(=clades).We excluded some numerical characters of Alvarado-Ortega et al ( 2008) and Murray and Wilson (2013) because they introduce subjective procedure of cut a continuum when data of new taxa are added to the matrix, always producing new arrangements to accomodate them.Missing characters or unclear states owing to the quality of preservation (including ambiguous data) were coded as "?" in the matrix.ACCTRAN was the optimization choice due to the preservation of primary homology (de Pinna 1991).The tree-building routine was traditional search with random stepwise addition, ten trees held at each iterative step, and TBR (tree bisection and reconnection) branch swapping routine.Branches without support were collapsed.Selected †Ellimmichthyiformes used in analysis are listed below together with literature review from which anatomical data are better known and discussed.Characters have been taken mainly from literature and checked (if possible), and in the case of changes they are indicated in our list (see Appendix 1).Additional fossil specimens housed in institutional collections from which we made original observations or confirmed previous data are also indicated.
The lateral ethmoid (Figs.3b, 4, LET) is represented by two incomplete and imperfect separated portions.One corresponds to a fragment of the main body of left ethmoid lateral which was slightly displaced to occupy the center of a broad gape (a probable condriferous space of nasal pit in life) covered by mesethmoid.Other portion corresponds to a well-developed fan-like shield of perichondral bone associated to a cup-like upper edge (for meeting frontal bone dorsally) of the right ethmoid lateral compounding the anterior limit of the orbit.Ventrally, this bone produces a wing-like outgrowth to touch the shaft of the parasphenoid.
The parasphenoid (Fig. 3b, PAS) is a long, low, and slightly curved bone.Only its orbital portion is visible.It is toothless.Due to state of preservation, we are unable to determine presence of a dermal basipterygoid process.The frontal (Figs. 3b,5,FR) is the largest bone of the cranial roof covering completely the orbit and finishing a little behind it.It is narrower at the level of the ethmoid region and broadens considerably at the posterodorsal orbital corner.Close to the contact zone with parietal this bone is ornamented with coarse rugae.Due to preservation it is difficult to determine presence of fontanels or fossae.The supraorbital sensory canal run bone enclosed so that its presence on surface is hardly noted through a faint tubular relief.Pores are not observed.
The meeting between frontal and parietal is through a long transverse suture.There is an ascending profile anterior to the well-developed supraoccipital crest as in the skull of the unnamed †Diplomystus from the Cenomanian of the English Chalk (see Forey 2004) in contrast with †Ellimma branneri and †Paraclupea.Deep grooves for supraorbital sensory canal separated by a medial bony bridge as seen in the unnamed †Diplomystus from the English Chalk and Scutatuspinosus itapagipensis are lacking.
The parietal (Figs.3b, 5, PA) is a large and subrectangular bone in lateral view.In the skull roof, judged by its placement and position of supraoccipital it meets its partner in the midline as commonly found in well-preserved †ellimmichthyiforms.As for frontal bone, there are rugose ornamentation near lateral border.
The commissural supratemporal sensory canal (Figs. 3b,5,stt.com) is entirely bone-enclosed, running within a curved tubular relief on parietal.The supraoccipital is placed outside this commissure.Pores are not visible on surface.
Like many clupeomorphs, the otico-occipital region is almost equal in size to orbital and ethmoid regions.Unfortunately, autosphenotic, prootic and pterotic bones are not preserved.There are any evidence of recessus lateralis and temporal fossa.Noteworthy the temporal fossa seems to be lacking in all †ellimmichthyiforms, as for the recessus lateralis, so that dermosphenotic are not reduced and opennings for preopercular and infraorbital branches of sensory canal are distantly positioned in otic neurocranium of certain well-preserved specimens (see Patterson 1970, Grande 1982, Forey 2004).
The epioccipital (Fig. 3b, EPO) is almost trapezoid and placed among parietal, supraoccipital, and an uninformative portion of pterotic.Its position in the skull resembles that found in †Triplomystus noorae (see Forey et al. 2003, p. 271, fig. 41).Posteriorly this bone shows a reduced smooth process to receive the upper arm of the posttemporal bone.Behind parietals, there is a smooth supraoccipital (Fig. 3b, SOC) bearing a well-developed median crest resembling that of †Diplomystus.This bone does not separate the parietals as usual in clupeiforms.Due to the state of preservation it is not possible to determine presence of pre-epioccipital fenestra and posttemporal fossa.But a so-called parietal excavation as described by Forey (2004) in a three-dimensional skull of †Diplomystus from the Upper Cretaceous of the English Chalk is very probable (Fig. 5).Such excavation is considered by Forey (2004, Fig. 13, Node E) derived featured shared by †Triplomystus, †Sorbinichthys, and †Diplomystus.The same author stated it is absent in †Ellimmichthys longicostatus, †Ellimma branneri and †Paraclupea.Otherwise we also noted its presence in †Scutatuspinosus and †Ellimma cruzae.
The orbit of †C.carnavalii is large.Its diameter is contained of about 3 1⁄2 in HL.The eyeball were supported by two large shields of sclerotic bones from which only one (anterior) is preserved (Figs. 3 and 4 The third infraorbital bone (Fig. 3b, IO3) is the largest of the set.It is subrectangular as in †Diplomystus dentatus (see Grande 1982, p. 10, Fig. 7) and show an anterior flange below the contact zone with second infraorbital.This bone forms most of the ventral rim of orbit extending from the level of the quadrate-mandibular joint to the posterior extremity of the symplectic, but not entirely recovering the cheek.The infraorbital sensory canal is bone-enclosed near the orbital rim.Its membranodermal component is very expanded.
The fourth and fifth infraorbital bones, although not preserved, judging by a great gap above third infraorbital, mighty be tubular flimsy bones as observed in †Diplomystus dentatus (see Grande 1982).
Jaws and suspensorium.-The premaxilla is partially preserved and displaced from the anteriormost region of skull (Fig. 3b, PMX).It is low, triangular and toothless.The maxilla (Figs.3b, 4, MX) exhibits an elongate anterior process finishing in a dilated articular head.There is a short and blunt autopalatine condyle upon this bone.The main body of the bone produces a slightly convex dentigerous lamina (as in †Diplomystus dentatus) on oral edge and extends backwardly to end at the level of the middle point of orbit.True teeth are lacking.
There are two smooth supramaxillae incompletely preserved.The first one, the anterior su-pramaxilla (Figs. 3b,4,ASMX), is an elongate and elliptical bone lying on the anterodorsal border of the maxilla.The posterior supramaxilla (Fig. 3b, PSMX) shows a large and ovoid main body, very fragmented in DGM 435-P, but its anterodorsal process is lost.
The lower jaw is relatively short, deep and well-ossified.Laterally it consists of dentary, anguloarticular and retroarticular (Fig. 3).
Most of the lower jaw is formed by the dentary (Fig. 3b, D) which contributes with at least 80% of length.This bone has a deep symphysis and the oral border ascends abruptly producing a high coronoid process.There is a short row of at least 20 minute conical teeth on oral border.The path of the mandibular sensory canal (Fig. 3b, md.c) is evident on surface through a slight and straight tubular relief.Pores are not visible.
The anguloarticular (Fig. 3b, AA) is a relatively short and deep bone.It contributes to form part of the coronoid process.Its articular facet for quadrate is well-developed.Posteriorly it shows a short and slightly rounded postarticular process.Its median portion is crossed by a extension of the mandibular sensory canal.
As for dentary, the anguloarticular is only a little incurved suggesting the presence of a low meckelian fossa and consequently few volume of adductor muscle in life.The posterior opening for the mandibular sensory canal is not observed laterally so that we interpreted it was placed medially.A small retroarticular (Fig. 3b, RAR) is visible at the posteroventral corner of the lower jaw upon anguloarticular.
The quadrate-mandibular articulation is placed below the middle point of the orbit.The triangular quadrate (Fig. 3b, Q) is well-ossified and slightly curved forwards.Its articular condyle for lower jaw is well-developed and the posteroventral process is short, sharp, and vertically oriented.The anterior border of preopercle is tightly attached to the posterior margin of posteroventral process.The dorsal margin of quadrate seems to be truncate.
A metapterygoid is not preserved but judged by a great gap on suspensorium this was of moderate size.The symplectic (Fig. 3b, S) is a short, narrow, and club-like bone inserted in a notch of posterodorsal margin of quadrate, between the main body of this bone and its posteroventral process.
The shape of the hyomandibula (Fig. 3b, HM) is inferred mainly from the imprint of this bone on the rock.It is vertically oriented in respect to braincase and shows an obliquely positioned articular head for otic neurocranium.The vertical process is elongate and the opercular process is very short and stout.
Anterior to quadrate there are a badly preserved ectopterygoid and an elliptical and a shieldlike endopterygoid (Figs.3b, 4, ENPT) closely associated.The latter is visible in the inner orbit and above the second and third infraorbitals.Both bones are toothless.
From the autopalatine remains only a prominent and robust facet for maxilla (Fig. 3b, APAL) placed anterior to lachrymal bone.

opercular SerieS
The usual complement of teleostean opercular bones is present in †Codoichthys.The preopercle (Fig. 3b, POP) is a L-shaped bone, with dorsal limb longer than the ventral one.The ventral limb has ventral margin slightly convex.The preopercular sensory canal (Fig. 3b, pop.c) runs into a bony tube in the midline of dorsal limb of the bone.We are unable to determine the presence of tubules in dorsal limb, but in the ventral limb the main canal gives off two conspicuous recumbent tubules (better seen in DGM 436-P) as in †Ellimma branneri (see Chang and Maisey 2003, p. 10, Fig. 5).
The interopercle (Fig. 3b, IOP) is a triangular and elongate bone underlying the entire length of the ventral limb of the preopercle.Its ventral edge is a slightly convex.
The opercle (Fig. 3b, OP) is a well-developed bone, deeper than long, with dorsal margin rounded and an oblique contact zone for subopercle ending in a protruding anteroventral corner.This bone is contained of about three times in the head length.The ventral half shows on surface fading ornamentation under form of parallel striations (better seen in DGM 436-P), a pattern shared with †Ellimma and †Tycheroichthys.
The subopercle (Fig. 3b, SOP) is a large and falcate bone.It shows digit-like anterior ascending process and smooth posteroventral margin.Its major depth is contained of about three times in the opercle depth.
paired FinS and GirdleS The pectoral girdle includes the posttemporal, supracleithrum, cleithrum, postcleithra, scapula, and coracoid.The posttemporal (Fig. 3b, PTM) exhibits a rounded and smooth-bordered main body and an elongate and sharp anterodorsal limb for contacting epioccipital.The ventral limb for intercalar is club-like, very slender and long.An oblique tubular relief for lateral line is seen along the main body of the bone.The supracleithrum (Fig. 3b, SCL) is an elliptical bone lying on the dorsal process of cleithrum.A short tubular relief for lateral line (Fig. 3b, lat.l) is visible on surface.
The cleithrum (Fig. 3b, CL) is a well-ossified L-shaped bone with expanded laminar posterior margin.Therefore it is not sigmoid as in most of †ellimmichthyiform fishes (see Forey 2004).The overall shape is similar to that of the clupeiform †Santanaclupea silvasantosi from the Araripe Basin, Northeastern Brazil (see Maisey 1993).The anteroventral process is elongate and projected forwards along all extension of the margin of subopercle.Regarding this aspect it is very similar to that of †Diplomystus dentatus.
There are two distinct postcleithra crossing over pectoral-fin rays in DGM 436-P.Grande (1985)  The pelvic bone is apparently hidden by abdominal scutes so that nothing is visible externally.The pelvic fin is in opposition to dorsal fin and remains in a middle point between insertions of pectoral and anal fins.There are five or six branched pelvic fin-rays.The dorsal fin origins in a vertical line spanning abdominal vertebrae 10 and 11 (Fig. 2B).There are two rudimentary fin-rays followed of at least 10 fin-rays in association with at least 10 pipe-like proximal pterygiophores.The first pterygiophore shows an anterior lamina with a notch between dorsal and ventral branches, forming an acute angle.It supports the three anteriormost fin-rays.The number of anal-fin rays is inferred of 15 preserved pterygiophores.There are one rudimentary followed of at least 15 fin-rays.

vertebral coluMn, pleural ribS and interMuScular boneS aSSociated
There are, at least, 33 preural vertebrae, 11 in caudal region.All vertebral centra are hour-glass shaped in lateral view.Each vertebral centrum is smooth, except for weak longitudinal ridge separating grooves (Fig. 8B).The precaudal abdominal centra are mostly deeper than long.From the origin of dorsal fin backwards they became gradually longer than deep.
The neural arches (Fig. 6, n.a) are co-ossified to centra and the parapophyses (Fig. 6, pap).Neural spines (Fig. 6, n.sp) are long and touch proximal end of the dorsal pterygiophores.We are unable to determine the presence or not of bifid neural spines in the abdominal region due to the state of preservation.But in the caudal region, only a single neural spine fused to arch is seen.
Hemal spines ( Fig. 6, h.sp) are as fine and long as the neural spines.There are any consistent differences in size or thickness among spines contributing to sustain the caudal fin.
The supraneural (=predorsal) bones form a series of eight slender, long, and almost sigmoid elements closely associated to predorsal scutes (Fig. 7).The space among supraneurals are unequal.The first supraneural is very short and positioned far from the other ones.The posteriormost one is the longest and closely associated to anterior lamina of the first pterygiophores of the dorsal fin.Apparently all supraneurals reach the level of the tips of abdominal neural spines.
The epineurals (Fig. 6, EPN) are long, thin, and laterally arched.The anteriormost ones are proximally fused to the base of abdominal neural arches.They are symmetrically aligned along the vertebral column spanning an area corresponding to five to six vertebrae.In the transition between abdominal to caudal regions there are high on the flank detached epineurals forming epimeral hypsiloid intermuscular bones (Fig. 6, EPM), in a framework backwards spanning an area spanning eight vertebrae.They continue backwardly on caudal region in opposition to some detached epipleurals (Fig. 6

, EPL).
There are at least 17 pleural ribs.They are long and slightly curved and reach the tip of abdominal scutes (Fig. 8).Each one shows a deep groove along most of its extension.The anteriormost pleural ribs are lodged in a groove of the abdominal vertebra whereas the remaining ones on short parapophyses.

predorSal and abdoMinal ScuteS
There are slight imprints of eight elliptical predorsal scutes bearing a median keel (Fig. 7).All scutes are smooth, subequal-sized and lack posterior spine.They touch the distal end of the supraneurals.
A row of pre-pelvic scutes composed of 10 elements originates closely to pectoral fin at the level of the last fin-ray (Fig. 2).Each scute is ventrally keeled and bears long and sharp triangular lateral processes.The abdominal postpelvic series shows nine scutes.They are triangular, ventrally keeled and show short posterior spine-like process unlike erected and pungent of the most of †ellimmichthyiform fishes.The series finishes at the origin of the anal fin.

caudal endoSkeleton and Fin
The caudal fin is forked and shows equal-sized lobes.It is supported by two preural and two ural centra.As usual in many basal teleosteans, there are 10 principal fin rays in the upper lobe and 9 in the lower lobe.The region of dorsal procurrent rays is very damaged but there are at least four preserved in DGM 435-P.There are at least seven ventral procurrent rays and a probable ventral caudal scute (Fig. 8b, c.sc).The vertebral axis bends gradually upwards as usually found in nonclupeiform clupeomorphs.The neural spine of the second preural centrum is slender and elongate.
The proximal end of parhypural (Fig. 8b, PH) is fused to the first preural centrum.The neural arch of first preural centrum is enlarged and leaf-like extending backwardly to cover the dorsal margin of the first ural centrum and to contact laterally the first uroneural.This condition differs from that of †Diplomystus dentatus in which only the neural arch of the first ural centrum is enlarged and backwardly expanded.
The first ural centrum (Fig. 8b, U1) is almost hour-glass shaped and equal-sized to preural centra.is hour-glass shaped and almost equal-sized to preural centra.However the second ural centrum (Fig. 8b, U2) is more reduced, cone-shaped with tubular posterior end.
We counted six hypurals (Fig. 8b).The first one (Fig. 8b, H1) is a large triangular plate, long and enlarged distally.Proximally it attaches the first ural centrum and contributes for the hypural foramen.The second hypural is slender and elongate, and fused to first ural centrum (Fig. 8b, U1+H2).The remaining four hypurals are associated to second ural centrum.The third hypural (Fig. 8b, H3) is triangular and large; its expanded spatulate portion invades ventrally space of lowermost hypurals.Probably it was attached to second ural centrum in life.There is a large diastema between third and second hypurals.This pattern differs from that of †Scutatuspinosus and †Rhombichthys whose third hypural has an upward process constraining the fourth hypural to project forwardly.This condition seems to be shared uniquely for these taxa.The other three hypurals are elongate and gradually decrease in size towards the dorsal margin of the upper lobe.
Our observations are in contrast with the original description.Santos (1994) pointed out the occurrence of four uroneurals.We consistently observed two uroneurals (Fig. 8b), all free from the centra.The existence of a third uppermost uroneural (Fig. 8b, ?UN3) is probable taking into account by a clear space in the region anterior to the tips of invading fin rays and the presence of a strut of bone in this place.The first uroneural (Fig. 8b, UN1) extends forwards for reaching the first preural centrum laterally.It is placed laterally to the neural arch from the first preural centrum.The second uroneural (Fig. 8b, UN2) is short and fusiform, and lies on the dorsal margin of the second ural centrum close to the origin of the second ural centrum.
There are three thin, elongate and curved epurals (Fig. 8b, EP, numbered) occupying a space between neural spine of the second preural centrum and dorsal border of neural arch from the first preural centrum.

reSultS oF the phyloGenetic analySiS
The hypothesis generated by the present phylogenetic analysis corresponds to a strict consensus of three equally parsimonious trees (Fig. 9).The majority consensus tree shows the same topology.The tree has 181 evolutionary steps, Consistency Index (CI)=0,387, Retention Index (RI)=0,632.
It is not possible to confirm if †Ornategulum belongs with clupeomorphs.In this analysis this taxon is excluded of Clupeiformes and †Ellimmichthyiformes and placed outside as the sister group of all clupeomorphs used in this analysis.
†Sorbinichthyidae appears in a basal polytomy together with the enigmatic †Horseshoeichthys and all other †ellimmichthyiforms.But we think that the position of †Horseshoeichthys is doubtful because it shows many missing data in the matrix and herein confirmed as a "wild-card" (sensu Murray and Wilson 2013).Thus its position among †Ellimmichthyiformes still depends on the collection of well-preserved specimens.If so, excluding †Horseshoeichthys, †Sorbinichthyidae becames the putative sister group of all other †ellimmichthyiforms.
We have found no diagnostic character for †Armigatus so that species of this genus appear in a basal polytomy below more advanced taxa.
Species of †Diplomystus (excluding † "D".solignaci) form a group supported by the second higher value of Bremer index (3).We obtained a different placement for this group so that it is separated of †Armigatus and †Sorbinichthyidae, dismantling †Armigatoidei ( †Diplomystus in part plus †Armigatus sensu Murray and Wilson 2013).†Diplomystus is placed in a more advanced position in the tree in comparison with †Sorbinichthys and †Armigatus.†"Diplomystus" solignaci is positioned within an advanced †Paraclupeidae but, in contrast with previous analyses, it appears as sister group of †Paraclupea chetunguensis.
A Node 4, linking †Diplomystus clade and †Paraclupeidae, is diagnosed by a special relation of pleural ribs and grooves of vertebral centra on abdominal region (C=18) and a conspicuous parietal excavation (C=60).
The Node 5 (Bremer index =3) corresponds to the †Diplomystus clade and is diagnosed by the  to 0).It contains †Codoichthys and all remaining †ellimmichthyiforms.A Node 7 (unnamed) includes †ellimmichthyiforms more advanced than †Codoichthys.It is divided in two subfamilies †Thorectichthyinae (Node 8, new name) and †Paraclupeinae (Node 9, new use).They share marked angle in front of the insertion of dorsal fin (C=1, secondarily lost is †Codoichthys and †Scutatuspinosus), and the size of scutes increasing backwards (C=44).
†Triplomystini (Node 10) contains all species of †Triplomystus clade (Node 12, diagnosed by the homoplastic features sharp end of first hypural and quantity of predorsal scutes, and an uniquely derived C=51, i.e., postdorsal scutes) that is sister group (Node 11) of †Tycheroichthys plus †Rhombichthys.This arrangement of Node 11 is new and shows the highest value of decay (i.e., Bremer index equals 4).Node 10 is characterized by two homoplastic features: derived number of uroneurals (C=29) and spatulate abdominal scutes (C=50).†Paraclupeini (Node 13) is linking †Scutatuspinosus and advanced taxa by sharing homoplastic condition of characters 12 (absence of supraorbital, state 1 to 0) and 53 (loss of postcleithra).We think these characters used to diagnosed both groups demands more study.Noteworthy †Scuta-tuspinosus shows reversions (characters 1 and 50), and a independent acquisition (character 56).
Within †Paraclupeini, a Node 14 is linking †Ezkutuberei and more advanced taxa (Node 15, †Ellimma, †Ellimmichthys, †Paraclupea and † "Diplomystus" solignaci).Only a reversion of character 30 (state 1 to 0) supported it.Better is the case of Node 15 characterized sculptured skull roof (C=5, state 0 to 1) and extension of epineurals and epipleurals on caudal region (C=19, state 0 to 1).Assuming the condition found in †Rhombichthys and †Triplomystus applegatei not ambiguous (as interpreted by the TNT program), it is a putative uniquely derived feature for the clade.
Alignement between †Ellimmichthys and †Paraclupea was yet detected in early efforts to decipher relationships within †Ellimmichthyiformes (e.g., Chang and Grande 1997, Chang and Maisey 2003, Forey 2004).It is confirmed in Node16 although species of †Ellimmichthys appear in a polytomy reflecting still bad knowledge of their structures.The node also includes the enigmatic †"Diplomystus" solignaci and is supported by three homoplastic features (characters 16, 20, and 52).In this we noted a well-supported clade (Node 17) formed by †Paraclupea and †"Diplomystus"solignaci.This node is the second highest value of bootstrap (i.e., 50) in our analysis although supported by homoplasies (characters 22, 32, and 40).Thus †"Diplomystus"solignaci is confirmed outside that genus.
We made an experimental analysis excluding the so-called 'wild-cards' or "swing taxa" (i.e., †Ornategulum, †Ellimmichthys longicostatus, †Ezkutuberezi carmenae, and †Horseshoeichthys, together with †Ellimmichthys goodi) and using command traditional search, we obtained seven trees (168 steps, CI 0,417, RI 0,611).Besides higher CI and lower steps, applying strict consensus we obtained a paraphyletic group for species of †Sorbinichthys and a great basal polytomy with remaining †Ellimmichthyiformes.In sum, in FRANCISCO J. DE FIGUEIREDO and DOUGLAS R.M. RIBEIRO this step we have found a paradoxical result with Murray and Wilson (2013).

DISCUSSION AND CONCLUSIONS
After Grande (1985), the classification of Ellimmichthyiformes, in a historical perspective, become a continuous effort to add or exclude taxa.It was obviously associated with many attempts of tie the name of the taxon to at least one character that is necessary for membership.In this context Grande's approach was based on drawing-by-hand analysis, with a monophyly accepted a priori with selected characters and omitting homoplasies.
Since then †Ellimmichthyiformes seems to be generally accepted, besides strikingly short of characters.Particularly, we think that the reality of the group is better supported mainly because any member demonstrates close affinities with taxa outside (e.g., clupeiforms, ostariophysans, and elopomorphs).
Chang and Maisey (2003) were the first to use parsimony analysis performed in a computer program (i.e., PAUP version 4.0) to test monophyly of †Ellimmichthyiformes. Their results were obtained on the basis of a data matrix including 30 characters and 11 taxa.The characters used were all taken of Grande (1982Grande ( , 1985) ) but they included new taxa too.As a result, they found a clade containing †Armigatus, †Diplomystus, †Ellimma, †Ellimmichthys, and †Paraclupea.This group was subdivided is two subclades, one for †Armigatus plus †Diplomystus and other containing †Ellimma, †Ellimmichthys and †Paraclupea.Noteworthy, these authors were the first to pointed out †Scutatuspinosus as showing some features indicating close affinities with †paraclupeines.
In an approach to known if Ellimmichthyiformes is a clade or a convenience group, Zaragueta-Bagils (2004) used a data matrix including 56 characters and 15 terminal taxa.He indicated that †Armigatus brevissimus, †Diplomystus birdi, and †D.dentatus form a group outside clupeiforms and †ellimmichthyiforms.His results dismantled †Ellimmichthyiformes. Forey (2004) furnished a drawing-by-hand phylogenetic scheme for †Ellimmichthyiformes diagnosing the clade by a S-shaped cleithrum.Even he recognized two great subclades: a node D (for †Armigatus, †Triplomystus, †Diplomystus) and node C (for †Paraclupea and †Ellimmichthys).He stated Node D diagnosed by enlarged endopterygoid teeth, elongate anal fin, and deep and narrow opercle.His node C was characterized by characters taken of Grande (1982Grande ( , 1985)).Unfortunately, he omitted informative taxa as †Ellimma, †Scutatuspinosus, †Codoichthys, all yet known in that time.
Alvarado-Ortega et al. ( 2008) using a data matrix of 28 clupemorphs and 58 characters excluded †Armigatus from the †Ellimmichthyiformes.They used mainly characters taken of Grande (1982Grande ( , 1985) ) and Zaragueta-Bagils (2004) and divided the group in two clades: †Sorbinichthyidae and †Paraclupeidae.According to them, †Sorbinichthys is a member of a †Diplomystus clade and †Paraclupeidae containing all other †ellimmichthyiforms.But certain taxa were omitted in the analysis of Alvarado-Ortega et al. (2008) as the case of †Codoichthys but other (e.g., †Tycheroichthys, †Rhombichthys, †Torectichthys) were only described in recent years.Other problem detected was about definition of characters and codification of states so that they were critically reassessed by Murray and Wilson (2013).
Murray and Wilson (2013) used 28 taxa and 62 characters in the most recent and complete cladistics analyses.Their results were condensed in a consensus tree of three shorest trees (176 steps, CI=0,44, RI=0,57) .It follows that †Armigatoidei ( †Armigatus plus †Diplomystus) was found as sister of all other †ellimmichthyiforms.The strict and majority-role consensus trees were as our results, identical.As in our analysis neither decay (Bremer support) nor bootstrap analyses indicated strong support for any branch except for certain pair of species.†Ellimmichthyoidei contained †Sorbinichthyidae and †Paraclupeidae (including †scutatuspinosines, †thorectichthyines, and †paraclupeines).During their analyses some taxa were excluded because introduce many problems (i.e., many missing data), for instance, †Ornategulum, †Ellimmichthys longicostatus, †Ezkutuberezi, and †Horseshoeichthys.Even, they indicated † Ellimmichthys could be paraphyletic.
Recently some preliminary approaches to decipher relationships of †ellimmichthyiforms were carried out by F.J. de Figueiredo (unpublished data, Ph.D dissertation, Museu Nacional-UFRJ, Rio de Janeiro, 2006), F.J. de Figueiredo and D. Barros (unpublished data, abstracts of the 14th Congresso Latino Americano de Paleontologia de Vertebrados, p. 172, San Juan, Argentina, 2011), D.R.M. Ribeiro and F.J. de Figueiredo (unpublished data, abstracts of the 23th Congresso Brasileiro de Paleontologia, p. 270, Gramado, Rio Grande do Sul, 2013) and A. Nunes and F.J. de Figueiredo (unpublished data, abstracts of the 23th Congresso Brasileiro de Paleontologia, p.255, Gramado, Rio Grande do Sul, 2013).F.J. de Figueiredo (unpublished data, Ph.D dissertation, Museu Nacional-UFRJ, Rio de Janeiro, 2006) furnished a review of articulated and complete clupeomorphs found in Lower Cretaceous and Tertiary deposits outcropping in many localities from the Northeastern Brazil.At first a data matrix containing 61 characters and 19 taxa were used in that analysis.The anatomy of many taxa were revisited in detail and a phylogenetic scheme was obtained.As a result †Ellimmichthyiformes appears as a weakly supported clade.Only a S-shaped cleithrum was indicated as uniquely derived feature, confirming hypothesis by Forey (2004).In addition, †Paraclupeidae only contained †Ellimmichthys longicostatus, †Ellimma cruzae, †Triplomytus noorae, †Paraclupea and a unnamed †Ellimmichthys-like taxon from the Lower Cretaceous of Reconcavo Basin, State of Bahia.†Codoichthys, †Scutatuspinosus, †Ezkutuberezi, and †Ellimma branneri appeared in a basal polytomy, and †Diplomystus (i.e., †D.shengliensis and †D.dentatus) plus †Armigatus formed a clade.
Like Alvarado-Ortega et al. (2008) this analysis become dated because did not include many taxa only described recently.F.J. de Figueiredo and D. Barros (unpublished data, abstracts of the 14th Congresso Latino Americano de Paleontologia de Vertebrados, p. 172, San Juan, Argentina, 2011) while revisiting the anatomy of †Scutatuspinosus preliminarly proposed a new hypothesis on phylogeny.They obtained eight trees with 128steps based on a data matrix including 15 putative ellimmichthyiforms, two basal clupeiforms (i.e., †Santanaclupea silvasantosi and †Pseudoellimma gallae, see Maisey 1993, Figueiredo 2009).And as for previous analyses, low index (CI=0.5;RI=0.6) indicating many missing data as the main problem.†Ellimmichthyiformes and †Paraclupeidae were confirmed.In addition Figueiredo and Barros suggested the exclusion of †Ellimma cruzae from the genus †Ellimma and put formally †Codoichthys within †Ellimmichthyiformes.But the inclusion of †Santanaclupea silvasantosi and †Pseudoellimma gallae as near outgroup in that analysis introduced many missing data and a doubtful result.
As a result, †Codoichthys appeared in basal position in a tree as sister group of remaining †Ellimichthyiformes.Once more †Diplomystus is indicated as paraphyletic taxon.†Rhombichthys revealed to be sister group of †Ezkutuberezi and appeared with †Tycheroichthys in a great polytomy together with †Ellimma branneri, †Diplomystus solignaci, †Ellimmichthys, and †Paraclupea plus †Triplomystus.They confirmed †Paraclupeidae with †Scutatuspinosus as sister group of the deepbodied †Ellimmichthyiformes. Noteworthy the unpublished results of Ribeiro and Figueiredo and Murray and Wilson (2013) appeared independently in the same year.Both works have not included or discussed data and results from each other.
Unlike results of D.R.M. Ribeiro and F.J. de Figueiredo (unpublished data, abstracts of the 23th Congresso Brasileiro de Paleontologia, p. 270, Gramado, Rio Grande do Sul, 2013) we took into account the corrections furnished by Murray and Wilson (2013), using new insights on characters and states (see Appendix 1, commented list of characters and states), and including new taxa.Thus we furnished an ultimate character list.
Although highlighting character conflict we feel again the urgency of revisiting characters and reassessing taxa.Clearly the situation with regard to relationships of taxa within ellimmichthyiforms is unresolved and most of lineages are supported only by homoplasies.Many problems remain.For instance, although †Horseshoeichthys has an incertae sedis position at the base of the group in polytomy with †Sorbinichthyidae and all other †ellimmichtyiforms, we think that more informative specimens are need to decipher its relationships.Thus the most basal group of †Ellimmichtyiformes is probably †Sorbinichthyidae.
In sum, unlike Zaragueta-Bagils ( 2004) but agreeing with Alvarado-Ortega at al. ( 2008) and Murray and Wilson (2013) we confirm the existence of a †Ellimmichthyiformes group, though weakly supported.There are major differences between the topology of trees of Chang and Maisey (2003), Zaragueta-Bagils (2004), Forey (2004), Alvarado-Ortega et al. (2008), and Murray and Wilson (2013) and the hypothesis herein proposed.It confirms the impact of new described taxa in the phylogenetic analysis of a problematic group whose definition, diagnose, and composition wait for better results.
Particularly we noted that the main divergence among most of the analyses is about the position of †Sorbinichthys, †Armigatus and †Diplomystus in relation to the remaining taxa, difficulting linking.They are "swing" groups, sliding from the base of tree to near †Paraclupeidae.This latter includes mostly deep-bodied taxa, but its composition until now remains unresolved.
†Codoichthys is a basal and primitive †Paraclupeidae.In our analysis this taxon appears as sister group of remaining †paraclupeids.
As far as we known most recent phylogenetic analyses were prejudiced by many homoplasies and missing data.All phylogenetic schemes after Chang and Maisey (2003) produced trees with weakly supported branches.Thus, as claimed by Murray and Wilson (2013) we need more complete and informative specimens as well as revisiting old material in a search for new characters.

APPENDIX 1 liSt oF characterS and StateS uSed in the phyloGenetic analySiS
The present cladistic analysis of †Ellimmichthyiformes is based on the characters below.[0] at first represents the assumed plesiomorphic state and [1], and [2], and [3] the apomorphic states.They are present according to anatomical region, in logical flow, from the head to tail, except for four last additional characters from the list.The outgroup used to polarize characters and rooting tree is Elops saurus.Additional taxa used in this analysis are the clupeomorphs †Ornategulum sardinioides, Chirocentrus dorab, Denticeps clupeoides, and Odaxothrissa vittata.Further explanation of the characters and their coding see publications of Chang and Maisey (2003), Zaragueta-Bagils (2004), Alvarado-Ortega et al. (2008), and Murray and Wilson (2013).These publications followed of character number are referred by abbreviations CM#, Z#, AO# and MW# respectively.Additional relevant authors are included in parenthesis.Comments in brackets detail changes we made or add explanation.
[The dermal basipterygoid process is present in many †ellimmichthyiform fishes (Chang and Maisey 2003, Zaragueta-Bagils 2004, Forey et al. 2003, Forey 2004, Alvarado-Ortega et al. 2008).But due to poor preservation it is not possible to confirm if †Scutatuspinosus, †Ezkutuberezi, †Ellimmichthys and †Codoichthys show that state, therefore this character is coded as missing data for them.](11) Teeth on parasphenoid (modified from AO11, Z12, Grande 1982Grande , 1985 (1) absent.[Althoughassumed as regularly present in many basal lineages of teleostean fishes (Arratia 1999), the presence or not this bone is very difficult to confirm in most of fossil fishes due to the flimsy condition.It commonly becomes easily lost or damaged.It is boomerang-like in †Scutatuspinosus (cf.DGM 1255-P) and quadrangular (and with canal) in †Horseshoeichthys (identified erroneously as a lachrymal by Newbrey et al. 2010) in which is very similar to Denticeps clupeoides (see Di Dario 2004).Gayet (1989) claimed the presence of an anamestic lachrymal as a synapomorphy of Clupeomorpha but as far is known it is not supported by any congruence of characters in foregoing analyses and, at worst, this bone clearly bears canal in Denticeps (see Di Dario and de Pinna 2006) and †Horseshoeichthys (see Newbrey et al. 2010).](14) Beryciform foramen on anterior ceratohyal bone (AO14, Z14, CM12, Grande 1985): (0) absent; (1) present.[It occurs in many taxa from early lineages of teleosts, many early euteleosteans, and early acanthomporhs.Otherwise it is absent in osteoglossomorphs, elopomorphs and advanced acanthopterygians.It is found in Scutatuspinosus (cf.DGM 1262-P).We coded this character mostly according to Zaragueta-Bagils (2004) assuming loss of this foramen in clupeiforms.](15) Foramen on the posterior ceratohyal (Z15, AO15): (0) absent; (1) present.[Although herein included in analysis, it is very difficult to verify this feature even in well-preserved specimens.It is a result of the enlargement of the groove for the hyoidean afferent artery.Apparently it is found in the hyoid apparatus of the unnamed †Diplomystus of the Cenomanian of English Chalk (Forey 2004).But as far as we known, among living clupeomorphs only Denticeps clupeoides possesses it.We found ceratohyal is widely covered by mandible and opercular bones in available specimens used in analysis.This feature contributes to increase the quantity of missing data in matrix.](16) Teeth on endopterygoid (CM9, AO16): (0) absent or minute; (1) present and well-developed.
indicated absence of postcleithra in †Ellimmichthyiformes.But since then postcleithra were registered in †Horseshoeichthys(Newbrey et al. 2010) and †Triplomystus applegatei (Alvarado-Ortega and Ovales-Damian 2008).Unfortunately endosteal bones of pectoral girdle are badly preserved and uninformative.An imprint of a small and quadrangular scapula is seen in DGM 435-P.The coracoid (Fig.3b, CO) is large, laminate and L-shaped.We counted at least 15 pectoral fin-rays and the uppermost fin-ray is the thickest of the set.
FRANCISCO J. DE FIGUEIREDO and DOUGLAS R.M. RIBEIRO presence of sub-rectangular scutes in the anterior (C=40) and posterior (C=41) portion of predorsal series, spines on posterior edge of predorsal scutes (C=42, homoplastic feature shared with †Sorbinichthys).Node 6, †Paraclupeidae new usage is diagnosed by another transformation of the character 11 (state 1 to 2), resulting in loss of teeth on parasphenoid (homoplasy shared with clupeiforms), and a derived number of predorsal scutes (C=46, state 2

DGM, formerly Divisão de Geologia e Mineralogia do Departamento Nacional de Produção Mineral, nowadays Museu de Ciências da Terra; D. UERJ, a collection of cleared and stained fishes in Departamento de Zoologia of Universidade do Estado do Rio de Janeiro; Pz.UERJ, paleozoological collection
of the Universidade do Estado do Rio de Janeiro.

DESCRIPTIVE MORPHOLOGY
General body ForM †Codoichthys carnavalii shows fusiform body, with predorsal profile rather arched(Figs.2a,b).All specimens are preserved in lateral view suggesting that this fish was laterally compressed in life.HL corresponds to 28% of SL and almost equal to HD. Snout is short and upper jaw not extending behind the orbit.Both dorsal and ventral profile are slightly convex.BD corresponds to 27% SL.PDL is of about half (54%) of SL.Dorsal fin is placed at the middle point of the body, in opposition to the insertion of pelvic fin.PANL is 72% of SL and the base of anal fin length is about 16% of SL.The caudal fin is forked, with deep notch, showing equal-sized dorsal and ventral lobes.