A new Cenomanian spiny-rayed fish from El Chango quarry, Chiapas, South Eastern Mexico is described and included, together with Zoqueichthys carolinae and Pepemkay maya, in the most recent phylogenetic analysis of acanthomorph fishes encompassing both extant and extinct taxa. The topologies recovered from the parsimony phylogenetic analysis using both multistate, or composite, and contingent, or reductive, coding are compared against each other and with a Bayesian estimation. All the analyses corroborate the placement of Z. carolinae as a member of the division Lampripterygii, while both Choichix alvaradoi gen. et sp. nov. and P. maya are considered as incertae sedis members into the division Acanthopterygii. Although no apomorphic character states for C. alvaradoi were found, the unique combination of features supports the description of a new genus and species for the Upper Cretaceous of Mexico. The species herein described is the earliest diverging acanthopterygian lineage. The most outstanding character found is the absence of a true pelvic-fin spine that, together with its placement in the phylogenetic analysis, suggests the independent acquisition of the pelvic-fin spine in both paracanthopterygians and acanthopterygians from a segmented first soft ray. This extinct taxon represents, thus, an important piece to understand both the fish diversity through time and some steps in the early diversification of acanthopterygians.

1. Description of methods used for collection/generation of data:

The comparative anatomy study included data from acanthomorphs housed at the National Collection of Palaeontology of Mexico and the Palaeontological Museum Eliseo Palacios Aguilera.

After the anatomical review, two phylogenetic analyses were performed. Both analyses are based on Davesne et al.´s (2016) matrix modified using the Mesquite software (Maddison & Maddison, 2018), to include the new fossil species, Zoqueichthys carolinae and Pepemkay maya, leaving a total of 29 terminals and an outgroup composed of Synodus Scopoli, 1777, Gymnoscopelus Günther, 1873, and †Ctenothrissa Woodward, 1899. The first analysis kept the composite (or multistate) coding of the original matrix. The second applied the reductive (or contingent) coding to reduce the use of multistate and polymorphic characters included in the original matrix, increasing the number of analysed features from the original 66 to 78 characters.

Parsimony and Bayesian methods were used to analyse both matrices. The parsimony analysis was performed in TNT (Goloboff et al. 2008), and the Bayesian inference was carried out in RevBayes (Höhna et al. 2016). For the parsimony analysis, a search with new technology using 10000 iterations of Ratchet with equal weights and 100 replacements, was carried out over the most parsimonious trees (MPT) recovered in the first analysis. The group analysis support was estimated via bootstrap resampling, which implemented 1000 replications and was performed in TNT (Goloboff et al. 2008). For Bayesian inference, the simple MK model (Wright et al., 2019) was used with 15000 generations. The convergence of both analyses was evaluated with Tracer (Rambaut et al. 2018).

2. Methods for processing the data:

Osteological description of the species studied; retrieved of the original matrix provided by Davesne et al. (2016); inclusion of the osteological of the species studied in the matrix; separation between both the composite and reductive dataset; running both the parsimony and Bayesian analysis; interpretation of the data; and mapping some specific character states.

3. Software-specific information needed to interpret the data:

Mesquite: a modular system for evolutionary analysis. Version 3.51

Tree analysis using new technology (TNT) version 1.5

RevBayes: Bayesian phylogenetic inference using graphical models and an interactive model-specification language version 1.1

Tracer: a software package for visualising and analysing the MCMC trace files generated through Bayesian phylogenetic inference version 1.7.1

Both composite and reductive data matrix are separate; however, the exact steps used in the command lines were performed with the matrix containing all characters (i.e. Cantalice_2020_11_24.nex). Each character was properly activated or deactivated to perform both analyses.