A distinct shoulder region, defined by endoskeletal and dermal girdles and associated pectoral musculature, is a major evolutionary adaptation of jawed vertebrates. In teleost model species the large (macromeric) pectoral dermal bones can be derived from multiple embryonic tissues, identifying the shoulder of osteichthyans as a developmentally complex area at the head-trunk boundary. The absence of bone in living chondrichthyans makes Palaeozoic stem groups capable of dermal ossification key to understanding the underpinnings of skeletal growth in the shoulder of crown gnathostomes (osteichthyans and chondrichthyans). Here, using synchrotron X-ray tomography we demonstrate that individual pectoral plates in the oldest unequivocal jawed vertebrate, the Silurian (c. 439 Mya) chondrichthyan Fanjingshania renovata, develop from five separate growth centres. These centres correspond to pectoral bony spines that fuse neighbouring dermal scales into a pinnal plate and their expansion is accompanied by cyclical resorption and remodelling of bone and dentine. Our phylogenetic analyses support an interpretation of these processes as crown and stem gnathostome characters that co-occur only in the shoulder girdle of stem chondrichthyans. The systematic hard tissue remodelling in Fanjingshania reveals an unexpected growth dynamic within chondrichthyans that relates to the formation of a macromeric skeleton through integration of modular elements.

(a) Material

The studied skeletal elements of Fanjingshania renovata are part of a larger fossil collection (more than 1000 fin spines, spines, tesserae, scales and dermal plates belonging to the species) recovered from the lower Silurian (Aeronian) beds of the Rongxi Formation, exposed near the village of Leijiatun, Shiqian county, Guizhou province, China. They are represented by a pinnal plate fragment (IVPP V27433.8) and other incomplete pinnal plates fused to prepectoral (IVPP V27433.9) and admedian (IVPP V27434.2) spines and pectoral fin spine fragments (IVPP V27433.10, IVPP V27433.11, IVPP V27433.12, IVPP V27433.4). All specimens were deposited at the Institute of Vertebrate Paleontology and Paleoanthropology (IVPP), Chinese Academy of Sciences.

(b) Synchrotron radiation X-ray tomographic microscopy

The fossil material was analysed with synchrotron radiation X-ray tomographic microscopy at the TOMCAT-X02DA beamline of the Swiss Light Source, Paul Scherrer Institute, Villigen, Switzerland. The scans were performed with a 10X and a 4X objectives, producing 1501 and 1201 projections respectively, acquired over a 180-degree rotation arc. The beam energy was set to either 25keV or 28keV and the beam was attenuated with 100μm Al, 10μm Cu and 10μm Fe filters. The projections were processed with tomographic reconstruction algorithms developed by Marone et al. 2017 which produced tomographic slices with pixel size of 1.625μm (4X objective) and 0.65μm (10X objective).

Analysis, segmentation and 3D visualisation of the tomographic slices was performed in Dragonfly software, Version 2022.2 (www.theobjects.com/dragonfly).

(c) Phylogeny

We performed a phylogenetic analysis under parsimony criteria based on a data matrix modified from Brazeau et al. 2023. The matrix was analysed in TNT version 1.6 employing the traditional search function. Search parameters were set to random seed value of 2, number of addition-sequence replicates to 1,000, a tree bisection reconnection (TBR) swapping algorithm limited to 100 trees per replication and a tree buffer of 10,000. The analysis was constrained by designating the** galeaspid Hanyangaspis guodingshanensis as an outgroup taxon and the Osteostraci as monophyletic according to the current phylogenetic framework of early vertebrates. This produced 1800 most parsimonious trees (MPTs) with a length of 1198 steps.

Bayesian analysis was also conducted for the same data matrix in MrBayes version 3.2.7. We ran a Markov chain Monte Carlo (MCMC) simulation for 10 million generations with Hanyangaspis guodingshanensis set as an outgroup** and jawed vertebrates constrained as monophyletic. Burn-in fraction was set to 25%, the sampling frequency of the chain to 1000 generations and the number of simultaneous MCMC chains to four.

Ancestral states for the parsimony and Bayesian trees were reconstructed with the MBASR toolkit in R version 4.4.2 running in RStudio (version 2024.09.0) integrated development environment. A random sample of 50 Run 1 and 50 run 2 tree outputs with posterior estimates (last 75% trees, burnin fraction 25%) were converted to nwk format in FigTree v.1.4.4. Similarly, 100 random most parsimonious trees containing branch lengths were exported to a nwk in FigTree version 1.4.4. Ancestral states estimates for the subsets of parsimony and Bayesian trees were generated by running in loop the MBASR toolkit with a custom script (see electronic supplementary material). Finally, we calculated the geometric mean of the ancestral state values in each subset of trees.