141-14 Machine learning reveals the evolution of the animal biomineralization toolkit from molecular fingerprints of modern and fossil tissues

Session: New Advances in Geobiology

Presenting Author:

Author:

Abstract:

Biomineralized tissues show a unique evolvability in the animal clade. The evolutionary drivers behind the emergence of templated biomineralization in spicules, shells, denticles, teeth, bone, and eggshells from non-biomineralized precursors are, however, obscured in the fossil record: original biominerals rarely survive without chemical alteration in deep time, and often experience dissolution; soft tissues, however, commonly attract secondary minerals resembling biominerals. Due to the lack of a proxy that allows for the distinction of biological and diagenetic mineralization, the mechanisms and timing of skeletal diversification and its potential covariance with environmental parameters are still considered enigmatic.

Here, we analyze n=150 modern and carbonaceous fossil metazoan hard and soft tissues that were sampled systematically across Phanerozoic clades in the search for a signature of original biomineralization. Due to the common alteration of the mineral phase, our proxy targets remnants of the macromolecular template: Among these samples, n=100 represent animal tissues with known biomineralization status, while n=50 represent (predominantly Paleozoic) samples with debated biomineralization status. A supervised Random Forest (RF) model trained on Raman spectral fingerprints (500-3000 cm^-1, 10 replicates; 532 + 785 nm) was complemented with hyperspectral 2-D maps, and assessed via simulated test data: the model classifies modern and fossil biominerals based on features of the macromolecular template and its fossilization products with >96% accuracy. Features diagnostic for original biomineralization are linked to an increased abundance of coordinating ligands in the original macromolecular template. Projection of the RF predictions on an unsupervised principal component analysis of the training data reveals distinct clustering of biocarbonate and biophosphate templates among modern and fossil taxa, regardless of phylogenetic affinity. Incorporation of fossil unknowns into the model illuminates the early Paleozoic diversification of skeletal materials and reveals convergent origins of a variety of carbonate and phosphate animal biominerals.

Our analyses show that independently co-opted template macromolecules experience convergent functionalization in order to precipitate certain mineralogies. Template optimization represents an evolutionary prerequisite for the emergence of skeletons and the complex body plans and behaviors that they facilitate.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-11024

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.