Classifying bivalve larvae using shell pigments identified by Raman spectroscopy.

ABSTRACT

Because bivalve larvae are difficult to identify using morphology alone, the use of Raman spectra to distinguish species could aid classification of larvae collected from the field. Raman spectra from shells of bivalve larvae exhibit bands that correspond to polyene pigments. This study determined if the types of shell pigments observed in different species could be unique enough to differentiate larvae using chemotaxonomic methods and cluster analysis. We collected Raman spectra at three wavelengths from 25 samples of bivalve larvae representing 16 species and four taxonomic orders. Grouping spectra within general categories based on order/family relationships successfully classified larvae with cross-validation accuracies ≥92% for at least one wavelength or for all wavelengths combined. Classifications to species were more difficult, but cross-validation accuracies above 86% were observed for 7 out of 14 species when tested using species groups within orders/families at 785 nm. The accuracy of the approach likely depends on the composition of species in a sample and the species of interest. For example, high classification accuracies (85-98%) for distinguishing spectra from Crassostrea virginica larvae were achieved with a set of bivalve larvae occurring in the Choptank River in the Chesapeake Bay, USA, whereas as lower accuracies (70-92%) were found for a set of C. virginica larvae endemic to the Northeast, USA. In certain systems, use of Raman spectra appears to be a promising method for assessing the presence of certain bivalves in field samples and for validating high-throughput image analysis systems for larval bivalve studies.