Discrimination of Camellia japonica cultivars and chemometric models: An interlaboratory study

Abstract

Camellia japonica is a valued plant since ancient times throughout the world mostly due to their ornamental flowers. It has a high number of cultivars, with very similar phenotypic and genotypic characteristics which are difficult to discriminate, being some of them often rare and with a high price at the market. Their discrimination is mostly done through visual inspection of the morphologic characteristics which is a hard and inefficient task. Spectroscopic techniques had already been used for taxonomic purposes at species and sub-species level with success and could be an alternative for accurate C. japonica cultivars discrimination. Despite the already recognized success of such techniques, most of the studies arises from a single laboratory and little is known about the robustness of these techniques regarding interlaboratory data transferability. In this context, the work developed herein presents a double aim: (I) to explore the ability of near infrared (NIR) spectroscopy and partial least square – discriminant analysis (PLS-DA) to discriminate C. japonica cultivars and (II) to evaluate data transferability between two independent laboratories (Lab A and Lab B). Air-dried leaves NIR spectra of 43 C. japonica plants (15 distinct cultivars) were acquired in both laboratories using two similar NIR instruments (same manufacturer and model). Spectra were further modelled by PLS-DA after exploratory analysis using principal component analysis (PCA): (I) individually for Lab A and Lab B; (II) using Lab A as calibration and Lab B as validation set and vice-versa and (III) with using Lab A and Lab B data together. The percentage of C. japonica cultivars discrimination for both laboratories was nearly the same (around 83%) indicating no significant differences between Lab A and Lab B analysis. However, the results were quite poor when spectra were modelled with data from a single laboratory and validated with the other (65.5 and 63.8%). When data were merged, 85.9% of correct cultivars assignments were obtained. The results herein obtained could benefit of including additional cultivars and plants; but demonstrated the ability of NIR spectroscopy for C. japonica cultivars discrimination. Regarding data transferability, even when dealing with a similar instrument, some issues arisen preventing easy and efficient spectral library transfers.