Hyperspectral imaging and metabolomics reveal color classification and anthocyanin profiles in Hydrangea macrophylla cultivars

Flower color is a key criterion for evaluating Hydrangea macrophylla germplasm and breeding programs. In this study, 95 accessions were classified into six color groups (white, yellow, pink, red, purple, and blue) using the CIELAB and Munsell color systems. Partial least squares discriminant analysis (PLS-DA) of color space values ( L *, a *, b *, C , and h ) and pigment contents (anthocyanins, carotenoids, flavonoids, phenols) showed that the C value was crucial for the red group, while the L * value was important for the white group. A total of 83 anthocyanin derivatives were identified, with cyanidins and delphinidins being the most abundant. The proportion of delphinidin-3-O-glucoside (Dp3g) was a key factor in sepal color diversity, accounting for less than 11 % of the variation in the white and yellow groups and more than 83 % in the other groups. Dp3g was upregulated in the red, pink, purple, and blue groups compared to the yellow group, whereas malvidin-3-O-arabinoside was upregulated in the red, pink, and blue groups relative to the purple group. Hyperspectral imaging combined with partial least squares regression (PLSR) accurately predicted pigment contents (R² = 0.767–0.939, RPD = 1.946–2.636), with the best performance for anthocyanins. The spectral models for delphinidin-3,5-O-diglucoside (D3g5g) and Dp3g presented the highest correlations among differentially accumulated metabolites (|r| > 0.7), with the D3g5g model achieving the best performance (R² = 0.944, RMSE = 0.101 μg·g⁻¹ FW, RPD = 3.697). This research provides a comprehensive classification of colors and anthocyanin profiles in H. macrophylla sepals and highlights the potential of hyperspectral imaging in breeding practices. • Comprehensive classification of H. macrophylla sepals based on CIELAB and Munsell color systems. • The proportion of Dp3g in anthocyanins is crucial for sepal color diversity. • Dp3g and D3g5g are identified as key metabolites highly correlated with sepal color and spectral characteristics. • Hyperspectral imaging combined with PLSR enables accurate prediction of pigment and metabolite contents.