Multimodal machine learning and Raman spectroscopy uncover biochemical pathways of autumnal leaf senescence
摘要
Quercus robur – the English oak, undergoes various complex biochemical changes during the process of senescence, one of the most significant being the catabolism of chlorophyll species. The full process of foliaceous ageing remains elusive and largely enigmatic. Here, we employed the use of Raman spectroscopy, density-functional theory calculation, spectrophotometry and colour channel analysis combined with an advanced artificial neural network, to explore senescence-induced biochemical changes to non-venous leaf tissue from Q. robur through the process of autumnal senescence. Our analysis demonstrates an increase of Raman-derived crystallinity from 0.185 in non-senescing leaves through to 0.594 in senescing leaves combined with a decrease of the intensity of the 854 cm− 1 peak and accompanying increase of the intensity of the 898 cm− 1 peak, which in totality suggests that pectin is moving from the α- to β- anomeric form as cellulose moves from an amorphous state into a more crystalline one. Further analysis indicates inconsistencies to the chlorophyll and carotenoid behaviours as leaves move from a non-senescent state to a fully senesced one, which combined with an unexpected increase to the 1215–1320 cm− 1 peak region in fully senesced tissue, suggests the presence of new Raman-active bond contributors in fully senesced tissue. Analysis of expected and predicted chlorophyll catabolites indicated the presence of unique bond vibration contributors at 1490 and 1495 cm− 1 in all senescence classes and 1510 cm− 1 in all senescence classes except in fully senesced tissue, highlighting both the progression of chlorophyll catabolism and potential further breakdown of late-stage catabolites. Classification via an advanced artificial neural network showed testing accuracies of 68.18% for non-senescent tissue, 83.40% for visually non-senescing tissue on minimally senesced leaves, 78.15% for visually non-senescing tissue on moderately senesced leaves, 94.15% for minimally senesced tissue, 99.20% for moderately senesced tissue and 100% for fully senesced tissue, yielding an overall differentiation accuracy of 88.50%. Insights obtained herein offer future avenues for both exploratory research and industrial application, namely the exploitation of foliar material for a high-value commercial and medicinal products.