Flavin fluorescence lifetime and autofluorescence optical redox ratio for improved visualization and classification of brain tumors

Abstract

Purpose: Modern techniques for improved tumor visualization have the aim to maximize the extent of resection during brain tumor surgery and thus improve patient prognosis. Optical imaging of autofluorescence is a powerful and non-invasive tool to monitor metabolic changes and transformation in brain tumors. Cellular redox ratios can be retrieved from fluorescence emitted by the coenzymes reduced nicotinamide adenine dinucleotide (phosphate) (NAD(P)H) and flavin adenine dinucleotide (FAD). Recent studies point out that the influence of flavin mononucleotide (FMN) has been underestimated.

Experimental design: Fluorescence lifetime imaging and fluorescence spectroscopy were performed through a modified surgical microscope. We acquired 361 flavin fluorescence lifetime (500-580 nm) and fluorescence spectra (430-740 nm) data points on freshly excised different brain tumors: low-grade gliomas (N=17), high-grade gliomas (N=42), meningiomas (N=23), metastases (N=26) and specimens from the non-tumorous brain (N=3).

Results: Protein-bound FMN fluorescence in brain tumors did increase with a shift toward a more glycolytic metabolism (R=-0.87). This increased the average flavin fluorescence lifetime in tumor entities with respect to the non-tumorous brain. Further, these metrics were characteristic for the different tumor entities and showed promise for machine learning based brain tumor classification.

Conclusions: Our results shed light on FMN fluorescence in metabolic imaging and outline the potential for supporting the neurosurgeon in visualizing and classifying brain tumor tissue during surgery.

Keywords: flavin mononucleotide; fluorescence guided surgery; fluorescence lifetime imaging; fluorescence spectroscopy; optical redox ratio.

Figure 1

(A):Average fluorescence spectra of low- and high grade gliomas (LGG, n=71, HGG, n=117), meningiomas (MNG, n=64), brain metastases (MET, n=93) and non-tumorous brain (CTL, n=16) for excitation at 405 nm. The spectral bands for integration of NAD(P)H and FAD fluorescence are colored in grey. Note the peaks at about 462 nm (free NAD(P)H), 495 nm (FMN) and 530 nm (FAD, side peak of FMN). (B): Optical redox ratios obtained from the integration of NAD(P)H and FAD fluorescence in spectra according to equation 1. (C): Ratio of the peaks at 495 nm and 530 nm reflecting the ratio of protein-bound FMN to FAD. (D): Flavin fluorescence lifetime for the spectral range from 500 nm to 580 nm. For (B–D), statistical significant differences of the tumor entities to the control group are indicated through asterisks (* p<0.005, ** p<1e-5).

Figure 2

(A): Higher protein-bound FMN fluorescence with respect to FAD was found to correlate with a reduced optical redox ratio (R = -0.87). All tumor entities predominantly showed reduced redox ratios and higher FMN fluorescence than non-tumorous brain (control, CTL). Only metastases (MET) were found to be heterogeneous and partly overlapped with the non-tumorous brain cluster. (B): The amount of FMN fluorescence plotted as a function of the flavin lifetime (500-580 nm) showed characteristic clusters for the different groups. While MET weren’t easily distinguishable from the CTL group in the redox ratio, flavin lifetime was predominantly increased beyond the lifetimes in non-tumorous brain. (C): Redox ratio plotted as a function of the flavin lifetime implied that the predominant metabolic strategy was characteristic for the different tumor entities. Note the high flavin lifetimes of MNG. Markers further indicate the predominant histopathological tissue classification for (A-C) (INF: tumor infiltrated brain, NEC: necrosis, REA: reactive tissue, TUM: core tumor). (D): Confusion matrix for classification of the different groups with a support vector machine. Input to the model were the optical redox ratio, the FMN/FAD ratio and the flavin lifetime. Autofluorescence was highly characteristic for CTL and meningiomas (MNG). MET were partly misclassified as high-grade gliomas (HGG) implying similarities with respect to the predominant metabolic strategy. Further, classification uncertainty existed between low-grade gliomas (LGG) and HGG. Multi-class sensitivity (SE) and specificity (SP) are given at the end of each row.

Figure 3

Selected clinical cases of (A) non-tumorous control (CTL) tissue, (B) a WHO grade II diffuse astrocytoma (low-grade, LGG), (C) a glioblasatoma (high-grade, HGG), (D) a bronchial carcinoma metastasized to the cerebellum and (E) a meningothelial meningioma (MNG). (I) shows a representative haematoxylin and eosin (H&E) stain of the respective sample, (II) the demodulated fluorescence intensity in mV_RMS, (III) the flavin fluorescence lifetime in ns (500-580 nm) and (IV) the spectroscopic measurements corresponding to the regions indicated in III.