HR-MAS NMR tissue metabolomic signatures cross-validated by mass spectrometry distinguish bladder cancer from benign disease

Abstract

Effective diagnosis and surveillance of bladder cancer (BCa) is currently challenged by detection methods that are of poor sensitivity, particularly for low-grade tumors, resulting in unnecessary invasive procedures and economic burden. We performed HR-MAS NMR-based global metabolomic profiling and applied unsupervised principal component analysis (PCA) and hierarchical clustering performed on NMR data set of bladder-derived tissues and identified metabolic signatures that differentiate BCa from benign disease. A partial least-squares discriminant analysis (PLS-DA) model (leave-one-out cross-validation) was used as a diagnostic model to distinguish benign and BCa tissues. Receiver operating characteristic curve generated either from PC1 loadings of PCA or from predicted Y-values resulted in an area under curve of 0.97. Relative quantification of more than 15 tissue metabolites derived from HR-MAS NMR showed significant differences (P < 0.001) between benign and BCa samples. Noticeably, striking metabolic signatures were observed even for early stage BCa tissues (Ta-T1), demonstrating the sensitivity in detecting BCa. With the goal of cross-validating metabolic signatures derived from HR-MAS NMR, we utilized the same tissue samples to analyze 8 metabolites through gas chromatography-mass spectrometry (GC-MS)-targeted analysis, which undoubtedly complements HR-MAS NMR-derived metabolomic information. Cross-validation through GC-MS clearly demonstrates the utility of a straightforward, nondestructive, and rapid HR-MAS NMR technique for clinical diagnosis of BCa with even greater sensitivity. In addition to its utility as a diagnostic tool, these studies will lead to a better understanding of aberrant metabolic pathways in cancer as well as the design and implementation of personalized cancer therapy through metabolic modulation.

Figure 1

Representative 600 MHz ¹H HR-MAS CPMG NMR spectra (area normalized) of benign (A), and various bladder cancer pathologic stages Ta/T1 (B and C) and ≥T2 (D). The vertical scale of all the spectra were kept constant and spectral regions containing spinning side bands (5.6 – 6.0 ppm) and OCT (3.6 – 3.8 ppm) were removed from the figure. The intensity of peaks in the chemical shift region 6.0 – 8.5 ppm was enhanced equally in all the spectra to show the low abundant metabolites. The triglycerides signals were indicated as ‘TG’. The corresponding H & E photomicrographs obtained from a section dissected from the same tissues are also shown alongside of each ¹H CPMG spectrum.

Figure 2

(A) Unsupervised 2D PCA score plot (B) supervised 2D PLS-DA score plot and (C) PC1 loadings of a total of 59 tissues generated from ¹H CPMG spectra of 26 benign, 17 bladder cancer Ta-T1 and 16 bladder cancer ≥T2 stage. Both PCA and PLS-DA score plots showed clear segregation of benign and BCa groups, predominantly due to PC1 loading. From PC1 loadings, TG levels were clearly decreased (negative loadings) while all other aqueous metabolites were elevated (positive loadings) in BCa compared to benign tissues. Few benign samples (indicated by black arrows) categorized under BCa samples were predominantly benign urothelium partially contaminated with urothelial carcinoma. Figure 2D and 2E show the average ¹H CPMG spectrum of all 26 benign spectra and 33 bladder cancer spectra, respectively. The spectral difference between the two average spectra is consistent with PC1 loadings.

Figure 3

Heat map showing z-score of 15 quantified metabolite entities in 26 benign and 33 bladder cancer tissues. An unsupervised hierarchical classification showing separate cluster for benign and bladder cancer tissues is also included in the heat map. Only few benign samples fall under BCa group those were partially contaminated with BCa tissues during resection. Heat map shows significant differences in the concentration of all metabolites between benign and cancer tissues. The difference in concentration of metabolites was further assessed using Mann-Whitney U test followed by Benjamini–Hochberg correction, and all the fifteen metabolites were found to be statistically significant (P < 0.001).

Figure 4

Box plot showing the levels of select metabolites that were measured in benign, Ta/T1 and ≥T2 tissues (normalized to the tissue weight) by targeted GC-MS analysis. Analysis was performed on the same tissues that were recovered from HR-MAS NMR analysis. The levels of metabolites were compared using Mann-Whitney U test followed by Benjamini–Hochberg correction and all the metabolites were found to be significantly elevated in Ta/T1 and ≥T2 cancer tissues, however, no significant changes were observed between Ta/Ta and ≥T2 tissues, consistent with HR-MAS NMR findings. Benign Vs Ta/T1: * P ≤ 0.05, ** P ≤ 0.01 and *** P ≤ 0.001 ; Benign Vs ≥T2: ■ P ≤ 0.05, ■■ P ≤ 0.01 and ■■■ P ≤ 0.001.

Figure 5

ROC curves generated from PC1 scores (A) and predicted Y-values of PLS-DA (B). AUC values of 0.978 and 0.979 respectively, represent a good predictability of PCA and PLS-DA model in differentiating benign and bladder cancer tissues.