Serum Metabolomics of Retinoblastoma: Assessing the Differential Serum Metabolic Signatures of Unilateral and Bilateral Patients

Retinoblastoma (Rb) is the most common pediatric eye cancer. To identify the biomarkers for early diagnosis and monitoring the progression of Rb in patients, mapping of the alterations in their metabolic profiles is essential. The present study aims at exploring the metabolic disparity in serum from Rb patients and controls using NMR-based metabolomics. A total of 72 metabolites, including carbohydrates, amino acids, and organic acids, were quantified in serum samples from 24 Rb patients and 26 controls. Distinct clusters of Rb patients and controls were obtained using the partial least-squares discriminant analysis (PLS-DA) model. Further, univariate and multivariate analyses of unilateral and bilateral Rb patients with respect to their age-matched controls depicted their distinct metabolic fingerprints. Metabolites including 2-phosphoglycerate, 4-aminobutyrate, proline, O-phosphocholine, O-phosphoethanolamine, and Sn-glycero-3-phosphocholine (Sn-GPC) showed significant perturbation in both unilateral and bilateral Rb patients. However, metabolic differences among the bilateral Rb cases were more pronounced than those in unilateral Rb cases with respect to controls. In addition to major discriminatory metabolites for Rb, unilateral and bilateral Rb cases showed specific metabolic changes, which might be the result of their differential genetic/somatic mutational backgrounds. This further suggests that the aberrant metabolic perturbation in bilateral patients signifies the severity of the disease in Rb patients. The present study demonstrated that identified serum metabolites have potential to serve as a noninvasive method for detection of Rb, discriminate bilateral from unilateral Rb patients, and aid in better understanding of the RB tumor biology.


Figure S1 :
Figure S1: Multivariate statistical analysis of human serum metabolites from Rb patients and controls: Representative PCA (principal component analysis) score plot depicting the discrimination between the two groups: Controls (pink) and Rb patients' groups (blue).

Figure S2 :
Figure S2: Performance measurements for the PLS-DA model generated for the concentration dataset of 75 metabolites from the serum of controls versus Rb patients: (A) Bar plots presenting the performance parameters (accuracy, goodness of fit (R 2 ) and predictability (Q 2 )) up to the 5 th component attained using the 10-fold cross validation.The best classifier of the model is marked as the red star symbol.(B) Cross validation performance measurement values are displayed for all the five components of the PLS-DA model.Validation parameters obtained for the best PLS-DA model include the accuracy of 0.93, R 2 and Q 2 values equal to 0.98 and 0.73 respectively.

Figure
Figure S4: (A) 2D PLS-DA score plot generated for the concentration dataset of 75 metabolites annotated from 1 H NMR spectra of serum samples from CL3 (pink) and CG3 (red).(B) VIP score plot generated on the basis of PLS-DA analysis of CL3 versus CG3 depicting the top 20 significantly altered metabolites among different sub-groups.(C) Bar plots representing the performance measurements: accuracy, goodness of fit (R 2 ), and predictability (Q 2 ) up to the 5 th component of PLS-DA model attained following the 10-fold cross validation analysis of multivariate data.The best classifier of the model is marked using the red star symbol.(D) Cross validation performance measurement values are displayed for all the five components of the PLS-DA model.

Figure S5 :
Figure S5: Bar plots representing the performance measurements (accuracy, goodness of fit (R 2 ) and predictability (Q 2 ) up to the 5 th component of PLS-DA model generated for the concentration dataset of 72 metabolites from the serum samples of (A) CL3 versus RL3, (C) CL3 versus RL3U, (E) CL3 versus RL3B, using the 10-fold cross validation analysis.The best classifier of the model is marked using the red star symbol.Cross validation performance measurement values are displayed for all the five components of the PLS-DA model for (B) CL3 versus RL3, (D) CL3 versus RL3U, (F) CL3 versus RL3B.A 10-fold cross validation of the best PLS-DA model for CL3 vs. RL3 displayed the accuracy value of 0.9, R 2 = 0.74 and Q 2 =0.56, for CL3 vs. RL3U: accuracy value=0.91,R 2 = 0.99, Q 2 = 0.75 and for CL3 vs. RL3B: accuracy value=0.95,R 2 = 0.98, Q 2 = 0.58.

Figure S6 :
Figure S6: Heat map displaying the z-scores of top 25 discriminatory metabolites altered in RL3 with respect to CL3.X-axis represents the serum samples from CL3 (CL3; lanes:1-11; pink) and RL3 (RL3; lanes: 12-31; purple).The color gradients signify the differential metabolite concentration in the serum samples, with dark red and dark blue implies the highest and lowest metabolite concentrations.

Figure S7 :
Figure S7:Heat map displaying the z-scores of top 25 discriminatory metabolites altered in RL3U with respect to CL3.X-axis represents the serum samples from CL3 (CL3; lanes:1-11; pink) and RL3U (RL3U; lanes: 12-24; blue).The color gradients signify the differential metabolite concentration in the serum samples, with dark red and dark blue implies the highest and lowest metabolite concentrations.

Figure S8 :
Figure S8:Heat map displaying the z-scores of top 25 discriminatory metabolites altered in RL3B with respect to CL3.X-axis represents the serum samples from CL3 (CL3; lanes: 1-11; pink) and RL3B (RL3B; lanes: 12-18); blue).The color gradients signify the differential metabolite concentration in the serum samples, with dark red and dark blue implies the highest and lowest metabolite concentrations.

Figure S9 :
Figure S9: (A) 2D PLS-DA score plot generated for the concentration dataset of 75 metabolites annotated from 1 H NMR spectra of serum samples from CL3 (CL3, pink), RL3U (blue), RL3B (green).(B) VIP score plot generated on the basis of PLS-DA analysis of CL3 versus RL3U and RL3B depicting the top 20 significantly altered metabolites among different sub-groups.(C) Bar plots representing the performance measurements (accuracy, goodness of fit (R 2 ) and predictability (Q 2 )) up to the 5 th component of PLS-DA model attained following the 10-fold cross validation analysis of multivariate data.The best classifier of the model is marked using the red star symbol.(D) Cross validation performance measurement values are displayed for all the five components of the PLS-DA model.Cross validation parameters for the best PLS-DA model include the accuracy, R 2 , and Q 2 equals to 0. 59, 0.75, and 0.56 respectively.

Figure S10 :
Figure S10: (A) 2D PLS-DA score plot generated for the concentration dataset of 75 metabolites annotated from 1 H NMR spectra of serum samples from healthy CG3 (red), with RG3U (yellow).(B) VIP score plot generated on the basis of PLS-DA analysis of CG3 versus RG3U depicting the top 20 significantly altered metabolites among different sub-groups.(C) Bar plots representing the performance measurements (accuracy, goodness of fit (R 2 ) and predictability (Q 2 )) up to the 5 th component of PLS-DA model attained following the 10-fold cross validation analysis of multivariate data.The best classifier of the model is marked using the red star symbol.(D) Cross validation performance measurement values are displayed for all the five components of the PLS-DA model.Cross validation parameters of the best PLS-DA model showed accuracy, R 2 , and Q 2 values equal to 0. 96, 0.98, and 0.64 respectively.

Figure S11 :
Figure S11: The fitting of 2-hydroxybutyrate and 4-Aminobutyrate peaks in the Chenomx are shown below.The NMR signals corresponding to 2-hydroxybutyrate and 4-aminobutyrate were identified and assigned by matching their specific chemical shifts and peak patterns using Chenomx software.Blue and red peak lines indicate the fitting of metabolites, and the black lines display experimental peak patterns.

Table S2 :
List of metabolites identified from 1D 1 H-NMR spectra attained for serum samples from Rb patients and controls (peak types: s = singlet, d= doublet, t= triplet, q= quartet, m=multiplet, td =triplet of doublet, tt = triplet of triplet, dd=doublet of doublet, ddd = doublet of doublet of doublet, quint = quintet, dtd = doublet of triplet of doublet).The fitting of 2hydroxybutyrate and 4-Aminobutyrate peaks in the Chenomx are presented in FigureS11.

Table S3 :
Comparative analysis of significantly altered metabolic pathways with -log 10(p) values > 0.5 and pathway impact values > 0.01 obtained via pairwise pathway impact analysis of all the Rb patients with respect to controls.

Table S4 :
Comparative analysis of top-most metabolites with VIP score greater than 1 based on PLS-DA model generated for different subgroups including: CL3 vs RL3, CL3 vs. RL3U, CL3 vs. RL3B.Green and red color denotes increased and decreased levels of metabolites in serum from Rb patients in contrast to controls.

Table S6 :
Pairwise pathway impact analysis of CL3 versus RL3U depicting the significantly altered metabolic pathways along with their -log 10(p) values and pathway impact values calculated based on the pathway enrichment and pathway topology analysis respectively.

Table S7 :
Pairwise pathway impact analysis of CL3 versus RL3B, depicting the significantly altered metabolic pathways along with their -log 10(p) values and pathway impact values calculated based on the pathway enrichment and pathway topology analysis respectively.