Skip to main content
main-content

01.12.2018 | Original research | Ausgabe 1/2018 Open Access

EJNMMI Research 1/2018

A novel genomic signature predicting FDG uptake in diverse metastatic tumors

Zeitschrift:
EJNMMI Research > Ausgabe 1/2018
Autoren:
Aurora Crespo-Jara, Maria Carmen Redal-Peña, Elena Maria Martinez-Navarro, Manuel Sureda, Francisco Jose Fernandez-Morejon, Francisco J. Garcia-Cases, Ramon Gonzalez Manzano, Antonio Brugarolas
Wichtige Hinweise

Electronic supplementary material

The online version of this article (https://​doi.​org/​10.​1186/​s13550-017-0355-3) contains supplementary material, which is available to authorized users.
Ramon Gonzalez Manzano and Antonio Brugarolas contributed equally to this work.

Abstract

Background

Building a universal genomic signature predicting the intensity of FDG uptake in diverse metastatic tumors may allow us to understand better the biological processes underlying this phenomenon and their requirements of glucose uptake.

Methods

A balanced training set (n = 71) of metastatic tumors including some of the most frequent histologies, with matched PET/CT quantification measurements and whole human genome gene expression microarrays, was used to build the signature. Selection of microarray features was carried out exclusively on the basis of their strong association with FDG uptake (as measured by SUVmean35) by means of univariate linear regression. A thorough bioinformatics study of these genes was performed, and multivariable models were built by fitting several state of the art regression techniques to the training set for comparison.

Results

The 909 probes with the strongest association with the SUVmean35 (comprising 742 identifiable genes and 62 probes not matched to a symbol) were used to build the signature. Partial least squares using three components (PLS-3) was the best performing model in the training dataset cross-validation (root mean square error, RMSE = 0.443) and was validated further in an independent validation dataset (n = 13) obtaining a performance within the 95% CI of that obtained in the training dataset (RMSE = 0.645). Significantly overrepresented biological processes correlating with the SUVmean35 were identified beyond glycolysis, such as ribosome biogenesis and DNA replication (correlating with a higher SUVmean35) and cytoskeleton reorganization and autophagy (correlating with a lower SUVmean35).

Conclusions

PLS-3 is a signature predicting accurately the intensity of FDG uptake in diverse metastatic tumors. FDG-PET might help in the design of specific targeted therapies directed to counteract the identified malignant biological processes more likely activated in a tumor as inferred from the SUVmean35 and also from its variations in response to antineoplastic treatments.
Zusatzmaterial
Additional file 1: Supplementary Methods. (DOCX 29 kb)
13550_2017_355_MOESM1_ESM.docx
Additional file 2: Table S4. Complete list of the 909 probes selected for the generation of the multivariable model along with their correspondent regression coefficient. (DOCX 110 kb)
13550_2017_355_MOESM2_ESM.docx
Additional file 3: Table S6. Detailed tumor histologies of the patients in the training and validation datasets. (DOCX 17 kb)
13550_2017_355_MOESM3_ESM.docx
Additional file 5: Figure S1. Selected clusters identified in the Protein Protein Interaction (PPI) subnetworks obtained from the signature genes in the STRING 10.0 PPI database. Figure S2. Goodness of fit of PLS-3 in the training set. a) Goodnes of fit including Pearson correlation of measured vs predicted SUV values b) Residuals of third component. No pattern is apparent in the residuals distribution c) Estimated goodness of fit after 10-fold CV. (ZIP 242 kb)
13550_2017_355_MOESM5_ESM.zip
Additional file 6: Table S2. Correlation coefficient (CC) on SUV of ssGSEA scores with the C2 subset from the MSigDB v5.1 in the training dataset (p < 0.05) (DOCX 31 kb)
13550_2017_355_MOESM6_ESM.docx
Additional file 7: Table S3. Characteristics of the patients in the validation set along with their measured and predicted (SUVPLS) SUV values. (DOCX 13 kb)
13550_2017_355_MOESM7_ESM.docx
Additional file 8: Table S5. List of PLS-3 probes with VIP values equal or greater than 1 along with their regression coefficients. (XLSX 34 kb)
13550_2017_355_MOESM8_ESM.xlsx
Literatur
Über diesen Artikel

Weitere Artikel der Ausgabe 1/2018

EJNMMI Research 1/2018 Zur Ausgabe