Distinguishing Obstruction and Reflux: A Pilot Study of the Urinary Proteome
Ted Lee, MD, MSc1, Dijana Vitko, PhD1, Ali H. Gheinani, PhD1, Tanya Logvinenko, PhD1, Shannon DiMartino, BS1, John W. Froehlich, PhD1, Richard S. Lee, MD2.
1Boston Children's Hospital, Boston, MA, USA, 2Boston Children's Hospital, Brooklin, MA, USA.
BACKGROUND: Determining the underlying cause of collecting system dilatation can be challenging. Current standard-of-care diagnostic methods that distinguish obstruction from reflux are invasive (e.g., voiding cystourethrography, radionuclide scans, antegrade fluoroscopy). A non-invasive method that can differentiate obstruction from reflux would be an immensely valuable clinical tool. In this study, we hypothesized that the urinary proteome can distinguish uretero-pelvic junction obstruction (UPJO) from vesico-ureteral reflux (VUR).
METHODS: In total, 122 children with UPJO and 122 children with VUR were identified from a long-standing, prospective pediatric biorepository. Urine samples were collected just prior to surgery and systematically processed. Urinary proteins were purified, digested with trypsin and analyzed by liquid chromatography coupled to mass spectrometer. Peptide spectral libraries were generated by pooling 30 representative samples from UPJO and VUR cohorts, fractionated into 12 fractions based on peptide isoelectric point (pI 3-10), and analyzed by data-dependent acquisition (DDA) method. All samples were then analyzed via data-independent acquisition (DIA) method. Proteome Discoverer 2.2 and Spectronaut 14 software were used to process DDA and DIA raw files, respectively. Final proteomes were identified by raw file search against DDA- and in silico-generated spectral libraries (FDR<0.01). Protein quantitation was inferred from area-under-the-curve (AUC) corresponding to intact peptides (MS1 spectra) and intensity of minimum 3 fragmented peptide ions (MS2 spectra). From 3372 proteins, 655 proteins were consistently identified across both cohorts and used for comparison. All samples were randomly divided into a training (n=170) and test (n=74) sets. When performing logistic regression to fit the prediction model, penalization techniques (least absolute shrinkage and selection operator (LASSO)) and dimensionality reduction techniques (Principal Components Analysis (PCA)) were used to prevent overfitting. LASSO regression and PCA regression were fitted in the training set and performance characteristics were obtained from the test set. The optimal regularization parameter in LASSO regression was determined using five-fold cross-validation. Only principal components explaining greater than 0.01 of the variance were used to fit the PCA regression.
RESULTS: The accuracy and AUC of the LASSO regression of the test set were 0.78 and 0.87, respectively. The optimal regularization parameter was 10,000 after five-fold cross validation. A total of 526 proteins were included in the regularized model. The accuracy and AUC of the PCA regression of the test set were 0.76 and 0.82, respectively. 15 principal components that were utilized in the PCA regression model explained 65% of the variability between UPJO and VUR subjects.
This pilot study demonstrates vastly differing urinary proteomes between patients with UPJO and VUR. The urinary proteome may be leveraged to distinguish urinary obstruction from reflux in a non-invasive manner. Increased sample size and appropriate handling of inconsistently identified proteins across patient cohorts will result in reliable predictive performance. Identification of key urinary proteins that contribute the most to prediction of obstruction or reflux will assist with clinical translation of our preliminary findings.
Back to 2021 Abstracts