Multi-omics Analysis for Prediction of Survival in Bladder Cancer Patients - Expert Commentary
The prospective study included 74 bladder cancer patients who had neoadjuvant treatment and radical cystectomy. Of these, 78% had muscle-invasive bladder cancer, 92% had urothelial carcinoma, and the remainder had variant histologies. The researchers performed deep sequencing (uCAPP-Seq) and ultra-low-pass whole genome sequencing (ULP-WGS) on patient urine cfDNA samples. Urine cfDNA samples from MIBC patients exhibited recurrent alterations in focal copy numbers in genes previously identified by The Cancer Genome Atlas (TCGA). Analysis of single nucleotide variants (SNV) revealed that the TERT promoter and TP53 were the most common mutations. Importantly, alterations in copy number and SNVs were not observed in urine cfDNA of healthy adults.
There was a correlation between pathologic complete response (pCR) and urine cfDNA features in bladder cancer patients, with those who achieved pCR having significantly lower variant allele frequency levels. As an assay, urine cfDNA significantly outperformed plasma circulating tumor DNA evaluation. Tumor mutational burden could also be inferred from urine cfDNA, with patients who did not exhibit pCR having significantly higher levels than patients with pCR. Copy number alterations were also significantly different depending on pCR status.
The different parameters in urine cfDNA that were found to be associated with pCR were subsequently integrated with pre-treatment clinical variables. The resulting machine learning model exhibited a sensitivity of 87%, a negative predictive value of 77%, and a positive predictive value of 65% for determining pCR. The urine cfDNA metrics were found to be the most important predictive features in the model. The model could also accurately predict survival among patients, with patients who were predicted to have MRD exhibiting significantly lower progression-free survival (PFS) and overall survival. A parallel analysis using univariate and multivariate Cox proportional hazards models validated the machine learning model’s predictions.
The integration of various sequencing techniques for evaluating molecular biomarkers of bladder cancer prognosis in urine cfDNA led to high accuracy compared to the gold standard for detecting MRD (pathologic analysis of surgically resected bladder). However, the accessibility and availability of specialized sequencing techniques may hinder the widespread use of this approach. Additional limitations of the study include the evaluation of urine cfDNA at a single time point and a relatively short follow-up period.
Written by: Bishoy M. Faltas, MD, Director of Bladder Cancer Research, Englander Institute for Precision Medicine, Weill Cornell Medicine
Reference:
Read the Abstract