Researchers have used machine learning to predict which patients with advanced bladder cancer are most likely to respond to immunotherapy with immune checkpoint inhibitors (ICIs).
ICIs were a major breakthrough in the treatment of advanced bladder cancer, researchers wrote in the journal Nature Communications, but were only effective in a subset of patients.
Bladder cancer also showed a high degree of heterogeneity, with five muscle-invasive subtypes, and it had been hypothesised that the subtypes with a high proportion of infiltrated immune cells were more likely to respond to ICI.
For this study, the authors integrated mutation and gene expression data from 707 patients from six cohorts of advanced bladder cancer patients treated with anti-programmed cell death protein 1 (PD-1) and anti-programmed death-ligand 1 (PD-L1) ICI to build computational predictive models.
Machine learning tools enabled them to identify key variables that influenced the success of immunotherapy treatment.
Analysis of the dataset found that, contrary to the hypothesis, patients with high immune filtration, such as those with basal-squamous and luminal-infiltrated tumour types, did not show an overall better response.
For non-immune-infiltrated subtypes, other biomarkers appeared to be particularly relevant, the researchers noted, which illustrated the importance of accounting for specific tumour types or the degree of immune inflammation.
Overall, the neuronal subtype of tumour was found to have the highest response rate to immunotherapy.
‘It has been hypothesised that this kind of tumour could be more responsive to ICI due to the expression of tissue-restricted neuronal or neuro-endocrine proteins, together with low [transforming growth factor β] TGF-β expression values,’ the study authors wrote, adding that their study ‘confirmed that TGF-β levels are particularly low in this class’.
In other findings, the study confirmed that the number of somatic mutations in the tumour was associated with immunotherapy treatment response, but researchers also found enrichment in the apolipoprotein B mRNA editing catalytic polypeptide-like (APOBEC) mutational signature and abundance of pro-inflammatory macrophages were major factors associated with response.
In addition, they found several unexplored markers of response to ICI, including the number of non-stop mutations.
‘The results have uncovered unexplored ICI associated variables and have shed light on the complex interplay between tumour biology, the immune microenvironment and treatment response,’ the study authors wrote.
‘Our study underlines the importance of subtype-specific factors for personalised treatment strategies and enhanced patient outcomes in the era of immunotherapy.’
The study was conducted by researchers at the Biomedical Informatics Research Programme (GRIB) and the Cancer Programme at the Hospital del Mar Research Institute in Barcelona, Spain, with the collaboration of the city’s Pompeu Fabra University.
Commenting on the findings, study lead author Lilian Marie Boll, a researcher for the GRIB, said: ‘The key to our study is understanding the response mechanisms within these subgroups, rather than treating all bladder cancer as a single entity.’
Study co-author Dr Joaquim Bellmunt, coordinator of the Urologic Cancer Research Group at the Hospital del Mar Research Institute and the Dana-Farber Cancer Institute in Boston, US, said the study expanded the current knowledge base of tumour heterogeneity, which was a limiting factor in immunotherapy efficacy.
‘It highlights the importance of identifying immune cell populations that facilitate immunotherapy response, while others have inhibitory effects,’ he said.
The findings add to other recent literature on bladder cancer, including a study that found multiparametric magnetic resonance imaging (mpMRI) was both feasible and beneficial in reducing treatment delays in the initial staging of muscle-invasive bladder cancer.
This prospective open-label, randomised study found using mpMRI in the muscle-invasive bladder cancer pathway reduced the time to correct treatment by 45 days compared to standard investigation.
The researchers concluded that introducing mpMRI ahead of transurethral resection of bladder tumour into the standard pathway was beneficial for all patients with suspected muscle-invasive bladder cancer as it could improve clinical decision making and accelerate the time to treatment.
Machine learning has also recently been used to improve the personalised prognostication of aggressive skin cancers, such as Merkel cell carcinoma.
