Inflammation drives bladder carcinogenesis. Chronic inflammation has been implicated in the initiation and progression of bladder cancer through multiple mechanisms including mutagenesis from reactive oxygen and nitrogen species, promotion of cell proliferation, inhibition of apoptosis, and stimulation of angiogenesis and tumor invasion.
Inflammatory cytokines, growth factors, and immune effectors collectively remodel the tumor microenvironment to favor cancer development. Genetic variants in inflammation-related genes may therefore modulate individual susceptibility to bladder cancer by altering the intensity or nature of inflammatory responses.
While epidemiological studies consistently link chronic infection, bladder inflammation, and immune dysregulation to bladder cancer risk, the specific genetic architecture of inflammatory gene variants contributing to risk has not been fully characterized, in part because most studies examined variants one at a time rather than capturing their joint effects.
Standard approaches miss joint genetic effects. Conventional genome-wide association studies and candidate gene studies test each single nucleotide polymorphism (SNP) independently using univariate logistic regression, treating each variant as if it acts in biological isolation from all others.
This approach ignores the reality that complex diseases like bladder cancer arise from the combined effects of many genetic variants with individually small effects, gene-gene interactions, and nonlinear contributions. Multiple testing corrections applied to handle the large number of individual tests further reduce statistical power to detect genuine but modest associations.
Multi-SNP methods that analyze all variants simultaneously can identify combinations of SNPs that jointly predict disease risk even when no single variant reaches significance alone. However, such methods must handle high-dimensional data where the number of predictors (hundreds of SNPs) exceeds or approaches the number of subjects, requiring specialized statistical frameworks.
Spanish Bladder Cancer Study with 886 inflammatory variants. The analysis used data from the Spanish Bladder Cancer (SBC)/EPICURO Study, a hospital-based case-control study comprising 1,047 bladder cancer cases and 988 controls recruited from 18 hospitals across five Spanish regions between 1998 and 2001.
A total of 886 SNPs in 886 probes within 144 inflammation-related genes were analyzed, selected based on prior biological knowledge of inflammatory pathways relevant to cancer. These genes included cytokines, chemokines, toll-like receptors, apoptosis mediators, and immune cell surface markers involved in regulating inflammatory responses.
Genotyping was performed using the Illumina HumanHap300 array with supplementary SNPs added to cover specific inflammatory genes of interest. Quality control excluded SNPs with call rates below 90%, minor allele frequencies below 1%, or significant deviation from Hardy-Weinberg equilibrium in controls. Replication was performed in the independent Texas Bladder Cancer Study.
Two complementary multi-SNP frameworks applied simultaneously. The Bayesian Threshold LASSO (BTL) uses a regularization prior that shrinks small SNP coefficients toward zero while allowing large coefficients to remain unconstrained, effectively selecting a sparse subset of variants with meaningful associations while controlling for all other variants simultaneously.
BTL reports a posterior inclusion probability for each SNP, representing the probability that the variant has a genuine non-zero effect given the data. A threshold of 50% posterior probability was used to define selected SNPs, with those above 80% considered strong hits. Bayesian credible intervals were derived from the posterior distribution of effect sizes.
The AUC-Random Forest (AUC-RF) method trains an ensemble of decision trees on bootstrap samples of the data, optimizing the area under the ROC curve rather than standard classification accuracy to handle the case-control design. Variable importance scores based on AUC decrease upon permutation were used to rank SNPs, with a permutation-based significance threshold applied to identify the top hits.
BTL identified 37 SNPs across 34 genes. The Bayesian Threshold LASSO selected 37 SNPs with posterior inclusion probabilities above 50%, with the top hit being CASP3-rs3087455 at 96.07% posterior probability. This caspase-3 variant, involved in apoptosis execution, showed a protective association consistent with enhanced apoptotic clearance of pre-malignant cells.
Other protective SNP associations identified by BTL included variants in CCR3 (chemokine receptor), CASP9 (initiator caspase), IL17A (pro-inflammatory cytokine), and IL6R (interleukin-6 receptor), suggesting that certain inflammatory pathway variants may enhance anti-tumor immune surveillance. Risk-increasing SNPs included variants in PRF1 (perforin), IL7R (IL-7 receptor), ABCA1 (cholesterol transporter), IFNAR2 (interferon receptor), TLR2, TLR4 (pattern recognition receptors), and ICAM1 (adhesion molecule).
AUC-RF detected 56 significant SNPs by permutation threshold. The two methods overlapped on 13 SNPs jointly identified: CASP3, PRF1, IL7R, ABCA1, IL6R, MASP1, SCARB1, TLR2, IL17C, MAP2K4, CD14_IK, FADD, and ICAM1. The concordance between two methodologically distinct approaches on these 13 SNPs provides stronger evidence for their genuine association with bladder cancer risk.
Multi-SNP signatures dramatically outperform smoking as a risk factor. The 37-SNP BTL signature combined into a multi-SNP risk score with a posterior median odds ratio of 123.5, representing the joint effect of carrying the combined risk allele profile. This dwarfs the odds ratio for smoking, approximately 5, the strongest established environmental risk factor for bladder cancer.
The magnitude of this combined genetic signature underscores that while each individual SNP has a modest effect, their collective contribution to bladder cancer risk can exceed that of known environmental exposures. This has implications for how genetic epidemiology should measure and report population-attributable risk for polygenic traits.
Approximately 30% of the SNPs identified by BTL replicated with consistent direction of effect in the independent Texas Bladder Cancer Study. While formal statistical replication requires larger sample sizes, directional consistency across independent populations supports the biological relevance of the multi-SNP signature rather than overfitting to the discovery cohort.
Distinct genetic signature in never-smokers. A stratified analysis restricted to non-smoking participants was performed to identify inflammatory gene variants associated with bladder cancer risk independent of tobacco carcinogen exposure. This population provides a cleaner window into the genetic architecture of inflammation-related risk without confounding from mutagenic tobacco compounds.
In never-smokers, BTL detected 9 SNPs with posterior probabilities above 50%, while AUC-RF identified 6 SNPs. Two SNPs were jointly identified by both methods in the non-smoker subset: NFKBIA-rs696 and BCL10-rs2647396. NFKB inhibitor alpha (NFKBIA) is a central regulator of the NF-kB transcription factor pathway, which controls expression of many inflammatory mediators, while BCL10 is involved in lymphocyte signaling and cell survival.
The identification of NF-kB pathway variants as top hits in never-smokers highlights the role of constitutive inflammatory signaling in bladder cancer etiology beyond tobacco-induced mutagenesis, suggesting that anti-inflammatory interventions targeting NF-kB signaling may have particular relevance for non-smoking bladder cancer patients.
First systematic comparison of multi-SNP methods for bladder cancer inflammatory genetics. This study was among the first to apply Bayesian regularization and ensemble machine learning approaches in parallel to inflammatory gene variants in a large bladder cancer case-control study, demonstrating that both BTL and AUC-RF can extract consistent multi-SNP signals from high-dimensional genetic data.
The biological plausibility of the identified variants, spanning apoptosis (CASP3, CASP9, FADD), pattern recognition (TLR2, TLR4, CD14), immune cell activation (IL7R, IL17A, IL17C, CCR3), and adhesion and trafficking (ICAM1, ABCA1, SCARB1), lends credibility to the genetic architecture revealed by both methods, connecting to known inflammation-cancer biology.
Limitations include the restriction to main SNP effects without modeling gene-gene interactions, the challenge of achieving full replication with current sample sizes, and the need for functional validation to establish mechanistic roles of identified variants. Nevertheless, multi-SNP approaches demonstrate substantial advantages over univariate methods for capturing the polygenic structure of inflammatory susceptibility in bladder cancer and may inform future genetic risk scores for bladder cancer screening.