A luminal non‐coding RNA‐based genomic classifier confirms favourable outcomes in patients with clinically organ‐confined bladder cancer treated with radical cystectomy

Abstract

Objective

To further evaluate a genomic classifier (GC) in a cohort of patients undergoing radical cystectomy (RC), as long non‐coding RNA (lncRNA)‐based genomic profiling has suggested utility in identifying a distinct tumour subgroup corresponding to a favourable prognosis in patients with bladder cancer.

Patients and Methods

Transcriptome‐wide expression profiling using Decipher Bladder was performed on transurethral resection of bladder tumour samples from a cohort of patients with high‐grade, clinically organ‐confined (cTa–T2N0M0) urothelial carcinoma (UC) who subsequently underwent RC without any neoadjuvant therapy (n = 226). The lncRNA‐based luminal favourable status was determined using a previously developed GC. The primary endpoint was overall survival (OS) after RC. Secondary endpoints included cancer‐specific mortality and upstaging at RC.

Results

In the study, 134 patients were clinical non‐muscle‐invasive bladder cancer (cTa/Tis/T1) and 92 patients were cT2. We identified 60 patients with luminal favourable subtype, all of which showed robust gene expression patterns associated with less aggressive bladder cancer biology. On multivariate analysis, patients with the luminal favourable subtype (vs without) were significantly associated with lower odds of upstaging to pathological (p)T3+ disease (odds ratio [OR] 0.32, 95% confidence interval [CI] 0.12–0.82; P = 0.02), any upstaging (OR 0.41, 95% CI 0.20–0.83; P = 0.01), and any upstaging and/or pN+ (OR 0.50, 95% CI 0.25–1.00; P = 0.05). Luminal favourable bladder cancer was significantly associated with better OS (hazard ratio 0.33, 95% CI 0.15–0.74; P = 0.007).

Conclusions

This study validates the performance of the GC for identifying UCs with a luminal favourable subtype, harbouring less aggressive tumour biology.

Abbreviations

c

clinical

CSM

cancer‐specific mortality

EMT

epithelial–mesenchymal transition

ESTIMATE

Estimation of STromal and Immune cells in MAlignant Tumours using Expression data

FGFR3

fibroblast growth factor receptor 3

GC

genomic classifier

GSC

genomic subtyping classifier

(s)HR

(subdistribution) hazard ratio

IQR

interquartile range

lncRNA

long non‐coding RNA

(N)MIBC

(non‐)muscle‐invasive bladder cancer

mRNA

messenger RNA

MVA

multivariate analysis

NOC

non‐organ‐confined (disease)

OR

odds ratio

OS

overall survival

p

pathological

PPARG

peroxisome proliferator‐activated receptor gamma

RC

radical cystectomy

TCGA

The Cancer Genome Atlas

TURBT

transurethral resection of bladder tumour

UVA

univariate analysis

Introduction

Bladder cancer is the 10th most common cancer worldwide and is recognised as the most expensive malignancy to treat [1]. Urothelial carcinoma is the most common histology of bladder cancer, but transcriptomic profiling has demonstrated there are multiple subtypes with different behaviour and response to therapy [2, 3]. At the highest level, bladder cancer can be divided into luminal and non‐luminal subtypes, with different classification models providing additional subtyping granularity [4]. While these classification models generally use messenger RNAs (mRNAs) to differentiate subtypes, most of the mammalian transcriptome is comprised of non‐coding RNAs.

Long non‐coding RNAs (lncRNAs) are mRNA‐like transcripts that range in length from 200 nucleotides to >100 kilobases and lack open reading frames. Although the biological functions of many lncRNAs remain unclear, their expression patterns have been associated with particular biological or disease states, facilitating opportunities for biomarker development [5]. In 2017, The Cancer Genome Atlas (TCGA) bladder cancer programme used the lncRNA transcriptome to divide their luminal‐papillary mRNA subtype into two subgroups with distinct prognosis, suggesting additional subtyping resolution [6].

In previous work, we expanded these initial TCGA findings and further explored the utility of lncRNA expression profiling, characterising a biologically distinct subset of luminal muscle‐invasive bladder cancer (MIBC) associated with favourable prognosis. To translate these findings into clinical applicability, we developed a single‐sample genomic classifier (GC) for identifying these cancers [7]. In the present study, we aimed to validate this GC in an independent, multi‐institutional cohort of patients with organ‐confined bladder cancer undergoing radical cystectomy (RC) without neoadjuvant therapy.

Patients and Methods

Patient Cohort

For the present study we analysed the MOLII cohort. This retrospective cohort has been described previously [8] and includes patients from selected United States tertiary care cancer centres, who were diagnosed with clinical (c)Ta–2N0M0 bladder cancer and were treated with RC and bilateral pelvic lymphadenectomy within 4 months of diagnosis. Eligible patients did not receive neoadjuvant systemic therapy, did not have variant histology (e.g., micropapillary) other than mixed urothelial with squamous or glandular differentiation, and did not have presence of cancer in a diverticulum. All patients underwent transurethral resection of bladder tumour (TURBT), bimanual examination and cross‐sectional imaging prior to RC to assist with clinical staging and had no sign of extravesical (cT3) disease.

Specimen Collection and Processing

Specimen collection and sample processing were conducted as described previously (PMID 31092337). In all patients, the TURBT prior to RC was used for molecular analyses. Decipher Bladder, a clinical‐grade whole‐transcriptome assay (Veracyte, San Diego, CA, USA), was used to generate gene expression profiles and genomic subtyping classifier (GSC) subtyping classes (‘Luminal’, ‘Basal’, ‘Claudin‐low’, ‘Infiltrated luminal’, and ‘Neuroendocrine‐like’) for all specimens based on validated signatures [9, 10]. R package ‘consensusMIBC’ was applied to classify tumours among TCGA 2017 and consensus molecular subtypes [4]. Our previously developed lncRNA‐based GC was applied separately for identifying luminal favourable bladder cancer [7]. Of note, the classification of luminal favourable is binary and based on whether a GC risk score surpasses a pre‐specified threshold on a distinct model that provides a continuous score. As such, tumours could meet the threshold of luminal favourable based on their lncRNA‐based GC score, while not being classified as luminal by the GSC. In addition, analyses were performed using a composite biomarker stratifying tumours that were not luminal favourable to other luminal or non‐luminal according to the GSC.

Gene Expression Analyses

Heatmaps and boxplots were used to visualise differences between subgroups, using MIBC biomarker genes, gene signatures, and hallmark gene sets from the molecular signature database hallmark gene set collection (MSigDB) [11]. Hedgehog signalling, fibroblast growth factor receptor 3 (FGFR3) signalling, luminal signature and basal signature scores were calculated as previously described [7]. Sample purity was calculated by using the ‘Estimation of STromal and Immune cells in MAlignant Tumours using Expression data’ (ESTIMATE) algorithm [12].

Statistical Analyses

Patient and tumour characteristics were compared between subgroups by Mann–Whitney U and chi‐squared tests for continuous and categorical factors, respectively. Kruskal–Wallis tests were applied when comparing continuous variables across more than two groups. The primary endpoint was overall survival (OS) calculated as the date from surgery till date of death from any cause. Secondary endpoints included cancer‐specific mortality (CSM) and upstaging to non‐organ‐confined (NOC) disease (pathological [p]T3+ and/or pN+) at RC. Exploratory endpoints of Stage ≥pT3, any upstaging (i.e., pT2+ for clinical non‐MIBC [cNMIBC] and pT3+ for cT2 patients), and any upstaging and/or pN+. Patients who were lost to follow‐up were censored at the date of last contact. Univariate analysis (UVA) and multivariate analysis (MVA) Cox proportional hazard models evaluated the association of the luminal favourable GC with OS, with covariable adjustment for age at RC, sex, and clinical stage (where appropriate) and hazard ratios (HRs) and their associated CIs reported. Similar Fine‐Gray models were evaluated for CSM with other cause mortality serving as a competing risk, with subdistribution HRs (sHR) and their associated CIs reported. Likelihood ratio and Gray's tests were reported when determining the overall statistical significance of luminal favourable classification as a three‐level categorical variable (luminal favourable vs remaining luminal vs non‐luminal) for OS and CSM, respectively. The Kaplan–Meier and cumulative incidence methods were used to estimate differences in survival outcomes between subgroups, with estimates of life months lost attributable to CSM and other cause mortality up to the 36‐month landmark (with bias‐corrected bootstrap CIs) reported by genomic subgroup [13]. Uni‐ and multivariate logistic mixed models were fit to evaluate the relationship between genomic signatures and pathological upstaging at RC, with similar covariable adjustment and an institution level random intercept and odds ratios (ORs) and their associated CIs reported. All statistical tests were two‐sided with P < 0.05 indicating statistical significance. All analyses were performed in R, version 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria). The study adhered to the ‘Reporting Recommendations for Tumor Marker Prognostic Studies’ (REMARK) recommendations [14].

Results

The study cohort (226 patients) was comprised of 131 (59%) patients with cNMIBC and 92 (41%) with cT2 bladder cancer. The cohort was predominantly male (81%), with a median (interquartile range [IQR]) patient age of 71 (61–76) years (Table 1). The median (IQR) follow‐up of the entire cohort was 33 (16–43) months.

Table 1.

Clinicopathological characteristics of the study cohort by lncRNA‐based luminal favourable status.

Characteristic	Luminal favourable (n = 60)	Not luminal favourable (n = 166)	Overall (n = 226)	P
Age at RC, years, median (IQR)	69 (62–75)	71 (61–76)	71 (61–76)	0.40 a
Sex, n (%)
Female	5 (8)	37 (22)	42 (19)	0.03 b
Male	55 (92)	129 (78)	184 (81)
Smoking status, n (%)
Never	9 (15)	42 (25)	51 (23)	0.23 b
Ever	44 (73)	103 (62)	147 (65)
Unknown	7 (12)	21 (13)	28 (12)
Clinical stage, n (%)
Ta	6 (10)	2 (1)	8 (4)	<0.001 b
Tis	3 (5)	13 (8)	16 (7)
T1	38 (63)	72 (43)	110 (49)
T2	13 (22)	79 (48)	92 (41)
GSC, n (%)
Luminal	57 (95)	81 (49)	138 (61)	<0.001 b
Non‐luminal	3 (5)	85 (51)	88 (39)

^a Mann–Whitney U‐test.

^b Chi‐squared test.

Application of the lncRNA‐Based GC to Identify Luminal Favourable Bladder Cancer

The lncRNA‐based luminal favourable GC identified 60 (27%) tumours as luminal favourable. Consistent with previous reports and with the biology of luminal tumours, these luminal favourable tumours robustly expressed luminal‐associated biomarker genes, including keratin 20 (KRT20), peroxisome proliferator‐activated receptor gamma (PPARG) and GATA binding protein 3 (GATA3) and showed reduced expression of basal‐associated biomarker genes (Figs. 1 and S1A). Most luminal favourable tumours were classified as luminal‐papillary by the TCGA 2017 subtyping model (56/60 [93%]) or GSC luminal (57/60 [95%]) by the Decipher subtyping model. Notably, luminal favourable bladder cancer was more common among patients with cNMIBC (47/134 [35%]) as compared to patients with cT2 (13/92 [14%]) disease (Table 1). cT2 disease was more prevalent in patients with other luminal (28/81 [35%]) and non‐luminal (51/72 [60%]) tumours as defined by GSC subtype compared to those with luminal favourable tumours (13/60 [22%]) (Table S1).

Fig. 1.

Biological characterisation, evaluating luminal signature, basal signature, FGFR3 pathway, Hedgehog pathway, EMT hallmark, and p53 pathway hallmark scores between luminal favourable, other luminal and non‐luminal tumours. BMP, bone morphogenetic protein; SHH, Sonic hedgehog.

The Biological Characteristics of Luminal Favourable Bladder Cancer Are Consistent with less Aggressive Disease

Evaluating hallmark scores for epithelial–mesenchymal transition (EMT) and the p53 pathway, we observed lower EMT and wild‐type p53 pathway scores for luminal favourable tumours compared to the rest of the cohort. Additionally, both the FGFR3‐ and Sonic hedgehog (SHH)‐signatures showed higher scores for luminal favourable tumours, which are typically associated with less aggressive disease. Of note, the luminal favourable tumours were found to have higher ESTIMATE purity scores, suggesting lower levels of immune or stromal infiltration (Fig. S1B). Importantly, these findings were consistent with previous reports describing luminal favourable bladder cancer in other cohorts [7].

Luminal Favourable NMIBC Showed Low Rates of Pathological Upstaging at RC

Luminal favourable tumours had the low rates of pathological upstaging, with 17% (10/60) of luminal favourable cases being upstaged to NOC disease at RC (pT3+ and/or N+) and 10% (six of 60) with lymph node positive disease at RC. Stratifying pathological upstaging outcomes for cNMIBC disease, we found only three of 46 patients with luminal favourable cNMIBC had upstaging to NOC disease at RC (Table 2). Table S2 lists pathological upstaging endpoints comparing luminal favourable, other GSC luminal, and GSC non‐luminal subtype stratifications. On MVA, patients with the luminal favourable subtype were significantly associated with lower odds of upstaging to NOC (OR 0.43, 95% CI 0.19–0.96; P = 0.04), pT3+ disease (OR 0.32, 95% CI 0.12–0.82; P = 0.02), any upstaging (OR 0.41, 95% CI 0.20–0.83; P = 0.01), and any upstaging and/or pN+ (OR 0.50, 95% CI 0.25–1.00; P = 0.05). Similar results were observed when stratifying patients without luminal favourable tumours as remaining GSC luminal and non‐luminal subtypes (Table S3).

Table 2.

Rates of pathological T stage upstaging, NOC disease, and node positivity by lncRNA status in the full cohort and within patients with cNMIBC and cT2.

Subset	Variable, n (%)	Luminal favourable	Not luminal favourable
Overall	Total	60 (27)	166 (73)
	pT0–1	40 (67)	67 (40)
	pT2	14 (23)	45 (27)
	pT3–4	6 (10)	54 (33)
	pN+	6 (10)	32 (19)
	NOC (pT3–4 and/or pN+)	10 (17)	64 (39)
	MIBC+ (pT2+ and/or pN+)	21 (35)	104 (63)
cNMIBC	Total	47 (35)	87 (65)
	pT0–1	36 (77)	49 (56)
	pT2	9 (19)	21 (24)
	pT3–4	2 (4)	17 (20)
	pN+	2 (4)	10 (11)
	NOC (pT3–4 and/or pN+)	3 (6)	22 (25)
	MIBC+ (pT2+ and/or pN+)	11 (23)	39 (45)
cT2	Total	13 (14)	79 (86)
	pT0–1	4 (31)	18 (23)
	pT2	5 (38)	24 (30)
	pT3–4	4 (31)	37 (47)
	pN+	4 (31)	22 (28)
	NOC (pT3–4 and/or pN+)	7 (54)	42 (53)

Luminal Favourable Tumours Have Better Survival Outcomes

Patients with luminal favourable tumours had better CSM and OS compared to the rest of the cohort, with only two CSM events observed among luminal favourable patients within 36 months (one each in the cNMIBC and cT2 cohorts) (Fig. 2). Fine‐Gray competing risks regression showed significantly lower rates of CSM in patients with the luminal favourable subtype on UVA (sHR 0.32, 95% CI 0.12–0.87; P = 0.03), with similar results on MVA that did not reach statistical significance (sHR 0.38, 95% CI 0.14–1.05; P = 0.06). Similarly, higher luminal favourable score was significantly associated with lower CSM in UVA (sHR per 0.1 increase 0.83, 95% CI 0.70–0.98; P = 0.03) but not MVA (sHR per 0.1 increase 0.89, 95% CI 0.75–1.06; P = 0.19) (Fig. 3, Table 3). Similar effect estimates were observed when analysis was performed within the cNMIBC and cT2 strata and when further stratifying patients as other GSC luminal and non‐luminal vs luminal favourable (Fig. S2, Table S4). A higher luminal favourable score was associated with better OS (UVA HR per 0.1 increase 0.83, 95% CI 0.72–0.96; P = 0.01 and MVA HR per 0.1 increase 0.84, 95% CI 0.72–0.97; P = 0.02), with patients with the luminal favourable subtype having significantly lower rates of all‐cause mortality (UVA HR 0.34, 95% CI 0.15–0.74; P = 0.006 and MVA HR 0.33, 95% CI 0.15–0.74; P = 0.007) (Table 4), with similar effect sizes observed in the cNMIBC and cT2 cohorts (Table S5).

Fig. 2.

(A) Kaplan–Meier for OS and (B) cumulative incidence of CSM comparing luminal favourable and not luminal favourable patients by clinical stage.

Fig. 3.

Model estimated 1‐, 2‐, and 3‐year rates of (A) OS and (B) CSM by luminal favourable score.

Table 3.

Uni‐ and multivariate Fine‐Gray competing risks regression models for CSM.

Variable	UVA		MVA (continuous)		MVA (categorical)
	sHR (95% CI)	P	sHR (95% CI)	P	sHR (95% CI)	P
Luminal favourable score (per 0.1)	0.83 (0.70–0.98)	0.03*	0.89 (0.75–1.06)	0.19	–	–
Luminal favourable vs not favourable	0.32 (0.12–0.87)	0.03*	–	–	0.38 (0.14–1.05)	0.06
Male vs female	0.83 (0.38–1.81)	0.64	0.96 (0.45–2.07)	0.92	1.03 (0.48–2.24)	0.93
Age at RC	1.03 (1.00–1.06)	0.07	1.02 (0.99–1.06)	0.22	1.02 (0.99–1.05)	0.26
cT2 vs cNMIBC	2.92 (1.57–5.43)	<0.001*	2.39 (1.21–4.74)	0.01*	2.47 (1.29–4.72)	0.006*

^*Statistically significant at P < 0.05.

Table 4.

Uni‐ and multivariable Cox proportional hazard model results for OS.

Variable	UVA		MVA (continuous)		MVA (categorical)
	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P
Luminal favourable score (per 0.1)	0.83 (0.72–0.96)	0.01*	0.84 (0.72–0.97)	0.02*	–	–
Luminal favourable vs not favourable	0.34 (0.15–0.74)	0.006*	–	–	0.33 (0.15–0.74)	0.007*
Male vs female	1.34 (0.66–2.71)	0.41	1.58 (0.78–3.22)	0.21	1.63 (0.81–3.32)	0.17
Age at RC	1.02 (0.99–1.04)	0.18	1.02 (0.99–1.04)	0.26	1.01 (0.99–1.04)	0.28
cT2 vs cNMIBC	1.61 (1.00–2.59)	0.05*	1.27 (0.76–2.11)	0.36	1.37 (0.84–2.23)	0.21

^*Statistically significant at P < 0.05.

When analysed with respect to the 36‐month benchmark, 1.6 (95% CI 0.2–3.4) life months were lost on average among patients with luminal favourable tumour types, compared to 6.3 (95% CI 4.7–8.1) months lost in the remaining cohort. This difference was driven largely by CSM, with comparable life months lost attributable to other cause mortality between lncRNA favourable and unfavourable categories. A similar decomposition of life months lost due to cancer or other cause mortality was observed within patients with cNMIBC and cT2 when analysed separately, albeit with a larger magnitude of difference in life months lost between lncRNA groups in patients with cT2 (Fig. S3, Table S6).

Discussion

Transcriptomic profiling of bladder cancer has revealed a range of molecular subtypes that have been studied in contexts of tumour biology [4] and treatment response [2, 3]. In both NMIBC and MIBC, lncRNA and mRNA expression appear to correlate with each other [6, 15] and while considerable molecular heterogeneity has been described for the luminal mRNA subtype [16], lncRNA expression profiles were previously shown to differentiate two distinct prognostic subgroups within luminal MIBC [7]. Here, we validated the performance of a previously developed lncRNA‐based GC within a multi‐institutional cohort of patients with clinically organ‐confined bladder cancer who underwent RC without neoadjuvant systemic therapy. The present study confirmed less aggressive tumour biology for the luminal favourable subtype, which resulted in more favourable patient outcomes.

Initial TCGA reports suggested the potential use for the non‐coding RNA transcriptome to divide their luminal‐papillary mRNA subtype into two subgroups, each with distinct patient prognosis. Whereas the biological function remains unknown for most lncRNAs, they have been reported to reflect certain biological states [5], suggesting that they may be suitable candidates for biomarker development [17]. In context, the luminal favourable subtype differed from remaining luminal cases in terms of biological signatures related to less aggressive bladder cancer biology, including wild‐type p53 activity and increased hedgehog signalling, which represents a clinically meaningful difference that is being captured by lncRNA profiling.

Using gene expression data, the GC previously captured a tumour subgroup that was enriched for FGFR3 mutations within the TCGA MIBC cohort [7] and within a cohort of trimodality therapy‐treated patients with MIBC [18], reflected by higher scores for an FGFR3 activation signature. In this cohort, we evaluated patients with cNMIBC and cT2N0 who underwent RC without neoadjuvant therapy. Similar to the prior cohorts, the present study found higher FGFR3 signalling activity for luminal favourable tumours identified by the GC, and although mutation status was not available for this cohort, luminal favourable (FGFR3‐active) tumours have been previously reported to be enriched with FGFR3 mutations and were significantly more common among patients with NMIBC, an observation that has also been reported for FGFR3‐mutated tumours [19].

Of interest, we could find meaningful biological nuances within the luminal mRNA‐based bladder cancer subtype by applying the lncRNA‐based GC, suggesting utility for a multi‐omics approach to assist in clinical staging and predicting patient prognosis. Furthermore, evaluating immune‐related signatures, we found luminal favourable tumours having lower immune signature scores and higher tumour purity scores, suggesting these luminal tumours to be immune cold, an observation that may be in line with the mechanism described in detail by Tate et al. [20], who suggested PPARG is likely to be important for inducing the immune cold phenotype in luminal bladder tumours.

With the luminal favourable subtype being almost exclusively luminal, it is tempting to suggest sparing neoadjuvant systemic chemotherapy from patients harbouring these type of tumours, as previous studies suggested less benefit from neoadjuvant chemotherapy in luminal bladder cancer, when compared to non‐luminal bladder cancer [2]. While luminal tumour biology may suggest a rationale for applying targeted systemic therapeutics onto luminal bladder cancer, we observed favourable survival outcomes and lowest rates of pathological upstaging for luminal favourable tumours that were taken to immediate RC. As such, it is tempting to suggest these clinically organ‐confined luminal favourable bladder cancers as candidates for immediate RC. Future trials are awaited evaluating these observations in a prospective fashion. In context, the ‘Gene expression subtypes of Urothelial carcinoma: Stratified Treatment and Oncological outcomes’ (GUSTO) trial spares neoadjuvant therapy in TCGA luminal‐papillary tumours [21], and future molecular profiling of tumours treated with neoadjuvant targeted therapies may shed light on whether luminal‐type tumours are suitable candidates for these novel therapeutic regimens.

Similarly, this study included many patients with NMIBC. While we recognise that patients with NMIBC are not currently recommended neoadjuvant therapies before RC, there is a substantial rate of upstaging to NOC disease among patients with NMIBC and non‐luminal subtype. Our previous study found that 30% of patients with cNMIBC had pT3–4 disease and this study found that 25% of patients with cNMIBC had NOC disease. This information may help guide future research efforts on application of neoadjuvant therapy for certain high‐risk NMIBC cases. On the other hand, it is possible that patients with NMIBC with luminal favourable subtype, with its overexpression of FGFR3 activity could be spared RC in the future. Of note, erdafitinib is presently being evaluated in the MIBC setting [22] and for NMIBC an erdafitinib intravesical delivery system is being tested in NMIBC as well [23].

Limitations include the retrospective nature of the study and the lack of detailed information on perioperative intravesical and adjuvant therapies. Notably, many of the centres are referral centres and the TURBT was performed outside the institutions for some of the patients. Lastly, there may exist a selection bias for patients included in this cohort as they represent the subsets of patients with NMIBC treated with RC and patients with MIBC treated with RC but without neoadjuvant chemotherapy.

Conclusions

This study validates the performance of a lncRNA‐based GC for identifying a subgroup of luminal tumours harbouring less aggressive tumour biology. Luminal favourable tumours, as identified by the GC, were significantly associated with lower rates of pathological upstaging in patients with cNMIBC undergoing RC and favourable survival outcomes in cT2 MIBC without neoadjuvant chemotherapy.

Funding

Veracyte Inc. provided Decipher Bladder transcriptome‐wide analyses for the study. Yair Lotan acknowledges the Wilson Foundation for financial support for study conduct.

Disclosures of Interests

Yair Lotan is a consultant for Nanorobotics, Photocure, Astra‐Zeneca, Merck, Fergene, Abbvie, Nucleix, Ambu, Seattle Genetics, Hitachi, Ferring Research, verity pharmaceutics, virtuoso surgical, Stimit, Urogen, Vessi medical, CAPs medical, Xcures, BMS, Nonagen, Aura Biosciences, Inc., Convergent Genomics, Pacific Edge, Pfizer, Phinomics Inc., CG oncology, Uroviu, On target lab, Promis Diagnostics, Valar labs, Uroessentials, NRx Pharmaceuticals, Vesica health, Janssen, Immunity Bio. Joep J. de Jong is a consultant for Veracyte Inc. Siamak Daneshmand is a consultant for Photocure, Pacific Edge, Ferring, Johnson and Johnson, Protara, Urogen, Pfizer, CG Oncology, Vesica Health, Immunitybio. Brant Inman: Clinical trials and research agreements with Genentech/Roche, FKD Therapies, Taris Biomedical, Seattle Genetics, Medtronic, Janssen, CG Oncology, Profound Medical, and Theralase. Advisor for Combat Medical, Johnson&Johnson, TerSera, and AbbVie. Robert S. Svatek is a consultant for CG Oncology and receives research support from Merck, Japanese BCG Laboratories, and Biodexa/Emtora, Valar Labs, and Veracyte. Roger Li: Research support: Predicine; Veracyte; CG Oncology; Valar labs; Merck. Clinical trial protocol committee – CG Oncology; Merck; Janssen. Scientific advisor/consultant – BMS, Merck, Fergene, Arquer Diagnostics, Urogen Pharma, Lucence, CG Oncology, Janssen, Thericon. James A. Proudfoot and Elai Davicioni are employees of Veracyte Inc. Ewan A. Gibb is a former employee of Veracyte Inc.

Supporting information

Table S1. Clinicopathological characteristics of the study cohort in luminal favorable, other luminal, and non‐luminal tumors.

Table S2. Rates of pathological T stage upstaging, non‐organ confined disease, and node positivity across luminal favorable, other luminal, and non‐luminal tumors in the full cohort and within clinical NMIBC and T2 patients.

Table S3. Univariable and multivariable logistic mixed model results for pathological upstaging to non‐organ confined disease (NOC, defined as pT3+ or N+ disease), pT3+, any upstaging (i.e., pT2+ for cNMIBC and pT3+ for cT2), and any upstaging or nodes (i.e., i.e., pT2+ for cNMIBC and pT3+ for cT2 and/or pN+) for luminal favorable score, a lncRNA status (i.e., luminal favorable vs. unfavorable), and lncRNA status combined with GSC (luminal favorable vs. luminal unfavorable vs. non‐luminal).

Table S4. Univariable and multivariable Fine and Gray regression results for cancer specific mortality (with other cause mortality as a competing risk) in the full, cNMIBC, and cT2 cohorts.

Table S5. Univariable and multivariable Cox regression results for overall survival in the full, cNMIBC, and cT2 cohorts.

Figure S1. Estimated life months lost (with bootstrapped 95% confidence intervals) attributable to cancer specific or other cause mortality restricted to a maximum of 36 months by lncRNA status and clinical stage.

Figure S2. (A) Forced order heatmap for five biological categories (luminal, basal, EMT, immune, and stromal) of selected bladder cancer marker genes.

Figure S3. (A) Kaplan‐Meier for overall survival and (B) cumulative incidence of cancer specific mortality comparing luminal favorable, other luminal, and non‐luminal by clinical stage.

Figure S4. Stacked incidence of death due to cancer specific mortality and other cause mortality by genomic subtype in (A) the full cohort, (B) cNMIBC patients, and (C) cT2 patients.