Fibroblast growth factor receptor 3 alterations and response to immune checkpoint inhibition in metastatic urothelial cancer: a real world experience

Abstract

Background

FGFR3-altered urothelial cancer (UC) correlates with a non-T cell-inflamed phenotype and has therefore been postulated to be less responsive to immune checkpoint blockade (ICB). Preclinical work suggests FGFR3 signalling may suppress pathways such as interferon signalling that alter immune microenvironment composition. However, correlative studies examining clinical trials have been conflicting as to whether FGFR altered tumours have equivalent response and survival to ICB in patients with metastatic UC. These findings have yet to be validated in real world data, therefore we evaluated clinical outcomes of patients with FGFR3-altered metastatic UC treated with ICB and investigate the underlying immunogenomic mechanisms of response and resistance.

Methods

103 patients with metastatic UC treated with ICB at a single academic medical center from 2014 to 2018 were identified. Clinical annotation for demographics and cancer outcomes, as well as somatic DNA and RNA sequencing, were performed. Objective response rate to ICB, progression-free survival, and overall survival was compared between patients with FGFR3-alterations and those without. RNA expression, including molecular subtyping and T cell receptor clonality, was also compared between FGFR3-altered and non-altered patients.

Results

Our findings from this dataset confirm that FGFR3-altered (n = 17) and wild type (n = 86) bladder cancers are equally responsive to ICB (12 vs 19%, p = 0.73). Moreover, we demonstrate that despite being less inflamed, FGFR3-altered tumours have equivalent T cell receptor (TCR) diversity and that the balance of a CD8 T cell gene expression signature to immune suppressive features is an important determinant of ICB response.

Conclusions

Our work in a real world dataset validates prior observations from clinical trials but also extends this prior work to demonstrate that FGFR3-altered and wild type tumours have equivalent TCR diversity and that the balance of effector T cell to immune suppression signals are an important determinant of ICB response.

Background

Treatment options for metastatic urothelial cancer (UC) have expanded dramatically, with the recent FDA approval of immune checkpoint antibodies, the antibody–drug-conjugate enfortumab vedotin, and the pan-fibroblast growth factor receptor (FGFR) inhibitor, erdafitinib [1–7]. Prior research shows that between 17 and 32% of UCs harbour alterations in FGFR family members [8, 9], which are also more common in upper tract and lower stage disease [10–12].

Groups have identified gene expression-based molecular subtypes of high-grade UCs and FGFR-altered tumours are enriched in the luminal or luminal papillary subtypes, which are less T cell-inflamed in humans [9, 13, 14] and in genetically engineered murine models [15]. Because of these characteristics, the luminal subtype and FGFR3-altered tumours were hypothesised to be less likely to respond to ICB. However, initial reports have been conflicting as to whether response and survival after ICB is associated with the presence of FGFR alterations in patients with metastatic UC. A retrospective analyses of phase II clinical trials of ICB treated advanced UC patients suggested no difference in ICB response between FGFR3 mutant versus wild type cancers [16], as did an analysis of pathologic response rates to neoadjuvant pembrolizumab in muscle-invasive UC [17, 18]. In contrast, only 5% of FGFR3-altered patients were reported to have responded to prior ICB in the BLC2001 study of erdafinitib, although only 22 patients had previously received ICB in that study [2]. These studies were performed using clinical trial data, which may select for a subset of healthier patients with fewer comorbidities, and potentially different underlying biology of their cancers. In this study, we have, therefore, analysed an independent cohort of patients with metastatic UC treated with ICB to assess the predictive value of FGFR3-alterations using real world data. Additionally, we performed DNA and RNA sequencing to evaluate immunogenomic features predicted to either increase or decrease responsiveness to ICB in FGFR-altered patients. The aim of this study is to add to the body of work assessing the role of FGFR3 in predicting outcome to ICB as well as the underlying immunobiology of FGFR3-altered tumours.

Methods

Patient samples

Patients with advanced UC were retrospectively identified as having received treatment with at least one dose of anti-programmed death (PD)-1 or anti-PD-ligand (PDL)-1 ICB within the University of North Carolina hospital system between January 2014 and June 2018. Eligible patients had available archived formalin-fixed paraffin embedded (FFPE) tumour tissue from a primary or metastatic site. There were 109 patients who met the initial criteria; 108 were evaluable for response (Supplementary Fig. 1). Successful DNA and RNA sequencing were performed on 103 and 89 patients respectively. Each tumour sample was reviewed by a genitourinary pathologist for presence of high-grade UC (S.E.W.). Only patients with available ICB response data and adequate DNA to evaluate FGFR status were included (103 patients) in the analysis.

Clinical annotation

Demographic and clinical variables were annotated by two separate investigators. Variables were recorded at the time of initiation of ICB therapy. Objective response rate was based on RECIST v1.1 criteria after retrospective review of patient images by a board-certified radiologist blinded to FGFR3 status (K.M.). Patients who died prior to radiologic assessment or with clear clinical progression on ICB without post-treatment imaging were considered non-responders. Clinical benefit was defined as complete response, partial response, or stable disease for > = 6 months per RECIST v1.1.

DNA and RNA sequencing

DNA and RNA were extracted from macrodissected FFPE sections using the truXTRAC FFPE total nucleic acid kit (Covaris). DNA was sequenced from a targeted mutation panel comprising 144 genes using the OmniSeq Comprehensive® assay (OmniSeq®) Buffalo New York. Sequencing was performed using the ION CHEF and Ion S5XL 540 NGS chip using 30 ng DNA input. A no template control (NTC), CNV negative (NA12878 FFPE DNA), SNV/Indel positive, CNV gain positive and CNV loss positive control sample was included in each run.

RNA was used to generate sequencing libraries using the SureSelect XT RNA direct prep kit (Agilent Technologies). Libraries were individually captured, reviewed for appropriate size using an Agilent Bioanalyzer or TapeStation trace, and quantitated using the KAPA library quantitation kit (Roche) prior to equal molar pooling. Samples were sequenced on the Illumina NovaSeq6000 to generate ≥50 million reads per sample. A NTC and positive control sample (NA12878 FFPE RNA) were included in each run.

RNA expression analyses

89 patients had adequate RNA quantity and quality for sequencing. Reads were aligned with STAR aligner to (GrCH38 version 22) human assembly using the STAR/Salmon pipeline [19]. Expression was quantified using the RSEM package [20] and the GrCH38 human transcriptome reference. Genes were filtered for a minimum expression count (at least 10 reads in at least 5 samples), and for a protein coding annotation by Ensemble (final set of genes = 16,901). Differential expression was assessed using the DESeq2 package [21] on this filtered set of genes. For all other analyses, all expression values were log2(1 + x) transformed, median centred and upper quartile scaled.

Subtyping

BASE47 PAM classifier [22] was applied using the normalised RNA expression. Other subtyping schemas (Baylor, MDA, CIT, Lund, TCGA), were obtained by applying the R-package released by Kamoun et al. [14] to the normalised RNA expression data. A new 60-gene classifier (GeneCentric (GC) classifier), was applied using normalised RNA expression data. This nearest centroid classifier was developed using publicly available TCGA BLCA RNAseq expression profiles on 129 TCGA BLCA samples from ref. [23]. An initial nearest centroid classifier was fit using expression profiles from the 129 tumours with an available cluster assignment and the 2708 variable genes that were used for clustering. This classifier was then applied to an expanded set of TCGA BLCA samples (n = 408) [9]. Cross-validation and ClaNC software [24] on the expanded set of samples were used to identify the genes in the final classifier (Supplementary Table 1).

Gene signatures

Signature scores were derived by averaging the normalised expression values over all of the genes in the signature. We normalised this further by subtracting out the median value across all samples and dividing by the standard deviation.

TCR clonality

T cell receptor (TCR) clonality was assessed using MiXCR version 2.1.6 [25]. IMGT TCR repertoire sequence from http://www.imgt.org/IMGTrepertoire was used to facilitate the assembly of TCRα and TCRβ repertoires [26].

Statistics

OS and PFS analysis was conducted using Cox-Proportional Hazards (CPH) modelling with right censored endpoints. Confidence intervals for covariates were determined individually (i.e. in a univariate model), as well as jointly while incorporating all of the variables at once. Chi-square testing was used to assess the association between FGFR3 alteration and subtype. Associations between categorical and continuous variables, for example, association between FGFR3 expression and subtype, were evaluated using a t-test or ANOVA depending on the number of categorical variable levels.

Results

Patients

109 patients met initial inclusion criteria of having received single agent ICB (all anti-PD1 or anti-PD-L1) for advanced UC. Of those, 108 patients had available response data and 103 had adequate tumour tissue to perform somatic sequencing for FGFR status (Supplementary Fig. 1). Successful targeted DNA sequencing and whole transcriptome profiling was achieved in 103 and 89 patients, respectively. In keeping with prior reports, 17% (18 patients) were found to have FGFR alterations (17 in FGFR3 and 1 in FGFR2) [9] Of the FGFR3-alterations, there were 13 mutations (n = 9: S249C, n = 2: R248C, n = 2: Y373C) and 4 fusions (all FGFR3-TACC3 fusions) (Table 1). Demographics were similar between FGFR-altered and unaltered patients, except that FGFR3-altered patients were more likely to be current/former smokers (p = 0.02) and less likely to have received prior platinum-based chemotherapy (p = 0.04) (Table 1).

Table 1.

Demographics of the study population.

Demographic	FGFR3-altered (n = 17)	FGFR3 unaltered (n = 86)	All (n = 103)	p value
Median age, years (range)	72 (41–89)	70 (44–89)	70 (41–89)	0.49
Male sex, n (%)	10 (59)	55 (64)	65 (63)	0.69
Race, n (%)				0.84
White	13 (76)	67 (78)	80 (77)
Black	3 (18)	16 (19)	19 (18)
Other/Unknown	1 (6)	3 (3)	4 (4)
Smoking status, n (%)				0.02
Current/Former	16 (94)	56 (65)	72 (70)
Never	1 (6)	30 (35)	31 (30)
Primary site, n (%)				0.72
Bladder	15 (88)	73 (85)	88 (85)
Upper tract	2 (12)	13 (15)	15 (15)
Line of therapy, n (%)				0.37
1	7 (41)	19 (22)	26 (25)
2	9 (53)	59 (69)	68 (66)
3+	1 (6)	8 (9)	9 (9)
Prior platinum-based chemo, n (%)				0.04
Yes	9 (53)	66 (76)	75 (73)
Setting of prior platinum chemo, n (%)				0.41
Localised	4 (24)	39 (45)	43 (42)
Metastatic	5 (29)	27 (31)	32 (31)
ECOG PS, n (%)				0.65
0	4 (23)	18 (21)	22 (21)
1	6 (35)	36 (42)	42 (41)
>=2	4 (24)	14 (16)	18 (17)
Unknown	3 (18)	18 (21)	21 (20)
Haemoglobin, n (%)				0.64
≥10 g/dL	12 (71)	58 (67)	70 (68)
<10 g/dL	5 (29)	23 (27)	28 (27)
Unknown	0	5 (6)	5 (5)
Liver metastases, n (%)				0.45
Yes	3 (18)	18 (21)	21 (20)
Median time from prior therapy, months				0.56
Median (95% CI)	7.7 (5.2–18.3)	8.8 (7.5–11.2)	8.8 (7.5–10.3)
Checkpoint inhibitor received				0.22
Atezolizumab	6 (35)	34 (40)	40 (39)
Avelumab	1 (6)	0	1 (1)
Durvalumab	0	2 (2)	2 (2)
Nivolumab	0	5 (6)	5 (5)
Pembrolizumab	10 (59)	45 (52)	55 (53)
Source of tumour for analysis, n(%)				0.73
Primary tumour	16 (94)	76 (88)	92 (89)
Regional lymph node	0	5 (6)	5 (5)
Distant metastasis	0	1 (1)	1 (1)
Unknown	1 (6)	4 (5)	5 (5)
FGFR3 alteration identified	17 (100)	0 (0)	17 (17)
S249C	9 (53)
FGFR3-TACC3 fusion	4 (24)
Y373C	2 (12)
R248C	2 (12)

The p values in italics are the values that are <0.05 and are therefore significant.

FGFR3-alterations do not predict ICB response or survival

Given the previously reported immune desert phenotype of FGFR3-altered UC, we hypothesised that they might not benefit from ICB in our real world dataset. We did not see any difference in overall (OS) or progression-free survival (PFS) between ICB treated patients harbouring FGFR3-altered and FGFR3 wild-type tumours (Fig. 1a, b). In keeping with these results, neither binary response rates or clinical benefit rates were statistically different between FGFR3-altered and FGFR3 wildtype samples (Fig. 1c, p = 0.73 and p = 0.71). Results were not different when the dataset was limited to patients that received at least 2 cycles of treatment. Multivariable CPH modelling showed a statistically significant association of presence of liver metastases with decreased survival, as well as an association of survival with tumour mutation burden, PDL1 expression, and IFN gamma-related gene expression (Fig. 1d). There was no association of line of therapy to overall survival on multivariable analysis. Both FGFR3-altered patients (n = 2) that had a response to therapy were treated with ICB as second-line therapy.

Fig. 1. FGFR3-altered patients treated with ICB have similar clinical benefit.

Patients with known FGFR3 status (n = 103) were classified as altered (n = 17) or wild-type (WT, n = 86) and compared for a OS (overall survival, Cox PH p = 0.38), b PFS (progression-free survival, Cox PH p = 0.95), and c Overall response rate. Response between FGFR3-altered and FGFR3 WT were compared by Responder (CR + PR) and Non-responder (SD + PD). Chi-square p = 0.73. d Interval estimates for coefficients in Cox PH model of OS for both univariate (black) and joint (blue) modelling. Variables significant at p < 0.05 marked with star. OS overall survival, CPH Cox-Proportional Hazards, PFS progression-free survival, WT wild type, TCC transitional cell carcinoma, LOT line of treatment, TMB tumour mutation burden, ECOG Eastern Cooperative Oncology Group, HR hazard ratio.

Median duration of response was not different based on FGFR3 status (Table 2), and there was no difference in median duration of treatment with ICB in FGFR3-altered patients compared with unaltered patients (Supplementary Fig. 2). Among non-responders to ICB, there was a slightly longer median duration of treatment in patients with FGFR-alterations compared with wild-type patients, despite a similar rate of clinical benefit (22% vs 28%, p = 0.71), progressive disease as best response to treatment (76% vs 71%, p = 0.73), and receipt of treatment greater than 30 days past the date of progressive disease (33% vs 27%, p = 0.23). Only 17% percent of all patients were able to receive subsequent cancer-directed therapy, and only two (12%) of the FGFR3-altered patients received a subsequent FGFR inhibitor (in the context of a clinical trial) with a partial response seen in one patient.

Table 2.

Outcomes and tumour characteristics by FGFR3 status.

Outcomes	FGFR3-altered (n = 17)	FGFR unaltered (n = 86)	p value
Best response, n (%)			0.73
CR	1 (6)	7 (8)
PR	1 (6)	9 (10)
SD	2 (12)	9 (10)
PD	13 (76)	61 (71)
ORR	2 (12)	16 (19)
Clinical benefita, n (%)			0.71
Yes	4 (22)	24 (28)
Median PFSb, mos (95% CI)	2.7 (2.0–5.0)	2.1 (1.7–2.8)	0.95
PFS at 6 months, (%)	18	28
PFS at 1 year, (%)	6	17
Median duration of response, mos (95% CI)	15.6 (2.7–28.5)	10.4 (7.1–15.9)	0.80
Median OSc, mos (95% CI)	9.5 (5.3-NR)	6.4 (4.1–11.1)	0.38
OS at 6 months, (%)	71	51
OS at 1 year, (%)	43	38
Received subsequent systemic therapy, n(%)	5 (29)	12 (14)	0.12
Tumour Mutation Burden, median mut/MB (95% CI)	6.2 (±2.5)	5.2 (±1.2)	0.55
PDL1 RNA expression, median (95% CI)	257 (±384)	320 (±322)	0.22

NR not reached.

^aClinical benefit defined as CR + PR + SD for > = 6 months.

^bPFS defined as time from immune checkpoint inhibitor initiation to progression or death.

^cOS defined as time from immune checkpoint inhibitor initiation to death.

The bold values were just to emphasize ORR numbers.

FGFR3-altered cancers have unique genomic characteristics

We next sought to assess the biological underpinnings of FGFR3-altered tumours. We examined whole transcriptome RNAseq data on 89 tumours (17 FGFR3-altered and 72 FGFR3 wild type) and noted that FGFR3-altered tumours have significantly higher FGFR3 RNA expression (Fig. 2a). Tumours with fusion alterations of FGFR3 had similarly high FGFR3 RNA expression as those with FGFR3 mutations (Fig. 2a) and while FGFR3-altered tumours had a higher tumour mutational burden (TMB), this did not reach statistical significance (Fig. 2b). Differential gene expression between FGFR3-altered and FGFR3 wild type (WT) tumours demonstrated many differentially expressed genes (148 Up and 802 Down) (Fig. 2c, Supplementary Tables 2, 3). Pathway analysis of the differentially expressed genes identified pathways related to cellular metabolism upregulated in FGFR3-altered tumours including pathways related to glutathione metabolism and steroid hormone biosynthesis (Fig. 2d). These pathways are well known to be regulated by peroxisome proliferator activator gamma (PPARG) [27]. In keeping with this hypothesis, we saw that a UC specific PPARG gene signature [13] (Supplementary Table 4) was significantly upregulated in FGFR3-altered tumours (Fig. 2e).

Fig. 2. FGFR3-altered tumours express higher FGFR3 and pathways associated with PPARG activation.

a Waterfall plot of relative expression of FGFR3 transcript from n = 89 tumours rank ordered by highest expression (right) to lowest expression (left). FGFR3-altered tumours are designated in dark red (mutations) and light red (fusions). b Violin plots of tumour mutational burden (TMB) by FGFR3 status (Mann–Whitney U test p = 0.055). c Volcano plot of differentially expressed genes identified by DEseq2 comparing FGFR3-altered and FGFR3 WT tumours. Significantly down (n = 802) or upregulated (n = 148) genes indicated in red. d Bar plot of −Log10 q-value of indicated pathways identified to be enriched in FGFR3-altered tumours relative to FGFR3 WT tumours. e Violin plot comparing levels of a previously published bladder cancer specific PPARG gene signature between FGFR3-altered and WT tumours. Dotted line = median. Solid lines = interquartile range (IQR).

FGFR3-altered patients are enriched in luminal and luminal-papillary subtypes

We applied a comprehensive panel of UC subtyping schema to our RNAseq expression data including the UNC [22], CIT [28], Lund [29], Baylor [30], MD Anderson [31], TCGA [9], Consensus [14], as well as the new GC classifier (Fig. 3a). Across different classification schema, FGFR3-alterations were consistently enriched in luminal or differentiated subtypes. Absolute numbers of patients with FGFR3-alterations in each subtype are shown in Supplementary Fig. 3. We also see stark differences in FGFR3 RNA expression levels between subtypes (Fig. 3b) with the luminal/differentiated subtypes tending to have a significantly higher FGFR3 expression. The pattern of FGFR3 RNA expression is mirrored by levels of a validated FGFR3 gene signature score [29] (Fig. 3c).

Fig. 3. FGFR3-altered tumours are enriched in luminal and differentiated molecular subtypes.

a The indicated molecular subtyping schema were applied to tumours using the R package from Kamoun et. al. Bar plots indicated the percentage of tumours within that subtype with FGFR3-alterations. P values for Chi-square test are indicated. b Violin plots of FGFR3 transcript expression by indicated molecular subtype. c Violin plots of previously published FGFR3 activation signature by Sjödhal et al. by indicated molecular subtype.

FGFR3-altered tumours do not show different TCR clonality

Prior work has shown that luminal UC are less immunologically inflamed [9, 13, 14] and that bladder and upper tract tumours with FGFR3-alterations express lower CD8 T cell gene signatures [10, 16]. It is possible that despite being less T cell-inflamed, FGFR3-altered tumours may still have a more effective immune response. T cell receptor (TCR) clonality is a surrogate measure for an antigen driven immune response, we, therefore, assessed the degree of TCR clonality using inferred TCR sequences from short read RNAseq using MiXCR [25]. We did not observe any significant difference in the clonality of TRA (TCR alpha chain) and TRB (TCR beta chain) between FGFR3-altered and WT tumours in either our real world dataset or in the IMvigor 210 dataset (Fig. 4a), suggesting the antigen driven immune response is not different between FGFR3-altered and wild type tumours.

Fig. 4. FGFR3-altered tumours have differential stromal and immunosuppression residuals.

a T cell receptor alpha (TRA) and T cell receptor beta (TRB) chain Shannon entropy was inferred from RNAseq data using MiXCR in both the current (UNC-108) and IMvigor210 datasets. All comparisons between FGFR3-altered and WT were not significant. b Fibroblast TGF beta Response Signature (FTBRS) from Mariathasan et al. [33] and the EMT_Stroma_core_18 from Wang et al. [32] signatures were evaluated in the UNC and IMvigor210 datasets comparing FGFR3-altered and WT tumours. c In FGFR3 wild type tumours from the UNC-108 dataset we tested whether the balance of effector (CD8 effector signature) to suppressor (FTBRS, EMT-Stroma or Immunosuppression signatures) gene signatures correlated with ICB response. A linear model was generated to determine the expected level of FTBRS, EMT-Stroma, and Immunosuppression signature for a given CD8 effector signature score (Supplementary Fig. 5). For each wild-type tumour we then used our model to calculate the ‘residual’ level of each suppressor signature versus its predicted level based upon its CD8 effector T cell signature. A higher stromal or immunosuppression residual score, therefore, represents a higher level of immune suppression than predicted. Non-responding patients had a higher level of all three of the EMT-stroma, FTBRS, and immunosuppression residuals than predicted.

FGFR3-altered tumours show lower expression of stromal related gene signatures

Stroma-related gene signatures appear to correlate with resistance to ICB in advanced UC tumours of the ‘excluded’ and ‘CD8-High’ subgroups [32, 33]. In keeping with the results from Wang et al. [16] we saw that both the fibroblast TGF-beta response signature (FTBRS) and the EMT-Stroma signatures were significantly downregulated in the FGFR3-altered tumours in our real world dataset (Fig. 4b) suggesting a potential for increased response to ICB. In keeping with the Mariathasan and Wang publications, overall survival and response rate did not correlate with the individual FTBRS and EMT-stroma gene signatures (Supplementary Fig. 4).

Balance between a T cell-inflamed phenotype and immunosuppressive stroma or immunosuppressive molecules determines ICB response

Wang and co-workers have previously hypothesised that while FGFR3-altered tumours were less T cell-inflamed, they also had less immunosuppressive stroma resulting in a relative balance of features that promote ICB response and resistance [16]. We extended these observations and formally tested whether the balance of effector (CD8 effector signature) to suppressor (FTBRS, EMT-Stroma, or Immunosuppression signatures) gene signatures correlated with ICB response in FGFR3 WT tumours. Specifically, we generated a linear model to determine the expected level of FTBRS, EMT-Stroma, and Immunosuppression signature for a given CD8 effector signature score (Supplementary Fig. 5). For each tumour we then used our model to calculate the ‘residual’ level of each suppressor signature versus its predicted level based upon its CD8 effector T cell signature. A higher stromal or immunosuppression residual score, therefore, represents a higher level of immune suppression than predicted. Correlating the immune suppression residual to ICB response, we saw that non-responding patients had a higher level of all three of the EMT-stroma, FTBRS, and immunosuppression residuals (Fig. 4c). These results validate Wang et al.’s hypothesis that a balance between effector and suppressor mechanisms are a key determinant of ICB response.

Discussion

FGFR3-altered UC has been previously reported to correlate with a non-T cell-inflamed phenotype [10, 34] and therefore predicted to be less responsive to ICB. There are however conflicting clinical data about the correlation between FGFR3 mutations and ICB response [2, 16, 18, 35]. Moreover, preclinical work supports the notion that FGFR3 signalling affects pathways that alter immune microenvironment composition [10, 16]. Therefore, further investigation is warranted to better understand the interaction between FGFR3 signalling, the immune microenvironment, and IC response. We present both our clinical and immunogenomic work in a real world dataset of patients. Our findings confirm that FGFR3-altered and wild type UC are similarly responsive to ICB [16]. Importantly our cohort of real world patients includes patients with poor performance status and poor prognosis disease seen by the majority of oncologists and many of these patients would have been excluded from trials.

Regardless of FGFR3’s lack of association with ICB response, it remains unclear if FGFR3 functionally promotes an immune desert phenotype, as all data to date merely associate FGFR3 mutations with a T cell depleted microenvironment [9, 13, 14, 34] or demonstrate that FGFR3 regulates expression of IFNG related genes in vitro [10, 16]. While prior work in lung cancer murine models demonstrated that FGFR2 inhibition promoted a more inflamed microenvironment, no studies in UC have shown that FGFR3 inhibition alters the immune microenvironment [36]. Moreover, precisely how FGFR3 signalling mediates IFNG related genes and a T cell depleted microenvironment has yet to be determined. We found that upregulated FGFR3 signalling significantly correlates with upregulated PPARG gene signatures. This may be one potential mechanism given that enhanced PPARG signalling has been both correlated with a non-T cell-inflamed phenotype [13, 34, 37] and functional upregulation of PPARG signalling appears to suppress pro-inflammatory cytokine signalling [38].

Wang and colleagues hypothesised that despite having a relatively non-T cell-inflamed phenotype FGFR3-altered tumours also had relatively low levels of immunosuppressive stromal signals that potentially permitted them to be equally response to ICB as their FGFR3 wild type counterparts. We confirmed these observations. In addition, our work importantly extends these observations by demonstrating that patients responsive to ICB do have a significantly lower predicted residual of all three of the suppressive signatures. In aggregate these findings suggest that despite not being highly T cell-inflamed, FGFR3 mutant tumours have lower levels of immunosuppressive stroma and immunosuppressive molecules that make them permissive to ICB response. Of note, we were unfortunately unable to compare the tumour microenvironment of metastatic sites in this study, which may potentially have an altered balance of immune microenvironment factors compared to primary tumours.

The response rates and median survival in patients treated with ICB were overall low in our cohort, reflective of the real world population of patients with metastatic UC. The median OS of our population, for example, was less than that of patients that received pembrolizumab in the Keynote-045 trial, and our study included more patients with ECOG PS >0 (71% vs 54%) and haemoglobin less than 10 (26% vs 16%) than Keynote-045. Similarly, compared to the IMvigor210 cohort, our real world population was more likely to have ECOG PS of 2 or higher (17% vs 6%) [3, 4]. Response rates in our study were similar to those previously reported, including the response rate of 19% in FGFR-wild type patients, similar to that seen in the reported clinical trials (21% in IMvigor210 and 21% in Checkmate-275) [16]. Although this is a retrospective cohort study and not a randomised study and therefore cannot comprehensively assess for presence of a predictive biomarker, FGFR3 status was not found to be predictive or prognostic in our dataset. Given the low response rates overall, it is possible that our study missed small differences in response by FGFR alteration status and statistical significance was not achieved due to a lack of power. Our study should be interpreted in the context of other published datasets of ICB-treated patients, but we were able to detect survival differences by presence of liver metastases, tumour mutation burden, and IFN-gamma-related pathway expression, which are known to be associated with survival in patients treated with ICB.

The FTBRS and EMT-stromal signatures were not prognostic or associated with response in our overall cohort, consistent with their lack of predictive value in all patients as described in the original publications. Our data do however support the notion that the balance of effector to suppressor components of the tumour immune microenvironment and not necessarily just their absolute levels are important determinants of ICB response and benefit further emphasising the need for an analysis of the balance of factors related to immune activation and suppression in metastatic UC. Finally, although there was a slight imbalance with more FGFR3-altered patients in the first-line treatment group and fewer receiving peri-operative prior chemotherapy, no patients with FGFR3-alterations treated in the first-line had a response to therapy, suggesting this is not likely to result in overestimation of the response rate in FGFR3-altered patients.

The optimal sequence of therapy remains unclear for patients with FGFR alterations, given the consistently low response rates with ICB and the numerically higher ORR of 40% with erdafitinib, and our study was not likely to miss an estimate of response rates in FGFR-altered patients that would exceed that of erdafitinib. It remains unclear if FGFR inhibition with erdafitinib would shift the counterbalance of pro- and anti-inflammatory factors that associated with FGFR3 alteration, and whether post-erdafitinib tumours would be more or less likely to respond to subsequent ICB. Only 17% of patients in our cohort received subsequent therapy, highlighting the importance of selecting the ideal order (or combination) of therapy for patients with metastatic UC.

In summary, our study is the first real world data that speaks to the impact of FGFR3-alterations on response to ICB and UC biology. In keeping with prior work, we observed that FGFR3-altered UC is enriched in the luminal or more differentiated molecular subtypes of UC and is associated with a T cell depleted phenotype. Our work supports the notion that FGFR3-altered UC does not have a differential objective response rate or survival to ICB although the optimal sequencing and the effects of combination of FGFR inhibition and ICB have yet to be defined.

Author contributions

Tracy L Rose: Conceptualisation, methodology, validation, formal analysis, data curation, writing—original draft, writing—review and editing, visualisation, project administration. William H Weir: Formal analysis, data curation, writing—original draft, writing—review and editing, visualisation. Greg M Mayhew: Methodology, formal analysis, data curation, writing—review and editing. Yoichiro Shibata: Software, validation, formal analysis, data curation, writing—original draft. Patrick Eulitt: Data curation. Josh M Uronis: Project administration, Mi Zhou: Data curation, writing—original draft. Matthew Nielsen: Resources, writing— review and editing. Angela Smith: Writing—review and editing. Michael Woods: Writing—review and editing. Michele C Hayward: Data curation, project administration, writing—review and editing. Ashley H Salazar: Data curation. Matthew I Milowsky: Writing—review and editing. Sara E Wobker: Formal analysis, data curation. Katrina McGinty: Formal analysis, data curation. Michael V Millburn: Conceptualisation, supervision. Joel R Eisner: Writing—original draft, supervision, writing—review and editing, supervision. William Y Kim: Conceptualisation, methodology, validation, writing—original draft, writing—review and editing, supervision. Tracy L Rose was placed first in the co-first authorship spot because she conceived the project. William H. Weir was placed as second in the co-first authorship because he made substantial contributions to computational biology analysis, figure generation, and intellectual insights.

Funding

This work was supported by the University Cancer Research Fund (UCRF) and NCI grant R01-CA241810 [WYK], a research collaboration agreement between Janssen Research & Development, LLC and GeneCentric Therapeutics, Inc. and a sponsored research agreement between GeneCentric Therapeutics, Inc and the University of North Carolina at Chapel Hill. TLR is supported by the National Cancer Institute K12 Career Development Award in Clinical Oncology (grant 5K12CA120780) and K08CA248967, as well as the Doris Duke Charitable Foundation (grant number 2015213).

Data availability

DNA sequencing data are available upon request. RNAseq data have been deposited to Gene Expression Omnibus under accession ID GSE176307.

Ethics approval and consent to participate

This study was reviewed and approved by the Institutional Review Board at UNC with an approved IRB 18-1478. All patient data were de-identified at the time of data abstraction from the electronic medical record. The study was performed in accordance with the Declaration of Helsinki.

Consent to publish

Not applicable. No individual patient data are included in the publication.

Competing interests

This work was funded in part through a research collaboration agreement between Janssen Research & Development, LLC and GeneCentric Therapeutics, Inc. and a sponsored research agreement between GeneCentric Therapeutics, Inc and the University of North Carolina at Chapel Hill. TLR is supported by the Doris Duke Charitable Foundation (grant number 2015213) and the National Cancer Institute of the National Institutes of Health (1K08CA248967-01 Clinical Investigator Award). WYK and TLR receive research funding from GeneCentric Therapeutics and Merck. TLR receives research funding from Genentech/Hoffman-La Roche and Bristol-Myers Squibb. GMM, YS, MVM, JMU and JRE are employees of GeneCentric Therapeutics, Inc. and have stock interest in the company. GM a patent holder of the GeneCentric bladder cancer subtype classifier.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-021-01488-6.