SPTAN1, APC, and FGFR3 Mutation Status and APOBEC Mutation Signatures are Predictive of Mitomycin C Response in Non-muscle-invasive Bladder Cancer

Take Home Message Four promising biomarkers were found to be predictive for intravesical mitomycin C in non–muscle-invasive bladder cancer patients. Prospective validation is needed.


Introduction
Non-muscle-invasive bladder cancer (NMIBC) is a heterogeneous disease with highly disparate prognoses [1,2]. Hence, generalised adjuvant treatment of this patient group will inevitably result in overtreatment in some while ineffective in others. Currently, we are unable to identify patients who will benefit from mitomycin C (MMC). The discovery of several distinct molecular signatures and subtypes in NMIBC provides a more accurate disease prognosis in addition to clinicopathological features [3]. Molecular characterisations could potentially explain why clinicopathologically similar tumours can display very different disease courses. However, so far, none of the molecular subtypes described in NMIBC have shown a predictive value. Optimally, we need predictive biomarkers in NMIBC to stratify individual patients to the appropriate adjuvant treatment: surveillance alone, MMC, or bacillus Calmette-Guérin (BCG).
To comply with the elderly and often multimorbid patients with NMIBC, we have investigated previously whether patients with intermediate-and high-risk Ta tumours could be spared from surgery [4]. We studied the effect of monotherapeutic chemoresection with MMC compared with surgery and adjuvant instillations. Chemoresection with MMC was found to be effective in 57% of patients and showed fewer side effects. Ideally, this treatment could be offered to patients with chemosensitive tumours, while surgery as monotherapy or in combination with BCG could be offered to patients with chemoresistant tumours. However, the molecular mechanisms of chemoresistance are currently poorly understood, and no research exists on predictive biomarkers in NMIBC patients treated with MMC [5].
The aim of the current study was to assess predictive tumour biomarkers of chemoresection with MMC in patients with recurrent NMIBC.

Patients and methods
The study population consisted of the intervention group of a randomised, controlled trial conducted between January 2018 and June 2019 [4]. The intervention group consisted of 58 chemo-naïve patients with a known history of Ta bladder tumours (low or high grade), who were included upon recurrence with multiple tumours smaller than 2 cm. All patients received short-term intensive chemoresection with MMC: three weekly instillations with MMC (40 mg/40 ml) for 2 wk. Treatment response was evaluated by flexible cystoscopy 4 wk after treatment completion by visual assessment. The study was approved by the regional ethics committee (1-10-72-148-17) and the Danish Medical Authorities (2017054019), registered with clinicaltrials.gov (NCT03348969), and monitored by Good Clinical Practice. REMARK guidelines were followed for reporting [6].

Biological material
The latest resected tumour prior to study inclusion was collected from all patients (chemo-naïve samples). Moreover, residual tumour tissue after chemoresection was collected from nonresponding patients (postchemo samples). All specimens were formalin fixed and paraffin embedded (FFPE). Carcinoma cells were marked on haematoxylin and eosinstained overview sections. Germline DNA was extracted from peripheral blood leucocytes or from smooth muscle cells in FFPE blocks using the QIAsymphony DSP DNA midi kit (cat#937255; Qiagen, Hilden, Germany).

Exome sequencing
A detailed description is published elsewhere [7]. In brief, the Twist

Statistical analyses
Genes associated with treatment response were analysed using Fisher's exact test. The mutational contribution and the mutational counts in a signature context were compared between groups by the Wilcoxon rank sum test. RFS was analysed using the Kaplan-Meier method and tested by the log-rank test. Patients lost to follow-up were censored. No correction for multiple testing was applied due to confinement of the analyses to candidate genes as mentioned above and due to the low sample size.

Results
In total, 47 patients contributed with 47 chemo-naïve tumour samples and 11 postchemo tumour samples ( Fig. 1A). See Figure 1B for a tumour flowchart. Patients with a complete response to chemoresection had significantly longer RFS than patients with no response (Fig. 1C Figure 2A outlines the most frequently mutated genes and genes where the mutation frequency differs between response groups, together with clinical and histopathological characteristics. The most frequently mutated genes in the chemo-naïve samples were well-established bladder cancer-associated genes: FGFR3 (70%), KMD6A (47%), PIK3CA (38%), KMT2D (26%), and ARID1A (23%). Overall, we observed no significant difference between chemosensitive and chemoresistant tumours regarding the number of SNVs or indels ( Fig. 2B and C). Chemo-naïve samples were then analysed to identify predictive biomarkers of chemoresection with MMC. When comparing the differences in gene mutation frequencies between responding and nonresponding patients, we found seven genes to be of interest: mutations in SPTAN1, a gene involved in DDR, and the tumour suppressor gene APC were significantly associated with chemoresistant tumours (p = 0.035 and 0.034, respectively). Wild-type FGFR3 was associated with chemosensitive tumours (p = 0.055). Furthermore, mutations in PARP4, AFDN, ZFHX3, and PTEN occurred in responding patients only (p = 0.056, p = 0.056, p = 0.1, and p = 0.1, respectively; Fig. 2A). The mutational landscape was further investigated by analysing previously identified bladder cancer-associated mutational signatures: single-base substitution 1 (SBS1), SBS2, SBS5, and SBS13 [9]. SBS2 and SBS13 are associated with APOBEC mutagenesis and have previously been linked to carcinogenesis in bladder cancer [10]. The mutational contribution from APOBEC signatures (SBS2 + SBS13) was significantly higher in nonresponding patients than in responding patients (p = 0.035; Fig. 2D). We also observed a higher absolute number of mutations in an APOBEC signature context in chemoresistant tumours than in chemosensitive tumours, but it reached statistical significance (p = 0.09). SBS5 is a signature related to ERCC2 mutations and proposedly other DDR pathways [11]. There was no difference in the absolute number of mutations contributed from SBS5 between responding and nonresponding patients (p = 0.4). In continuation, there was no association between protein damaging mutations in DDR-related genes and response to chemoresection ( Fig. 2A).

Independent validation
No associations between 24-mo RFS and the mutation status of the candidate genes were found ( As the effect of chemotherapy has been linked to the patient's immune system, we investigated the correlation with an immune infiltration score that estimates the presence of immune cells in the tumour microenvironment [3,12,13]. Importantly, we observed that wild-type FGFR3 tumours in the UROMOL validation cohort were significantly associated with a higher RNA-derived immune infiltration score (n = 129; Wilcoxon rank sum test, p = 0.003; Fig. 3H), indicating a tumour microenvironment with an immune system supportive of response. Trinucleotide signatures in the UROMOL cohort were called based on RNAsequencing data, for which the methodology was different from that for DNA-based signature calls. The resulting APOBEC-like signatures from the RNA-sequencing data showed no association with RFS in the UROMOL cohort.

Clonal evolution following MMC treatment
Eleven nonresponding patients had paired chemo-naïve and postchemo samples ( Supplementary Fig. 1). Overall, we observed no difference in the median number of SNVs or indels between chemo-naïve and postchemo samples. Subsequently, we investigated whether subclonal tumour cell populations sensitive to MMC would disappear during chemoresection, while subclonal tumour cell populations resistant to MMC would persist. All mutations were therefore categorised as shared mutations or mutations private to either chemo-naïve or postchemo samples. ID 35 experienced a tumour burden progression during chemoresection; we observed a change in the composition of mutations shared between samples before and after chemoresection. The change in allele frequency observed in a selected group of mutations ( Fig. 4A; red circle) demonstrates how these mutations constitute only a small fraction of mutations in the chemo-naïve sample (low allele frequency) but a large fraction in the postchemo sample (high allele frequency), indicating that a subclone had become    Figure 4C and D, where no substantial changes are seen. Here, the allele frequencies are similar, and a large fraction of mutations are shared between chemo-naïve and postchemo samples, although the postchemo sample is somewhat less heterogeneous. These patterns were, however, not persistent in all patients.

Discussion
Bladder cancer is a heterogeneous disease, and large patient cohorts are required for identifying robust biomarkers [14]. Here, we set out to delineate biomarkers based on a relatively small, but clinically well-defined cohort. The strength and novelty of this study are the   direct response measures, which allowed us to investigate true predictive biomarkers.
In the present study, we observed an association between MMC treatment response and the mutation status of SPTAN1, APC, FGFR3, and APOBEC-related signatures. The results on FGFR3 did not meet statistical significance (p = 0.055), but independent validation showed consistent results for patients treated both with and without MMC (p = 0.036). Biomarkers found in the UROMOL validation cohort could be both predictive and prognostic; we do not know whether patients are recurrence free because of the intravesical MMC or because they harbour tumours of good prognosis. This could explain the lack of consistency with the validation cohort, which however could also be due to type I statistical error, even though only candidate genes were investigated in the discovery cohort. The validation is furthermore challenged by the low number of patients who have received MMC in the validation cohort. Based on the imperfectly matched validation cohort, the results need further validation.
Interestingly, no association was observed between the mutation status of DDR-related genes and treatment response, even though MMC exerts its cytotoxicity through DNA cross-linkage. However, both SPTAN1 and APC, which were significantly associated with response, and PARP4, which was mutated in chemosensitive tumours only, have been described as DDR-involved genes [15,16]. Zhang et al. [15] showed that repressing SPTAN1 led to increased MMC-induced chromosomal instability and hence treatment response in lung cancer. This supports our results, implying that the mutations found in SPTAN1 are activating.
APC mutations have been shown to induce chemoresistance in numerous cancer forms through several mechanisms within the cancer cell. Among other mechanisms, APC, in in vivo studies, is shown to regulate the nonhomologous end joining repair system that is involved in the repair of DNA double-strand breakage. Furthermore, APC mutations have been shown to induce chemoresistance through angiogenesis by interacting with the tumour microenvironment. This is supportive of a connection between APC mutation status and MMC resistance, considering that MMC works through reduction that is inhibited by oxygen and hence blood supply [17,18]. Lastly, in vivo studies showed that loss of APC was associated with chemoresistance through alterations of the immune cell infiltration [17].
A corresponding interaction between the immune system and the response to chemotherapy could be responsible for the MMC sensitivity seen in wild-type FGFR3 tumours in the present study. In the UROMOL cohort, wild-type FGFR3 was associated with tumours infiltrated by immune cells, which is in line with other studies [19,20]. Further support is found in muscle-invasive bladder cancer (MIBC), where immune cell infiltration was found to be associated with a response to chemotherapy [7]. In continuation, a growing body of evidence indicates that the antitumour activities of chemotherapy are reliant on the patient's immune system [13,21].
The mechanism of chemoresistance is certainly complex, and additional information could be gained from analysing mutational signatures to the mutation status of candidate genes. The published data on APOBEC-related signatures  and treatment response are conflicting [22]. Tumours with APOBEC-induced mutagenesis are suggested to be mutated widely, more heterogeneous, and hence more treatment resistant than tumours with no APOBEC-related mutagenesis [23]. In MIBC, a high APOBEC mutational contribution has been associated with a poor response to neoadjuvant chemotherapy [7]. Furthermore, an in vitro study found increased MMC resistance in cell lines with APOBEC mutagenesis activity [24]. On the contrary, APOBEC-related T1 high-grade tumours have been reported to have a good outcome [25]. Furthermore, APOBEC-related tumours have been shown to have a higher tumour mutation burden [26], which in MIBC has been associated with a good response to immune therapy [27]. In the present study, a high level of ABOBEC-related signatures could be a marker for chemotherapy resistance rather than for MMC resistance specifically. We found no substantial change in the number of mutations between chemo-naïve and postchemo samples, indicating that MMC treatment did not result in any substantial clonal selection or increased mutagenesis. The proportion of shared mutations between chemo-naïve and postchemo samples, however, ranged widely, which could rely on both sampling issues and tumour evolution prior to or during MMC treatment. We would, however, expect the mutations of chemosensitive cells to be present in chemo-naïve samples only. In continuation, we would expect that tumours developing during chemoresection would be more homogeneous: only harbouring mutations of chemoresistant cells, as depicted in the postchemo sample of ID 35. Furthermore, we would expect the mutations of a tumour unaffected by treatment to be generally unchanged, as seen for ID 41. The small decrease in heterogeneity in the postchemo sample could rely on sampling or that some chemosensitive cells disappeared during treatment after all.
Owing to a high mutation frequency, bladder cancer can exhibit considerable temporal heterogeneity, which imposes a major limitation to the study, as no immediate pretreatment biopsy was taken. For maximum comparability, the latest resected tumour prior to study inclusion was therefore investigated. We chose not to take a pretreatment biopsy to avoid tumour ablation not derived from chemoresection.   Another limitation is the validation of the present study. We were not able to confirm the MMC treatment status for all patients in the UROMOL cohort, and the response measure was RFS. The biomarkers observed could therefore be both predictive and prognostic. However, we hypothesise that RFS is a good proxy for MMC efficacy in this patient cohort. Further validation is however needed from a prospective cohort treated with chemoresection.

Conclusions
APC, SPTAN1, and FGFR3 mutation status and the level of mutational contribution from APOBEC-related signatures were identified as potential predictive biomarkers for chemoresection with MMC in NMIBC patients. A prospective validation study is however needed.