Background
Bladder cancer was estimated to account for 81,190 new cases and 17,240 deaths in the United States in 2018 [
1]. Muscle invasive bladder cancer (MIBC) is characterized by a high risk of relapse and metastasis [
2]. The standard treatment consists of neoadjuvant chemotherapy followed by radical cystectomy. Nevertheless, neoadjuvant chemotherapy is a cisplatin-based schedule that is associated with significant toxicity. Some patients do not benefit from this approach, with their tumors progressing despite the administration of chemotherapy. Therefore, these patients are receiving a toxic and unnecessary treatment, as well as delaying a potentially curative treatment, such as surgery. Unfortunately, there are no reliable biomarkers to guide the selection of patients for these therapies. Several translational studies have aimed to identify subgroups of patients with different clinical behavior.
Choi et al. identified three groups of MIBC (luminal, basal and p53-like) with different response to neoadjuvant chemotherapy [
3]. The Cancer Genome Atlas (TCGA) developed a molecular classification of MIBC based on RNAseq data and hierarchical cluster analysis [
4]. In this work, five different groups were established: luminal-papillary (which included luminal tumors with papillary histology), luminal-infiltrated (characterized by lymphocyte infiltration), luminal, basal/squamous (also with lymphocyte infiltration) and a small neuronal group.
Seiler et al. associated TCGA molecular subtypes with response to neoadjuvant chemotherapy in a new cohort of patients [
5]. Basal tumors appeared to benefit most from neoadjuvant chemotherapy, whereas luminal immune infiltrated tumors had a worse prognosis. However, these findings are not compelling enough to drive clinical decisions, so further insight into the molecular biology of MIBC is needed.
Data were analyzed using three mathematical methods that have proven to be very useful in other fields. Probabilistic graphical models (PGM) can identify differences in biological process among tumors [
6‐
10]. Mathematical classification methods, such as sparse k-means [
11] and Consensus Cluster [
12], previously demonstrated their usefulness in the establishment of tumor subtypes [
6]. On the other hand, Flux Balance Analysis (FBA) is a widely used approach for modeling biochemical networks. FBA could be used to calculate tumor growth rate [
13].
In this study, data from the TCGA cohort were analyzed through PGM and computational analysis to characterize MIBC at the functional level.
Discussion
MIBC has a poor prognosis, with over 50% relapses in spite of appropriate therapy [
26]. Therefore, it is still necessary to characterize MIBC in order to propose new therapeutic targets. Molecular, functional and metabolic characterization by PGM, layer analyses and FBA were performed in this study to provide insight into the molecular features of MIBC.
The PGM unveiled the functional structure of tumors, which has been previously described by our group [
6,
8,
9]. This allows for the study of gene expression data from biological and functional points of view. The functional structure obtained by the PGM analyses is a useful visual tool to determine differential biological processes between tumors. As an example of consistency in this regard, cytochrome P450 and UDP-glucuronosyl transferase were related to androgen receptor in the same node, and androgens are known to modulate the expression of these enzymes [
27].
Layer analyses provided different information about the molecular features of the tumors, as for example, about the cellular adhesion process and about immune status. This new approach, based on well-known classificatory mathematical algorithms, allows us to study and interpret the molecular information separately from the adhesion and the immune information, which are grouped together within the expression data. The first layer divided MIBC tumors into Luminal and Basal. Dadhania et al. validated Luminal and Basal biomarkers across three different cohorts of patients with MIBC by gene expression and immunohistochemistry [
28].
It is remarkable that Luminal tumors presented higher steroid metabolism node activity. Indeed, AR gene presents higher expression in Luminal tumors, suggesting the usefulness of AR as a possible therapeutic target. AR was previously associated with bladder cancer progression [
29] and in vitro studies showed that a siRNA against AR decreased proliferation of AR-positive bladder cancer cell lines but had no effect on AR-negative cells [
30]. However, the role of AR in bladder cancer remains unclear and further characterization is still necessary. Li et al. showed that more than 30–50% of bladder tumors have a detectable presence of AR [
31]. However, they did not relate AR to any molecular subtype. Our proposal is that patients with Luminal MIBC tumors are characterized by a high expression of AR and they could be candidates for therapy with AR inhibitors.
Luminal tumors showed a higher flux activity of androgen and steroid metabolism pathway as well, which agrees with the results found in node activity. Luminal tumors also showed a higher flux activity of the glycolysis pathway so they may respond to drugs targeting metabolism, such as metformin, which has been shown to reduce growth in bladder cancer cells [
32].
Moreover, a group of luminal tumors with extracellular matrix-high characteristics was also described. This group is not related to the papillary histology because it had both papillary and non-papillary tumors. As far we know, a group with these features has not been described before.
On the other hand, FBA predicted that, as has been seen in breast cancer [
8], basal tumors are more proliferative than luminal tumors. It has been established that basal breast cancer tumors have a good response to chemotherapy because they are more proliferative [
33,
34]. Based on the FBA results, the previous knowledge of basal breast tumors, and taking into account that this cohort is chemotherapy-naïve, basal patients may be good responders to chemotherapy as was previously suggested by Seiler et al. [
5]. Proliferation has been determined in other tumor types through gene expression, but data in bladder carcinoma are scarce. FBA allows not only to study proliferation but also to compare metabolic pathways.
The third layer identified an immune high expression group with elevated expression of CTLA4 and CD274, which may be a group of patients who are candidates for receiving immunotherapy, given that CD274, also known as PD-L1, and CTLA4 are the basis of current immunotherapy [
35]. Interestingly, immune high tumors had a worse prognosis than immune low tumors, consistent with the results reported by Seiler et al., which suggested that luminal immune infiltrated tumors had a worse prognosis [
5].
Percentage distribution between luminal and basal tumors was comparable in the TCGA classification and the layer classification. However, the TCGA classification mixes immune, histological and molecular information. Our approach, on the contrary, establishes two independent informative layers: molecular and immune; and, consequently, two complementary classifications. It rendered some interesting findings that complement the TCGA classification: for instance, 10% of basal tumors had an immune-low status, whereas in the TCGA classification all basal tumors had an immune-positive status. Additionally, the TCGA luminal-papillary group, which is defined solely by histological features, had immune-high characteristics when the layer classification was applied. With the arrival of immunotherapies to the clinic, it could be useful to characterize the immune status of tumors and establish groups with differential immune features to drive treatment decisions.
The study has some limitations. Our findings should be validated in an independent cohort. Publications including expression and clinical data are scarce, so validation should be performed in a prospective study. On the other hand, the existence of a neuronal group could not be confirmed. As the neuronal group accounted for a minority of cases in the TCGA study, maybe we should have analyzed a larger population. Finally, although our results suggest that some drugs may work better in specific groups, this should be prospectively validated. Response to chemotherapy, for instance, has been related to multiple factors and the proliferation profile may not be enough to identify responders.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.