Loss of canonical sex steroid signaling correlates with prostate cancer progression and induces tumor escape in Drosophila

Reviewer #1 (Public review):

Summary:

In this article, Vialat and his colleagues examine the early stages - which remain largely unknown - of the tumor escape process, particularly the basal extrusion of tumor cells following endocrine therapy for prostate cancer.

They first used the "Prostate Cancer Atlas" database, which provides access to a vast amount of transcriptomic data, to perform high-throughput analyses. Interestingly, analyzing a series of Androgen Receptor (AR) target genes in castration-resistant prostate cancers, they concluded that the loss of the canonical AR signaling pathway may contribute to tumor resistance.

Using a well-established model in Drosophila, they then replicated in vivo an endocrine therapy targeting the accessory gland by genetically inhibiting the expression of ecdysone, the only sex steroid present in Drosophila. These experiments induced basal extrusion similar to the mechanism observed in tumor escape in humans.
These results suggest that the deprivation of sex steroids may play an important role in tumor progression.

However, although the data from the "Prostate Cancer Atlas" constitutes a powerful tool that serves as the basis for this new concept, clinical validation using carefully selected human tumor samples would help strengthen the authors' conclusions.

Strengths:

(1) The Prostate Cancer Atlas is a comprehensive collection of clinical data derived from RNA sequencing and serves as a powerful tool for conducting high-throughput analyses in this paper.

(2) The Drosophila model used in this article is well established and has already been the subject of publications by the team. In addition to being an in vivo model, Drosophila offers a threefold advantage for this study: i) the presence of an accessory gland, similar to the prostate, which allows for the simulation of tumor formation and, in particular, extrusion mechanisms; ii) its regulation by a single sex steroid, ecdysone; iii) the genetic ability to modulate or inactivate ecdysone expression, which allows for a parallel to be drawn with hormonal deprivation in humans.

(3) This study presents interesting and original findings. The data are, for the most part, of high quality.

Weaknesses:

(1) The Prostate Cancer Atlas, which is an essential tool in this study, was described only briefly - if at all - in both the introduction and the "Materials and Methods" section. The selection criteria used to distinguish CRPC or NEPC from adenocarcinoma in the Atlas or as determined by the authors, as well as the analytical methods, were not specified. It is therefore difficult to be convinced by the results, particularly those presented in Figures 1 and 2.

(2) Although the hypothesis put forward by the authors - that the deprivation of sex hormones contributes to tumor progression - is strongly supported by the Drosophila model and by the in silico analysis of transcriptomic data from the Atlas, this concept still needs to be clinically validated by analyzing a series of prostate cancer samples, either through transcriptomic analysis or by tracking gene expression in histological sections.

(3) With regard to the cells responsible for tumor escape, stem cells have been described as "candidates for the initiating resistant tumor growth" (lanes 50-55), but it is also essential to address the recent concept of "persistent cells". Indeed, these cells have been primarily associated with their tolerance to treatment (chemotherapy) and are referred to as "drug-tolerant cells". However, persistent cells could also correspond to cells that evade hormone therapy in the case of prostate cancer. This possibility should be discussed in the article.

Reviewer #2 (Public review):

Summary:

In this study, Vialat and collaborators study the role of steroid hormone signalling on the development of prostate cancer (patients) and of accessory gland tumours in Drosophila, a tissue functionally equivalent to the prostate. Mining publicly available prostate cancer expression data and using gene expression signatures, they uncover that androgen signalling is actually down-regulated in castration resistant prostate cancers (CRPC) compared to "primary" cancers, leading the authors to wonder whether down-regulation of canonical androgen signalling could represent an important event increasing tumour aggressiveness. They then take advantage of their recently published tumour model in the accessory gland of Drosophila adult males, in which cells are primed for tumorigenesis by the constitutive activation of the EGFR receptor, to test directly this hypothesis. They show that the genetic invalidation of ecdysone reception and signalling increases the aggressiveness of the "pre-cancerous" lesions, and that ecdysone-insensitive tumours present higher proliferation and initiate basal extrusion.

Strengths:

The authors bring original observations on the role of ecdysone signalling to prevent male accessory gland tumour development in Drosophila

Weaknesses:

(1) The link between the human data mining and Drosophila model is not straightforward.
(2) Important information, in particular clinical information, is missing in the presentation of the cancer patients' data, making it complicated to grasp the solidity of the claims.
(3) Data-mining insights should be validated by orthogonal approaches.
(4) Ecdysone signalling activity should be monitored.

While the two parts of the study both investigate the role of steroid signalling on tumour growth, the link remains slightly artificial. I think starting with Drosophila and then opening with some patient data would be better suited to the level of proof reached here, implying that the anti-tumoral role of steroids observed experimentally in the fly might be conserved based on data mining in patients, rather than trying to prove in the fly the hints gained from public data mining. Indeed, there are many important differences between the mammalian prostate and the fly accessory gland, as well as between sex hormone androgen signalling and developmental timing ecdysone signalling.

The prostate cancer data mining and re-evaluation brings some interesting observations that appear to challenge the androgen driver, contrary to the vast amount of literature. Indeed, the authors observe an apparent decrease in androgen signalling in the more advanced states of the disease, in particular CRPC. In order to better evaluate its clinical relevance, more background on the tumours analysed should be provided.

What treatments were received by the patients? Hormonotherapy? LH/RH analogues? +/- anti-androgens? Are these treatments still given when CRPC emerge and tissues were banked? Metastatic disease? Are these only primary tumours in situ? Are there metastases included in the analyses?

Frequently, castration resistance is associated with alternatively spliced variants of the AR (AR-V7) that become constitutive and could bind to new AR-sensitive enhancers, even in the absence of androgen. Is the splice variant status of patients known, or could it be inferred from the expression data? Would there be different responses according to AR-V7 status?

Regarding the signature used. Why not monitor PSMA, one of the major prostate cancer markers, which is regulated by AR?

Finally, to consolidate the surprising observation that AR signalling is repressed in CRPCs, the authors should back these in silico predictions with orthogonal approaches such as histochemistry on patients' TMA or tissues from mouse models, monitoring AR activity.

Regarding the fly experiments, the observation that ecdysone signalling depletion cooperates with EGFR-lambda activation to generate big overgrowths that delaminate basally without passing through the muscular sheet is interesting. However, several important controls need to be provided in order to support the claims:
a) The authors should use an ecdysone reporter (ERE-LacZ, ERE-GFP...) to monitor and show that Ecdysone signalling is indeed lower in the tumours after genetic manipulations, or that it is higher in EGFR-lambda small clones.
b) EcR is normally a repressor, which is turned into an activator in the presence of 20-hydroxyecdysone. The removal of EcR could lead to de-repression of genes and thus slightly activate the pathway. Monitoring ecdysone signalling activity is thus critical.
c) The authors should also monitor the expression of Phantom, Shadow, Shade, and EcR in the different accessory glands (wild-type, EGFR-lambda, EGFR-lambda & EcR-RNAi). It is extremely surprising that systemic ecdysone has so little role since Phantom, Shadow, or Shade RNAi appear as potent as EcR-RNAi. This quantification has actually been performed for Sad in Figure S5, which is not even mentioned in the text. It should be done for Phtm.

A UAS-yellow-RNAi (or similarly irrelevant RNAi) rather than UAS-GFP should be used as a control for the EcR, Phtm, Sad, Shd, and Tub RNAi. Indeed, loading the RNAi machinery could have some unexpected effects not controlled by the UAS-GFP.

The authors should not use the term "sex steroid" when referring to ecdysone. It is a steroid hormone important for developmental timing and rate of growth, but is not a sex hormone, as sex is cell autonomously genetically determined in the fly.

Author response:

eLife Assessment

This valuable study provides insights into the role of steroid signaling during tumorigenesis in the adult male drosophila accessory gland (functional equivalent of the prostate gland in mammals), hinting at a possible counterintuitive anti-tumoral role of sex hormones during prostate cancer in certain patients. While the Drosophila model provides an elegant way to study the hypothesis derived from the Cancer Atlas analysis, the analyses of public prostate cancer expression data are incomplete and critical knowledge on patients' treatment modalities and normalization across different datasets is missing. This work would be of interest to prostate cancer researchers as it suggests that the absence of androgen receptor signaling in humans could constitute a mechanism promoting tumor escape.

We thank the reviewers for the time they have spent on the manuscript, the production of a public review and their useful recommendations. As a general goal for the corrected version, we will try to provide more insight on the data (especially the human data), and more controls, to strengthen our conclusions. We are aware of the lack of a definitive proof of the role of the apparent decrease in AR signaling on tumour progression, but hope that this manuscript will encourage medical scientists to test/challenge its counterintuitive results in large cohorts of tissues and mouse/human models.

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this article, Vialat and his colleagues examine the early stages - which remain largely unknown - of the tumor escape process, particularly the basal extrusion of tumor cells following endocrine therapy for prostate cancer.

They first used the "Prostate Cancer Atlas" database, which provides access to a vast amount of transcriptomic data, to perform high-throughput analyses. Interestingly, analyzing a series of Androgen Receptor (AR) target genes in castration-resistant prostate cancers, they concluded that the loss of the canonical AR signaling pathway may contribute to tumor resistance.

Using a well-established model in Drosophila, they then replicated in vivo an endocrine therapy targeting the accessory gland by genetically inhibiting the expression of ecdysone, the only sex steroid present in Drosophila. These experiments induced basal extrusion similar to the mechanism observed in tumor escape in humans.

These results suggest that the deprivation of sex steroids may play an important role in tumor progression.

However, although the data from the "Prostate Cancer Atlas" constitutes a powerful tool that serves as the basis for this new concept, clinical validation using carefully selected human tumor samples would help strengthen the authors' conclusions.

Strengths:

(1) The Prostate Cancer Atlas is a comprehensive collection of clinical data derived from RNA sequencing and serves as a powerful tool for conducting high-throughput analyses in this paper.

(2) The Drosophila model used in this article is well established and has already been the subject of publications by the team. In addition to being an in vivo model, Drosophila offers a threefold advantage for this study: i) the presence of an accessory gland, similar to the prostate, which allows for the simulation of tumor formation and, in particular, extrusion mechanisms; ii) its regulation by a single sex steroid, ecdysone; iii) the genetic ability to modulate or inactivate ecdysone expression, which allows for a parallel to be drawn with hormonal deprivation in humans.

(3) This study presents interesting and original findings. The data are, for the most part, of high quality.

Weaknesses:

(1) The Prostate Cancer Atlas, which is an essential tool in this study, was described only briefly - if at all - in both the introduction and the "Materials and Methods" section. The selection criteria used to distinguish CRPC or NEPC from adenocarcinoma in the Atlas or as determined by the authors, as well as the analytical methods, were not specified. It is therefore difficult to be convinced by the results, particularly those presented in Figures 1 and 2.

As the tool has been published in different articles, we chose to limit its description. However, we agree that explanation are necessary, that will be added in the new version. First, we initially used here just basic categories, in order to avoid any possible bias; so mCRPC includes rare DNPC and NECP patients. We will also put the data with true ARPC, with essentially the same results.

For human data, we also expect to use transcriptomic data from an independent cohort to check whether the same loss of AR signaling occurs during progression. Furthermore, we consider to add the data showing that decrease in canonical AR signaling also (logically with the previous results) correlates with castration status or ADT exposure (these info are available on ProstateCancerAtlas). Interestingly, and this can be put in supplementary data, prostatecanceratlas detects changes in EMT genes or proliferation genes that are coherent with what is known about cancer progression, indicating that the apparent decrease in AR signaling should correspond to a real phenomenon.

(2) Although the hypothesis put forward by the authors - that the deprivation of sex hormones contributes to tumor progression - is strongly supported by the Drosophila model and by the in silico analysis of transcriptomic data from the Atlas, this concept still needs to be clinically validated by analyzing a series of prostate cancer samples, either through transcriptomic analysis or by tracking gene expression in histological sections.

There are many indirect evidences that loss of AR signaling induces tumor progression in mouse (as stated in the intro or the discussion of the manuscript). However, as suggested in the introduction of the letter, we believe that medical scientists are the most qualified to prove that sex steroid deprivation indeed induces tumor progression in human. We will add in any case data to at least reinforce this puzzling finding of a decrease in AR canonical signaling during progression.

(3) With regard to the cells responsible for tumor escape, stem cells have been described as "candidates for the initiating resistant tumor growth" (lanes 50-55), but it is also essential to address the recent concept of "persistent cells". Indeed, these cells have been primarily associated with their tolerance to treatment (chemotherapy) and are referred to as "drug-tolerant cells". However, persistent cells could also correspond to cells that evade hormone therapy in the case of prostate cancer. This possibility should be discussed in the article.

This is an interesting suggestion, which can be discussed: on the one hand, intrabasal cells may not have accumulated mutations to survive the loss of EcR signaling, as would do persistent cells. On the other hand, they strongly proliferate, and show no sign of senescence, behaving more like resistant cells. So, it does not look to us that we induced the appearance of persistent cells in the Drosophila accessory gland, except if these cells are quickly reactivating to give rise to intrabasal cells.

Reviewer #2 (Public review):

Summary:

In this study, Vialat and collaborators study the role of steroid hormone signalling on the development of prostate cancer (patients) and of accessory gland tumours in Drosophila, a tissue functionally equivalent to the prostate. Mining publicly available prostate cancer expression data and using gene expression signatures, they uncover that androgen signalling is actually down-regulated in castration resistant prostate cancers (CRPC) compared to "primary" cancers, leading the authors to wonder whether down-regulation of canonical androgen signalling could represent an important event increasing tumour aggressiveness. They then take advantage of their recently published tumour model in the accessory gland of Drosophila adult males, in which cells are primed for tumorigenesis by the constitutive activation of the EGFR receptor, to test directly this hypothesis. They show that the genetic invalidation of ecdysone reception and signalling increases the aggressiveness of the "pre-cancerous" lesions, and that ecdysone-insensitive tumours present higher proliferation and initiate basal extrusion.

Strengths:

The authors bring original observations on the role of ecdysone signalling to prevent male accessory gland tumour development in Drosophila

Weaknesses:

(1) The link between the human data mining and Drosophila model is not straightforward.

(2) Important information, in particular clinical information, is missing in the presentation of the cancer patients' data, making it complicated to grasp the solidity of the claims.

(3) Data-mining insights should be validated by orthogonal approaches.

(4) Ecdysone signalling activity should be monitored.

While the two parts of the study both investigate the role of steroid signalling on tumour growth, the link remains slightly artificial. I think starting with Drosophila and then opening with some patient data would be better suited to the level of proof reached here, implying that the anti-tumoral role of steroids observed experimentally in the fly might be conserved based on data mining in patients, rather than trying to prove in the fly the hints gained from public data mining. Indeed, there are many important differences between the mammalian prostate and the fly accessory gland, as well as between sex hormone androgen signalling and developmental timing ecdysone signalling.

This is an interesting suggestion. Actually, we first wrote the manuscript by starting with Drosophila data and then going to patients data, and previous reviewers said that this was not possible to directly go from Drosophila to human. So, we suppose that the real way to solve this will be by the validation or refutation of the data by other teams in different models.

The prostate cancer data mining and re-evaluation brings some interesting observations that appear to challenge the androgen driver, contrary to the vast amount of literature. Indeed, the authors observe an apparent decrease in androgen signalling in the more advanced states of the disease, in particular CRPC. In order to better evaluate its clinical relevance, more background on the tumours analysed should be provided.

This point is also important to Reviewer 1, and will be implemented.

What treatments were received by the patients? Hormonotherapy? LH/RH analogues? +/- anti-androgens? Are these treatments still given when CRPC emerge and tissues were banked? Metastatic disease? Are these only primary tumours in situ? Are there metastases included in the analyses?

Most CRPC come from metastatic sites. Most of the CRPC were treated by ADT. We chose to have an approach that included all the samples; but we will provide insight, whenever available, on these absolutely relevant questions.

Frequently, castration resistance is associated with alternatively spliced variants of the AR (AR-V7) that become constitutive and could bind to new AR-sensitive enhancers, even in the absence of androgen. Is the splice variant status of patients known, or could it be inferred from the expression data? Would there be different responses according to AR-V7 status?

As a first approach, from the cohorts that were used, it seems that in the PCA patients, there are around or less than 15% of patients harboring the AR-V7 driver. It could be of interest to test their behavior regarding the same set of genes, and it will be done if we can identify the patients.

Regarding the signature used. Why not monitor PSMA, one of the major prostate cancer markers, which is regulated by AR?

It will be done. PSMA behaves as the others, even though the drop between primary samples and ARPC samples is very limited and just statistically significant.

Finally, to consolidate the surprising observation that AR signalling is repressed in CRPCs, the authors should back these in silico predictions with orthogonal approaches such as histochemistry on patients' TMA or tissues from mouse models, monitoring AR activity.

As said previously, we believe that this specific work will be better done by medical scientists.

Regarding the fly experiments, the observation that ecdysone signalling depletion cooperates with EGFR-lambda activation to generate big overgrowths that delaminate basally without passing through the muscular sheet is interesting. However, several important controls need to be provided in order to support the claims:

Considering the comments regarding the fly experiments, we agree that, if experiences are taken individually, controls are lacking. However, we have to explain our strategy and why the results taken in their entirety have a significance. In our model of epithelial tumorigenesis, we have started to explore the EcR pathway after years of work on other pathways. At the first experiment (with the EcR RNAi line), we were struck by the intrabasal phenotype that did not occurred in our previous experiments, and especially for the 14 RNAi lines that we published in two independent articles on Ras/MAPK, Pi3K/Akt pathways and cholesterol metabolism. As justly said in the review, many unexpected effects can happen, so we decided to explore the role of five other genes of the same pathway to be sure of the reproducibility of the phenotype when we block the EcR pathway. The odds of having the same specific phenotype for 6 lines of the EcR pathway when there is always another phenotype for 14 lines targeting other pathways can be calculated: p = 0.00000494. So, the best control we offer, and it is largely significant, is the repetition of the experiments intended to downregulate the EcR pathway, that produce the same phenotypes independently of the target.

Furthermore, all lines we used were previously tested, validated and most of the time published in other scientific works. This is essential to us, as one complexity of doing rare clones in an otherwise normal tissue is that decreasing an mRNA in less than 5% of the cells of course difficultly leads to a detectable drop in overall expression in the whole gland. This is also the reason why we always tried pairs of fly lines to block the receptor activity itself (RNAi EcR, RNAi Shd), the receptor's downstream targets (RNAi HR3, RNAi HR4), and the production of ecdysone (RNAi Sad, RNAi Phtm). As we validated RNAi Sad, we can try anyway to validate at least another RNAi of another category. Furthermore, we did use a RNAi White control: it behaves in the same way as the GFP control. We will put the results comparing the two lines in supplementary data.

(a) The authors should use an ecdysone reporter (ERE-LacZ, ERE-GFP...) to monitor and show that Ecdysone signalling is indeed lower in the tumours after genetic manipulations, or that it is higher in EGFR-lambda small clones.

This would be of interest to validate that the 6 lines are behaving in the same way (at least, they give similar phenotypes). However, in the adult accessory gland, ERE activity is largely lower than during development (DOI: 10.1016/j.jinsphys.2011.03.027), and to be able to decrease it, authors had to express notoriously strong dominant-negative EcR-DN. We can try the experiment but are really not persuaded that we will be able to see a drop of activity with only a decrease of expression of the gene. If we can think of another solution that could be more efficient, we will try it as the idea is of course interesting.

(b) EcR is normally a repressor, which is turned into an activator in the presence of 20-hydroxyecdysone. The removal of EcR could lead to de-repression of genes and thus slightly activate the pathway. Monitoring ecdysone signalling activity is thus critical.

Actually, there are different EcR isoforms. EcR-B1 is generally considered as the main activator of the pathway, as EcR-A is a repressor of the pathway. The EcR RNAi line which was used does not target a specific isoform.

(c) The authors should also monitor the expression of Phantom, Shadow, Shade, and EcR in the different accessory glands (wild-type, EGFR-lambda, EGFR-lambda & EcR-RNAi). It is extremely surprising that systemic ecdysone has so little role since Phantom, Shadow, or Shade RNAi appear as potent as EcR-RNAi. This quantification has actually been performed for Sad in Figure S5, which is not even mentioned in the text. It should be done for Phtm.

The levels of ecdysone are tenths of times lower in adult compared to the peaks during embryogenesis or metamorphosis. And one source of production is the epithelial cells of the accessory glands themselves. Considering that EcR is expressed in all the cells of the accessory gland (epithelial cells and muscle cells), it seems plausible that there is only a very little amount of ecdysone that can in fact be available for the other epithelial cells.

A UAS-yellow-RNAi (or similarly irrelevant RNAi) rather than UAS-GFP should be used as a control for the EcR, Phtm, Sad, Shd, and Tub RNAi. Indeed, loading the RNAi machinery could have some unexpected effects not controlled by the UAS-GFP.

The NLSGFP line we used here is the one we already published twice (and we compared it to RNAi lines), and this is the reason why we used this already validated control. However, we tested a RNAi White line, and it behaves in the same way.

The authors should not use the term "sex steroid" when referring to ecdysone. It is a steroid hormone important for developmental timing and rate of growth, but is not a sex hormone, as sex is cell autonomously genetically determined in the fly.

In human, sex hormones control sexual differentiation (up to adult characteristics) and sexual reproduction. In Drosophila, ecdysone controls sexual reproduction in both sexes and sexual differentiation at least in the female (DOI: 10.1007/s004270050186). From these results, we do not think that saying it is a sex steroid (not a sex hormone) is ill suited. We intend to precise what we put in this term in the introduction to avoid overinterpretation from our part.