Introduction

The incidence of brain metastasis continues to increase, yet current therapies available for patients with disseminated cancer cells in their central nervous system (CNS) have a limited efficacy and fail to improve survival (Valiente et al, 2018; Moravan et al, 2020; Suh et al, 2020).

Consequently, during the past years, there have been recurrent efforts to improve clinical trial design and management specifically concerning this patient population (Lin et al, 2013a, 2013b, 2015; Le Rhun et al, 2021). However, the inclusion of patients with active CNS disease has been limited in the trials of the past, and this represents an unsolved issue (Arvold et al, 2016). As a result, information regarding CNS clinical efficacy of most anti-cancer agents that are FDA-approved or in clinical trials is limited. Thus, exploring therapeutic vulnerabilities and corresponding pharmacological agents with high CNS activity in preclinical models are crucial to promote urgently needed prospective clinical trials that include patients with brain metastases (Camidge et al, 2018).

In vivo drug-screening using mouse models that faithfully recapitulate the clinical phenotype imposes high demand of economic costs, time, and resources (Gao et al, 2015) that are unaffordable by most academical research institutions. On the other hand, cell-based assays lack the contribution of the tumor-associated microenvironment, which has gained relevance in the context of response to therapy during recent years (Hirata & Sahai, 2017). In this regard, the brain microenvironment is a key aspect in the biology of CNS metastasis (Boire et al, 2020) that has been demonstrated to limit therapeutic benefits of systemic therapy (Chen et al, 2016).

To overcome limitations of both in vivo and in vitro approaches, we report an ex vivo organotypic culture-based drug-screening system: METPlatform. We use this strategy to evaluate the impact of different therapeutic agents on metastases growing in situ (i.e., the brain), thus identifying biologically relevant drug candidates in a rapid and cost-effective manner.

Brain organotypic cultures have been used in cancer research due to their ability to mimic the progression of metastatic disease locally (Zhu & Valiente, 2021). They resemble both early (Valiente et al, 2014; Er et al, 2018) and advanced stages of the disease (Priego et al, 2018). Their versatility allows exploring diverse functional and mechanistic insights of brain metastasis, including the interaction between cancer cells and different components of the microenvironment using genetic or pharmacologic approaches (Valiente et al, 2014; Er et al, 2018; Priego et al, 2018). However, to the best of our knowledge, their use for drug-screening has not been reported. We describe here the use of brain organotypic cultures for performing a medium-throughput screening using an in-house library of anti-cancer agents, FDA-approved, or under clinical development (Bejarano et al, 2019), with unknown or limited information regarding their activity in the CNS.

In addition to other hits, METPlatform identified inhibitors of heat shock protein 90 (HSP90) as a potential target to increase the vulnerability of brain metastasis. HSP90 is a molecular chaperone required for correct protein folding, intracellular disposition, and proteolytic turnover of its client proteins, and therefore essential for cellular proteostasis (Schopf et al, 2017). It is heavily exploited by cancer cells not only to maintain numerous pro-survival oncoproteins and transcription factors, but also to buffer proteotoxic stress induced during oncogenic transformation and progression (Whitesell & Lindquist, 2005) as well as to regulate mechanisms of immune evasion (Fionda et al, 2009; Kawabe et al, 2009). High HSP90 expression levels have been correlated with poor prognosis in all subtypes of breast cancer patients (Pick et al, 2007; Dimas et al, 2018), several independent cohorts of non-small cell lung cancer (NSCLC) patients (Gallegos Ruiz et al, 2008), and in colorectal cancer (Kim et al, 2019).

Following METPlatform identification of HSP90 as a potential target, we show the potent anti-metastatic activity of a second-generation HSP90 inhibitor, DEBIO-0932, in experimental and human metastases. Furthermore, we use METPlatform to dissect the underlying biology downstream of HSP90 inhibition using unbiased proteomics to identify novel mediators of brain metastasis, biomarkers of the disease, and combination strategies to overcome resistance.

As a final proof-of-concept, we show that METPlatform could be additionally exploited as a clinically compatible “avatar” to predict the therapeutic response of patients with brain tumors.

Results A chemical library applied to METPlatform identifies potential vulnerabilities of brain metastasis

Given our interest in targeting clinically relevant stages of brain metastasis, we used METPlatform to study vulnerabilities of macrometastases. The human lung adenocarcinoma brain metastatic (BrM) cell line H2030-BrM (Nguyen et al, 2009) was injected intracardially into athymic nude mice to obtain fully established brain metastases at clinical endpoint of the animals. Brains were processed into organotypic cultures, and the efficacy of the anti-tumoral library (Table EV1) was evaluated at a concentration of 10 µM (Fig 1A). Of note, established methods to assess the viability of this preparation such as LDH detection from dead cells showed a slight increase during the initial stages of culture preparation, which could be associated with sample processing since it gets stabilized during culture (Appendix Fig S1A). Given the expression of luciferase and GFP in the H2030-BrM model (Nguyen et al, 2009), the impact of specific inhibitors on the viability of brain metastases in organotypic cultures was assessed by bioluminescence imaging (BLI) and immunofluorescence against GFP in comparison with DMSO-treated cultures. We used a PI3K inhibitor, BKM120, as an internal positive control in our experiments due to the known involvement of this signaling pathway and therapeutic benefit in brain metastasis (Nanni et al, 2012; Brastianos et al, 2015; Pistilli et al, 2018). In addition to reproduce the efficacy of BKM120, METPlatform identified additional compounds that are superior in their ability to compromise the viability of established brain metastasis (Fig 1B and C). Top hits were selected by reducing in 80% or more the bioluminescence values that correspond to controls treated with DMSO (Fig 1B). This threshold was confirmed to be a good correlate of compromised viability based on a complementary histological analysis (Fig 1C). The analysis of the drug-screen provided us with 17 hits: carfilzomib (#1), dovitinib (#9), trametinib (#22), mitomycin C (#39), GSK2126458 (#44), AT7519 (#52), CNIO-DUAL (#56), sorafenib (#59), geldanamycin (#60), SN-38 (#72), bortezomib (#84), KU-57788 (#87), CNIO-TRIPLE (#104), crizotinib (#106), CNIO-ATR (#107), pazopanib (#110), and linifanib (#113) out of 114 compounds tested (Fig 1B and C, Table EV1).

Figure 1. A chemical library applied to METPlatform identifies potential vulnerabilities of brain metastasis

Schema of the experimental design. Quantification of the bioluminescence signal emitted by established H2030-BrM brain metastases in each organotypic culture at day 3 normalized by their initial value at day 0 (before the addition of DMSO or any compound). The final value in the graph is normalized to the organotypic cultures treated with DMSO. Blue: DMSO-treated organotypic cultures; red: hits, compounds with normalized BLI ≤ 20%; green: BKM120 and compounds with similar efficacy to BKM120; gray: compounds that do not reduce BLI values. Values are shown in box-and-whisker plots where the line in the box corresponds to the mean. Boxes extend from the minimum to the maximum value (n = 28 DMSO; n = 21 BKM120-treated organotypic cultures; each experimental compound of the library was assayed by duplicate, 8 independent experiments). Hits highlighted in bold are common to those obtained in the in vitro screening (Fig EV1A). Gray dashed line indicates the minimum decrease in BLI (25%) that we considered as a positive phenotype. The black dashed line represents 80% decrease in BLI, which identifies top hits. Representative images of bioluminescence (BLI) and histology of organotypic cultures with established brain metastases from H2030-BrM treated with DMSO, BKM120 or the indicated hits. Cancer cells are in green (GFP) and proliferative cells are in red (BrdU). Scale bar: 75 µm. Venn diagram showing the number of hits ex vivo (17) and in vitro (14) and common to both approaches (7). Compounds tested in additional screens (screen#3: H2030-BrM spheroids; screen#4: established MDA231-BrM breast cancer brain metastasis; and screen#5: metastasis initiation H2030-BrM) only include those considered as hits ex vivo in panel B. Number of hits in each screen are indicated over the total number of hits obtained in screen#1 (B). Schema of the experimental design. Organotypic cultures with H2030-BrM cells mimicking the early steps of colonization were used to perform dose-response optimization with DEBIO-0932. Representative BLI and histology of organotypic cultures with H2030-BrM cancer cells treated with DMSO or decreasing concentrations of DEBIO-0932. Scale bar: 100 µm; high magnification: 50 µm. Quantification of the bioluminescence signal emitted by each condition shown in (F) at Day 3 normalized by the initial value obtained at Day 0 and normalized to the organotypic cultures treated with DMSO. Day 0 is considered 12–16 h after the addition of cancer cells and treatment or DMSO. Values are shown in box-and-whisker plots where each dot is an organotypic culture and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (n = 8 DMSO, n = 8 BKM120 and n = 7 per concentration of DEBIO-0932-treated organotypic cultures, 2 independent experiments). P value was calculated using two-tailed t-test. Schema of the experimental design. Organotypic cultures with H2030-BrM established metastases were used to test the efficacy of DEBIO-0932. Quantification of the bioluminescence signal emitted by H2030-BrM established metastases in organotypic cultures at Day 3 normalized by the initial value obtained at Day 0 and normalized to the organotypic cultures treated with DMSO. Day 0 is considered right before addition of the treatment or DMSO. Values are shown in box-and-whisker plots where each dot is an organotypic culture and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (n = 4 organotypic cultures per experimental condition, 2 independent experiments). P value was calculated using two-tailed t-test. Quantification of the concentration of DEBIO-0932 reached in animals harboring H2030-BrM established brain metastases 6 h after oral administration of DEBIO-0932 at 160 mg/kg. The concentration was measured in both the plasma and the brain for each mouse. Values are shown as mean + s.e.m. (n = 3 mice per experimental condition). P value was calculated using two-tailed t-test.

To compare METPlatform with a traditional cell-based assay as a drug-screening platform, we applied the same chemical library to H2030-BrM cells cultured in vitro (Fig EV1A). Interestingly, after applying the same criteria based on luminescence, only 7 out of 14 hits obtained in vitro were part of the 17 hits obtained with METPlatform (Figs 1D and EV1C, Table EV1). Even if these hits were applied to H2030-BrM spheroids, only 7 out of 17 also scored (Figs 1D and EV1C, Table EV1). Thus, METPlatform selected hits that would not have been considered as such with other established approaches.

Click here to expand this figure.

Figure EV1. A chemical library applied to METPlatform identifies potential vulnerabilities of brain metastasis

Quantification of the proliferation of H2030-BrM cells at day 3 normalized to the cells treated with DMSO measured with CellTiter-Glo®. Green: hits, compounds with ≤ 20% proliferation; gray: compounds with > 20% proliferation. Values are shown in box-and-whisker plots where the line in the box corresponds to the mean. Each experimental compound of the library was assayed by duplicate. Hits highlighted in bold were common to the ex vivo screening (Fig 1B). Quantification of the bioluminescence signal from MDA231-BrM established brain metastases in organotypic culture after 3 days in culture. Values were normalized by the level of bioluminescence at Day 0 for each culture (before the addition of DMSO or any compound). Final data is shown in percentage respect to reference, the organotypic cultures treated with DMSO. Blue: DMSO-treated organotypic cultures; red: hits, compounds with normalized BLI ≤ 20% (dashed line); green: BKM120; gray: compounds with normalized BLI > 20%. Values are shown in box-and-whisker plots where the line in the box corresponds to the mean. Boxes extend from the minimum to the maximum value (n = 14 DMSO; n = 13 BKM120-treated organotypic cultures; each experimental compound was assayed by duplicate, 4 independent experiments). Detailed representation of the data shown in Figs 1B, EV1A and Table EV1 indicating relative viability using bioluminescence generated by H2030-BrM cells ex vivo (established brain metastases, light red), in vitro 2D (green) and in vitro 3D (spheroids, yellow) treated with compounds of the anti-tumoral library (compounds were assayed by duplicate in each assay). All hits for any condition are shown. The rectangles of the top indicate whether a given compound was effective (< 20% luminescence respect to control) ex vivo (light red rectangle), in vitro 2D (green rectangle), in vitro 3D (yellow rectangle). Representative wild-type brain slices treated with DMSO or the HSP90 inhibitor geldanamycin stained with anti-Col.IV (endothelial cells) and anti-NeuN (neurons). Scale bar: 50 µm. Representative wild-type liver slices treated with DMSO or the HSP90 inhibitor geldanamycin and stained with anti-Ki67 to score proliferation. BB: bisbenzamide. Scale bar: 50 µm. Quantification of GI50 values of geldanamycin in a panel of BrM cell lines in vitro from various primary origins and oncogenomic profiles. Nine serial concentrations of geldanamycin were assayed by duplicate and GI50 was calculated from a viability curve normalized to DMSO-treated cells of the corresponding cell line. Values are shown as mean + s.e.m. (each concentration was assayed by technical duplicates for each cell line and the experiment was performed twice). Quantification of the bioluminescence signal emitted by MDA231-BrM established metastases in organotypic cultures incubated in the presence of DEBIO-0932 (1 µM) during 3 days. Bioluminescence at Day 3 is normalized by the initial value obtained at day D and quantified relative to the organotypic cultures treated with DMSO. Day 0 is considered right before addition of the treatment or DMSO. Values are shown in box-and-whisker plots where each dot is an organotypic culture and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (n = 6 organotypic cultures per experimental condition, 1 experiment). P value was calculated using two-tailed t-test.

We extended our ex vivo drug-screen to a triple-negative breast cancer brain metastasis model, MDA231-BrM (Bos et al, 2009), to identify vulnerabilities regardless the primary tumor origin. Out of the 17 hits tested, 15 of them decreased the viability of cancer cells in 80% or more as measured by BLI (Figs 1D and EV1B, Table EV1). In addition, we used METPlatform to analyze whether any hit also scored not only against advanced stages of the disease when metastases are fully established (Fig 1B), but also against the initial steps of organ colonization, which could be mimicked ex vivo by plating cancer cells on top of tumor-free organotypic brain cultures (Valiente et al, 2014). Interestingly, 14 out of 17 hits inhibited both early and advanced stages of brain metastasis (Fig 1D, Table EV1), which suggests that these compounds may not only be effective treating, but also preventing metastasis outgrowth by acting on the initiation of organ colonization. On the other hand, reported differences in the biology of initial and established brain metastases (Valiente et al, 2014; Priego et al, 2018) could be exploited therapeutically by interrogating those hits only scoring in one or another stage of colonization (dovitinib (#9), pazopanib (#110), and linifanib (#113)) (Table EV1).

Finally, METPlatform also allows simultaneous evaluation of the potential toxicity derived from selected compounds on non-cancer cell types and in different organs. For instance, the use of specific markers for various brain cell types, such as neurons and endothelial cells, allowed us to discard a major unselective cytotoxicity in this organ (Fig EV1D). In contrast, evaluation of reported sensitive organs confirmed the ability of the drug-screening platform to reproduce clinical toxicity (i.e., hepatotoxicity) (Fig EV1E; Supko et al, 1995).

Altogether, our results support METPlatform as a comprehensive and more informative drug-screening platform in the context of metastasis compared to conventional cell-based assays (Fig 1D, Table EV1).

In order to select compounds for further validation, we focused on those targeting not only established metastasis from different cancer types but also initial stages of organ colonization (Fig 1D, Table EV1). Out of this selection, we then focused on those that, although with inhibitory activity 2D and 3D in vitro (Fig EV1F), did not score as hits in this condition (Fig EV1A and C). With this selection criterion, we wanted to evaluate the potential of METPlatform to select hits working in vivo. Six hits fulfilled the criteria: trametinib (#22), AT7519 (#52), sorafenib (#59), geldanamycin (#60), KU-57788 (#87), and CNIO-ATR (#107). Unfortunately, METPlatform has no capacity to score blood–brain barrier (BBB)/blood–tumor barrier (BTB) permeability, and indeed, we failed to recognize this property among these compounds, suggesting that, when METPlatform is applied to metastasis in the brain, a previous step to prioritize BBB/ BTB-permeable compounds should be incorporated to design the library (Saxena et al, 2019). Given the improved efficacy of brain permeable compounds to target metastasis in this organ (Osswald et al, 2016), we looked for alternative inhibitors focused on the targets identified. DEBIO-0932, a second-generation HSP90 inhibitor, has an improved toxicity profile in comparison with geldanamycin, increased bioavailability and, more importantly, a remarkable ability to cross the BBB (Supko et al, 1995; Bao et al, 2009). As geldanamycin, DEBIO-0932 blunted the viability of initial and established brain metastases from lung (H2030-BrM) and breast (MDA231-BrM) cancer models in ex vivo assays (Figs 1 E–I and Fig EV1G, Table EV10). Furthermore, the concentration reached by DEBIO-0932 in a brain affected by metastases (Fig 1J) is above the therapeutic levels as determined ex vivo (Fig 1E–I).

Given the importance of the metastasis-associated microenvironment for local disease progression (Boire et al, 2020), we evaluated in more detail this aspect in METPlatform (Fig 2A). First, we determine that the vehicle used was not influencing the brain microenvironment at the concentration used (Appendix Fig S1B). Second, we introduced inhibitors previously reported to influence glial cells such as methotrexate (MTX) (Gibson et al, 2019) and BKM120 (Blazquez et al, 2018). In comparison with DEBIO-0932, MTX massively induced tumor-associated microglia/macrophages and reactive astrocytes (Fig 2B) markers, although this was not translated into a compromise of metastasis viability as assessed by histology and bioluminescence (Fig 2B and C). Finally, although established methods for assessing major toxicity effects (i.e., LDH) did not reflect any major impact from any compound (Fig 2D), high concentrations of BKM120 and DEBIO-0932 showed incipient signs of their impact on the tumor-associated microenvironment (Fig 2B). Given that low concentrations used for DEBIO-0932 had a major effect on the viability of metastatic cells (Figs 1F–I and 2C), we conclude that METPlatform not only identified potential vulnerabilities but it also allows to evaluate the differential sensitivity of cancer cells versus tumor-associated microenvironment to a given drug. Given the limited efforts to test drugs currently available or under clinical trials in patients with brain metastasis, METPlatform provides an additional strategy to generate initial data on this potential application. As such, we identified DEBIO-0932 as a potent inhibitor of brain metastases viability ex vivo that is able to accumulate in the brain at therapeutic concentrations.

Figure 2. METPlatform is compatible with the evaluation of the metastasis-associated microenvironment

Schema of the experimental design. Representative images of organotypic cultures with established metastases with various glial components of the microenvironment labeled. Scale bar: 75 µm. Each individual condition was evaluated in several organotypic cultures (3–6 slices). Quantification of the bioluminescence signal emitted by established H2030-BrM brain metastases in each organotypic culture at Day 3 normalized by their initial value at Day 0 (before the addition of DMSO or any compound). The final value in the graph is normalized to the organotypic cultures treated with DMSO. Values are shown in box-and-whisker plots where the line in the box corresponds to the mean. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (n = 5–6 organotypic cultures, 1 independent experiment). P value was calculated using two-tailed t-test. Quantification of LDH levels in the conditioned media of organotypic slices cultured during 3 days relative to a lysate of the same preparation. Values are shown as mean + s.e.m. (n = 3 organotypic cultures per experimental condition, 1 independent experiment). Brain metastases are positive for HSP90

Before testing the potential benefits of DEBIO-0932 in vivo, we evaluated the presence of its target in brain metastases. To evaluate HSP90 levels in situ, we performed tissue immunofluorescence in four experimental brain metastasis models from both human and mouse origin, characterized by different oncogenic drivers and derived from breast cancer, lung cancer, and melanoma, which are the most frequent sources of brain metastasis (Valiente et al, 2020). Established brain metastases obtained at experimental endpoint showed high HSP90 levels in cancer cells (Fig 3B). In sharp contrast, the unaffected brain did not show any positivity with the exception of specific neuronal nuclei, such as the medial habenula (Fig 3A). Of interest, metastasis-associated Iba1+ microglia/macrophages showed high intensity of HSP90; however, they were outnumbered by HSP90high cancer cells (Fig 3C). Thus, we focus our efforts on the characterization of the drug target in metastatic cells.

Figure 3. Brain metastases are positive for HSP90

A–C. Immunofluorescence against HSP90 in mouse brains with established metastases. (A) HSP90 positive structures in areas not affected by the metastasis includes the medial habenula, where neurons co-localize with the chaperone. Scale bars: 100 µm (low magnification), 50 µm (medial habenula nucleus), 12 µm (high magnification neurons). (B) Established metastases from different primary origins and oncogenomic profiles stained with HSP90. Dotted lines delineate the metastasis (cc: cancer cells). Scale bars: 75 µm. (C) Iba1 colocalizes with HSP90 within areas affected by metastases. BB: bisbenzamide. Scale bar: 75 µm (low magnification), 12 µm (high magnification). D. Immunohistochemistry against HSP90 was performed in human brain metastases (n = 60) from lung (40 cases) and breast cancer (20 cases). E. Representative human brain metastases showing different intensities or scores for HSP90. Scale bar: 50 µm. F. Quantification of HSP90 in human brain metastases. 59 out of 60 (98%) showed positive staining of HSP90 in the tumor, 15 (25%) scored with 3 (strong), 36 (60%) with 2 (moderate), and 8 (13%) with 1 (weak) according to the signal intensity of HSP90 in the cytoplasm of cancer cells. G. Human brain metastases (n = 30) and their matched primary tumors (n = 28 lung and n = 2 breast) were evaluated and compared for HSP90 expression by immunohistochemistry. H. Quantification of HSP90 in human primary tumors. 29 out of 30 (97%) showed positive staining of HSP90 in the tumor, 6 (20%) scored with 3 (strong), 10 (34%) with 2 (moderate), and 13 (43%) with 1 (weak) according to the signal intensity of HSP90 in the cytoplasm of cancer cells. I. Schema showing HSP90 scores in matched pairs of primary tumor and brain metastasis. Red: increase of HSP90 score from primary to brain metastasis; green: decrease of HSP90 score; gray: no changes in HSP90 score. J, K. Representative human brain metastases showing different percentages of nuclear HSP90. Scale bars: (J) 50 µm; (K) low magnification: 100 µm; high magnification: 10 µm. Black arrows point to cancer cells positive for HSP90 in the nucleus. L. Quantification of nuclear HSP90 in human brain metastases. 54 out of 60 samples (90%) showed positive nuclear HSP90 in the tumor. 27 (45%) showed 1–5% (moderate) and 27 (45%) showed > 5% (high) of nuclear HSP90. M. Quantification of nuclear HSP90 in human primary tumors. 19 out of 30 (63%) showed positive nuclear HSP90 in the tumor. 9 (30%) showed 1–5% (moderate) and 10 (33%) showed > 5% (high) of nuclear HSP90.

60 paraffin-embedded human brain metastases from NSCLC (40 samples) and breast adenocarcinoma (20 samples) were stained with anti-HSP90 by immunohistochemistry and blindly evaluated and scored by a pathologist (Fig 3D, Table EV2). 98% of brain metastases were positive for HSP90, with 85% of them showing moderate or strong staining of the protein (score ≥ 2, HSP90high) (Fig 3E and F), which is a value higher than previous reports on primary tumors (Pick et al, 2007; Gallegos Ruiz et al, 2008; Kim et al, 2019). To investigate this possibility, we scored 30 matched primary tumors (Fig 3G) and confirmed a lower percentage (54%) of samples scoring as HSP90high in comparison to brain metastases (Fig 3H). When comparing matched pairs of a primary tumor and a brain metastasis, 13/30 (43%) brain metastases had increased HSP90 levels compared to the primary tumor, from which 10/13 (77%) switched from HSP90low (score ≤ 1) to HSP90high (score ≥ 2). 12/30 (40%) matched pairs showed equal HSP90 levels; however, 8/12 (67%) cases were HSP90high in the primary tumor to start with. Out of the 5/30 (17%) brain metastases with lower HSP90 than the corresponding primary tumor, 3/5 (60%) cases still remained within the HSP90high category and only 2/5 (40%) switched from HSP90high to HSP90low (Fig 3I).

Although HSP90 is primarily a cytoplasmic protein, several studies have described its role in nuclear events such as transcriptional processes, chromatin remodeling, and DNA damage (Trepel et al, 2010; Antonova et al, 2019). Moreover, increased nuclear HSP90 correlated positively with poor survival and distant metastasis in NSCLC patients (Su et al, 2016). Interestingly, we found nuclear staining of HSP90 in 90% of brain metastasis samples (Fig 3J–L), with 45% of them scoring as HSP90high (> 5% of positive nuclei out of total tumor) according to a previously described criteria (Su et al, 2016) (Fig 3L). Similar to the previous analysis, we found fewer primary tumors (63%) positive for nuclear HSP90, with 33% of them scoring HSP90high (Fig 3M). Nevertheless, due to the prevalent low percentage of positive nuclei observed in most samples (Fig 3J), we were not able to accurately assess a potential enrichment of nuclear HSP90 in brain metastases compared to their paired primary tumor.

Taken together, our results demonstrate that high levels of HSP90 in cancer cells are a frequent finding among human brain metastasis independently of the primary tumor. Indeed, a clear tendency to maintain or further increase the levels of this protein is evident when compared to matched primary tumors. Overall, these results support potential functional implications of HSP90 in human brain metastasis.

Inhibition of HSP90 is effective to treat established brain metastasis

We used DEBIO-0932 in preclinical models to study whether the results obtained with METPlatform could be translated in vivo.

Brain metastases were induced by intracardiac inoculation of H2030-BrM cells (Nguyen et al, 2009). Two weeks after injection, we confirmed the presence of established metastases in the brain using BLI, histology, and magnetic resonance imaging (MRI) (Fig 4A). DEBIO-0932 administration at 160 mg/kg during the following 3 weeks significantly impaired the growth of both brain metastases and extracranial lesions (Figs 4B–G and EV2F–H) by targeting HSP90 in cancer cells (Fig EV2A–D; Bagatell et al, 2000). We did not observe similar effects of DEBIO-0932 in the microenvironment (Fig EV2E). These results were confirmed by brain and thorax ex vivo BLI (Figs 4B and EV2G and H) as well as histological quantification of dissected brains at the endpoint of the experiment, 5 weeks after cancer cell inoculation, including a reduction of metastases (Fig 4E and F) with an increased in cancer cell death (Fig 4E and G). Of note, we did not observe significant weight loss (Fig EV2I), food intake (Fig EV2J), or any other sign of toxicity after detailed multi-organ histological analysis by an expert pathologist (Fig EV2K) in treated animals compared to the control group, ruling out major toxicities of DEBIO-0932. Indeed, DEBIO-0932 monotherapy increased survival of treated mice (Fig EV2L). However, rather than overinterpreting this significant but limited survival benefit, we use it as an added value reinforcing the need for further characterization of this therapeutic strategy derived from METPlatform. In this sense, treatment of established melanoma brain metastases (Fig EV2N) in an immunocompetent background (Priego et al, 2018) with DEBIO-0932 confirmed the anti-metastatic phenotype (Fig EV2M–P).

Figure 4. Inhibition of HSP90 is effective to treat established brain metastasis

Schema of the experimental design. H2030-BrM cells were inoculated intracardially into nude mice and established brain metastases were detected 2 weeks after by BLI, MRI (arrows) and histology (GFP+ cancer cells). DEBIO-0932 was administered orally at 160 mg/kg for 3 weeks (daily during the first week and every 48 h during the two following weeks) and ex vivo BLI of brains and thoracic regions were analyzed. Brains were processed for histological analysis. Scale bar: 100 µm. Representative in vivo and ex vivo images of vehicle and DEBIO-0932-treated mice 5 weeks (experimental endpoint) after intracardiac inoculation of H2030-BrM cells. Quantification of metastatic progression as measured by in vivo BLI of head of animals. Values are shown as mean ± s.e.m. (n = 23 vehicle and n = 25 DEBIO-0932-treated mice, 3 independent experiments). P value was calculated using two-tailed t-test (P values: *P < 0.05, **P < 0.01, ***P < 0.001). Quantification of ex vivo BLI of brains at the endpoint of the experiment. Values are shown in box-and-whisker plots where every dot represents a different animal and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (n = 21 vehicle and n = 24 DEBIO-0932-treated mice, three independent experiments). P value was calculated using two-tailed t-test. Representative sections of brains from vehicle and DEBIO-0932-treated mice in (B–D). The dotted lines surround the metastases (GFP+). Representative field of view of metastasis stained with GFP and cleaved caspase 3. Scale bars: slices, 1 mm; cleaved caspase 3, 50 µm. Quantification of established metastases found in vehicle and DEBIO-0932-treated brains from panel (E). Values are shown in box-and-whisker plots where every dot represents a different brain and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles and the whiskers go from the minimum to the maximum value (vehicle: n = 10 brains; DEBIO-0932: n = 14 brains). P value was calculated using two-tailed t-test. Quantification of number of cleaved caspase 3 (CC3+) in cancer cells found in vehicle and DEBIO-0932-treated brains from panel (E). Values are shown in box-and-whisker plots where every dot is a metastatic lesion and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles, and the whiskers go from the minimum to the maximum value (n = 8 metastatic lesions from 4 brains per condition). P value was calculated using two-tailed t-test. Schema of the experimental design. Fresh surgically resected human brain metastases (n = 19) from various primary origins were used to perform patient-derived organotypic cultures (BrM-PDOC) and treated with DEBIO-0932 at 10 µM and 1 µM for 3 days. Representative BrM-PDOC stained with proliferation markers (BrdU) and markers of the microenvironment (GFAP for astrocytes, Iba1 for microglia/ macrophages). Scale bar: 50 µm. Quantification of the relative number of BrdU+ cancer cells found in DMSO DEBIO-0932-treated BrM-PDOC respect to the corresponding PDOC treated with DMSO. Values are shown in box-and-whisker plots where every dot represents a patient (mean value obtained from all PDOC from the same condition and patient) and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles, and the whiskers go from the minimum to the maximum value (n = 19 patients with DMSO-treated PDOC, n = 14 DEBIO-0932 10 µM and n = 15 DEBIO-0932 1 µM, each patient is an independent experiment). P value was calculated using two-tailed t-test. Dots are colored according to the primary source of the metastasis. Pie chart showing all BrM-PDOC in (J) classified according to the specific dose tested and the type of response observed. Partial responder means that the response was different depending on the dose of DEBIO-0932, with PDOC not responding at 1 µM. Click here to expand this figure.

Figure EV2. Inhibition of HSP90 is effective to treat established brain metastasis

A. Representative images showing HSP70 levels in brain metastases (generated by intracardiac inoculation of H2030-BrM) found at endpoint of vehicle and DEBIO-0932-treated animals. Scale bar: 75 µm. B. Quantification of HSP70 levels shown in (A) in arbitrary fluorescent units (A.F.U.). Values are shown in box-and-whisker plots where each dot is a metastatic lesion and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles, and the whiskers go from the minimum to the maximum value (n = 6–12 metastatic lesions from 3 to 6 brains per condition). P value was calculated using two-tailed t-test. C, D. HSP90AA1 (C) and HSPB2 (D) expression levels obtained by qRT–PCR of H2030-BrM brain metastases obtained at endpoint of vehicle and DEBIO-0932-treated animals. Values are shown in box-and-whisker plots where every dot represents a different animal and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles, and the whiskers go from the minimum to the maximum value (n = 4 mice per experimental condition). P value was calculated using two-tailed t-test. E. Representative images of HSP90+ non-cancer cell compartments including the medial habenula and the Iba1+ microglia/macrophages in the metastasis-associated microenvironment from vehicle and DEBIO-0932-treated brains at the endpoint of the experiment (Fig 4A). Scale bars: Medial habenula low magnification (nucleus): 50 µm; Medial habenula high magnification (cells): 12.5 µm; Metastasis: 32 µm. F. Quantification of metastatic progression as measured by in vivo BLI of extracranial region of animals. Values are shown as mean ± s.e.m. (n = 23 vehicle and n = 25 DEBIO-0932-treated mice, 3 independent experiments). P value was calculated using two-tailed t-test (P values: **P < 0.01). G. Representative images of thorax from vehicle and DEBIO-0932-treated mice at the endpoint of the experiment. H. Quantification of ex vivo BLI of thoracic regions at the endpoint of the experiment. Values are shown in box-and-whisker plots where every dot represents a different animal and the line in the box corresponds to the median. The boxes go from the upper to the lower quartiles, and the whiskers go from the minimum to the maximum value. (n = 21 vehicle and n = 24 DEBIO-0932-treated mice, three independent experiments). P value was calculated using two-tailed t-test. I. Animal weight from vehicle and DEBIO-0932-treated mice during the treatment period. DEBIO-0932 treatment started 2 weeks (day 14) after inoculation of cancer cells and was maintained for 3 weeks, once every 24 h during the first week and once every 48 h during the two following weeks. Values are shown as mean ± s.e.m. (n = 9 vehicle and n = 10 DEBIO-0932-treated mice). J. Quantification of mean food consumption during the interval of ti