Genetic manipulation of LKB1 elicits lethal metastatic prostate cancer

Prostate cancer (PCa) is primarily diagnosed as a localized disease that is considered curable with first-line therapies. However, two major limitations exist in its clinical management (Chang et al., 2014). First, a considerable proportion of men will relapse after such treatments. These patients will frequently fail to respond to subsequent lines of treatment (androgen pathway inhibitors and chemotherapy) and succumb to metastatic PCa. Second, a fraction of PCa patients will exhibit metastatic disease at diagnosis. Whereas the incidence of this phenomenon is estimated as <10% in populations heavily prescreened for prostate-specific antigen, it is thought to account for ≤50% of PCa mortality (James et al., 2015). The clinical management of lethal metastatic disease is the ultimate challenge in PCa. However, the success in recapitulating this phenomenon in experimental models has been limited (Arriaga and Abate-Shen, 2019).

LKB1 (the protein encoded by the gene STK11) is a master kinase with a prominent role as a tumor suppressor in various tissue types. This gene is mutated and deleted in lung, cervix, uterine, skin, and pancreatic cancers (Momcilovic and Shackelford, 2015). The study of LKB1 in cancer has been boosted by the development of mouse models with varying doses of the tumor suppressor, which result in a reduction of its levels of 50–100% (for heterozygous and homozygous knockout mice, respectively; Pearson et al., 2008) or 25–75% (through the combination of hypomorphic alleles with WT or knockout alleles, respectively; Huang et al., 2008).

Tumor suppressors are clinically and experimentally defined based on gene dosage. Knudson (1971) proposed that genes with such functions would follow the two-hit hypothesis, according to which cancer would emerge upon the loss of the second allele in cells with carrying a mutated gene. However, we know that the relationship between gene dosage and tumor-suppressive activity is more complex, which led to the introduction of concepts such as haploinsufficient tumor suppressors (Berger et al., 2011). However, further experimental effort is required for a comprehensive classification of tumor suppressors, based on evidence produced through the analysis of mouse models with hypomorphic alleles (Alimonti et al., 2010).

In this study, we generate an unprecedented model of lethal metastatic PCa by introducing perturbations in Lkb1 in a precancerous setting and demonstrate the high competence of this kinase to suppress biological features of PCa.

To ascertain the role of LKB1 in PCa, we engineered a series of Lkb1 prostate-specific mutant mice based on a conditional mouse model (Fig. S1, A–D; see Materials and methods; Nardella et al., 2010; Pearson et al., 2008). The survival of mice with prostate-specific Lkb1 (Lkb1^pc−/−) deficiency was indistinguishable from that of control counterparts (Fig. 1 A). Although Lkb1 deletion led to increased tissue mass (Fig. S1 E), it did not result in any histological alteration (Fig. 1 B). These data are in contrast to results obtained using other Cre recombinases such as p450 CYP1A1-driven Cre, which show a broader tissue distribution in the urogenital track (Pearson et al., 2008). The discrepancies might be associated with the impact of the non-prostate epithelial deletion of Lkb1 (Pearson et al., 2008). Also, the original description of this Cre driver reported expression in multiple tissues and cell types beyond the prostate (Ireland et al., 2004). Our results using a well-established prostate epithelium–restricted Cre driver support the notion that Lkb1 loss is not an initiating event in prostate tumorigenesis.

Next, we evaluated whether deletion of Lkb1 would contribute to PCa pathogenesis in the presence of an oncogenic event. To this end, we intercrossed prostate-conditional Lkb1 knockout mice with a well-established driver of murine PCa, Pten knockout mice (Fig. S1, C and D). Whereas Pten^pc−/− mice develop nonmetastatic PCa, the phenotype of heterozygous mice is limited to prostate intraepithelial neoplasia (PIN; Nardella et al., 2010). Surprisingly, the deletion of Lkb1 resulted in a remarkable reduction in survival when combined with one allele deletion of Pten (Fig. 1 C). Since the combination of Pten heterozygosity with Lkb1 loss (hereafter referred to as PTLK) resulted in early lethality, we performed histological characterization of all genotypes at 10 mo of age. PTLK mice exhibited PCa at full penetrance, whereas none of the other genotypes (Pten^pc+/− Lkb1^pc+/+, Pten^pc+/− Lkb1^pc+/−) presented lesions beyond PIN at the same age (Fig. 1 D). Moreover, the macroscopic and histopathological analysis of distant organs unexpectedly revealed that PTLK mice developed lung metastasis with an incidence >80% (justifying their reduced survival rate) and inguinal lymph node colonization to a lower extent (Fig. 1, E and F). Importantly, lung lesions were histologically identical to the prostatic tumor and exhibited positive staining for androgen receptor (AR), supporting the notion that the metastases were of prostatic origin (Fig. S1 F). No lesions were observed in heart, liver, spleen, kidney, or bone (Table S1).

PCa disseminates predominantly to bone. However, dissemination to lungs is well recognized to occur in 10–20% of metastatic disease (Arriaga and Abate-Shen, 2019). We took advantage of publicly available PCa transcriptomics and genomics studies to ascertain the alteration of LKB1 in primary and metastatic lesions. We started by evaluating point mutations in a series of PCa datasets (Taylor et al., 2010, Cancer Genome Atlas Research Network, 2015; Kumar et al., 2016; Abida et al., 2017, 2019; Fraser et al., 2017; Armenia et al., 2018; Gerhauser et al., 2018) using cBioPortal (Cerami et al., 2012; Gao et al., 2013). The results revealed a lack of this type of alteration in PCa compared with a compendium of datasets from lung adenocarcinoma (Cancer Genome Atlas Research Network, 2013; Imielinski et al., 2012; Jordan et al., 2017), a tumor type with frequent alterations in LKB1 (Fig. S1 G). We next focused on genomic aberrations using a large-scale study of PCa (n = 1,013), in which primary tumor biopsies and metastatic specimens where sequenced and tropism was annotated (Armenia et al., 2018). In this dataset, LKB1 heterozygosity was observed in 24.6% of cases (249 out of 1,013), whereas complete loss was restricted to 3.36% of specimens (34 of 1,013; Table S2), in turn suggesting that LKB1 is not a commonly aberrant tumor suppressor in the prostate. However, we noted that metastatic tumors had higher rates of LKB1 genomic aberrations, with 4.8% of homozygous deletions (16 of 333) compared with 2.65% in primary tumors (18 of 680), and 29.1% of heterozygous loss (97 of 333) compared with 22.35% in primary tumors (152 of 680; χ² P = 0.007). We decided to analyze whether perturbations in this gene could be associated with any specific metastatic behavior. 66% of metastatic cases presented site annotation (220 of 333). Metastatic cases within this subset with one or two intact alleles of LKB1 exhibited similar distribution of metastasis, with a predominance of bone and lymph node dissemination and negligible lung metastasis (Fig. 1 G and Table S3). Among the limited number of metastatic cases with homozygous deletion of LKB1, 50% belonged to lung metastatic specimens (2 of 4; Fig. 1 G). The incidence of LKB1 homozygous deletion was 28.6% (2 of 7, Fisher’s exact test P < 0.01) in lung metastasis compared with 0.9% (2 of 213) of all other metastatic specimens, whereas we found no enrichment in bone metastasis or lymph node dissemination (Fig. 1 H, Fig. S1 H, and Table S3).

As a complementary approach, we analyzed an independent cohort (Kumar et al., 2016) in which we compared the expression of LKB1 in the two predominant metastatic sites of interest, bone versus lung. In agreement with our previous results, lung metastatic lesions exhibited a significant reduction in the expression of the kinase (Fig. 1 I).

Taking advantage of the aforementioned large genomic dataset (Armenia et al., 2018), we explored the genetic events that could co-occur with LKB1 in metastatic specimens. To this end, we extracted from this dataset the most frequent copy number alterations (Table S4) and evaluated their association to LKB1 deletion in metastasis. We noted that none of the classic PCa oncogenic mutations were cooccurring with LKB1 deletion (Table S5). These data led us to speculate that the aggressive features enabled upon LKB1 loss are not associated with a single oncogenic hit, but rather to the process of transformation itself. In this regard, we envision the perturbation in Pten used in the mouse model as an illustrative oncogenic signal that would unleash prostate cell proliferation and enable the emergence of Lkb1 loss–dependent phenotypes.

LKB1 phosphorylates 13 different effector kinases (Lizcano et al., 2004). To ascertain which of the potential targets of LKB1 could account for its prostate tumor–suppressive activity, we asked whether any of these targets would phenocopy the pattern of genomic loss of the master regulator using the same patient dataset (Armenia et al., 2018). When looking at the loci of the aforementioned kinases, we found genomic aberrations in only nine of them. Among these genes, we identified the strongest pattern of lung dissemination in SIK3 (Fig. S1 I). Similar to LKB1, SIK3 homozygous deletion was present in 28.6% (2 of 7, Fisher’s exact test P < 0.01) of all lung metastatic cases (Fig. S1 J and Table S6). Interestingly, accounting for metastatic specimens with homozygous deletion in LKB1 and/or SIK3 increased the incidence in lung metastasis in this group to nearly 50% (3 of 7; Fisher’s exact test P < 0.01; Fig. S1 K and Table S7). It is worth noting that the low number of specimens with alterations in these genes precluded a more conclusive result. The potential involvement of SIK3 in the prostate tumor–suppressive activity of LKB1 is coherent with recent reports in non–small cell lung cancer (Hollstein et al., 2019; Murray et al., 2019), in turn suggesting that the LKB1-SIK tumor suppressive axis might be relevant in different cancer types.

The aggressive nature of PCa in PTLK mice prompted us to characterize the molecular and pathological features of these tumors. PTLK prostate tissues exhibited reduced AMPK phosphorylation and elevated activity of mTOR complex 1 (by means of ribosomal protein S6 phosphorylation), in line with the reported inhibitory activity of LKB1 on this pathway (Momcilovic and Shackelford, 2015; Fig. S2, A and B). The histological analysis of PTLK lesions revealed that these tumors presented squamous features, with elevated p63 positivity and Krt5 expression, in line with observations in other tumor types (Momcilovic and Shackelford, 2015; Fig. S2, B and C). We then compared the molecular features in PCa with intact Lkb1 (Pten^pc−/−, PIN or adenocarcinoma as indicated) or Lkb1 complete deletion (PTLK mice, squamous cell carcinoma). PIN lesions retained a well-defined layer of basal prostate cells, characterized by cytokeratin 5/6 (CK5/6) and p63, which were lost in invasive lesions (Fig. 2 A), in line with the notion that these tumors emerge from the aberrant proliferation of luminal cells (Arriaga and Abate-Shen, 2019). In contrast, prostate tumors and metastases in PTLK mice were predominantly composed of CK5/6- and p63-positive basal prostate cells and exhibited transcriptional signatures of squamous cell carcinoma in prostate tumor microdissected specimens (Fig. 2 A, Fig. S2 D, and Table S8). In agreement, prostate squamous cell carcinoma lesions from PTLK mice exhibited a 10-fold increase in basal cell proliferation, shown by Ki67, compared with adenocarcinoma lesions (Fig. S2 E). Our results support the notion that PCa in PTLK mice develops predominantly from the aberrant proliferation of basal cells, leading to highly metastatic squamous cell carcinoma.

The vast majority of available PCa high-throughput molecular studies focus on adenocarcinoma lesions, since this histological cancer type is predominant. However, according to PCa guidelines, squamous cell carcinoma of prostate presents an incidence of 0.5–1% and exhibits highly aggressive characteristics (Brunnhoelzl and Wang, 2018; Munoz et al., 2007). We searched for LKB1-deficient PCa specimens that exhibited squamous pathological features. Interestingly, we identified three such cases (Fig. 2 B). Two of the three cases exhibited positivity for 34βE12, a basal cell marker, whereas one case showed weak positivity. Interestingly, two of these three specimens harbored combined LKB1 and PTEN alterations, analogous to our murine model (Fig. 2 B).

To approach the mechanism by which LKB1 suppresses PCa, we characterized a panel of PCa cells. In line with previous reports, DU145 cells lack LKB1 protein expression, due to a frameshift homozygous mutation (Ikediobi et al., 2006; Fig. 3 A). We took advantage of this cell line and reintroduced LKB1 at expression levels comparable to other prostate cells. As a technical control, we also reexpressed a reported kinase activity-defective mutant, LKB1^K78I (Fig. 3 Band Fig. S2 F; Shaw et al., 2004). Whereas reintroduction of LKB1 did not affect two-dimensional cell growth (Fig. S2 G), we observed a remarkable inhibition of migration, invasion, invasive growth, and anchorage-independent growth in both WT and K78I mutant LKB1-expressing cells (Fig. 3, C–E; and Fig. S2 H). We extended our studies to xenograft assays. At the level of both subcutaneous tumor growth and metastatic dissemination, LKB1 was tumor suppressive regardless of its kinase activity (Fig. 3, F–I; and Fig. S2, I and J).

The results with LKB1^K78I were suggestive of an unprecedented kinase-independent tumor-suppressive activity. We ruled out this possibility by different means. On the one hand, we evaluated the correct and equivalent formation of the LKB1 complex in cells (Boudeau et al., 2003; Momcilovic and Shackelford, 2015). By both coimmunoprecipitation coupled to Western blot and mass spectrometry analysis, we could demonstrate that LKB1^WT and LKB1^K78I presented equivalent protein interaction profiles with core components of its functional complex, including MO25, STE20-related adaptor, HSP90 and CDC37 (Fig. S3, A and B; and Table S9). On the other hand, the kinase-independent effect could be due to the sequestration of MO25 away from other STK family proteins. MO25A was the predominant isoform in these cells (Fig. S3 C). We hypothesized that, if the tumor suppressive activity of both LKB1 constructs relied on the sequestration of MO25, silencing this gene would mimic LKB1 expression and elicit tumor suppression. However, the silencing of this adaptor protein did not reduce two-dimensional or anchorage-independent growth (Fig. S3, D–F).

The lack of evidence supporting a kinase-independent tumor suppressive function of LKB1 prompted us to carefully monitor the impact of K78I mutation on the activity of the kinase. Surprisingly, LKB1^K78I retained 10% of the WT activity (Fig. 4 A). We evaluated the impact of LKB1 expression on the activity of its targets. In line with our clinical observations, the phosphorylation of SIK1-3 was increased by LKB1^WT and to a lower extent by the K78I mutant (Fig. 4 B), and this effect was retained, albeit with lower amplitude, when monitoring AMPK activity (Fig. S3, G and H). To ascertain whether the residual activity of LKB1^K78I is sufficient to exert tumor suppression, we generated a second mutant, harboring a nucleotidic change in the magnesium-binding domain (D194A; Lizcano et al., 2004) that we confirmed to be fully devoid of kinase activity (Fig. 4 Cand Fig. S3 I). Importantly, compared with the deleted parental cells, the expression of LKB1^D194A neither activated SIK1-3 nor influenced the signaling pattern or the biological response in vitro and in vivo, as opposed to LKB1^WT or LKB1^K78I mutant (Fig. 4, D–G; and Fig. S3, J–O). Altogether, our results demonstrate that loss of LKB1 is associated with the emergence of squamous features in murine PCa and that the kinase requires minimal activity to suppress aggressive PCa features in cellular systems.

PCa is a highly prevalent disease in men. Although it is generally considered curable with first-line therapies, a fraction of patients will exhibit biochemical recurrence associated with the failure on subsequent therapeutic alternatives (Chang et al., 2014). These men will eventually develop metastatic disease, thus accounting for the mortality of the disease, together with those individuals that exhibit metastatic PCa at presentation (James et al., 2015). Bone is the predominant site of metastasis in this disease, although a subset of highly aggressive PCa disseminates to the lung and soft tissues (Arriaga and Abate-Shen, 2019). The biology of lethal metastatic PCa is poorly understood, and there are few murine models to date that recapitulate this pattern of aggressiveness (Arriaga and Abate-Shen, 2019). Our study provides the field with an invaluable model to study the process of PCa dissemination in an immune-competent and tissue-specific context. Importantly, our patient dataset analysis using the largest cohort available (Armenia et al., 2018) shows that homozygous deletion in LKB1 is rare in PCa (2.6% in primary tumors and 4.8% in metastasis), and point mutations are negligible. Despite this low frequency of LKB1 homozygous deletion, we find an association of this event with metastasis in the lung. It is important to note that our data in murine models and human specimens is insufficient to postulate a role for LKB1 in metastatic tropism. However, together with the in vitro assays, our data indicate that LKB1 deletion results in the acquisition of aggressiveness features, which would enable prostate tumor cells to gain access to the lung instead of the long metastatic process that underlies bone colonization.

Lkb1 deletion is associated with the onset of metastatic disease in other tumor types when combined with perturbation of different cancer genes (Momcilovic and Shackelford, 2015). Overall, experimental evidence suggests that loss of LKB1 is not an initiating event in cancer, but rather an enhancer of aggressive features once oncogenic signaling is present. Following this line of reasoning, and in line with our patient analysis, it is tempting to speculate that the oncogenic signaling provides a proliferative advantage to cells, regardless of the cancer gene perturbed, that will unleash the metastasis-suppressor potential of LKB1.

Squamous cell carcinoma is included in the clinical guidelines as a highly aggressive subtype of PCa, with an incidence of 0.5–1% and average survival of 14 mo after diagnosis (Brunnhoelzl and Wang, 2018; Munoz et al., 2007). The combination of Pten and Lkb1 deletion in the prostate epithelium performed in this study results in aggressive squamous cell prostate carcinoma. In line with the notion that basal prostate cells have reduced androgen requirements (Xie et al., 2017), PTLK tumors exhibit delocalized or negative AR expression. It is worth highlighting the different cellular compartments that contribute to PCa aggressiveness upon loss of tumor suppressors. Lkb1 loss affects basal cell proliferation. In contrast, deletion in other tumor suppressors such as Tp53 predominantly favors luminal cell renewal and expansion (Agarwal et al., 2015). Importantly, whereas the incidence is low in human PCa, mouse models of PCa are more prone to exhibit squamous features (Ittmann et al., 2013; Martin et al., 2011). In addition, several studies report that Lkb1 deletion in different tissues predisposes to squamous lesions (Momcilovic and Shackelford, 2015). Specifically, combined deletion of Pten and Lkb1 in the lung was recently reported as the first bona fide mouse model of lung squamous cell carcinoma (Xu et al., 2014). Interestingly, the existence of compound LKB1 and PTEN deletions in human prostate tumors with squamous pathological and molecular features opens an avenue of research toward better understanding of this disease.

The introduction of hypomorphic alleles in mouse models has enabled the fine-tuning of dosage beyond the use of gene knockouts. Taking advantage of this technology, we previously reported that Pten is a weak breast tumor suppressor, since the introduction of a hypomorphic allele in heterozygosity (70–75% of WT Pten expression) resulted in breast cancer (Alimonti et al., 2010). The combined use of conditional mouse models and LKB1 mutant in cells in this study provides strong support for the notion that this kinase is a potent prostate tumor suppressor that can operate at very low doses. Moreover, our analysis of the mutant LKB1^K78I provides a feasible explanation for the discordance of our results with the elegant study performed by Huang et al. (2008). Using a whole-body hypomorphic Lkb1 mutant mouse combined with systemic Pten heterozygous deletion, they observed that there was Lkb1 dose-dependent acceleration of various cancerous lesions. However, Pten^+/−;Lkb1^Hy/Hy mice did not exhibit alterations in prostate pathology compared with Pten^+/−;Lkb1^+/+. Our report, together with past results from our group and others, suggests that classic genetic dosage distributions or definitions such as haploinsufficiency are unable to correctly define the extent of their role in cancer.

We put forward the concept of tumor suppressive “competence” as a defining feature of tumor suppressors, taking PTEN and LKB1 as illustrative genes (Fig. 4 H). We define standard competence as a continuum of reduction in tumor suppression capacity along with the dosage, which is recapitulated in the prostate phenotype of Pten mutant mice (Nardella et al., 2010; Chen et al., 2005; Di Cristofano et al., 1998; Trotman et al., 2003). We define low competence as a reduction of <50% when a tumor suppressor gene dose results in cancerous lesions, as reported in the mammary tissue with Pten mutants expressing 75% of Pten (Alimonti et al., 2010) and at lower gene doses by others (Stambolic et al., 2000; Li et al., 2002). The integration of our results with the study of the group of Alessi (Huang et al., 2008) justifies the definition of LKB1 as a prostate tumor suppressor with high competence. In this context, reduction of the gene dosage of 50% (heterozygous) or 70% (hypomorphic) is inconsequential for PCa development. Moreover, retention of LKB1 activity as low as 10% is sufficient for the kinase to exert tumor suppression in vitro and in vivo. Only complete deletion of this gene is associated with the pathogenesis of PCa, a genetic condition that turns a premalignant disease (PIN, typical of Pten^+/− mice) into lethal metastatic PCa. It is worth noting that the introduction of competence in the definition of tumor suppressors remains a conceptual implementation, based on the integration of results derived from conditional and germline mouse models of two exemplifying genes, PTEN and LKB1. Further research is warranted to explore the relevance of this concept for other tumor suppressors, but it could harbor therapeutic relevance, since it would help establish the extent of pathway reactivation required for targeted therapies to show clinical effectiveness.

The conditional tissue-specific phosphatase and tensin homologue (Pten) knockout (C57/BL6/129sv; Pb-Cre4; Pten lox/lox) model was kindly provided by Pier Paolo Pandolfi (Beth Israel Deaconess Cancer Center, Boston, MA). The conditional tissue-specific Lkb1 null homogeneous background model (FVB;Lkb1lox/lox) was from Mouse Models of Human Cancer Consortium in a pure FVB background. First, we backcrossed this line into C57BL/6J for four generations to obtain animals with genetic background enriched in C57/BL6 (>90% C57/BL6). Next we crossed Lkb1lox/lox mice with Pb-cre4 and Pten lox/lox mice for at least four generations to obtain a founder colony with mixed homogeneous background, and experimental animals were generated. Littermates where analyzed when possible. However, due to the complexity of the allele combination, the different genotypes of interest with Pten and/or Lkb1 alteration were obtained from parallel contemporary breeding pairs. Probasin Cre was always retained in male mice, since in females Pb-Cre4 expression in utero can lead to recombination in embryos during pregnancy. The Cre recombinase expression, under the control of androgen-dependent ARR2B probasin promoter (Pb-Cre4), allowed the deletion of Pten in the prostate epithelium at puberty. Prostate Pten/Lkb1-deleted male mice were termed pc^−/+ (heterozygous) or pc^−/− (homozygous knockout) for each gene.

Patients were identified from a cohort of men with advanced castrate-resistant PCa treated at the Royal Marsden National Health Service Foundation Trust. All patients had given written informed consent and were enrolled in institutional protocols approved by the Royal Marsden Hospital (London, UK) ethics review committee (reference no. 04/Q0801/60).

DU145 cells expressing WT LKB1 (LKB1^WT), LKB1 K78I (LKB1^K78I), and LKB1 D194A (LKB1^D194A) mutants were generated by retroviral gene transfer. pBABE-puro (Addgene plasmid 1764; RRID:Addgene_1764) was a gift from Hartmut Land (Imperial Cancer Research Fund, Lincoln's Inn Fields, London, UK), Jay Morgenstern, and Bob Weinberg (Whitehead Institute for Biomedical Research, Cambridge, MA; Morgenstern and Land, 1990). pBABE-FLAG-LKB1 (Addgene plasmid 8592; RRID:Addgene_8592) and pBABE-FLAG-KD LKB1 (Addgene plasmid 8593; RRID:Addgene_8593) were a gift from Lewis Cantley (Meyer Cancer Center, New York, NY; Shaw et al., 2004). pBabe-FLAG-LKB1-D194A was a gift from Dr. Bin Zheng (Columbia University, New York, NY).

Lentiviral shRNA constructs targeting MO25α (TRCN0000044254; TRCN0000044256) were purchased from Sigma-Aldrich, and a scramble shRNA (hairpin sequence: 5′-CCG​GCA​ACA​AGA​TGA​AGA​GCA​CCA​ACT​CGA​GTT​GGT​GCT​CTT​CAT​CTT​GTT​G-3′) was used as control. pLKO.1 puro was a gift from Bob Weinberg (Addgene plasmid 8453; RRID:Addgene_8453; Stewart et al., 2003).

HEK293FT cells were transfected with retroviral or lentiviral vectors following standard procedures, and viral supernatant was used to infect cells. Selection was done using puromycin (2 µg/ml; Sigma-Aldrich).

Human prostate carcinoma cell lines PC3, DU145, and LNCaP were purchased from Leibniz-Institut Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, who provided an authentication certificate. Human prostate cell line PWR1E and human prostate carcinoma cell lines 22RV1 and VCaP were purchased from ATCC. HEK293FT and C4-2 were generously provided by the laboratories of Rosa Barrio (CIC bioGUNE, Derio, Spain) and Pier Paolo Pandolfi, respectively. Cell lines were periodically subjected to microsatellite-based identity validation. The cell lines used in this study were not found in the database of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee and National Center for Biotechnology Information Biosample. All cell lines were routinely monitored for mycoplasma contamination and quarantined while treated if positive. DU145, PC3, VCaP, and 293FT cell lines were maintained in DMEM supplemented with 10% (vol/vol) FBS (Gibco) and 1% (vol/vol) penicillin-streptomycin (Gibco). LNCap, C4-2, and 22RV1 cell lines were maintained in RPMI supplemented with 10% (vol/vol) FBS and 1% (vol/vol) penicillin-streptomycin. PWR1E cell line was maintained in Keratinocyte Serum Free Medium (Gibco) supplemented with 0.05 mg/ml bovine pituitary extract (Gibco) and 5 ng/ml epidermal growth factor (Gibco).

1 × 10⁴ cells were plated in triplicate in 12-well dishes. Quantification was performed as reported (Torrano et al., 2016). Anchorage-independent growth was tested by the ability to grow in soft agar as described previously (Torrano et al., 2016). The cell migration and invasion assay was performed in a 24-well-size Transwell plate of 8-µm pores (BD Falcon 351185). The invasive growth assay was performed as referenced (Martín-Martín et al., 2018) using the hanging drop method.

A 6-well tissue culture dish was coated with the bottom agar layer (complete DMEM containing 0.6% agar; SeaKem LE agarose, Lonza) and stored at 4°C for ≥30 min to let the agar solidify. Cells (2,500 cells/well) were resuspended in a 0.3% low-melting-point agar (Agarose LM, Pronadisa, Conda) and plated on top of the bottom agar. The plates were incubated (37°C and 5% CO₂) for 4 wk for growth of colonies. Colonies were counted using Fiji software.

Cells (50,000 cells/well) were resuspended in final volume of 500 µl 0.1% FBS DMEM and seeded in the upper part of the chamber. The bottom part was filled with 1.4 ml of complete DMEM. Plates were maintained at 37°C and 5% CO₂ for 24 h. Nonmigrated cells from the upper surface of the membrane were carefully removed by rinsing with PBS and using a cotton swab. The membrane was fixed with 10% formalin (15 min at 4°C) and stained with crystal violet. Transwell invasion was performed using Matrigel-coated chambers (Corning BioCoat Matrigel Invasion Chamber). After an incubation period of 48 h (37°C and 5% CO₂), membrane washing and fixation were performed under the same conditions as in the migration assay. Both Transwell assays were evaluated by photomicrographs from five randomly chosen fields (100×) per membrane. The number of migrated or invaded cells was counted using Fiji.

28,000 cells were resuspended in a final volume of 500 µl DMEM with 6% methylcellulose (Sigma-Aldrich, M0387). Drops (25 µl/drop) were pipetted on the cover of a 100-mm culture plate. The inverted plates were incubated at 37°C and 5% CO₂ for 48 h. Once formed, spheroids were collected, resuspended in collagen I solution (1.7 mg/ml in DMEM), and seeded into 24-well plates that were maintained at 37°C and 5% CO₂ for 48 h. The area of the invading spheroids was measured using Fiji. Relative invasive growth was quantified as area difference on day 2 minus day 0.

All mouse experiments were performed following the ethical guidelines established by the Biosafety and Welfare Committee at the Center for Cooperative Research in Biosciences as previously described (Martín-Martín et al., 2018; Torrano et al., 2016). The procedures employed were performed following the recommendations from Association for Assessment and Accreditation of Laboratory Animal Care. DU145 cells (pBABE, LKB1^WT, LKB1^K78I, and LKB1^D194A; 4 × 10⁶ cells per site) were resuspended in a final volume of 100 µl cold PBS supplemented with glucose (5 µM) mixed with 50 µl of Matrigel (Corning) and injected subcutaneously in two flanks per 6–12-wk-old male athymic nude mouse. Measurement of tumor size was performed three times a week, and tumor volume was estimated using the following formula: volume = length × width² × 0.526. After euthanasia, tumors were weighed, and tissue was fresh frozen or fixed overnight in 10% neutral buffered formalin, embedded in paraffin, sectioned 3-µm thick, and dried. Proliferation was assessed in dewaxed and rehydrated slides using Ki67 antibody (MA5-14520, Thermo Fisher Scientific).

All animal studies were performed according with the institutional animal care and use committee of Institute for Research in Biomedicine, Barcelona, Spain. For these studies, male BALB/c nude mice (Harlan) were used (12 wk old). Before cell injection, mice were anaesthetized as previously described (Torrano et al., 2016). Using a 26-G needle, 500,000 cells were injected into the left cardiac ventricle of mice resuspended in 1× PBS. Luciferase activity from the bone legs was detected by IVIS imaging. Data were processed using Living Image software. Photon Flux was measured every week for 38–42 d and normalized to values obtained on day 0. Ex vivo tibia analysis was performed in those mice with evidence of bioluminescence signal above background and that survived for the entire experiment.

Tissue sample collection (prostate gland, lymph nodes, long bones from lower limbs, lungs, and liver) was performed at 18 mo of age (LKB1^pc+/+ and LKB1^pc−/− mice) or 10 mo of age (Pten^pc+/− LKB1^pc+/+, Pten^pc+/− LKB1^pc+/−, and Pten^pc+/− LKB1^pc−/−). Tissue samples were fixed overnight in 10% neutral buffered formalin, embedded in paraffin, sectioned 3-µm thick, and dried. Slides were dewaxed and rehydrated through a series of graded ethanol to water and stained with the required antibody and/or H&E.

Histological observations on H&E-stained tissues were performed using an Olympus DP73 digital camera. Prostatic lesions were histologically classified according to the criteria of the Consortium Prostate Pathology Committee (Shappell et al., 2004) and scored as 0, no lesion observed; 1, focal or multifocal low-grade PIN; 2, focal or multifocal high-grade PIN; 3, carcinoma affecting <50% of tissue; 4, carcinoma affecting >50% of tissue. SPCs originate from squamous metaplasia of neoplastic prostatic epithelial cells. Lungs metastasis were described as micrometastasis when neoplastic cell growth in a small focus or multifocal or severe metastasis when neoplastic cells progressed involving multiple or large lung regions. Lymph node lesions were scored as follows: 0, no lesion observed; or 1, metastasis.

A standard immunohistochemistry protocol was applied. Briefly, slides were deparaffinized, and antigen retrieval was performed in a pressure cooker in citric acid buffer, pH 6 (phospho-S6, p63) or EDTA, pH 8 (phospho-AMPK). Sections were then immersed in 3% hydrogen peroxide to block endogenous peroxidase activity and washed in PBS. Incubation with specific blocking serum for 30 min was followed by incubation with primary antibodies at 4°C overnight using the following concentrations: phospho-AMPK (CST, 2535, 1:50), phospho-S6 (CST, 4858, 1:500), p63 (Abcam, ab735, 1:100), and AR (Santa Cruz Biotechnology, N-20, 1:200). The next day, after washing in PBS, signal was amplified with avidin/biotin technology (Vectastain Elite ABC system) or ImmPress HRP IgG (Peroxidase) Polymer Detection Kit (Vector Laboratories) and visualized with ImmPact DABSubstrate (Vector Laboratories). Finally, tissues were counterstained with H&E.

CK5/6 (Ventana, 790-4554, ready to use, cytoplasmic staining), p63 (Ventana, 790-4509, ready to use, nuclear staining), and Ki67 (Ventana, 790-4286, ready to use, nuclear staining) stains were performed in automated immunostainers (BenchMark Ultra, Ventana Medical Systems) following routine methods. Tris-EDTA was used for antigen retrieval. Negative controls were slides not exposed to the primary antibody, and these were incubated in PBS and then processed under the same conditions as the test slides. The analysis was performed using a Nikon Eclipse 80i microscope.

High molecular weight CK immunohistochemistry was performed on the Bond RX automated platform (Leica Microsystems) using a mouse monoclonal anti–high molecular weight CK antibody (Abcam, ab776, clone 34βE12). Heat-induced epitope retrieval was achieved with Bond Epitope Retrieval Solution 2 (Leica Microsystems, AR9640) for 20 min at 100°C before incubation with primary antibody (1:100 dilution) for 15 min at room temperature. The reaction was visualized using the Bond Polymer Refine Detection system (DS9800, Leica Microsystems).

PTEN immunostaining and scoring criteria was done following the procedure in Ferraldeschi et al. (2015). Proliferative basal epithelium cells were identified by staining with Ki67 (Thermo Fisher Scientific, 1/200, 14-5698-82) and CK5 (BioLegend, 1/500, PRB-160P) primary antibodies. For quantification of double-positive cells, five independent sections from five independent mice were counted using ImageJ 1.52a (National Institutes of Health).

LKB1 kinase activity was measured as described previously (Lizcano et al., 2004). Cells stably overexpressing LKB1 WT, K78I, or D194A mutants (FLAG-tagged), were used to immunoprecipitate anti-FLAG-agarose resin, in triplicate. Reactions were performed in FLAG-agarose immunoprecipitates from 1.5 mg protein, using 200 µM NuaKtide as peptide and 100 µM ³²P-ATP as substrates in a 60-min reaction. 1 mU of LKB1 kinase activity is the amount that catalyzed the phosphorylation of 1 pmol of NuaKtide in 1 min in the standard assay at 30°C.

RNA was extracted using NucleoSpin RNA isolation kit (Macherey-Nagel, 740955.240C). For animal tissues, a Trizol-based implementation of the NucleoSpin RNA isolation kit protocol was used as reference (Ugalde-Olano et al., 2015). For all cases, 1 µg of total RNA was used for cDNA synthesis using Maxima H Minus cDNA Synthesis Master Mix (Thermo Fisher Scientific, M1682). qRT-PCR was performed as previously described (Carracedo et al., 2012). Universal Probe Library primers (Roche) and probes (Roche and Thermo Fisher Scientific) are detailed in Table S10. All qRT-PCR data presented were normalized using GAPDH/Gapdh (Applied Biosystems, Hs02758991_g1, Mm99999915_g1).

Western blot was performed as previously described (Carracedo et al., 2012). Briefly, cells were lysed in radioimmunoprecipitation assay buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.1% SDS, 1% Nonidet P40, 1% sodium deoxycholate, 1 mM sodium fluoride, 1 mM sodium orthovanadate, 1 mM β-glycerophosphate, and protease inhibitor cocktail; Roche). Antibodies used were phospho-AMPK substrate motif (CST 5759), phospho-ACC (CST 11818), phospho-AMPK (CST 2535), LKB1 (CST 3050), MO25α/CAB39 (CST 2716s), PTEN (CST 9559), phospho-Raptor (CST 2083), Raptor (CST 2280), β-actin (CST 3700S), and HSP90 (CST 4874). All antibodies were used at a 1:1,000 dilution except β-actin at 1:2,000. Mouse and rabbit secondary antibodies were purchased from Jackson Immunoresearch. After standard SDS-PAGE and Western blotting techniques, proteins were visualized using the enhanced chemiluminescence.

Cells were lysed in immunoprecipitation buffer (40 mM Tris-HCl, pH 7.5, 1 mM EDTA, 150 mM NaCl, 0.5% Triton X-100, 150 mM NaCl, 1 mM MgCl₂, 1 mM sodium fluoride, 1 mM sodium orthovanadate, 1 mM β-glycerophosphate, and protease inhibitor cocktail; Roche). An equal amount of each protein lysate was incubated with anti-FLAG M2 Affinity Gel (Sigma-Aldrich) for 1 h at 4°C. The immune complexes and protein lysates were analyzed by Western blot with anti-LKB1 and anti-MO25 antibody. The immune complexes were in parallel subjected to liquid chromatography/tandem mass spectrometry (LC–MS/MS) analysis.

For LC–MS/MS analysis, protein samples were boiled for 5 min and resolved in 12.5% acrylamide gels, using a Mini-Protean II electrophoresis cell (Bio-Rad). A constant voltage of 150 V was applied for 45 min. Gels were fixed in a solution containing 10% acetic acid and 40% ethanol for 30 min and stained overnight in Sypro Ruby (Bio-Rad). Gels were then washed in a solution containing 10% ethanol and 7% acetic acid for 30 min, and the image was acquired using a Typhoon Trio scanner (GE Healthcare). Each lane was cut in slices and subjected to tryptic digestion, followed by LC–MS/MS analysis.

Gel bands were sliced into three small pieces, which were washed in milli-Q water. Reduction and alkylation were performed using 10 mM dithiothreitol in 50 mM ammonium bicarbonate at 56°C for 20 min, followed by iodoacetamide (50 mM iodoacetamide in 50 mM ammonium bicarbonate) for another 20 min in the dark. Gel pieces were dried and incubated with trypsin (12.5 µg/ml in 50 mM ammonium bicarbonate) for 20 min on ice. After rehydration, the trypsin supernatant was discarded. Gel pieces were hydrated with 50 mM ammonium bicarbonate and incubated overnight at 37°C. After digestion, acidic peptides were cleaned with trifluoroacetic acid 0.1% and dried in an RVC2 25 speedvac concentrator (Martin Christ). Peptides were resuspended in 10 µl of 0.1% fluoroacetic acid and sonicated for 5 min before analysis.

LC was performed using an NanoAcquity nano-HPLC (Waters), equipped with a Waters BEH C18 nano-column (200 mm × 75 µm; internal diameter, 1.8 µm), A chromatographic ramp of 30 min (5–60% ACN) was used with a flow rate of 300 nl/min. Mobile phase A was water containing 0.1% vol/vol formic acid, and mobile phase B was ACN containing 0.1% vol/vol formic acid. A lock mass compound [Glu1]-fibrinopeptide B (100 fmol/µl) was delivered by an auxiliary pump of the LC system at 500 nl/min to the reference sprayer of the NanoLockSpray (Waters) source of the mass spectrometer.

LC was connected online to a Synapt G2Si electrospray ionization Q-Mobility-TOF mass spectrometer (Waters) equipped with an ion mobility chamber (T-Wave-IMS) for high-definition data acquisition analyses. All analyses were performed in positive-mode electrospray ionization. Data were postacquisition lock mass corrected using the double-charged monoisotopic ion of [Glu1]-fibrinopeptide B. Accurate mass LC-MS data were collected in hexadehydro-Diels–Alder mode, which enhances signal intensities using the ion mobility separation step.

Database searching was performed using Mascot 2.2.07 (Matrixscience) against a Uniprot-Swissprot database filled only with entries corresponding to Homo sapiens (without isoforms). For protein identification, the following parameters were adopted: carbamidomethylation of cysteines (C) as fixed modification, oxidation of methionines (M) as variable modification, 15 ppm of peptide mass tolerance, 0.2 D fragment mass tolerance, ≤3 missed cleavage points, and peptide charges of +2 and +3.

Progenesis LC-MS (v2.0.5556.29015, Nonlinear Dynamics) was used for the label-free differential protein expression analysis. One of the runs was used as the reference to which the precursor masses in all other samples were aligned. Only features comprising charges of 2+ and 3+ were selected. The raw abundances of each feature were automatically normalized and logarithmized against the reference run. Samples were grouped in accordance to the comparison being performed, and ANOVA was performed. A peak list containing the information of all the features was generated and exported to the Mascot search engine (Matrix Science). This file was searched against a Uniprot/Swissprot database under the conditions stated in the previous section, and the list of identified peptides was imported back to Progenesis LC-MS. Protein quantitation was performed based on the three most intense nonconflicting peptides (peptides occurring in only one protein), except for proteins with only two nonconflicting peptides. The significance of expression changes was tested at the protein level, and proteins identified with at least two peptides, an ANOVA P value ≤0.05, and a ratio >1.5 in either direction were selected for further analyses.

Gene expression analysis of tumors from different mouse models (the original Pten^pc−/− mouse model [Chen et al., 2005] versus the Pten^pc+/− Lkb1^pc−/− generated in this study) was performed using GeneChip Mouse Transcriptome Array 1.0 (Affymetrix). Differential expression was obtained using Transcriptome Analysis Console v4.0. Expression data were normalized and background- and batch-corrected using the Signal Space Transformation-Robust Multi-Chip Analysis implemented in that software. Data are deposited in GEO (GSE133837).

Total RNA was extracted from formalin-fixed and paraffin-embedded (FFPE) tissue using RNAstorm FFPE RNA extraction kit (Cell Data Sciences) according to the manufacturer’s instructions. cDNAs from a total of 12 ng RNA were generated, fragmented, biotinylated, and hybridized to the GeneChip Mouse Transcriptome Array 1.0. The arrays were washed and stained on a GeneChip Fluidics Station 450 (Affymetrix); scanning was performed with the GeneChip Scanner 3000 7G and image analysis with the Affymetrix GeneChip Command Console Scan Control. Gene set enrichment analysis was performed using the MSigDB database (Subramanian et al., 2005).

DNA was extracted from FFPE tumor blocks positive for tumor content using the FFPE Tissue DNA kit (Qiagen). DNA was quantified with the Quant-iT high-sensitivity PicoGreen double-stranded DNA Assay Kit (Invitrogen), and quality control was performed with the Illumina FFPE quality control kit (WG-321-1001). Libraries for next-generation targeted sequencing were constructed from 40 ng of DNA using a customized panel (Generead DNaseq Mix-n-Match Panel v2; Qiagen) as previously described (Mateo et al., 2015). CNVs were assessed from the aligned sequence files (.bam) using CNVkit v0.7.3 (Seed et al., 2017; Talevich et al., 2016; Turajlic et al., 2018).

No statistical method was used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. n values represent the number of independent experiments performed, the number of individual mice, or the number of patient specimens. Unless otherwise stated, data analyzed by parametric tests are represented by the mean ± SEM of pooled experiments and the median ± interquartile range for experiments analyzed by nonparametric tests. For each independent in vitro experiment, at least three technical replicates were used, and a minimum number of three experiments was performed to ensure adequate statistical power. In the in vitro experiments, normal distribution was assumed, and one-sample t tests were applied for one-component comparisons with control and Student’s t test for two-component comparisons. For in vivo experiments, D’Agostino–Pearson, Shapiro–Wilk, and Kolmogorov–Smirnov normality tests were applied. Student’s t test was used to compare data with normal distribution, and nonparametric Mann–Whitney U exact test was used to analyze samples not following a normal distribution. The confidence level used for all the statistical analyses was 95% (α = 0.05). Two-tailed statistical analysis was applied for experimental design without predicted result, and one-tailed for validation or hypothesis-driven experiments. GraphPad Prism 8.0.2 software was used for statistical calculations.

We are grateful to the Carracedo laboratory for valuable input and to Diego Barriales, Elisabet Megias-Roda, Juan Anguita, David Barneda, Len Stephens, Ana M. Aransay, and Felix Elortza for technical advice.

I. Hermanova and F. Salvador were funded by the Juan de la Cierva program from the Spanish Ministry of Research. V. Torrano is funded by Fundación Vasca de Innovación e Investigación Sanitarias, Bioef (BIO15/CA/052), the Ministerio de Ciencia, Innovación y Universidades (RTI2018-097267-B-100), the Fundación Científica Asociación Española Contra el Cáncer J.P. Bizkaia, and the Basque Department of Health (2016111109). The work of A. Carracedo is supported by the Basque Department of Industry, Tourism and Trade (Elkartek), the Department of Education (IKERTALDE IT1106-16, with A. Gomez-Muñoz), the Department of Health (2019222031), the Fundación BBVA, the Ministerio de Ciencia, Innovación y Universidades (SAF2016-79381-R, FEDER [European Regional Development Fund, EU]; Severo Ochoa Excellence Accreditation SEV-2016-0644; Excellence Networks SAF2016-81975-REDT), European Training Networks Project (H2020-MSCA-ITN-308 2016 721532, with G. Velasco), the Fundación Científica Asociación Española Contra el Cáncer (IDEAS175CARR; GCTRA18006CARR, with M. Graupera and R.R. Gomis), “La Caixa” Foundation (HR17-00094), and the European Research Council (Starting Grant 336343, PoC 754627, and Consolidator Grant 819242). We are grateful for the support of Mondravember and Movembergara. J.M. Paramio was funded by Ministerio de Ciencia, Innovación y Universidades (SAF2015-66015-R and CB16/12/00228). J.I. López was funded by the Ministerio de Ciencia, Innovación y Universidades (SAF2016-79847-R). R. Barrio and J.D. Sutherland acknowledge Ministerio de Ciencia, Innovación y Universidades/European Regional Development Fund (BFU2017-84653-P, SEV-2016-0644). G. Velasco’s research is supported by grants integrated into the State Plan for R & D + I2013-2016 and funded by the Instituto de Salud Carlos III (PI18/00442 and PI15/00339), the European Regional Development Fund, and Fundació la Marató de TV3 (20134031). J.M. Lizcano was supported by the Ministerio de Ciencia, Innovación y Universidades (SAF2015-64237-R) and cofunded by the European Regional Development Fund. R.R. Gomis is supported by the Ministerio de Ciencia, Innovación y Universidades, European Regional Development Fund (CIBERONC and SAF016-76008R), and “La Caixa” Foundation (HR17-00092). CIBERONC was cofunded with European Regional Development funds and funded by Instituto de Salud Carlos III. L. Valcarcel-Jimenez was funded by a Basque Government predoctoral grant.

Author contributions: In vitro studies: I. Hermanova, P. Zúñiga-García, and A. Caro-Maldonado, with support from S. Fernandez-Rui, N. Martín-Martín, A. Arruabarrena-Aristorena, L. Valcarcel-Jimenez, and A. Talamillo. In vivo studies: I. Hermanova, P. Zúñiga-García, A. Caro-Maldonado, and M. Nuñez-Olle, with support from N. Martín-Martín, A. Zabala-Letona, A. Carracedo, M. Lorente, M. Guiu, and V. Torrano. In vivo bioluminescence assays: F. Salvador and R.R. Gomis. Genotyping and cloning: I. Astobiza and S. Fernandez-Rui, with support from J.D. Sutherland and R. Barrio. Pathologic and immunohistochemical analysis of murine tissues: J.M. Flores and J.I. López, with participation of I. Hermanova, I. Lodewijk, C. Suárez-Cabrera, S. Fernandez-Rui, and L. Camacho for immunohistochemistry. Pathologic, genetic, and immunohistochemical analysis of human PCa specimens: B. Gurel and S. Carreira, supervised by J.S. de Bono. Immunofluorescence analysis of tissues: A. Martinez-Romero, supervised by M. Graupera. LKB1 kinase radiochemical assay: J.M. Lizcano. Bioinformatics and statistical analysis: A.R. Cortazar. Murine tissue microdissection and gene expression analysis: C. Suárez-Cabrera, I. Lodewijk, and J.M. Paramio. Other contributions: G. Velasco (supervision in xenograft experiments), A. Gomez-Muñoz (shared supervision of L. Camacho), J. Trka (shared supervision of I. Hermanova), V. Torrano (support with human cancer dataset analysis). General supervision of data analysis, project direction, and manuscript writing: A. Carracedo.