NKX3.1 Localization to Mitochondria Suppresses Prostate Cancer Initiation

Prostate cancer is the most prevalent type of cancer and a leading cause of cancer-related death in men (1, 2). These days, most new prostate tumors are diagnosed with early-stage disease, which is usually not life-threatening (5-year survival of >90%; refs. 3, 4). Consequently, it has become common practice to monitor men by “active surveillance,” which avoids treatment unless there is evidence of progression. Although active surveillance minimizes overtreatment, it may miss the opportunity for early intervention of tumors that are likely to progress to advanced prostate cancer, which, unlike locally invasive disease, is often lethal (5-year survival of <30%; refs. 3, 4).

A distinguishing feature of prostate cancer is its long latency, because preinvasive lesions, termed prostatic intraepithelial neoplasia (PIN), are common in young adults, whereas clinically detectable disease is usually not manifested until the sixth decade or later (2, 5, 6). Risk factors for prostate cancer include aging and inflammation, and incidence and aggressiveness are influenced by genetic factors and race (2, 6, 7). Given this long latency, men diagnosed with or at risk for prostate cancer may benefit from early intervention to delay progression, which underscores the importance of elucidating mechanisms of cancer initiation that may provide opportunities for precision prevention (8, 9).

A key factor associated with prostate carcinogenesis is oxidative stress (10–12), which leads to excessive production and accumulation of reactive oxygen species (ROS; refs. 13–15). Most cellular ROS is produced by mitochondria as a by-product of cellular respiration (15–17). In normal cellular contexts, ROS levels and oxidative phosphorylation (OXPHOS) are tightly balanced, which is important for mitochondrial homeostasis. However, aberrant oxidative stress may lead to excess ROS and thereby impaired mitochondrial function (13, 14). The electron transport chain (ETC), which generates mitochondrial ROS, is located on the inner mitochondrial membrane and is comprised of several protein complexes whose subunits are encoded by both nuclear and mitochondrial genes (18). Consequently, communication between the nucleus and mitochondria is essential for sustaining ETC function and appropriate ROS levels in response to oxidative stress (19).

Although the role of ROS in cancer is complex (13), substantial evidence indicates that the prostate is particularly susceptible to elevated ROS as a consequence of oxidative stress associated with inflammation, aging, or other factors (11, 12, 20). Furthermore, alterations of mitochondrial function or mutational burden, which may affect ROS levels or response to oxidative stress, are prevalent in prostate cancer and associated with adverse disease outcome (21–25). Among genes that have been attributed to protect against oxidative stress in prostate is NKX3.1, a prostate-specific homeobox gene that regulates prostate differentiation and suppresses prostate cancer initiation (26). In mice, loss of function of Nkx3.1 leads to PIN, which, in collaboration with loss of other tumor suppressor genes, progresses to adenocarcinoma (27–30). In humans, allelic imbalance of NKX3.1 is among the most prevalent and earliest genetic alterations in prostate cancer, whereas reduced expression of NKX3.1 is associated with increased risk of disease aggressiveness, particularly among African American men (31–37).

NKX3.1 is expressed in the prostatic epithelium, where it functions to safeguard prostatic specification and maintenance of luminal prostatic stem cells and to protect against DNA damage and inflammation (38–46). Because NKX3.1 encodes a homeoprotein, which are known regulators of nuclear gene transcription, its functions in prostate have long been attributed to regulation of nuclear gene expression. Here, we uncover a nonnuclear role for NKX3.1 in mitochondria, where it regulates expression of mitochondrial genes to promote mitochondrial homeostasis and suppress tumorigenesis. Notably, we find that NKX3.1 expression levels and subcellular localization are associated with clinical outcome in patients with prostate cancer. We propose that analyses of NKX3.1 expression and localization may enhance precision monitoring of patients with prostate cancer to identify those at risk of progression.

On the basis of our previous study showing that loss-of-function of Nkx3.1 in mouse prostate leads to increased expression of genes associated with oxidative stress (40), we performed RNA sequencing comparing prostate tissues from Nkx3.1 wild-type (Nkx3.1^+/+) or mutant (Nkx3.1^−/−) mice (27) with human RWPE1 prostate cells expressing or lacking exogenous NKX3.1 (39) in normal contexts or in response to oxidative stress (Supplementary Datasets S1 and S2). We found that NKX3.1-dependent biological pathways are highly conserved in mouse and human prostate, particularly following oxidative stress (Supplementary Figs. S1A and B; Supplementary Datasets S3 and S4). Conserved pathways include those associated with regulation of oxidative stress, mitochondrial function, and glucose metabolism (Supplementary Figs. S1C and S1D). Therefore, we have investigated the functions of NKX3.1 in response to oxidative stress in both mouse models and human prostate cells.

First, we studied the consequences of oxidative stress for prostate carcinogenesis in Nkx3.1-mutant mice, which develop PIN indicative of preinvasive prostate cancer by 12 months of age (Fig. 1A; Supplementary Table S1; refs. 27, 28). We treated adult Nkx3.1^+/+ or Nkx3.1^−/− mice continually for 1 to 9 months with paraquat (N,N′-dimethyl-4,4′-bipyridinium dichloride [(C₆H₇N)₂]Cl₂), a redox-active compound that produces superoxide anions (ref. 47; Fig. 1A–H; Supplementary Figs. S2A–E). As a control, we analyzed seminal vesicle (Supplementary Figs. S2F–J; Supplementary Table S1), which, like prostate, is an androgen-regulated male secondary sexual organ, but unlike prostate does not express Nkx3.1 and rarely develops cancer (27, 39).

After 1 month (i.e., in 4-month-old mice), paraquat-treated Nkx3.1^−/− prostates displayed increased ROS, which was not observed in Nkx3.1^+/+ prostates or seminal vesicles of either genotype (Fig. 1B; Supplementary Fig. S2A, S2B, and S2G). Furthermore, after 9 months (i.e., in 12-month-old mice), we observed a significant acceleration of PIN in all prostatic lobes of Nkx3.1^−/− mice, but not in any prostatic lobes of Nkx3.1^+/+ mice or seminal vesicles from either genotype (Fig. 1C and D; Supplementary Figs. S2C and S2F). Acceleration of prostate carcinogenesis was evident by the significant increase in high-grade PIN (PIN3/4) compared with low-grade PIN (PIN1/2) in paraquat-treated versus control Nkx3.1^−/− prostate (Fig. 1D; Supplementary Table S1). This was accompanied by increased DNA damage, as evident by immunostaining for γH2AX (Fig. 1C). These phenotypes were not due to general paraquat toxicity, because no other major organs were grossly affected by long-term paraquat treatment (Supplementary Fig. S3), consistent with previous studies showing that prostate is particularly susceptible to stressful stimuli (46).

To find out whether oxidative stress resulted in abnormalities of mitochondrial morphology, we performed electron microscopy (EM). These analyses revealed a significant reduction in number and area of mitochondria in the paraquat-treated Nkx3.1^−/− prostate, but not the Nkx3.1^+/+ prostate or seminal vesicle (Fig. 1E–G; Supplementary Fig. S2D, S2H, and S2I; Supplementary Table S2). Aberrations of mitochondrial morphology included condensation, lysis, and membrane dissolution (Fig. 1E). We also observed a significant reduction in active mitochondria, specifically in the paraquat-treated Nkx3.1^−/− prostate, evidenced by measuring mitochondrial membrane potential using tetramethylrhodamine, ethyl ester (TMRE; Fig. 1H; Supplementary Fig. S2E and S2J). Therefore, NKX3.1 protection against oxidative stress in vivo is associated with suppression of cancer initiation and protection of mitochondrial integrity.

We next investigated the functions of NKX3.1 for protection against oxidative stress in human prostate cells in vitro (Fig. 1I–W; Supplementary Figs. S4A–S4K and S5A–S5F; Supplementary Tables S3 and S4). In particular, we silenced endogenous NKX3.1 in LNCaP cells (Fig. 1J), one of the few human prostate cancer cell lines that expresses endogenous NKX3.1 (41), and in BPH1 cells (Supplementary Fig. S5A), which are immortalized benign prostatic hyperplasia cells that express endogenous NKX3.1 (48). Alternatively, we expressed exogenous NKX3.1 in RWPE1 cells (Fig. 1K), an immortalized human prostate epithelial cell line with negligible levels of endogenous NKX3.1 (39). We induced oxidative stress by treating cells with paraquat and measuring accumulation of cellular superoxide (O⁻₂) or hydrogen peroxide (H₂O₂) ROS, using dihydroethidium (DHE) or 6-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate, acetyl ester (CM-H2DCFDA), respectively (Fig. 1I).

In these cancerous and benign contexts, cells expressing or lacking NKX3.1 had similar ROS levels in standard culture conditions (i.e., with vehicle), whereas oxidative stress was associated with increased ROS in cells lacking but not those expressing NKX3.1 (Fig. 1N–W; Supplementary Figs. S4A–S4K and S5A–S5F). This was accompanied by increased DNA damage as evident by expression of γH2AX (Fig. 1L and M). In particular, treatment of NKX3.1-silenced LNCaP or BPH1 cells with paraquat significantly increased ROS, which was not observed in the corresponding control cells (Fig. 1N and O; Supplementary Figs. S4B and S5B and S5C). Notably, the effects of silencing NKX3.1 in LNCaP cells were rescued by exogenous NKX3.1 (Supplementary Figs. S4C–E). Conversely, paraquat treatment resulted in increased ROS in control RWPE1 cells (which lack NKX3.1), but not in the NKX3.1-expressing cells (Fig. 1S and T; Supplementary Fig. S4H). Similar results were obtained by inducing oxidative stress with hydrogen peroxide or hypoxia (Fig. 1O and T; Supplementary Figs. S4B, S4H, and S5C).

Given the Nkx3.1-dependent effects of oxidative stress on mitochondria in vivo, we asked whether NKX3.1 protects specifically against mitochondrial ROS (compared with general cellular ROS) by measuring mitochondrial-specific superoxide (O⁻₂) or hydrogen peroxide (H₂O₂) using MitoSOX or MitoPY1, respectively (Fig. 1I). As observed for general cellular ROS, mitochondrial-specific ROS was significantly increased following paraquat treatment in NKX3.1-silenced LNCaP and BPH1 cells, but not the corresponding control cells (which express NKX3.1; Fig. 1P; Supplementary Fig. S5D). Conversely, oxidative stress led to increased mitochondrial ROS in control but not NKX3.1-expressing RWPE1 cells (Fig. 1U).

Induction of oxidative stress with MitoParaquat, which has a mode of action similar to paraquat but induces oxidative stress selectively in mitochondria (49), resulted in increased ROS in cells lacking NKX3.1 (i.e., in NKX3.1-silenced BPH1 cells and control RWPE1 cells), which was fully abrogated in the corresponding NKX3.1-expressing cells (i.e., control BPH1- and NKX3.1-expressing RWPE1 cells; Supplementary Figs. S4I, S5E, and S5F). Furthermore, increased ROS in cells lacking NKX3.1 was rescued by the mitochondrial-specific antioxidant MitoQ (ref. 50; Fig. 1I; Supplementary Figs. S4F and S4J), further demonstrating that NKX3.1 protects primarily against mitochondrial ROS. In addition, oxidative stress resulted in the significant reduction of active mitochondria in cells lacking NKX3.1, as evident by measurement of mitochondrial mass and/or membrane potential (Fig. 1Q, R, V, and W; Supplementary Fig. S4G and S4K). These findings demonstrate that NKX3.1 protects against mitochondrial ROS, which is associated with protection of mitochondrial integrity.

Although Nkx3.1 displayed the expected nuclear localization in vehicle-treated mouse prostate, its subcellular localization in paraquat-treated prostate did not appear exclusively nuclear (Fig. 1C, inset). This prompted us to investigate the subcellular localization of NKX3.1 in response to oxidative stress. Under standard conditions (i.e., in the vehicle-treated contexts) NKX3.1 was localized primarily to nuclei in mouse prostate tissues and human prostate cells; in contrast, following oxidative stress, NKX3.1 became localized to mitochondria as well as nuclei (Fig. 2A–D; Supplementary Figs. S5G, S5H, and S6A). This was evident by colocalization of NKX3.1 with the mitochondrial-specific marker ATP synthase subunit beta (ATPB) as visualized by confocal imaging (Fig. 2A and C; Supplementary Fig. S5G), and Western blot analysis after biochemical fractionation to isolate nuclei or mitochondria (Fig. 2B and D; Supplementary Figs. S5H and S6A).

To more precisely study localization of NKX3.1 in mitochondria, we performed immuno-EM on LNCaP and RWPE1 cells (Fig. 2E). This revealed that, following oxidative stress, NKX3.1 was localized within the interior of mitochondria, where it is associated with the cristae of the inner mitochondrial membrane (Fig. 2E). Localization within the interior of mitochondria was further supported by biochemical fractionation analyses, in which NKX3.1 was resistant to proteolysis in intact mitochondria but not after disruption by sonication (Supplementary Fig. S6B).

To gain insights regarding the relevance of these findings for NKX3.1-related cancer risk, we examined the localization and activities of two germline polymorphisms encoding NKX3.1^T164A and NKX3.1^R52C (Fig. 2F–L; Supplementary Fig. S6C–S6I; Supplementary Table S5; refs. 51, 52). NKX3.1^T164A occurs in 14% of the population and is associated with early-onset prostate cancer (52), whereas NKX3.1^R52C is present in 11% of the population and is associated with increased cancer aggressiveness (51). NKX3.1^T164A corresponds to a threonine-to-alanine substitution in the homeodomain that renders the protein inactive for DNA binding (39), whereas NKX3.1^R52C corresponds to an arginine-to-cysteine substitution in the N-terminal region that does not impair DNA binding (Fig. 2F; Supplementary Fig. S6C; ref. 51).

NKX3.1^T164A and NKX3.1^R52C were located only in the nucleus of RWPE1 cells in standard culture conditions, as evident by confocal imaging and Western blot analysis; however, following oxidative stress, NKX3.1^T164A was localized to mitochondria as well as nuclei, whereas NKX3.1^R52C was only in the nucleus (Fig. 2G; Supplementary Fig. S6D). This deficiency of mitochondrial localization of NKX3.1^R52C was specific for the arginine-to-cysteine substitution, because a protein with a substitution of arginine-to-alanine (NKX3.1^R52A) localized to mitochondria following oxidative stress (Fig. 2G).

Interestingly, neither NKX3.1^T164A nor NKX3.1^R52C reduced general cellular or mitochondrial-specific ROS (Fig. 2H and I; Supplementary Fig. S6E–S6I), suggesting that multiple regions of NKX3.1 are required for protection against ROS. Indeed, only full-length NKX3.1, but not various truncated proteins, localized to mitochondria and alleviated ROS in response to oxidative stress (Supplementary Fig. S7A–S7E).

To further study their localization and functions and to assess the relevance for tumorigenicity, we asked whether NKX3.1^T164A or NKX3.1^R52C could rescue the consequences of silencing NKX3.1 in LNCaP cells, as we had observed for native NKX3.1 (see Supplementary Figs. S4C–E). As in RWPE1 cells, exogenous NKX3.1, NKX3.1^T164A, and NKX3.1^R52C were located in nuclei in NKX3.1-silenced LNCaP cells in standard conditions, while following oxidative stress, native NKX3.1, and NKX3.1^T164A, but not NKX3.1^R52C, localized to mitochondria (Fig. 2J). In addition, only native NKX3.1, but neither NKX3.1^T164A nor NKX3.1^R52C, could alleviate ROS in response to oxidative stress in NKX3.1-silenced LNCaP cells (Figs. 2K and L). Notably, oxidative stress resulted in increased cellular proliferation, colony formation, and invasion in NKX3.1-silenced LNCaP cells, which was abrogated by native NKX3.1, but not NKX3.1^T164A or NKX3.1^R52C (Fig. 2M–P).

Together, these findings demonstrate that protection from mitochondrial ROS and suppression of tumorigenicity in response to oxidative stress require NKX3.1 binding to DNA, which is impaired for NKX3.1^T164A, and localization to mitochondria, which is impaired for NKX3.1^R52C. Furthermore, these observations raise the possibility that increased risk of aggressive prostate cancer associated with these germline variants of NKX3.1 is due, at least in part, to a loss of protection of mitochondria from oxidative stress.

We consistently observed that localization of NKX3.1 to mitochondria was not accompanied by its reduced nuclear expression; rather, relative levels of nuclear NKX3.1 were similar in standard conditions and following oxidative stress (Fig. 2B and D; Supplementary Figs. S5H and S6A). Time-course analysis revealed localization of NKX3.1 to mitochondria beginning 6 hours after oxidative stress and peaking at 24 hours, during which time the levels of nuclear NKX3.1 remained relatively constant (Supplementary Fig. S8A); this paralleled the time course of protection from oxidative stress (Supplementary Fig. S8B; Supplementary Table S6). Notably, mitochondrial localization and protection from oxidative stress required new protein synthesis, because they were blocked by cycloheximide (Supplementary Figs. S8A and B). Furthermore, we found that localization of NKX3.1 to mitochondria, as well as protection from oxidative stress, were not affected by Leptomycin B, an inhibitor of nuclear export, but were blocked by Valinomycin, an inhibitor of mitochondrial import (Supplementary Fig. S9A and S9B; Supplementary Table S6). These findings suggest that mitochondrial NKX3.1 corresponds to newly synthesized protein imported to mitochondria directly rather than first exported from the nucleus.

Interrogation of the primary sequence of NKX3.1 using the MitoMiner resource tool (53) did not reveal an evident mitochondrial localization domain; however, the lack of a three-dimensional structure for the full-length NKX3.1 protein may preclude identification of such domain. Indeed, our findings showing that only full-length NKX3.1, but not various truncated proteins, is localized to mitochondria (Supplementary Fig. S7B) suggest that its tertiary structure is necessary for import.

To gain insight regarding how NKX3.1 is imported to mitochondria, we performed mass spectrometry to identify proteins associated with NKX3.1 in mitochondria after oxidative stress (Fig. 3A and B; Supplementary Dataset S5). Among these, we focused on HSP family A member 9 (HSPA9, also known as Mortalin), a chaperone protein known to import proteins to mitochondria (54) that has various functions in cancer (55–57). Notably, HSPA9 is a nonessential gene, making it feasible to study the consequences of its silencing for NKX3.1 localization to mitochondria and protection from oxidative stress.

Coimmunoprecipitation assays performed using NKX3.1-expressing or control RWPE1 cells demonstrated association of native NKX3.1 with HSPA9 in mitochondria after oxidative stress (Fig. 3B and C). Notably, NKX3.1^T164A, but not NKX3.1^R52C, interacts with HSPA9 in mitochondria (Fig. 3C), consistent with our findings that NKX3.1^R52C does not localize to mitochondria. Furthermore, shRNA-mediated silencing of HSPA9 in NKX3.1-expressing RWPE1 cells abolished localization of NKX3.1 to mitochondria and protection from oxidative stress (Fig. 3D and E). We observed similar findings after HSPA9 silencing in LNCaP cells, as evident by Western blot and immunofluorescence analyses (Fig. 3F–I). Moreover, HSPA9 silencing in LNCaP cells impaired the ability of NKX3.1 to protect from oxidative stress–induced acceleration of tumorigenicity in vitro (Fig. 3J–M). These findings demonstrate that import of NKX3.1 to mitochondria by HSPA9 is necessary for protection from ROS and suppression of tumorigenicity.

The mitochondrial genome encodes 37 genes, 13 of which are essential subunits of the ETC complexes (Fig. 4A–C), which are necessary for OXPHOS and generate ROS as a byproduct (15–18). Strikingly, in response to oxidative stress, all 13 mitochondrial-encoded ETC genes were upregulated in NKX3.1-expressing RWPE1 cells, whereas none were upregulated in cells expressing either NKX3.1^T164A or NKX3.1^R52C (Fig. 4B–D; Supplementary Dataset S2). Similar results were observed in LNCaP cells, in which reduced expression of ETC genes in NKX3.1-silenced cells after oxidative stress could be rescued by native NKX3.1, but not NKX3.1^T164A or NKX3.1^R52C, and also in Nkx3.1^−/− versus Nkx3.1^+/+ mouse prostate tissues (Supplementary Figs. S10A and S10B; Supplementary Dataset S1). Furthermore, NKX3.1-dependent upregulation of mitochondrial-encoded ETC genes was abrogated by HSPA9 silencing and rescued treatment with MitoQ (Supplementary Fig. S10C–S10E).

In considering mechanisms by which NKX3.1 upregulates mitochondrial ETC genes, we noticed that the D-loop, which is the regulatory region in the mitochondrial genome that controls expression of these genes (18), has a perfect NKX3.1 consensus DNA-binding sequence, namely TAAGTA (Fig. 4E; ref. 58). In addition, transcription of ETC genes is necessary for protection from oxidative stress by NKX3.1 because balapiravir, an inhibitor of mitochondrial-specific RNA polymerase (59), abrogated NKX3.1-dependent protection from oxidative stress (Supplementary Fig. S11A and S11B).

Indeed, using chromatin immunoprecipitation followed by quantitative PCR (ChIP-qPCR), we found that, in response to oxidative stress, NKX3.1 binds specifically to the D-loop in the mitochondrial genome (Fig. 4F; Supplementary S12A). Furthermore, mouse Nkx3.1 bound to the corresponding site in the D-loop of the mouse mitochondrial genome (Supplementary S12B). Interaction of NKX3.1 with the D-loop was dependent on binding to DNA and localization to mitochondria, because neither NKX3.1^R52C nor NKX3.1^T164A bound to the D-loop, and because binding was abrogated by silencing HSPA9 (Fig. 4F; Supplementary S12C). Therefore, in response to oxidative stress, NKX3.1 regulates transcription of mitochondrial ETC genes, which requires import to mitochondria and binding to a regulatory region of the mitochondrial genome.

The mitochondrial-encoded genes regulated by NKX3.1 are essential subunits of the ETC complexes (ETC complexes I–V), which act in concert to regulate OXPHOS functions, beginning with the conversion of NADH to NAD⁺ and culminating with the generation of ATP, and in the process generating mitochondrial ROS (ref. 17; Fig. 4A). To evaluate whether NKX3.1 affects ETC complex activity, we isolated mitochondria from NKX3.1-expressing or control RWPE1 cells after treatment with paraquat and measured the activity of each complex as follows: (i) complex I, NADH ubiquinone oxidoreductase activity; (ii) complex II, succinate-coenzyme Q reductase activity; (iii) complex III, conversion of oxidized cytochrome c to its reduced form; (iv) complex IV, cytochrome c oxidase activity; and (v) complex V, ATP synthase activity (see Methods). The activity of all five ETC complexes was significantly reduced in mitochondria isolated from paraquat-treated control RWPE1 cells; however, activity was fully restored in mitochondria isolated from the NKX3.1-expressing cells, but not from cells expressing either NKX3.1^T164A or NKX3.1^R52C (Fig. 5A). Restoration of ETC complex activity by NKX3.1 was accompanied by increased expression of the ETC proteins (Fig. 5B); in addition, the requirement for NKX3.1 was overcome by the mitochondrial-specific antioxidant MitoQ (Supplementary Fig. S13A).

To further investigate the consequences of NKX3.1 for ETC activity, we measured the ratio of NADH and NAD⁺ as well as the level of ATP, which correspond to the first and last steps of the ETC, respectively (Fig. 4A), although not exclusively produced via OXPHOS. Paraquat treatment of control RWPE1 cells resulted in an increase in the NADH/NAD⁺ ratio and a corresponding reduction in ATP levels relative to vehicle-treated cells, whereas both NADH/NAD⁺ and ATP levels were restored in NKX3.1-expressing RWPE1 cells, but not cells expressing NKX3.1^T164A or NKX3.1^R52C (Fig. 5C and D). Similarly, in mice, paraquat treatment resulted in an increased NADH/NAD⁺ ratio and reduced ATP levels in Nkx3.1^−/− relative to Nkx3.1^+/+ prostate (Supplementary Fig. S13B and S13C).

Furthermore, paraquat treatment of control RWPE1 cells or NKX3.1-silenced LNCaP cells resulted in a significant reduction in OXPHOS [oxygen consumption rate (OCR)] and a corresponding increase in glycolytic metabolism [extracellular acidification rate (ECAR)], which were restored in the corresponding NKX3.1-expressing cells (Fig. 5E–H). This was accompanied by increased production of l-lactate in the control RWPE1 cells, which was fully restored in NKX3.1-expressing cells, but not in those expressing NKX3.1^T164A or NKX3.1^R52C, and rescued by MitoQ (Supplementary Figs. S13D–S13F). Therefore, NKX3.1 restores energy balance in conditions of oxidative stress, promoting mitochondrial homeostasis.

To investigate the relevance of these findings for human prostate cancer, we established organotypic cultures from freshly collected prostate tumor specimens obtained directly from the surgical suite (Fig. 6A). We cultured tissues for ≤5 days, during which time they retained the histologic features of the corresponding primary tumors, including expression of NKX3.1 (Fig. 6B). Notably, whereas NKX3.1 was localized primarily to nuclei in standard culture conditions, paraquat treatment resulted in increased nonnuclear protein (Fig. 6B); confocal imaging showed colocalization of nonnuclear NKX3.1 with mitochondrial markers (Fig. 6C). Localization of NKX3.1 to mitochondria following oxidative stress was accompanied by binding to the D-loop of the mitochondrial genome (Fig. 6D), indicating that mitochondrial NKX3.1 is active in the organotypic cultures.

We and others have shown that expression of NKX3.1 in prostate cancer is heterogenous at both the mRNA and protein levels (34, 36, 37, 44). In particular, human patients with prostate cancer can be segregated on the basis of having low or high levels of NKX3.1, which is associated with their differing response to a chemopreventive agent (36). We now find that the organotypic cultures can also be separated on the basis of having low or high NKX3.1 mRNA expression levels (Fig. 6E; Supplementary Table S7), thereby making it feasible to evaluate response to oxidative stress relative to NKX3.1 expression levels. We found that paraquat treatment resulted in upregulation of ETC gene expression in organotypic cultures having high versus low levels of NKX3.1 (Fig. 6F); correspondingly, we observed an increased NADH/NAD⁺ ratio and reduced ATP levels in the paraquat-treated cultures with low versus high levels of NKX3.1 (Fig. 6G and H).

To identify biological pathways enriched in tumors with high versus low levels of NKX3.1, we queried expression profiles from a cohort of primary prostate tumors (Taylor cohort, n = 131; Supplementary Table S8; ref. 60). Enriched pathways included those associated with oxidative stress, metabolism, and mitochondrial homeostasis (Fig. 6I; Supplementary Dataset S6), several of which are conserved between mouse and human prostate (see Supplementary Fig. S1). To examine the association of expression levels of NKX3.1 and those of mitochondrial ETC genes with clinical outcome, we queried expression profiles from a cohort of primary tumors from The Cancer Genome Atlas (TCGA; n = 426; Supplementary Table S8; ref. 61), which has available expression data for nuclear as well as mitochondrial genes (62), including the 13 ETC genes regulated by NKX3.1 (Fig. 4). We grouped the patients based on having low (25% lowest, n = 107) versus high (25% highest, n = 107) expression levels of NKX3.1 or combined expression levels of the 13 ETC genes, such that we compared 4 patient groups with high–high (n = 20), high–low (n = 68), low–high (n = 18), or low–low (n = 62) expression levels of NKX3.1 and ETC genes. Patients with low expression of NKX3.1 and ETC genes had worse clinical outcome than any of the other combinations (log-rank P = 5.8 × 10⁻⁴; Fig. 6J), which underscores the prognostic relevance of NKX3.1 and mitochondrial ETC expression for prostate carcinogenesis. Interestingly, NKX3.1 expression levels also correlated with those of HSPA9 in this cohort (P = 1.6 × 10⁻⁵; Supplementary Fig. S14A).

To investigate whether NKX3.1 protein expression levels and subcellular localization are associated with clinical outcome, we used two independent human prostate cancer tissue microarrays (TMA-1 and TMA-2). TMA-1 is comprised of low Gleason score primary tumors (Gleason 6 and 7) with a limited clinical follow-up period (n = 194; Supplementary Table S8), and TMA-2 is comprised of primary tumors in the full range of Gleason scores (Gleason 5 to 9) and has long-term clinical follow-up data (n = 118; Supplementary Table S8; ref. 30). We performed immunostaining for NKX3.1 to evaluate staining intensity (high, low, or negative) for nuclear and nonnuclear NKX3.1 and to correlate staining levels with Gleason score, PSA levels, and clinical outcome (Fig. 7A–G; Supplementary Figs. S14B–S14E). Immunostaining for NKX3.1 revealed both nuclear and nonnuclear expression (Fig. 7B), and confocal imaging showed colocalization of nonnuclear NKX3.1 with mitochondrial markers (Fig. 7C), consistent with its localization to mitochondria. Expression of NKX3.1 was positively correlated with HSPA9 and negatively correlated with 4HNE, a marker of lipid oxidation (Fig. 7D).

We separated patients based on their having high or low levels of NKX3.1 considering total protein (e.g., both nuclear and nonnuclear), only nuclear protein, or only nonnuclear protein (Fig. 7A and B). Although expression levels of NKX3.1 varied across Gleason scores, consistent with previous studies (e.g., refs. 34, 37), we found that patients with higher Gleason scores tended to have low levels of nonnuclear NKX3.1, which was not the case when we considered total or only nuclear NKX3.1 protein (Supplementary Fig. S14B). Furthermore, patients with low levels of nonnuclear NKX3.1 had significantly higher PSA levels, whereas no difference in PSA levels was observed in patients based on total or nuclear NKX3.1 protein (Fig. 7E; Supplementary Fig. S14C).

Most notably, we observed a significant association of NKX3.1 protein expression levels and subcellular localization with clinical outcome for both TMA-1 and TMA-2 (Fig. 7F and G; Supplementary Fig. S14D and S14E). In particular, Kaplan–Meier survival analyses revealed that reduced expression of NKX3.1 is associated with adverse disease outcome when estimated on the basis of biochemical recurrence for TMA-1 (log-rank P = 0.0232) or overall survival for TMA-2 (log-rank P = 0.0063; Fig. 7F and G). Furthermore, a multivariable Cox proportional hazards model, adjusted for Gleason score only or for all clinical variables, showed that the relationship of NKX3.1 expression to outcome is independent of these variables (HR = 4.7, P = 0.019; HR = 7.4, P = 0.032, respectively; Supplementary Fig. S14D). Most notably, the association of NKX3.1 protein levels with outcome was primarily accounted for by reduced expression of nonnuclear NKX3.1 protein (Fig. 7F and G; Supplementary Fig. S14E). In particular, low levels of nonnuclear, but not nuclear, NKX3.1 protein were significantly associated with adverse outcome, estimated on the basis of biochemical recurrence for TMA-1 (log-rank P = 0.0170, Fig. 7F; log-rank P = 0.027, Supplementary Fig. S14E) and overall survival for TMA-2 (log-rank P = 0.0063, Fig. 7G; log-rank P = 0.0079, Supplementary Fig. S14E). These findings demonstrate the significance of nonnuclear NKX3.1 for clinical outcome in prostate cancer.

Mitochondria provide a first line of defense for protecting cells from the adverse consequences of oxidative stress in normal and abnormal contexts. Mitochondrial homeostasis is controlled by a series of protein complexes that comprise the ETC, which are made up primarily of nuclear genes, although a few essential ETC genes are encoded by the mitochondrial genome. We now show that, in response to oxidative stress, the nuclear transcriptional factor NKX3.1 is recruited to mitochondria, where it regulates expression of these mitochondrial-encoded ETC genes, thereby promoting mitochondrial homeostasis and suppressing cancer initiation.

We propose a model in which NKX3.1 localization to mitochondria provides protection from the adverse consequences of oxidative stress, which contributes to its suppression of cancer initiation (Supplementary Fig. S15). Conversely, in scenarios in which NKX3.1 localization to mitochondria is low or impaired, loss of such protection leads to aberrant mitochondrial integrity and acceleration of cancer initiation (Supplementary Fig. S15). Thus, our findings elucidate a hitherto unappreciated role of NKX3.1 in mitochondria for suppression of prostate cancer initiation and provide new insights regarding how communication between the nucleus and mitochondria might sustain ETC function and ROS levels in response to oxidative stress.

Indeed, relatively few nuclear transcription factors have been shown to function in mitochondria, and even fewer to directly regulate the expression of mitochondrial genes (63, 64). Notable among these is the androgen receptor (AR), which, like NKX3.1, is a key regulator of prostate differentiation and cancer. However, in contrast to NKX3.1, which increases ETC gene expression and restores ETC function, AR negatively regulates expression of ETC complexes, which is associated with mitochondria dysfunction (65). This raises the interesting possibility that NKX3.1 and AR counterbalance each other in regulating mitochondrial homeostasis in response to oxidative stress. Indeed, prostate epithelial cells, which express both NKX3.1 and AR (2), are highly susceptible to oxidative stress that arises as a consequence of cellular senescence, inflammation, DNA damage, aging, or other factors (7, 11).

Unexpectedly, mitochondrial NKX3.1 does not originate from the nucleus; rather, in response to oxidative stress, newly synthesized NKX3.1 is imported to mitochondria by HSPA9. Nonetheless, the transcriptional properties of mitochondrial NKX3.1 appear similar to those of nuclear NKX3.1, because transcription of mitochondrial ETC genes is mediated by binding via the homeodomain to an NKX3.1 consensus DNA-binding site in the mitochondrial genome. The significance of these biochemical data is underscored by our finding that patients with low expression levels of NKX3.1 and mitochondrial ETC genes have less favorable clinical outcomes compared with counterparts having high expression levels of NKX3.1 and/or ETC genes.

Interestingly, regulation of ETC gene expression by the prostate-specific homeoprotein NKX3.1 occurs via binding to an essential regulatory region of the mitochondrial genome called the D-loop. The occurrence of an NK-class homeodomain consensus binding site within this critical and highly conserved region of the mitochondrial genome raises the question of whether similar regulation occurs in other tissue-specific contexts. Indeed, other tissue-specific NK-class homeoproteins have been shown to control ROS levels and affect mitochondrial function, as well as to exhibit nonnuclear localization (66–68), although they have not been reported to be localized to mitochondria.

Although association of NKX3.1 expression status with prostate cancer progression has been well studied (26, 34, 36, 37, 43), most previous studies have focused on nuclear NKX3.1. Our current findings unveil a new dimension of NKX3.1 regulation, because we demonstrate association of nonnuclear NKX3.1 with cancer outcome. We propose that analyses of NKX3.1 expression status combined with analyses of its subcellular localization may inform precision prevention monitoring, particularly for men on active surveillance.

Furthermore, NKX3.1 is associated with hereditary prostate cancer, and its loss of expression has been linked to increased risk of aggressive prostate cancer, especially among African American men (35). We find that two polymorphisms of NKX3.1, each of which is associated with increased risk of aggressive prostate cancer (51, 52), encode proteins that are incapable of functioning in mitochondria to protect against oxidative stress or preserve mitochondrial function. These data suggest a mechanism directly linking NKX3.1 status to patients with prostate cancer who are at higher risk, opening new avenues for genetic risk assessment in prostate cancer.

All experiments using animals were performed according to protocols approved by the Institutional Animal Care and Use Committee. Nkx3.1^+/+ and Nkx3.1^−/− mice (27) were administered paraquat dichloride/methyl viologen dichloride hydrate (10 mg/kg/day; Sigma-Aldrich) or vehicle for 1 to 9 months. Histopathologic analysis and IHC staining were done as described (39). A list of all antibodies used in this study is provided in Supplementary Table S9.

RWPE1 and LNCaP prostate cancer cells were obtained from ATCC; BPH1 cells were obtained from EMD Millipore. Cells were maintained under Mycoplasma-free conditions (30-1012K; ATCC) and only low-passage cells were used. All procedures for lentiviral studies were approved by the Office of Environmental Health and Safety. Lentiviruses were generated using second-generation packaging vectors (psPAX2 and pMD2.G; Addgene). Procedures for their preparation and transduction were as described previously (39). A list of all primers and shRNAs used in this study is provided in Supplementary Table S10.

To induce oxidative stress, cells were treated with 100 μmol/L paraquat dichloride/methyl viologen dichloride hydrate (PQ; Sigma-Aldrich) or 50 μmol/L MitoParaquat (MitoPQ; Cayman Chemical). Measurement of ROS levels was done using (i) 10 μmol/L DHE (Invitrogen); (ii) 10 μmol/L CM-H2DCFDA (Invitrogen); (iii) 5 μmol/L MitoPY1 (Tocris Bioscience); or (iv) 10 μmol/L MitoSOX (Invitrogen). FACS analysis was done using excitation/emission spectra of (i) A566/A616 for DHE, (ii) A595/A519 for CM-H2DCFDA, (iii) A595/519 for MitoPY1, or (iv) A596/A576 for MitoSOX. Proliferation, colony formation, and invasion assays were performed as described previously (28).

RNA sequencing was performed as described previously (39) on anterior prostate of Nkx3.1^+/+ or Nkx3.1^−/− mice or NKX3.1-expressing (or control) human RWPE1 cells. Differential expression signatures were ranked by t values from a two-tailed two-sample Welch t test. Pathway enrichment was done using gene set enrichment analysis (GSEA) to query pathways collected from the C2 database. Heat-map depiction of the differential expression of ETC genes was done using https://www.wikipathways.org/ (RRID:SCR_002134).

EM was performed as described previously (69). Immuno-EM was done on cells fixed in 4% paraformaldehyde plus 1% glutaraldehyde (Electron Microscopy Sciences), which were embedded in Lowicryl HM20 resin (Electron Microscopy Sciences). Biochemical fractionation of organelles was done as described (70). Measurement of mitochondrial membrane potential using TMRE was done following the manufacturer's instructions (Abcam). Measurement of mitochondrial mass was performed using MitoTracker Red CMXRos (M7512; Invitrogen) or MitoView Green (70054; Biotium) as per the manufacturer's instructions.

Gel retardation assays were done as described previously (39), using protein extracts from nuclear or mitochondrial cell fractions. ChIP-qPCR was done on mitochondria genomic DNA from cells or tissue (as above).

For measurement of ETC complex activity, a Mitochondria Isolation Kit (ab110169; Abcam) was used to isolate mitochondria, and complex I to V enzymatic activity was measured using assay kits for complex I (ab109903; Abcam), complex II (700940; Cayman Chemical), complexes II and III (ab109905), complex IV (ab109906; Abcam), and complex V (ab109907; Abcam), following the manufacturer's instructions. OCR and ECAR were quantified using a Seahorse XF96 (Agilent Technologies). Quantitative analyses of ATP, NADH/NAD⁺ ratio, and lactate levels were performed according to the manufacturers' recommendations (ATP Assay kit, Sigma-Aldrich; NADH/NAD quantitation kit, Sigma-Aldrich; l-Lactate Assay kit, Abcam).

All studies using human prostate specimens were done with written informed consent from the patients. Studies were conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, CIOMS, Belmont Report, U.S. Common Rule) and protocols approved by the Institutional Human Research Protection Office and Institutional Review Board.

For in vivo studies, the number of mice needed to achieve statistical significance was determined using standard power analysis. All cell culture analyses were done using a minimum of 3 biological replicates, each done in triplicate. Statistical analyses were performed using two-sample unpaired Welch t test, χ² test, Fisher exact test, or Spearman correlation test as indicated in each figure legend. For the box plots, boxes indicate the 25th–75th percentiles, center lines show the median, and whiskers show the minimum to maximum values. Statistical analysis was performed using GraphPad Prism software (v8.4.3, RRID:SCR_002798) and R Studio (0.99.902, R v4.0.2). Cox proportional hazards and survival analysis were performed using survival and survminer packages in R Studio.

The RNA-sequencing expression profiling data have been deposited in the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/, RRID:SCR_005012) with accession code GSE115338.

Complete details of all materials and methods are provided in the Supplementary Materials.

A. Rodriguez-Calero reports personal fees from University of Bern during the conduct of the study; and personal fees from University of Bern outside the submitted work. M.M. Shen reports grants from NIH during the conduct of the study. A. Dutta reports grants from NIH, USDA, and UDRF during the conduct of the study. No disclosures were reported by the other authors.

A. Papachristodoulou: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Rodriguez-Calero: Formal analysis, validation, investigation, visualization, methodology, writing–review and editing, evaluated the TMA data.S. Panja: Data curation, formal analysis, investigation, methodology. E. Margolskee: Resources, data curation, formal analysis, designed and generated the tissue microarray used in this study. R.K. Virk: Data curation, formal analysis, investigation, pathological evaluation of the mouse prostate phenotype. T.A. Milner: Data curation, formal analysis, visualization, methodology, performed the EM studies.L. Pina Martina: Data curation, formal analysis, methodology. J.Y. Kim: Data curation, methodology. M. Di Bernardo: Data curation, formal analysis, methodology. A.B. Williams: Data curation, formal analysis. E.A. Maliza: Data curation. J.M. Caputo: Data curation.C. Haas: Data curation. V. Wang: Data curation. G. De Castro: Data curation. S. Wenske: Data curation, supervision. H. Hibshoosh: Data curation, supervision. J.M. McKiernan: Data curation, supervision. M.M. Shen: Data curation, formal analysis, supervision, writing–review and editing. M.A. Rubin: Resources, data curation, formal analysis, supervision, writing–review and editing. A. Mitrofanova: Conceptualization, resources, formal analysis, supervision, visualization, methodology, writing–review and editing. A. Dutta: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing, co-corresponding and cofirst author.C. Abate-Shen: Conceptualization, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We are grateful to Michael Murphy for the gift of MitoQ, and Ed Reznik and Michael Espey for helpful discussions. We thank Christine Chio, Sabrina Diano, Edward Gelmann, Clementine Le Magnen, Eric Schon, and Eileen White for comments on the manuscript. Fig. 4A was created with BioRender.com using an institutional license sponsored by Columbia University's VP&S Office for Research. These studies were supported by the Herbert Irving Comprehensive Cancer Center Flow Core Facility, and the Proteomics, Genomics, and the Molecular Pathology Core facilities which are funded in part through Center Grant P30 CA013696, and by the Neuroanatomy EM Core and the EM Core at Weill Cornell Medicine. This work was supported by NIH grants CA196662, CA193442, CA183929, CA173481 (to C. Abate-Shen) and CA238005 (to M.M. Shen) and LM013236 (to A. Mitrofanova). A. Papachristodoulou was supported by the Swiss National Science Foundation Early Postdoc Mobility Fellowship, grant number P2ZHP3 181557. A. Dutta received support from NIH NCATS grant UL1TR000040, USDA NIFA grant 2021-70410-32903, and UDRF Strategic Initiative Award 2021. S. Panja is supported by the New Jersey Commission on Cancer Research Pre-doctoral Fellowship, grant number DCHS20PPC028. T.A. Milner was supported by NIH grants DA08259, HL098351. C. Abate-Shen is an American Cancer Society Research Professor supported in part by a generous gift from the F.M. Kirby Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.