Distinct microRNA Expression Profiles in Prostate Cancer Stem/Progenitor Cells and Tumor-Suppressive Functions of let-7

Most tumors contain a dynamic population of less differentiated and highly tumorigenic cells operationally defined as cancer stem cells (CSC) or tumor-initiating cells (1–10). CSCs may be phenotypically purified using surface markers. CD44 is one such marker widely used to enrich tumor-initiating cells, for example, in cancers of the breast (2), pancreas (5), head and neck (8), colon (9), and the prostate (6, 7). Our previous work has shown that CD44⁺ cells from prostate cancer cell cultures or prostate cancer xenografts exhibit high proliferative and clonogenic potential in vitro. Moreover, using limiting dilution assays in NOD/SCID (nonobese diabetic/severe combined immunodeficient) mice, we find that CD44⁺ prostate cancer cells possess 6 to 30 times higher tumor-regenerating capacity than CD44⁻ cells (6, 7). CD133 has similarly been used to enrich CSCs in brain (3), colon (10), and other cancers. Several surface marker–independent strategies have also been used to enrich tumor-initiating cells (1). Side population assay is a flow cytometry–based method initially developed to enrich hematopoietic stem cells owing to their expression of high levels of drug-detoxifying surface transporter proteins such as ABCG2 and MDR1 that efficiently efflux the Hoechst dye 33342 (11). Using the side population technique, we have shown that the side population cells in LAPC9 xenografts, although representing only approximately 0.01% of the total tumor cell population, are more than 500-fold more tumorigenic than the isogenic non–side population cells (12).

With the preponderant evidence for CSCs and our increasing knowledge of CSC heterogeneity (1), it becomes apparent that we need to understand how tumorigenic cancer cells are regulated at the molecular level so that we can design CSC-specific therapeutics. MiRNAs are small noncoding RNAs that regulate many biologic processes by inhibiting the target mRNA translation or stability (13). Deregulation of miRNAs has been observed in a variety of human tumors (14, 15). In prostate cancer, several groups have conducted miRNA expression profiling studies using either miRNA microarray (16–20) or whole-genome deep sequencing (21) in prostate cancer cell lines, xenografts, or patient samples. These studies, although reporting prostate cancer–related miRNA alterations and shedding light on differential miRNA expression in prostate cancer (relative to benign tissues), have all been conducted in bulk tumor cells and thus fail to address alterations of miRNA expression and functions specifically in tumorigenic prostate cancer cell subsets. We recently conducted, for the first time, an miRNA expression profiling in 6 highly purified prostate cancer stem/progenitor cell populations and reported that miR-34a, a p53 target, was underexpressed in all these populations (22). We further showed that miR-34a negatively regulated prostate CSC (PCSC) activity and inhibited prostate cancer metastasis by directly repressing CD44 (22). Herein, we present detailed miRNA expression profiling procedures and results and report the miRNAs that are commonly and differentially expressed in prostate cancer stem/progenitor cell populations. We further investigate the biologic functions of 2 commonly altered miRNAs, that is, let-7, and miR-301, in the context of regulating CSCs and prostate cancer regeneration. Finally, using miR-34a as an example, we explore potential mechanisms that may be responsible for the differential miRNA expression in prostate cancer stem/progenitor cells. Our results converge with the emerging theme that distinct miRNAs coordinately and distinctively regulate CSC properties (23).

Many basic experimental procedures have been described in our earlier publications (6, 7, 12, 22, 24–26). Some experimental procedures are described in Supplementary Methods. Primary human prostate tumors (HPCa) used in this study are presented in Supplementary Table S1.

PPC-1, PC3, LNCaP, and Du145 cells were obtained from American Type Cell Culture and cultured in RPMI-1640 plus 7% heat-inactivated FBS. Human xenograft prostate tumors, LAPC9 [bone metastasis; androgen receptor (AR)⁺ and prostate-specific antigen (PSA)⁺], LAPC4 (lymph node metastasis; AR⁺ and PSA⁺), and Du145 (brain metastasis; AR⁻ and PSA⁻) were maintained in NOD/SCID mice. NOD/SCID mice were produced mostly from our own breeding colonies and purchased occasionally from the Jackson Laboratories and maintained in standard conditions according to the Institutional Guidelines. All animal experiments were approved by our Institutional Animal Care and Use Committee. All these 6 prostate cancer cell types were routinely checked to be free of mycoplasma contamination using the Agilent MycoSensor QPCR Assay Kit (cat. #302107). Cell authentification by DNA fingerprinting is under way.

Prostate cancer cells were transfected with 30 nmol/L of miR-34a, let-7a, let-7b, or miR-301 mirVana mimics, or nontargeting negative control miRNA (miR-NC) oligos (Ambion) using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions (22). MirVana mimics are synthetic double-stranded oligonucleotides (oligos) that mimic mature miRNAs. In some experiments, mirVana miRNA inhibitors, chemically modified antisense oligos against let-7b, miR-301, or miR-NC (Ambion) were introduced into prostate cancer cells using the same conditions. After culturing overnight for 48 hours, transfected cells were harvested for in vitro and in vivo studies.

pLL3.7-let-7a and pLL3.7 control vector were kindly provided by Dr. J. Lieberman (Harvard University, Cambridge, MA; ref. 27). Lentiviruses were produced in 293FT packaging cells and titers determined for GFP using HT1080 cells (22). Prostate cancer cells were infected with the lentiviral supernatant [multiplicity of infection (MOI), 5–10] in the presence of 8 μg/mL polybrene and harvested 48 to 72 hours after infection for experiments.

In general, the unpaired 2-tailed Student t test was used to compare differences in cell numbers, cumulative population doublings, percentages of CD44⁺ cells, percentage of bromodeoxyuridine (BrdUrd)⁺ cells, percentage of cell-cycle phases, cloning and sphere-formation efficiency, and tumor weights. The Fisher exact test and χ² test were used to compare incidence and latency. In all these analyses, a P < 0.05 was considered statistically significant.

We first used the quantitative real-time-PCR (qRT-PCR; ref. 22) to determine the expression levels of 310 mature human miRNAs (Supplementary Table S2) in bulk prostate cancer cells purified from 3 xenografts, that is, LAPC9 (bone metastasis, AR⁺/PSA⁺), LAPC4 (lymph node metastasis, AR⁺/PSA⁺), and Du145 (brain metastasis, AR⁻/PSA⁻; Supplementary Fig. S1, step I). We then chose 136 miRNAs (Supplementary Table S3) including the top 120 abundantly expressed miRNAs and 16 less abundant miRNAs of interest (including 2 miRNAs, i.e., miR-24 and miR-103 that were used as internal controls). We measured the levels of these 136 miRNAs in CD44⁺ (i.e., cells expressing high levels of CD44) and CD44⁻ cells purified from LAPC9, LAPC4, and Du145 tumors; CD133⁺ and CD133⁻ cells from LAPC4 tumor, and integrin α2β1⁺ and α2β1⁻ cells from Du145 tumor (Supplementary Fig. S1, step II). The LAPC9, LAPC4, and Du145 tumors contain approximately 20%, 0.1%, and 30% CD44^hi cells, respectively (6), whereas the LAPC4 tumors contain approximately 1% CD133⁺ cells. The CD44⁺ prostate cancer cells are enriched in tumor- and metastasis-initiating cells (6, 7), whereas CD133⁺(CD44⁺α2β1^hi) cells purified from primary prostate cancer samples are highly clonogenic (28). In addition to these 5 (i.e., 3 CD44⁺, 1 CD133⁺, and 1 α2β1⁺) prostate cancer cell populations, we also purified, from the LAPC9 tumor, the side population, which harbors great tumor-regenerative activity (12). Because the side population represents less than 0.1% of the total population in LAPC9 tumor (12), we manually curated 57 miRNAs (Supplementary Table S4) that could be reliably detected and compared with their expression levels in the side population versus non–side population cells (Supplementary Fig. S1, step III). Comparisons of 6 marker-positive and -negative prostate cancer cell populations revealed interesting and informative differences in miRNA expression patterns.

We first compared the expression levels of 134 miRNAs between the CD44⁺ and CD44⁻ populations and observed cell type–related differential miRNA expression patterns (Supplementary Fig. S2A–S2C and Supplementary Table S3). The CD44⁺ LAPC4 and LAPC9 cells had significantly more underexpressed than overexpressed miRNAs compared with the corresponding CD44⁻ cells, whereas CD44⁺ and CD44⁻ Du145 cells had roughly similar numbers of overexpressed and underexpressed miRNAs (Supplementary Fig. S2). When we analyzed the miRNA expression patterns common to all 3 populations of CD44⁺ prostate cancer cells, we found that 3 miRNAs, that is, miR-452, miR-19a, and miR-301, were commonly overexpressed and 37 miRNAs were commonly underexpressed (Table 1; Supplementary Table S3). Among the 37 underexpressed miRNAs, miR-34a was most dramatically downregulated, representing 2% of the level in CD44⁻ cells. We have recently shown that miR-34a acts as a critical negative regulator of PCSC properties by directly targeting CD44 (22). In addition to miR-34a, 4 let-7 members (let-7a, let-7b, let-7e, and let-7f) were underexpressed in the 3 CD44⁺ populations (Table 1). Moreover, miR-141, a miR-200 family member, was also expressed at lower levels in CD44⁺ than in CD44⁻ prostate cancer cells (Table 1). miR-34, let-7, and miR-200 families of miRNAs are well-established tumor-suppressive miRNAs (22, 23, 27, 29, 30).

miR-199a*, which is downregulated in many cancers (in particular, hepatocellular carcinoma) and possesses tumor-suppressive functions by targeting oncogenic molecules such as c-MET, versican, PAK4, Brm, mTOR, and AKT (31–33), was expressed in CD44⁺ prostate cancer cells at only approximately 4% levels of the CD44⁻ cells (Table 1). Strikingly, in other cancer cells, miR-199a* has been shown to target CD44, leading to its deficiency in CD44⁺ cancer cells (32, 33). Of interest, miR-214 is in a cluster with miR-199a* (∼6 kb apart) within human dynamin-3 gene intron (DNM3os) and was co-downregulated with miR-199a* in CD44⁺ cells (Table 1). Similarly, miR-10a and miR-196a are embedded in the HoxB gene cluster and both were underexpressed in CD44⁺ prostate cancer cells (Table 1). Several other clusters of miRNAs, including let-7e/miR-99b (19q13.33), miR-183/182 (7q31-34), miR106a/19b/92a (in the Chr-X mir-106a-363 cluster), and miR-193b/365 (16p13.12), were also coordinately downregulated in the CD44⁺ prostate cancer cells (Table 1). miR-193b targets multiple oncogenic molecules including uPA, cyclin D1, 14-3-3ζ, c-Kit, and Mcl-1 and is important for cellular differentiation (34). Many other miRNAs commonly underexpressed in CD44⁺ prostate cancer cells (Table 1), including miR-218 (35), miR-148a (36), miR-181b (37), miR-203 (38), miR-183 (39), miR-24 (40), and miR-335 (41), all possess tumor/metastasis-inhibitory functions.

Together, our profiling results indicate that multiple tumor-suppressive miRNAs are coordinately downregulated in CD44⁺ prostate cancer cells. We then used the online database Diana mir-Path (42) to probe the potential signaling pathways that might be engaged by differentially expressed miRNAs. The software conducts an enrichment analysis of multiple miRNA target genes against all known Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. When we input the set of miRNAs commonly underexpressed in CD44⁺ cells, the top hits were TGFβ, Wnt, and mitogen-activated protein kinase (MAPK) signaling pathways (not shown).

The CD44⁺ prostate cancer cells are generally less differentiated (e.g., expressing less AR; ref. 6). Consistent with this notion, many of the miRNAs identified here to be underexpressed in the CD44⁺ LAPC9, LAPC4, and Du145 cells, including miR-34a, miR-141, let-7 members, miR-10a, miR-214, miR-203, miR-183, miR-365, miR-193b, miR-24, and miR-30c (Table 1) are generally depleted in (cancer) stem cells and preferentially expressed in differentiated progeny. In further support, several miRNAs underexpressed in CD44⁺ prostate cancer cells, such as miR-148a (43) and miR-141 (44), have been shown to be androgen-responsive.

We then analyzed the expression levels of 134 miRNAs in LAPC4 CD133⁺ and Du145 α2β1⁺ cells and 57 miRNAs in LAPC9 side population cells in comparison to their corresponding marker-negative populations and we observed miRNA expression patterns unique to each tumor cell population (Fig. 1; Supplementary Tables S3 and S4). Interesting, the top overexpressed miRNA in CD133⁺ LAPC4 cells was miR-21, one of the best-characterized oncomiRs widely overexpressed in human cancers (45). Among the top 10 downregulated miRNAs were many miRNAs that were also underexpressed in CD44⁺ prostate cancer cells and several tumor-suppressive miRNAs including miR-133a, miR-126, miR-15a, and miR-200a (Fig. 1A). In general, the magnitude of downregulation (i.e., to ∼10⁻⁶) was much more pronounced than that of upregulation (up to ∼10² for most), although surprisingly, there were more miRNAs overexpressed than underexpressed in CD133⁺ LAPC4 cells (Fig. 1A). When we compared the 134 miRNA expression in CD133⁺ versus CD44⁺ LAPC4 cells, we observed 25 commonly overexpressed and 29 commonly underexpressed miRNAs (Supplementary Fig. S3A).

In contrast to CD133⁺ prostate cancer cells, there were significantly more miRNAs underexpressed than overexpressed in LAPC9 side population cells when compared with the isogenic non–side population cells and, again, the levels of downregulation were higher than those of upregulation (Fig. 1B). The top overexpressed miRNA was miR-451, which was recently shown to regulate the self-renewal and tumorigenicity of colorectal CSCs (46). Among the top 10 underexpressed miRNAs in side population were miR-15a/15b and several oncosuppressive miRNAs downregulated in CD44⁺ prostate cancer cells. We observed 6 commonly overexpressed and 31 commonly underexpressed miRNAs in side population versus CD44⁺ LAPC9 cells (Supplementary Fig. S3B). Finally, roughly similar numbers of up- and downregulated miRNAs were observed in α2β1⁺ and α2β1⁻ Du145 cells (Fig. 1C). We observed 44 commonly overexpressed and 22 commonly underexpressed miRNAs in α2β1⁺ versus CD44⁺ Du145 cells (Supplementary Fig. S3C). Surprisingly, among the top 10 upregulated miRNAs were miR-30a-5p, let-7a, and miR-196a (Fig. 1C), which were commonly underexpressed in CD44⁺ prostate cancer cells (Table 1). These observations are consistent with our earlier conclusions that the α2β1⁺ prostate cancer cell population overlaps with but is also distinct from the CD44⁺ population (7).

Subsequently, we tried to identify commonly changed miRNAs. We first compared the common CD44 profiles (Table 1) with the profiles generated from CD133⁺ or α2β1⁺ populations (Supplementary Tables S2 and S3) and uncovered the miRNAs that were commonly over- or underexpressed in the 4 (i.e., 3 CD44⁺ together with CD133⁺ or α2β1⁺) cell populations (Table 2). When we combined 5 populations (i.e., 3 CD44⁺ together with CD133⁺ and α2β1⁺), only 4 miRNAs, that is, let-7b, miR-106a, miR-141, and miR-34a, were commonly underexpressed and 2 miRNAs, that is, miR-301 and miR-452, were commonly overexpressed (Fig. 2A; Table 2). When we further included the expression profile from the LAPC9 side population, only one miRNA, that is, miR-34a, was commonly underexpressed and one miRNA, miR-452, was commonly overexpressed in all 6 prostate cancer cell populations (Table 2; Supplementary Fig. S4A).

The preceding miRNA library expression profiling was conducted in cells purified from 3 xenograft models. To validate the miRNA expression data, we purified CD44⁺ and CD44⁻ prostate cancer cells from 21 primary HPCa samples (Supplementary Table S1) and measured the levels of 4 commonly underexpressed (miR-34a, let-7b, miR-141, and miR-106a) and 2 commonly overexpressed (miR-301 and miR-452) miRNAs. This strategy has an additional advantage of establishing the potential clinical relevance. We previously verified miR-34a underexpression in all HPCa-purified CD44⁺ prostate cancer cells (22). let-7b also showed underexpression in the majority (18 of 21) of samples in the CD44⁺ HPCa cells (Fig. 2B). Likewise, miR-141 was detected at much lower levels in CD44⁺ than in CD44⁻ cells derived from most HPCa samples (data not shown). In contrast, miR-106a was underexpressed in 3 of the 5 xenograft-derived populations (Supplementary Fig. S4B) and in only approximately 50% of 21 HPCa-derived CD44⁺ prostate cancer cells (Supplementary Fig. S4C). With the 2 commonly overexpressed miRNAs, we detected an overrepresentation of miR-301 in the CD44⁺ cells in 18 of 21 HPCa samples (Fig. 2C). Unexpectedly, although miR-452 was dramatically upregulated in 4 of the 5 xenograft populations (Supplementary Fig. S4D), it was downregulated in most CD44⁺ HPCa cells (Supplementary Fig. S4E). Altogether, of the 6 miRNAs commonly changed in the 5 prostate cancer cell populations, we could corroborate the underexpression of miR-34a, let-7b, and miR-141 and overexpression of miR-301 (i.e., 4 of 6 or 67%) using primary tumor–derived CD44⁺ HPCa cells.

To investigate the biologic functions of commonly and differentially expressed miRNAs, we first focused on 2 underexpressed miRNAs, that is, miR-34a and let-7, mainly because both had been shown to possess strong tumor-suppressive functions in other systems (27, 29, 30). Our earlier studies showed that miR-34a functioned as a negative regulator of PCSCs and prostate cancer metastasis (22). Herein, we focused on let-7 as 4 let-7 miRNA family members were underexpressed in CD44⁺ cells (Table 1) and let-7b was commonly underexpressed in 5 prostate cancer cell populations (Fig. 2A) as well as in CD44⁺ HPCa cells (Fig. 2B). Overexpression of let-7b in Du145 cells by transfection of a let-7b mimicking oligonucleotide reduced cell number (Fig. 3A) due to inhibition of proliferation as assessed by BrdUrd incorporation assays (Fig. 3B). In addition, let-7b oligos, when compared with the negative control (NC) oligos that contain a scrambled sequence, inhibited the establishment of Du145 holoclones (Fig. 3C–E) and spheres (Fig. 3F). Prostate cancer cell holoclones contain self-renewing tumor-initiating cells (24) and prostate cancer cell spheres formed under anchorage-independent conditions harbor tumor-initiating cells (6, 12, 25). Finally, when we infected Du145 cells with a lentivirus (i.e., pLL3.7-let-7a; ref. 27) that encodes let-7a (which recognizes the same seed sequence as let-7b), both clonal development (Supplementary Fig. S5A) and sphere formation (Supplementary Fig. S5B and S5C) were inhibited. We observed similar inhibitory effects of let-7b oligos in another prostate cancer cell type, PPC-1 (Fig. 3G; Supplementary Fig. S6A–S6C). It should be noted that all miRNA mimicking oligos used in our previous (22) and present studies are mature miRNAs, which mimic the dicer cleavage product loaded into the RNA-induced silencing complex (RISC) in the cytoplasm (22).

Overall, let-7b mimicking oligos showed similar inhibitory effects to miR-34a overexpression on prostate cancer cell holoclones and spheres (Fig. 3C–G; Supplementary Fig. S6B and S6C). However, when we analyzed cell-cycle profiles in PPC-1 cells treated with miR-34a or let-7b oligos, we observed that miR-34a caused G₁ cell-cycle arrest, whereas let-7b led to prominent G₂–M phase arrest (Fig. 3H and I). Fully consistent with the differential effects between miR-34a and let-7b on cell cycle, miR-34a overexpression induced significantly increased cell senescence assessed by staining of prostate cancer cells for senescence-associated β-gal (SA-βgal) activity (Fig. 3J; Supplementary Fig. S6D). It is well documented that G₁ cell-cycle arrest generally precedes cell senescence. In contrast, let-7b oligos, which did not cause G₁ arrest, did not induce PPC-1 cell senescence (Fig. 3J; Supplementary Fig. S6D). These results altogether suggest that let-7b and miR-34a exert differential mechanisms in prostate cancer cells with respect to their effects on cell cycle and senescence.

Next, we investigated the let-7 effects on tumor regeneration. We first conducted the “positive” control experiments by s.c. implanting A549 lung cancer cells that had been transfected with the let-7b or NC oligos in NOD/SCID mice. As reported earlier by others (30), let-7b overexpression suppressed A549 tumor development (Supplementary Fig. S7A). Surprisingly, in multiple similar tumor experiments carried out in Du145 (Supplementary Fig. S7B and S7C) or LAPC9 (Fig. 4A; Supplementary Fig. S7D) cells, let-7b oligos did not manifest obvious tumor-suppressive effects whether cells were implanted s.c. or in the dorsal prostate (DP; Supplementary Fig. S7B and S7C). Similar to let-7b oligos, let-7a oligos also did not inhibit LAPC9 tumor regeneration, although miR-34a oligos significantly retarded tumor growth (Fig. 4A). These surprising results suggested that (i) let-7 miRNAs might exert differential effects on lung (A549) versus prostate (Du145 and LAPC9) cancer cells; (ii) transfected let-7 oligos might be turned over faster in prostate cancer cells compared with lung cancer cells; (iii) let-7 and miR-34a might exert divergent regulatory roles in prostate cancer cells; and/or (iv) let-7 oligos might become degraded or turned over faster than miR-34a oligos in prostate cancer cells.

To start addressing these possibilities, we first measured let-7a/b and miR-34a levels in both freshly transfected cells and endpoint tumors (Fig. 4B–D). Du145 (Fig. 4B) and LAPC9 (Fig. 4C) cells transfected with let-7 oligos had several 100-fold higher levels of let-7 than the same cells transfected with NC oligos at 48 hours. Unexpectedly, however, A549 cells transfected with the same amount (i.e., 30 nmol/L) of let-7b possessed much higher levels of intracellular let-7b than either Du145 or LAPC9 cells (Fig. 4B and C). More surprisingly, at 48 hours after transfection of the same amount of miR-34a or let-7b (30 nmol/L for each), LAPC9 cells retained significantly higher levels of miR-34a than let-7b (Fig. 4C). As expected, the endpoint tumors all expressed similarly low levels of let-7a/b or miR-34a (Fig. 4D). These results suggest that transfected let-7 oligos, in contrast to miR-34a oligos, were rapidly degraded in prostate cancer cells, in contrast to A549 cells. Consistent with this suggestion, when we infected LAPC9 cells with pLL3.7-let-7a, the continuously delivered let-7a significantly slowed tumor growth (Fig. 4E) and inhibited tumor regeneration (Fig. 4F). Impressively, pLL3.7-let-7a also inhibited tumor development of the purified CD44⁺ Du145 cells (Fig. 4G).

The let-7 family miRNAs repress many oncogenic molecules including Ras, c-Myc, HMG, and Bcl-2 (27, 29, 30). We observed that prostate cancer cells freshly transfected with the let-7b oligos exhibited significantly reduced c-Myc and K-Ras, both at the mRNA (Fig. 4H) and protein (Fig. 4I) levels. Luciferase reporter assays confirmed K-Ras as a direct let-7 downstream target (Supplementary Fig. S8). In contrast, the Bcl-2 mRNA and protein levels were not affected by let-7b (Fig. 4H and I).

We also probed for the biologic functions of one commonly overexpressed miRNA, that is, miR-301 (Supplementary Figs. S9 and S10). Unexpectedly, enforced miR-301 expression via oligo-transfection in purified CD44⁻ Du145 cells (Supplementary Fig. S9A) or knocking down endogenous miR-301 in CD44⁺ Du145 cells (Supplementary Fig. S9B) did not significantly affect sphere formation. Similar negative results were obtained in holoclone assays (Supplementary Fig. S9C and S9D). miR-301 overexpression and knockdown were verified by quantitative PCR (qPCR; Supplementary Fig. S9E). Manipulation of miR-301 levels also did not affect the tumor regeneration of CD44⁺/CD44⁻ Du145 cells (Supplementary Fig. S9F–S9I). Similarly, anti-miR-301 oligos did not alter the clonal and tumorigenic properties of PC3 cells (Supplementary Fig. S10A–S10C). In contrast, enforced miR-301 expression promoted, whereas anti-miR-301 reduced the clonal and sphere-forming capacities of xenograft-derived LAPC9 cells (Supplementary Fig. S10D and S10E).

How tumor-suppressive miRNAs such as miR-34a and let-7 might be underexpressed in tumorigenic subpopulations is an interesting question. We attempted to address this question by focusing on miR-34a, whose expression is regulated in both p53-dependent and -independent mechanisms (47). The miR-34a levels in the 4 prostate cancer cell types with null or mutant p53 were significantly lower than those in the 6 prostate (cancer) cell types with wild-type (wt) p53 (22). To explore whether the lower levels of miR-34a in tumorigenic prostate cancer cells might be related to lower p53 expression/activity, we treated p53-wt LNCaP cells with paclitaxel and 3 DNA-damaging agents, that is, doxorubicin, etoposide, and γ-irradiation (X-ray). p53 was activated by etoposide and X-ray, as evidenced by both p53 protein accumulation (Fig. 5A) and increased protein and mRNA levels of p21 (Fig. 5A and B), a p53 transcriptional target. When miR-34a (1p36.22) and miR-34b/c (11q23.1) levels were measured in treated LNCaP cells, we observed that miR-34a levels did not significantly change except a slight increase at 48 hours (Fig. 5C). In contrast, both etoposide and X-ray increased miR-34b and miR-34c levels by several fold (Fig. 5C). These observations suggest that underexpression of miR-34a in CD44⁺ prostate cancer cells might not be related to p53 expression or activity. In support, the miR-34a mRNA levels in the CD44⁺ cells freshly purified from 14 primary HPCa cells did not correlate with p53 (Supplementary Fig. S11A and S11B) or p21 (not shown) mRNA levels. Previous studies suggest that c-Myc may positively regulate miR-34a (48). However, the miR-34a mRNA levels in CD44⁺ HPCa cells also did not correlate with c-Myc mRNA (Supplementary Fig. S11C).

For the first time, we have profiled the miRNA expression patterns in purified subpopulations of prostate cancer cells that possess stem/progenitor cell properties. Among the CD44⁺, side population, CD133⁺, and α2β1⁺ cells studied, the CD44⁺ prostate cancer cells are best characterized and have been consistently shown to enrich for tumor-initiating and prometastatic cells (6, 7, 22, 25). The side population is also enriched in tumorigenic cells, although it is more rare (<0.1%) and detectable only in LAPC9 cells (12, 25). The CD133⁺ prostate cancer cells are clonogenic and may also harbor CSCs (28). In contrast, the α2β1⁺ prostate cancer cells are proliferative progenitors that do not enrich for CSCs (7). Our current miRNA profiling substantiates the heterogeneous nature of prostate cancer stem/progenitor cells as CD44⁺, side population, CD133⁺, and α2β1⁺ cell populations exhibit overall distinct miRNA expression profiles. This is perhaps best illustrated by comparing CD44⁺ and CD133⁺ populations—the CD44⁺ prostate cancer cells predominantly downregulate, whereas the CD133⁺ LAPC4 cells significantly upregulate multiple miRNAs. Some of the underexpressed miRNAs such as miR-34a and let-7 are also downregulated in prostate tumors in comparison with benign tissues (16–20).

On the other hand, different prostate cancer stem/progenitor cells also commonly over- and underexpress certain miRNAs. One of the most striking observations is that the 3 populations of CD44⁺ prostate cancer cells commonly and coordinately downregulate 37 miRNAs, many of which exist in genomic clusters and most of which possess tumor-suppressive functions. This observation is remarkably similar to the findings that multiple tumor-suppressive miRNAs are coordinately “depleted” in other CSCs, for example, the underexpression of let-7 family, miR-200 family, and miR-30 in breast CSCs and of miR-34a, miR-128, miR-451, and others in glioblastoma stem cells (reviewed in ref. 23). More remarkably, Shimono and colleagues also reported that 37 miRNAs were differentially expressed in the CD44⁺CD24^−/lo breast CSCs including the downregulation of 3 clusters, miR-200c/141, miR-200b/200a/429, and miR-183/96/182 (39), which are also downregulated in CD44⁺/α2β1⁺ prostate cancer cells. Because these CSC-depleted miRNAs generally target potent oncogenic molecules involved in regulating cell cycle and proliferation [e.g., E2F, HMGA2, Ras, cyclins, cyclin-dependent kinases (CDK)], cell survival (e.g., Bcl-2, Mcl-1, Bcl-XL), self-renewal (e.g., Bmi, Notch, Myc), and cell migration/invasion (e.g., CD44, c-MET, ZEB; ref. 23), it is conceivable that lack of these miRNAs would confer many stem cell properties. Many of these tumor-suppressive miRNAs are also deficient in embryonic and adult stem cells and preferentially expressed in differentiated progeny (23). Consequently, their lower expression in CD44⁺ prostate cancer cells further supports the stem-like features of these cells and is consistent with earlier observations that the CD44⁺ prostate cancer cells are less differentiated expressing little AR (6). In this regard, it is interesting that at least 2 androgen-responsive miRNAs, that is, miR-148a (43) and miR-141 (44), are underexpressed in the CD44⁺ prostate cancer cells. In contrast, another androgen-regulated miRNA, miR-21 (49), is the most highly expressed miRNA in CD133⁺ LAPC4 cells, emphasizing the difference between these 2 populations of prostate cancer cells.

More miRNAs are downregulated than upregulated in the 3 CD44⁺ prostate cancer cell populations in common with either CD133⁺ or α2β1⁺ cells (Table 2). When these 5 populations are combined for analysis, 4 miRNAs (i.e., miR-34a, let-7b, miR-106a, and miR-141) are commonly downregulated and 2 miRNAs (i.e., miR-301 and miR-452) are commonly upregulated. Using the CD44⁺ HPCa cells freshly purified from patient tumors, we have confirmed the differential expression of 4 of the 6 (i.e., miR-34a, let-7b, miR-141, and miR-301) miRNAs in marker-positive versus -negative cells. It is presently unclear why the underexpression of miR-106a and overexpression of miR-452 observed in 3 xenograft prostate cancer cells are not borne out in CD44⁺ HPCa cells.

To establish whether the miRNAs identified in our miRNA library screening are functionally relevant, we have by far thoroughly studied 2 commonly underexpressed (i.e., miR-34a and let-7b) and 1 commonly overexpressed (i.e., miR-301) miRNAs. Our earlier studies have uncovered a powerful role of miR-34a in restricting PCSC activity and prostate cancer regeneration/metastasis via repressing CD44 itself (22). In the present study, we report similar prostate cancer–suppressive functions of let-7b/a. Our observations are in line with the widely recognized tumor-inhibitory effects of let-7a/b (27, 29, 30) and suggest that like miR-34a, the underexpressed let-7 normally functions to inhibit certain PCSC properties. An intriguing finding is that in prostate cancer cells, the transfected mature let-7a/b oligos seem to be degraded much more rapidly than miR-34a oligos, explaining why the former do not manifest obvious tumor-inhibitory effects. In fact, even prostate cancer cells infected with the pLL3.7-let-7a lentiviral vectors, which do manifest prostate cancer–inhibitory effects, keep low steady-state levels of let-7a (Liu and colleagues, unpublished observations). Coupled with the observations in A549 lung cancer cells, our work suggests that in prostate cancer, let-7 miRNAs have a faster turnover rate than other miRNAs such as miR-34a. Future work will further explore this potentially interesting phenomenon. Another interesting finding is that let-7 and miR-34a possess mechanistic differences in suppressing prostate cancer stem/progenitor cells: miR-34a induces G₁ cell-cycle arrest followed by cell senescence, whereas let-7 causes a prominent G₂–M phase arrest without inducing senescence. Furthermore, miR-34a, but not let-7, induces apoptosis in some prostate cancer cells (22). In support, let-7 overexpression does not affect the prosurvival molecule Bcl-2 (Fig. 4I).

In contrast to consistent prostate cancer–inhibitory effects of miR-34a and let-7, miR-301, which is commonly overexpressed in prostate cancer stem/progenitor cells and recently shown to promote breast cancer cell proliferation and invasion (50), seems to exhibit cell type–dependent effects. Although manipulation of miR-301 levels does not affect Du145 and PC3 cells, its overexpression promotes, whereas its knockdown inhibits the clonogenic properties of LAPC9 cells.

It will be of general interest to understand how certain miRNAs are differentially expressed in CSCs versus non-CSCs. Because the 2 populations are isogenic, it stands to reason that the differential expression results from epigenetic events rather than genetic mutations. Indeed, many tumor-suppressive miRNAs (e.g., miR-34a) are downregulated in cancer because of promoter hypermethylation or aberrant histone modifications. When we treated CD44⁺ prostate cancer cells with 5-aza-deoxycytidine and/or trichostatin A (an inhibitor of histone deacetylase), we did not observe any significant increase in miR-34 (Liu and colleagues, unpublished observations). The miR-34a levels also do not correlate with the 2 known upstream transcriptional regulators, that is, p53 and c-Myc. In fact, even in p53-wt bulk LNCaP cells, p53 activation does not consistently result in significant upregulation of miR-34a. Altogether, these observations argue that some other mechanisms might be operating to dampen miR-34a expression in PCSC-enriched cells.

In summary, we have successfully conducted an miRNA expression profiling study in several prostate cancer stem/progenitor cell populations, which has revealed both distinctively and commonly expressed miRNAs in tumorigenic prostate cancer cells. While shedding important light on how PCSCs may be regulated by miRNAs, our results converge with the emerging theme that distinct miRNAs both distinctively and coordinately regulate CSC properties (23). Finally, our study establishes that tumor-suppressive miRNAs identified herein, such as miR-34a and let-7b/a, may represent novel therapeutics to specifically target CSCs and can be used in replacement therapy regimens.

K. Kelnar is the fulltime employee of Mirna Therapeutics and A. Vlassov is a fulltime employee of Life Technologies. D. Brown has ownership interest (including patents) and is employed as the Director of Research in Mirna Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Liu, D. Brown, J. Wang, D.G. Tang

Development of methodology: C. Liu, D.G. Tang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Liu, D.G. Tang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Liu, K. Kelnar, D. Brown, D.G. Tang

Writing, review, and/or revision of the manuscript: C. Liu, D. Brown, D.G. Tang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Kelnar, A.V. Vlassov, D.G. Tang

Study supervision: D.G. Tang

Coordinating several collaborating groups: D.G. Tang

The authors thank P. Whitney, B. Liu, and X. Liu for their technical assistance, Dr. J. Lieberman for providing pLL3.7-let-7 lentivector, and other Tang laboratory members for helpful discussions.

This work was supported, in part, by grants from the NIH (R01-ES015888, R21-CA150009), Department of Defense (W81XWH-08-1-0472, W81XWH-11-1-0331), CPRIT funding (RP120380), and the MD Anderson Cancer Center Bridge fund and Center for Cancer Epigenetics and Laura and John Arnold Foundation RNA Center pilot grant (to D.G. Tang), and by two Center Grants (CCSG-5 P30 CA016672 and ES007784). C. Liu was supported in part by a predoctoral fellowship from the Department of Defense (W81XWH-10-1-0194).

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.