Main

A large body of evidence indicates that alterations in the expression of microRNAs (miRNAs) contribute to cancer pathologies.1 miRNAs function as agents of the RNA interference pathway to silence their cognate coding target genes either by cleaving mRNA molecules or by inhibiting their translation.2 By silencing tumor-suppressive and oncogenic mRNAs, miRNAs themselves can function as oncogenes or tumor suppressors, respectively.3 The let-7 family, for instance, limits lung tumorigenesis through inhibition of several oncogenes including members of the Ras family or HMGA2.4, 5

Initial efforts to link miRNAs with cancer were based on expression analyses in which tumors were compared with normal tissues. Such expression profiling analyses revealed the characteristic miRNA signatures of human cancers.6 Surprisingly, they also highlighted an overall downregulation of mature miRNAs in several types of mouse and human cancers.7 This observation raised two important questions: (i) one of causality – is there a causative link between global downregulation of miRNAs and cancer progression or is the downregulation a byproduct of tumor development? and (ii) one of mechanism – what are the molecular mechanisms and the genetic events underlying such widespread changes in tumors? The first question was addressed by RNAi targeting of factors involved in miRNA maturation, such as DICER1 and DROSHA, showing that global downregulation of miRNA processing increased the transforming properties of a lung adenocarcinoma cell line in in vitro culture assays and in xenograft experiments.8 Widespread silencing of miRNA expression was proposed to be, at least partly, a consequence of Myc-mediated transcriptional repression;9 however, the data also raised the possibility that DICER1 might be a target of genetic disruption in human cancers. Surprisingly, however, although reduced levels of DICER1 have been reported in tumors,10, 11 no loss-of-function mutations in DICER1 have been reported to date. There have, however, been reports of truncating mutations in TARBP2, encoding an integral component of a DICER1-containing complex, in sporadic and hereditary carcinomas with microsatellite instability.12 Frameshift mutations in TARBP2 diminished TRBP protein expression and caused a partial defect in the processing of miRNAs. Importantly, the TRBP impairment is associated with a partial destabilization of the DICER1 protein. These data raised the possibility that Dicer downregulation rather than its complete loss of function is selected for during tumorigenesis.

To directly address this question, we manipulated Dicer1 (referred hereafter as Dic) gene dosage in a mouse model of retinoblastoma. As germline inactivation of Dic in mice causes an early embryonic lethal phenotype,13 we specifically inactivated Dic in retinoblasts by combining a conditional floxed allele of Dic14 with the retinal Chx10Cre transgenic line.15 We chose this genetic model system for two main reasons. First, Chx10Cre-mediated inactivation of one Dic allele decreases the production of mature miRNAs without affecting retinogenesis.16 Second, the Chx10Cre mice have been used to create the first preclinical mouse model of retinoblastoma.17, 18 Chx10Cre-mediated inactivation of the Retinoblastoma (Rb) gene leads to inappropriate exit from the cell cycle of retinal progenitor cells and a block in rod maturation.19 These mice do not develop retinoblastoma as expression of another Rb family member, p107, increases in a compensatory manner.18 However, on a p107-null background, retinal Rb inactivation (Chx10Cre; Rblox/lox p107−/−) leads to the formation of early hyperproliferative lesions,17 which are often referred to as retinomas.20 Importantly, these lesions rarely go on to become aggressive and invasive tumors and only do so in older animals, suggesting that additional oncogenic lesions are required for full-blown tumorigenesis. For example, conditional inactivation of p53 in mice is one mechanism through which these lesions can progress into aggressive retinoblastoma.17 Accordingly, amplification and overexpression of MDMX, a key negative regulator of p53, is a frequent event that is selected during human retinoblastoma formation.21, 22

Results

We first confirmed that specific loss of one Dic allele in retinal progenitor cells does not affect normal retinal development. As expected, histological analysis of the retinae of several Chx10Cre; Diclox/+ mice at various stages of postnatal development did not reveal any morphological abnormalities (Figure 1). Strikingly, however, on the retinoblastoma-sensitized background, loss of one Dic allele dramatically accelerated tumor formation (Figure 2a). Virtually all Chx10Cre; Rblox/lox p107−/−; Diclox/+ mice develop aggressive and invasive intraocular retinoblastoma (Figure 2b and c), with an average time to visible tumors of 180 days (Figure 2a). Half of these mice had anterior chamber invasion that was clearly visible by 10 weeks of age. By contrast, Chx10Cre; Rblox/lox p107−/−; Dic+/+ developed retinoblastoma only with slow and inconsistent kinetics (Figure 2a). Moreover, although more than 30% of Chx10Cre; Rblox/lox p107−/−; Diclox/+ mice had developed bilateral retinoblastoma with clear evidence of anterior chamber invasion, Chx10Cre; Rblox/lox p107−/−; Dic+/+ mice only ever developed unilateral tumors (data not shown). Finally, metastatic tumors that had invaded local tissues outside the eye through the optic nerve were only observed in Chx10Cre; Rblox/lox p107−/−; Diclox/+ mice (data not shown).

Figure 1
figure 1

Loss of one copy of Dicer1 does not affect retinogenesis. Hematoxilin–eosin stain of P35 and P200 retinae show normal retinal lamination in Chx10Cre; Diclox/+ mice. Abbreviations: ONL, outer nuclear layer; RPE, retinal pigment epithelium; INL, inner nuclear layer; GL, ganglion layer. Scale bars=40 μm

Figure 2
figure 2

Dicer1 (Dic) heterozygosity enhances tumorigenesis on a retinoblastoma-sensitized background. (a) Kaplan–Meier curve showing the time to first observation of externally visible retinoblastoma. This time was respectively markedly decreased in Chx10Cre; Rblox/loxp107−/−; Diclox/+ mice and increased in Chx10Cre; Rblox/loxp107−/−; Diclox/lox mice (log rank test, P=0.0015 and 0.0036) relative to Chx10Cre; Rblox/loxp107−/−; Dic+/+ littermates. (b) A 3-month-old Chx10Cre; Rblox/loxp107−/−; Diclox/+ mouse with aggressive retinoblastoma. (c) Hematoxylin and eosin stain with the three retinal nuclear layers (GL: ganglion layer; INL: inner nuclear layer; ONL: outer nuclear layer) indicated. Early retinoblastoma lesions at P35 and invasive tumors seeding the vitreous at P60 are found in Chx10Cre; Rblox/loxp107−/−; Diclox/+ mice. P200 shows late stage retinoblastoma that had filled the vitreous and the anterior chambers (arrow). Inset: a representative Homer–Wright rosette found in Chx10Cre; Rblox/loxp107−/−; Diclox/+ tumors. Scale bars in the top panels=40 and=200 μm in the lower panels

We next examined the retinae histologically at postnatal days P35 and onwards. As previously described,18 the retinal cytoarchitecture of Chx10Cre; Rblox/lox p107−/−; Dic+/+ at P35 is slightly disrupted because of the focal expansion of immature cells from the inner nuclear layer (INL) and their protrusion through the outer plexiform layer (OPL) (Figure 1c). In addition, defects in the maturation of the rod photoreceptors lead to a hypocellular outer nuclear layer (ONL), although the three nuclear layers are still detected in these mice. In contrast, the three nuclear layers can no longer be distinguished in the Dic heterozygous mutants (Figure 2c). The laminar organization in Chx10Cre; Rblox/lox p107−/−; Diclox/+ retinae was focally disrupted severely, with immature cells from the INL invading the OPL and extending up to the apical surface of the retina. This resulted in the disruption of the interaction between photoreceptor outer segments and the retinal pigment epithelium (RPE) and focally to dramatic ONL hypocellularity, presumably as a consequence of photoreceptor cell death. The focal nature of the phenotype is consistent with the previously reported mosaic expression pattern of Cre in Chx10Cre transgenic mice.15, 23 Beyond P35, larger dysplastic lesions, found mainly at the periphery (6/7 eyes analyzed), seeded the vitreous (Figure 2c, P60) and eventually invaded the anterior chamber of the eye (Figure 2c, P200). The lesions contained Homer–Wright rosettes (Figure 2c), which consist of a radial arrangement of cells around a central tangle of neuronal processes. Interestingly, these histological structures are often found in a subset of human retinoblastomas.24

This histopathological analysis indicated that the phenotype observed in Chx10Cre; Rblox/lox p107−/−; Diclox/+ retinae is similar in nature but is significantly more severe than that observed in Chx10Cre; Rblox/lox p107−/− retinae. We therefore hypothesized that the dysplasia and the partial degeneration observed in Dic heterozygous retinae might result from severe expansion of a pool of retinal progenitor cells that normally reside in the INL. Accordingly, immunostaining showed that the cells that disrupted synaptogenesis in the OPL and extended all the way to the apical surface of the retinae expressed the progenitor cell markers Syntaxin and Chx1025, 26, 27 (Figure 3). The early tumors also stained for Calretinin, which labels a subset of amacrine and ganglion cells. However, Calretinin expression was less abundant than Syntaxin, more scattered and variable from animal to animal (data not shown). Calbindin, which labels horizontal cells and a subset of amacrine cells weakly, was either expressed at very low levels or undetectable in Chx10Cre; Rblox/lox p107−/−; Diclox/+ early lesions (data not shown). Together, these data indicate that the lesions are composed of retinal progenitor cells biased toward the amacrine cell fate. Importantly, these immature cells were GFP-positive and therefore Cre-positive (Figure 3). Cre was fused to GFP in the Chx10Cre transgenic mice,15 so that Cre-positive cells can be identified using anti-GFP antibodies.

Figure 3
figure 3

Immunostaining of the Chx10Cre; Rblox/loxp107−/−; Diclox/+ retinoblastoma lesions. Syntaxin, Chx10 and GFP immunostaining of Chx10Cre; Rblox/loxp107−/−; Diclox/+ (Diclox/+) and Chx10Cre; Rblox/loxp107−/−; Dic+/+ (Dic+/+) retinae at P45

As inappropriate expansion of the immature retinal cells could be a consequence of increased cell proliferation and/or survival, we measured cell proliferation and apoptosis using immunohistochemical assays. As previously reported, we found extensive BrdU incorporation in Chx10Cre; Rblox/lox p107−/− retinae at P14, a time when retinogenesis is normally complete. This phenotype was significantly exacerbated in Chx10Cre; Rblox/lox p107−/−; Diclox/+ retinae (Figure 4a and b). Importantly, BrdU-positive cells were localized to the regions containing the GFP immunopositive cells and to where lamination was disrupted. These regions were also strongly positive for the proliferation marker Ki67 (Figure 4a and c). BrdU- and Ki67-positive cells were found in all early dysplastic lesions and late (i.e., P200) retinoblastoma tumors (data not shown). Inappropriate cell proliferation leads to increased apoptosis in Chx10Cre; Rblox/lox p107−/− retinae.17 Accordingly, active capase-3-positive cells could be detected in the Chx10Cre; Rblox/lox p107−/− retinae (Figure 4d). There was no significant decrease per se in the number of cleaved caspase-3-positive cells in the Chx10Cre; Rblox/lox p107−/−; Diclox/+ retinae. However, as the number of GFP-positive cells is increased in these retinae compared with the Chx10Cre; Rblox/lox p107−/− retinae, the data indicate that the number of dying Cre-positive cells is reduced proportionally in the Dicer1 heterozygous retinae (Figure 4d and e).

Figure 4
figure 4

Proliferation and apoptosis in Chx10Cre; Rblox/loxp107−/−; Diclox/+ (Diclox/+) and Chx10Cre; Rblox/loxp107−/−; Dic+/+ (Dic+/+) at P14 and P21. (a, c and d) BrdU incorporation, Ki67 and active Caspase3 immunostaining. Scale bars=40 μm. Quantification of BrdU (b) and Caspase3 (c) staining was performed on horizontal serial sections at the optic nerve level. (b) The number (nb) of BrdU-positive cells were normalized to the nb in Chx10Cre; Rblox/loxp107−/−; Dic+/+ (Dic+/+), which was set at 1. (e) The number of immunoreactive cells to active Caspase3 staining were normalized to the number of GFP-positive cells in serial consecutive sections. The data are presented relative to the observed ratio of GFP/Caspase3 in Chx10Cre; Rblox/loxp107−/−; Dic+/+ (Dic+/+), which was set at 1. Error bars represent standard deviation. P-values (Student's t-test; N=3)

Collectively, these data indicate that heterozygosity for a Dicer1 mutation promotes the switch from benign retinoma lesions to aggressive and invasive retinoblastomas. Monoallelic loss of Dic is sufficient to promote the expansion of retinal progenitor cells (with a bias toward the amacrine cell fate), which ultimately leads to the formation of early neoplastic lesions that progress into aggressive and metastatic tumors. The ability of these progenitor cells to form these aggressive tumors seems to be a consequence of both increased cell proliferation potential and resistance to apoptosis.

These data strongly support the view that Dicer1 functions in vivo as a haploinsufficient tumor suppressor gene. Considering that the dysplastic lesions in Chx10Cre; Rblox/lox p107−/−; Diclox/+ retinae are observed as early as P35, it is very unlikely that active selection against complete loss of the remaining wild-type Dic allele is required for tumorigenesis. However, this possibility had to be formally excluded experimentally. As expected, we obtained evidence of Cre-mediated recombination of the conditional Dic allele in Chx10Cre; Rblox/lox p107−/−; Diclox/+ normal retinae (Ctr Retinae, P20) and isolated tumors (Figure 5a and b). Importantly, the wild-type allele was retained in all tumors carefully dissected and genotyped (Figure 5b). To ensure that the remaining Dic allele was not functionally inactivated through mutations, we assessed the expression levels of all mature miRNAs in five different Chx10Cre; Rblox/lox p107−/−; Diclox/+ tumors by RT-qPCR analysis. Consistent with a decrease of Dicer1 function, we observed a global decrease in steady-state miRNA levels in all five tumors analyzed compared with the levels in Chx10Cre; Rblox/lox p107−/−; Dic+/+ (Figure 5c, left panel). Importantly, all mature miRNAs that are expressed at high levels in Chx10Cre; Rblox/lox p107−/−; Diclox/+ P20 retinae are expressed at comparable levels in all five tumors (Figure 5c, right panels). These analyses indicate that microRNA processing is only partly impaired, but not completely disrupted, in these tumors. Finally, we directly assessed the consequences of complete Dicer1 inactivation in the retinoblastoma mouse model. We confirmed that complete ablation of Dicer1 in the retina is well tolerated and only leads to progressive retinal degeneration.16 Similar results were obtained on the retinoblastoma-sensitized background. The retinae of adult Chx10Cre; Rblox/lox p107−/−; Diclox/lox were partly dis-organized (Lambertz et al., manuscript in preparation). Surprisingly, these mice were protected from tumor formation Figure 2a). Together, the data argue that although partial loss of Dicer function favors tumor formation, complete loss of Dicer is deleterious to retinoblastoma development, further arguing in favor of a haploinsufficient tumor suppressor function of Dicer1.

Figure 5
figure 5

Chx10Cre; Rblox/loxp107−/−; Diclox/+ retinoblastoma tumors retain a wild-type and functional Dic allele. (a) Schematic representation of the Dic wild-type, floxed and Cre-excised alleles. (b) DNA was prepared from two Chx10Cre; Rblox/loxp107−/−; Diclox/+ P20 retinae (Ctr Retina) and 10 isolated tumors (results from 4 different tumors are shown) and examined by PCR using the primers depicted in the top panel. (c) RT-qPCR analyses of mature miRNA expression. Left panel: hierarchical clustering of miRNAs significantly downregulated in five Chx10Cre; Rblox/loxp107−/−; Diclox/+ tumors (black, dKO; Diclox/+) compared with Chx10Cre; Rblox/loxp107−/−; Dic+/+ P20 retinae (gray, dKO; Dic+/+). Middle panel: heat map of all miRNAs expressed at higher levels than the mean in Chx10Cre; Rblox/loxp107−/−; Diclox/+ P20 retinae shows comparable expression levels in both Chx10Cre; Rblox/loxp107−/−; Diclox/+ P20 retinae and tumors. Right panel shows the mean cq values for all miRNAs analyzed (522 different murine miRNAs) in both Chx10Cre; Rblox/loxp107−/−; Diclox/+ P20 retinae and tumors

These data also imply that a set of miRNAs, the function of which requires Dicer1 function, might act to prevent full-blown retinoblastoma formation on the Chx10Cre; Rblox/loxp107−/− background. To identify such candidate tumor suppressor miRNAs, we determined the expression profile of the entire miRNome in P20 retinae from wild-type (Cre-negative), Chx10Cre; Rblox/lox p107−/−; Dic+/+ and Chx10Cre; Rblox/loxp107−/−; Diclox/+ using LNA-based microarray and RT-qPCR approaches. Both types of analyses identified a common set of 11 miRNAs that are consistently upregulated between wild-type and Chx10Cre; Rblox/lox p107−/−; Dic+/+ (Figure 6a). Most interestingly, among them are two members of the let-7 family (let-7c and let7i), the upregulation of which and of another let-7 member, let-7b, was confirmed by independent Q-RT-PCR analysis (Figure 6b). Given that this upregulation was significantly attenuated in the Dic heterozygous retinae and the recognized role of let-7 family members in tumor suppression,28 this observation raises the possibility that let-7 family members have a causal role in retinoblastoma formation as critical regulators of the switch from retinomas to retinoblastoma. Our list of 11 differentially expressed miRNAs also included miR-34c (Figure 6a and b). There was also a clear upregulation of miR-34b-3p in both microarray and RT-q-PCR analyses, and a moderate, but reproducible, upregulation of miR-34a was evident from the microarray data (data not shown). Interestingly, the miR-34 family members have been identified as p53 targets and key mediators of its tumor suppressor function.29

Figure 6
figure 6

Search for putative tumor suppressor miRNAs in retinoblastoma. (a) Heat map of selected differentially expressed miRNAs in Chx10Cre-negative mice (light gray, wild-type), Chx10Cre; Rblox/loxp107−/−; Dic+/+ (dKO; Dic+/+) and Chx10Cre; Rblox/loxp107−/−; Diclox/+ P20 retinae (black, dKO; Diclox/+). (b) Expression analysis by RT-qPCR of let-7 family members (let-7b, let-7c and let-7i) and miR-34c

Discussion

To address the importance of Dicer1 gene dosage in cancer development, we used a preclinical mouse model of retinoblastoma.17 Our data provide clear genetic evidence that Dicer1 functions as a haploinsufficient tumor suppressor in vivo. In keeping with this observation, information in the public domain, for example, Cancer Genome Project at the Sanger Institute,30 indicates that hemizygous deletions of DICER1 occur in 27% (207/761) of tumors derived from tissues of diverse origins, such as the central nervous system, lung, pancreas, soft tissues, breast, bone, hematopoietic or lymphoid. Importantly, consistent with our findings, homozygous deletions have never been observed in any of these 761 tumors. Very recently, heterozygous point mutations in DICER1 were reported in patients with pleuropulmonary blastoma.31 However, DICER1 expression was retained in the mesenchymal tumor cells from these patients, again arguing against strong selective pressure for complete loss of Dicer function in human tumors. Thus, although this study focuses on the role of Dicer1 in a mouse model of retinoblastoma, there is evidence supporting a broad role for DICER1 as a haploinsufficient tumor suppressor in human cancer.

Kumar et al. had attempted to investigate an etiological role for Dicer1 in a mouse model of cancer, in which Dicer1 was inactivated in a K-Ras-induced mouse model of lung cancer using intranasal infection with adenovirus expressing Cre.8 Although it highlighted an important role for Dicer1 in tumor suppression, this elegant study did not resolve the important issue of gene dosage. The authors had shown that tumorigenesis was enhanced on both Diclox/+ and Dicloxlox genetic backgrounds, but they did not analyze the extent of Cre-mediated Dicer1 inactivation in these tumors. It therefore remained formally possible that lung tumors developed in the Dicloxlox mice only as a result of incomplete inactivation of Dicer1 function. This possibility has just very recently been confirmed in a study that was published while our manuscript was being processed.32 This last report, together with the data presented herein, clearly emphasizes that only partial, rather than complete, inactivation of Dicer1 enhances tumorigenesis in vivo (Figure 7).

Figure 7
figure 7

Monoallelic loss of Dicer enhances retinoblastoma formation. Our data support a model in which monoallelic loss of Dicer1 cooperates with Rb inactivation in the formation of aggressive and invasive retinoblastoma. In contrast, biallelic loss of Dicer1 leads to retinal degeneration on a retinoblastoma-sensitized background and inhibition of tumor formation

It is generally accepted that complete loss of tumor suppressor function through mutations and loss of heterozygosity is a prerequisite for tumor development. To date, however, several haploinsufficient tumor suppressor genes have been identified,33 mostly through the use of genetically engineered mouse models. Our mouse genetic data identify DICER1 as yet another member of this group. A general theme in haploinsufficient genes is that tumors generated in a haploinsufficient context are frequently of later onset and less severe than the corresponding tumor carrying the homozygously mutated gene. This is not the case for Dicer1. Complete loss of Dicer1 did not enhance but rather inhibited retinoblastoma formation. In agreement with this, biallelic loss of Dicer1 was strongly selected against in K-Ras-induced lung tumors.32 To our knowledge, Dicer1 is therefore the first example of a gene that functions as a tumor suppressor only in the context of haploinsufficiency and not on homozygous loss of function. Regardless, as the list of haploinsufficient tumor suppressors grows, hemizygous deletions at loci such as DICER1 should be considered as potential key pro-tumorigenic lesions.

Our data further establish a causative link between global downregulation of miRNA-processing and cancer development. It is possible that the reduction of expression of only a subset of tumor suppressor miRNAs is the event that promotes tumorigenesis (Figure 7). In the context of retinoblastoma, our profiling data identified the let-7 family members as potential candidate tumor suppressors. It may be noted that let-7b has previously been shown to be downregulated in human retinoblastoma.34 Other sets of miRNAs that can account for the observed tumor-suppressive activity of Dicer1 in retinoblastoma are the miR-34 family members. Given that these genes have been identified as p53 targets and are critical mediators of its activity29 and the importance of the p53 tumor suppressor function in retinoblastoma,17, 21 it will be interesting to further assess genetically their functional relevance in the retinoblastoma mouse model. It has been shown that miR-34 family members are differentially expressed in human retinoblastoma cell lines and tumors compared with normal retina samples.35 Exogenous miR-34a inhibited cell growth and/or increased apoptosis in retinoblastoma cell lines (Y79, Weri-Rb1).35 Together, these data identify the miR-34 family members as potential therapeutic targets for retinoblastoma. Interestingly, among the common putative mRNA targets of both let-7 and miR-34 family members is the N-myc oncogene. N-myc is indeed a validated miR-34a target (http://mirecords.umn.edu/miRecords/) and is a predicted target of let-7i (http://www.targetscan.org/). This gene is amplified in another mouse model of retinoblastoma36 and in approximately 10% of human retinoblastomas.37 These observations raise the possibility that members of both let-7 and miR34 families cooperate to restrain N-Myc oncogenic function in retinoblastoma.

Finally and importantly, our data argue against the development of DICER1-inactivating molecules as anticancer drugs. Indeed, even if complete Dicer1 inactivation seems to be deleterious for cancer development its partial inactivation promotes, rather than inhibits, tumor formation.

Materials and Methods

Mice

All animal experiments were performed in accordance with the guidelines of the University of Ghent Animal Care and Use Ethics Committee. BrdU (100 μg/g of body weight) was injected intraperitoneally 1 h before killing.

Immunohistochemistry

Eyes were fixed overnight in 4% paraformaldehyde/PBS, and paraffin embedded. Sections of 5 μm thickness were immunostained with the following antibodies: GFP (Santa Cruz Biotechnology, Santa Cruz, CA, USA, 1/100); BrdU (BD Pharmingen, San Jose, CA, USA, 1/10); Ki-67 (DAKO, Glostrup, Denmark, 1/30); cleaved caspase-3 (Cell Signaling, Danvers, MA, USA, 1/100); syntaxin (Sigma, St Louis, MO, USA, 1/20 000); calretinin (Millipore, Billerica, MA, USA, 1/700); Chx10 (Exalpha Biologicals, Shirley, MA, USA, 1/100).

Recombination analysis

DNA was isolated from dissected retinae and isolated tumors using the DNeasy Blood & Tissue Kit (Qiagen, Germantown, MD, USA). Dicer1 recombination was analyzed by PCR using the following primers: primer a 5′-ATTGTTACCAGCGCTTAGAATTCC; primer c 5′-TCGGAATAGGAACTTCGTTTAAAC; and the reverse b primer 5′-GGGAGGT GTACGTCTACAATT. PCR conditions were as follows: 1 × precycle at 94 °C for 3 min and 30 cycles of 94 °C, 30 s; 60 °C, 30 s; 72 °C, 45 s.

microRNA expression analyses

Total RNA was prepared from dissected retinae or isolated tumors using the miRNeasy kit (Qiagen) according to the manufacturer's instructions. Profiling of these samples was performed by both microarray (miRCURY LNA Array, Exiqon, Vedbaek, Denmark) and qPCR. The RT-qPCR expression profiling of 522 murine miRNAs and subsequent data normalization were performed as described previously.38, 39 Differential miRNA expression was evaluated using Student's t-test. Hierarchical clustering was performed using Ward's method and Manhattan distance using R Bioconductor software (Oxford, UK).

Conflict of interest

The authors declare no conflict of interest.