Tumorigenesis involves multistep genetic alterations. To elucidate the microRNA (miRNA)–gene interaction network in carcinogenesis, we examined their genome‐wide expression profiles in 96 pairs of tumor/non‐tumor tissues from hepatocellular carcinoma (HCC). Comprehensive analysis of the coordinate expression of miRNAs and mRNAs reveals that miR‐122 is under‐expressed in HCC and that increased expression of miR‐122 seed‐matched genes leads to a loss of mitochondrial metabolic function. Furthermore, the miR‐122 secondary targets, which decrease in expression, are good prognostic markers for HCC. Transcriptome profiling data from additional 180 HCC and 40 liver cirrhotic patients in the same cohort were used to confirm the anti‐correlation of miR‐122 primary and secondary target gene sets. The HCC findings can be recapitulated in mouse liver by silencing miR‐122 with antagomir treatment followed by gene‐expression microarray analysis. In vitro miR‐122 data further provided a direct link between induction of miR‐122‐controlled genes and impairment of mitochondrial metabolism. In conclusion, miR‐122 regulates mitochondrial metabolism and its loss may be detrimental to sustaining critical liver function and contribute to morbidity and mortality of liver cancer patients.
Hepatocellular carcinoma (HCC) is one of the most aggressive human malignancies, common in Asia, Africa, and in areas with endemic infections of hepatitis‐B or ‐C viruses (HBV or HCV) (But et al, 2008). Globally, the 5‐year survival rate of HCC is <5% and about 600 000 HCC patients die each year. The high mortality associated with this disease is mainly attributed to the failure to diagnose HCC patients at an early stage and a lack of effective therapies for patients with advanced stage HCC. Understanding the relationships between phenotypic and molecular changes in HCC is, therefore, of paramount importance for the development of improved HCC diagnosis and treatment methods.
In this study, we examined mRNA and microRNA (miRNA)‐expression profiles of tumor and adjacent non‐tumor liver tissue from HCC patients. The patient population was selected from a region of endemic HBV infection, and HBV infection appears to contribute to the etiology of HCC in these patients. A total of 96 HCC patients were included in the study, of which about 88% tested positive for HBV antigen; patients testing positive for HCV antigen were excluded. Among the 220 miRNAs profiled, miR‐122 was the most highly expressed miRNA in liver, and its expression was decreased almost two‐fold in HCC tissue relative to adjacent non‐tumor tissue, confirming earlier observations (Lagos‐Quintana et al, 2002; Kutay et al, 2006; Budhu et al, 2008).
Over 1000 transcripts were correlated and over 1000 transcripts were anti‐correlated with miR‐122 expression. Consistent with the idea that transcripts anti‐correlated with miR‐122 are potential miR‐122 targets, the most highly anti‐correlated transcripts were highly enriched for the presence of the miR‐122 central seed hexamer, CACTCC, in the 3′UTR. Although the complete set of negatively correlated genes was enriched for cell‐cycle genes, the subset of seed‐matched genes had no significant KEGG Pathway annotation, suggesting that miR‐122 is unlikely to directly regulate the cell cycle in these patients. In contrast, transcripts positively correlated with miR‐122 were not enriched for 3′UTR seed matches to miR‐122. Interestingly, these 1042 transcripts were enriched for genes coding for mitochondrially localized proteins and for metabolic functions.
To analyze the impact of loss of miR‐122 in vivo, silencing of miR‐122 was performed by antisense inhibition (anti‐miR‐122) in wild‐type mice (Figure 3). As with the genes negatively correlated with miR‐122 in HCC patients, no significant biological annotation was associated with the seed‐matched genes up‐regulated by anti‐miR‐122 in mouse livers. The most significantly enriched biological annotation for anti‐miR‐122 down‐regulated genes, as for positively correlated genes in HCC, was mitochondrial localization; the down‐regulated mitochondrial genes were enriched for metabolic functions. Putative direct and downstream targets with orthologs on both the human and mouse microarrays showed significant overlap for regulations in the same direction. These overlaps defined sets of putative miR‐122 primary and secondary targets. The results were further extended in the analysis of a separate dataset from 180 HCC, 40 cirrhotic, and 6 normal liver tissue samples (Figure 4), showing anti‐correlation of proposed primary and secondary targets in non‐healthy tissues.
To validate the direct correlation between miR‐122 and some of the primary and secondary targets, we determined the expression of putative targets after transfection of miR‐122 mimetic into PLC/PRF/5 HCC cells, including the putative direct targets SMARCD1 and MAP3K3 (MEKK3), a target described in the literature, CAT‐1 (SLC7A1), and three putative secondary targets, PPARGC1A (PGC‐1α) and succinate dehydrogenase subunits A and B. As expected, the putative direct targets showed reduced expression, whereas the putative secondary target genes showed increased expression in cells over‐expressing miR‐122 (Figure 4).
Functional classification of genes using the total ancestry method (Yu et al, 2007) identified PPARGC1A (PGC‐1α) as the most connected secondary target. PPARGC1A has been proposed to function as a master regulator of mitochondrial biogenesis (Ventura‐Clapier et al, 2008), suggesting that loss of PPARGC1A expression may contribute to the loss of mitochondrial gene expression correlated with loss of miR‐122 expression. To further validate the link of miR‐122 and PGC‐1α protein, we transfected PLC/PRF/5 cells with miR‐122‐expression vector, and observed an increase in PGC‐1α protein levels. Importantly, transfection of both miR‐122 mimetic and miR‐122‐expression vector significantly reduced the lactate content of PLC/PRF/5 cells, whereas anti‐miR‐122 treatment increased lactate production. Together, the data support the function of miR‐122 in mitochondrial metabolic functions.
Patient survival was not directly associated with miR‐122‐expression levels. However, miR‐122 secondary targets were expressed at significantly higher levels in both tumor and adjacent non‐tumor tissues among survivors as compared with deceased patients, providing supporting evidence for the potential relevance of loss of miR‐122 function in HCC patient morbidity and mortality.
Overall, our findings reveal potentially new biological functions for miR‐122 in liver physiology. We observed decreased expression of miR‐122, a liver‐specific miRNA, in HBV‐associated HCC, and loss of miR‐122 seemed to correlate with the decrease of mitochondrion‐related metabolic pathway gene expression in HCC and in non‐tumor liver tissues, a result that is consistent with the outcome of treatment of mice with anti‐miR‐122 and is of prognostic significance for HCC patients. Further investigation will be conducted to dissect the regulatory function of miR‐122 on mitochondrial metabolism in HCC and to test whether increasing miR‐122 expression can improve mitochondrial function in liver and perhaps in liver tumor tissues. Moreover, these results support the idea that primary targets of a given miRNA may be distributed over a variety of functional categories while resulting in a coordinated secondary response, potentially through synergistic action (Linsley et al, 2007).
A moderate loss of miR‐122 function correlates with up‐regulation of seed‐matched genes and down‐regulation of mitochondrially localized genes in both human hepatocellular carcinoma and in normal mice treated with anti‐miR‐122 antagomir.
Putative direct targets up‐regulated with loss of miR‐122 and secondary targets down‐regulated with loss of miR‐122 are conserved between human beings and mice and are rapidly regulated in vitro in response to miR‐122 over‐ and under‐expression.
Loss of miR‐122 secondary target expression in either tumorous or adjacent non‐tumorous tissue predicts poor survival of heptatocellular carcinoma patients.
Hepatocellular carcinoma (HCC) is one of the most aggressive human malignancies, common in Asia, Africa, and the areas with endemic infections of hepatitis‐B or ‐C viruses (HBV or HCV) (But et al, 2008). It has the third highest mortality rate among cancers worldwide and since the 1990s, the second highest cancer mortality rate in China. Globally, the 5‐year survival rate of HCC is <5% and ∼600 000 HCC patients die each year (Parkin et al, 2005; Hao et al, 2009). The high mortality associated with this disease is mainly attributed to the failure to diagnose HCC patients at an early stage and a lack of effective therapies for patients with advanced stage HCC. Although surgery remains the most effective treatment for HCC, the majority of patients are inoperable at presentation because of late diagnosis (Lee et al, 2007; Sun et al, 2007). The consequent improvement in long‐term survival of post‐surgery patients is only modest because of a high recurrence rate of intrahepatic metastases that develop through invasion of the portal vein or spread to other parts of the liver (Poon et al, 2001). Understanding the relationships between phenotypic and molecular changes in HCC is of paramount importance to develop new diagnosis and treatment of HCC and improve the prognosis of diagnosed patients.
Although expression of many mRNAs and microRNAs (miRNAs) has been shown to differ between tumor and non‐tumor tissue (Croce, 2008), the direct relevance of such differential regulation is unclear. Changes in chromosomal DNA, in expression of miRNAs, or other transcriptional/post‐transcriptional regulators may contribute to the coordinate regulation of groups of transcripts, driving tumor progression. Expression of additional miRNAs and mRNAs may subsequently change in further driving the process of carcinogenesis or in responding to it. Driver miRNAs may be distinguished from responder miRNAs by characteristics of miRNA targeting (Lim et al, 2005) in coordinately regulated mRNAs (Linsley et al, 2007). In the simplest case, miRNAs driving differential expression should be anti‐correlated with their targets, that is over‐expressed in tissues with decreased target transcript expression and under‐expressed in tissues with increased target transcript expression.
We examined mRNA‐ and miRNA‐expression profiles of tumor and adjacent non‐tumor liver tissue from HCC patients. The patient population was selected from a region of endemic HBV infection, and HBV infection appears to contribute to the etiology of HCC in these patients. About 88% of the patients in this study tested positive for HBV antigen; patients testing positive for HCV antigen were excluded. In these HCC patients, who were found to be generally under‐expressing miR‐122 relative to the control population and in mice treated with miR‐122 antagomir, miR‐122 expression was both inversely correlated with expression of genes bearing potential target sites for miR‐122 and positively correlated with expression of pathways related to liver metabolic function. These results suggest that miR‐122 regulation of transcripts may be a driver of differences in metabolic function between tumor and adjacent non‐tumor tissue, and within each tissue type and extend literature reports of the association of miR‐122‐expression levels with lipid metabolism (Jopling et al, 2005; Esau et al, 2006; 2008a, 2008b; Coulouarn et al, 2009) and suppressing the metastatic properties of HCC (Tsai et al, 2008; Coulouarn et al, 2009).
Integrating miRNA‐ and mRNA‐expression profiling
We conducted mRNA‐ and miRNA‐expression profiling on tumor and adjacent non‐tumor tissues from 96 HCC patients. Among 220 miRNAs profiled, miR‐122 is the most highly expressed miRNA in tumor and in adjacent non‐tumor liver tissue (Figure 1A), agreeing well with published work (Lagos‐Quintana et al, 2002). On average, adjacent non‐tumor liver tissue contained ∼150 000 copies of miR‐122 per 10 pg (approximately one cell) of input RNA. In addition, miR‐122 expression was decreased in tumor tissue relative to non‐tumor tissue with an average level almost two‐fold lower than in adjacent liver tissue (Figure 1B), ranking 14th out of 220 miRNAs for decreased expression in HCC by rank‐sum P‐value, and showing the largest absolute change in expression between HCC and adjacent non‐tumor tissue, confirming earlier observations (Kutay et al, 2006; Budhu et al, 2008). miR‐122 was expressed more variably in tumors than in corresponding non‐tumor tissues (the s.d. of log10 expression level is 0.30 in tumor versus 0.13 in non‐tumor). miRNAs whose expression was reduced comparably or even more significantly than miR‐122 in HCC include miR‐139, miR‐99a, miR‐10a, miR‐199a/miR‐199a*, miR‐450, miR‐378, miR‐125b, miR‐214, miR‐422b, miR‐424, miR‐451, and miR‐101. In 34 patients suffering from cirrhosis without HCC, miR‐122 was reduced approximately three‐fold relative to levels in non‐tumor tissue from HCC patients, and across in all tissues examined miR‐122 was reduced ∼25% in the presence versus in the absence of hepatitis‐B antigen expression (data not shown), suggesting that liver diseases other than HCC may also decrease miR‐122 expression.
To identify transcripts whose expression was associated with expression of particular miRNAs, we scored transcripts for correlation of their expression levels with the expression levels of 220 human miRNAs in the 96 paired tumor and adjacent non‐tumor tissues. Transcripts correlated with miRNAs with correlation coefficients >0.5 or <−0.5 were identified. Over 1000 transcripts were correlated and over 1000 transcripts were anti‐correlated with miR‐122 at this level. Among the 220 miRNAs profiled, only 13 miRNAs were found correlated with ⩾1000 transcripts and anti‐correlated with ⩾1000 transcripts.
Identification of driver miRNA
To determine which of the miRNAs might be driving expression of its correlated transcripts, we examined the sets of correlated or anti‐correlated genes for miRNA seed region enrichment in their 3′UTRs (Lim et al, 2005). We calculated the hypergeometric P‐value for enrichment of the seed sequences of each miRNA in the 3′UTRs of its correlated and anti‐correlated genes, and found 13 miRNAs with expectation or E‐value <0.01 (P‐value corrected for testing of 4096 hexamers) that remained significant after further correction for examination of 241 miRNA‐correlated gene sets (Figure 2). As a control, for each set of mRNAs correlated to a particular miRNA, we also examined enrichment for seed matches to all other miRNAs. Only miR‐122 negatively correlated genes showed more significant enrichment for seed matches to the correlated miRNA itself than for seed matches to all other miRNAs, and retained significance after correcting for examination of 241 gene sets. The miR‐122 correlated and anti‐correlated genes are displayed in Supplementary Figure 1.
Characterization of miR‐122‐associated transcripts
Transcripts anti‐correlated with miR‐122 are potential miR‐122 targets. The transcripts most highly anti‐correlated with miR‐122 expression were highly enriched for the presence of the miR‐122 central seed hexamer, CACTCC, in the 3′UTR. Among the 1106 negatively correlated transcripts, 321 of 892 annotated transcripts contained the 3′UTR hexamer CACTCC, out of 4097 total transcripts containing that hexamer. This hexamer was ranked most highly by enrichment test with a hypergeometric E‐value of enrichment of 5e−33. In general, the miR‐122 negatively correlated transcripts were enriched for cell‐cycle genes (among KEGG Pathways, ‘cell‐cycle’ ranked most highly with an E‐value of 4e−18, 35 of 199 annotated genes overlapped with 109 ‘cell‐cycle’ genes). Surprisingly, those negatively correlated CACTCC seed‐matched genes were not significantly enriched for cell‐cycle annotation (E>0.01) and did not significantly overlap with the negatively correlated cell‐cycle genes. The data suggests that miR‐122 is unlikely to directly regulate the cell cycle in these patients. Thus, the seed‐matched genes had no significant KEGG Pathway annotation.
Transcripts positively correlated with miR‐122 were not enriched for 3′UTR seed matches to miR‐122. Interestingly, these 1042 transcripts were enriched for genes coding for mitochondrially localized proteins (among GO Cellular Components, ‘mitochondrion’ ranked most highly with an E‐value=9e−33, 117 of 562 annotated genes overlapped with 785 ‘mitochondrion’ genes) and for metabolic functions (among KEGG Pathways, ‘fatty‐acid metabolism’ ranked most highly with an E‐value=5e−16, 27 of 307 annotated genes overlapped with 49 ‘fatty‐acid metabolism’ genes; ‘valine, leucine, and isoleucine degradation’ ranked No. 2, E‐value=9e−13, 23 of 307 annotated genes overlapped with 44 ‘valine, leucine, and isoleucine degradation’ genes). The great majority of transcripts involved in fatty‐acid and amino‐acid metabolism were mitochondrially localized, with highly significant overlap between all three pairwise comparisons (single‐test hypergeometric P‐values—mitochondrion/fatty‐acid metabolism overlap: P=7.4e−12; mitochondrion/valine–leucine–isoleucine degradation overlap: P=6.2e−22; fatty‐acid metabolism/valine–leucine–isoleucine degradation overlap: P=3.1e−9). Thus, miR‐122 expression appeared positively associated with mitochondrially related metabolic functions in human liver.
In vivo validation of the miR‐122‐regulated genes
We wanted to test whether the seed‐matched genes were directly targeted by miR‐122, and further, whether the transcripts encoding mitochondrially localized proteins were regulated as a consequence of miR‐122 expression independent of tumor status. To test the direct and indirect effects of loss of miR‐122 alone, in vivo silencing of miR‐122 was performed by antisense inhibition of miR‐122 (anti‐miR‐122) in wild‐type mice. Either 1 week or 4 weeks (Esau et al, 2006) post‐treatment, livers were subjected to microarray profiling (Figure 3A). We examined the consensus of multiple animals to minimize issues associated with animal‐to‐animal and array‐to‐array variations in gene expression. A total of 689 up‐regulated transcripts from the consensus of both time points were enriched for 3′UTR matches to the central miR‐122 seed hexamer CACTCC in both human and mouse miR‐122 (3′UTR hexamer ‘CACTCC’ ranked most highly by enrichment test, with an E‐value=5e−83, 350 of 639 annotated genes overlapped with 3904 CACTCC hexamer‐containing genes). As with the genes negatively correlated with miR‐122 in HCC patients, no significant biological annotation was associated with the seed‐matched genes up‐regulated by anti‐miR‐122 in mouse livers. The most significant biological annotation for 1865 consensus down‐regulated genes was mitochondrial localization (among GO Cellular Components, ‘mitochondrion’ ranked most highly, E‐value=2.5e−72, 219 of 1180 annotated genes overlapped with 663 ‘mitochondrion’ genes). Enrichment for mitochondrially localized gene products was not observed in genes down‐regulated in treatments by antagonists to other miRNAs (data not shown). The mitochondrial genes down‐regulated by anti‐miR‐122 treatment were enriched for metabolic functions (among KEGG Pathways, ‘oxidative phosphorylation’ ranked most highly, E‐value=5e−20, 53 of 540 annotated genes overlapped with 116 ‘oxidative phosphorylation’ genes; ‘valine, leucine, and isoleucine degradation’ ranked No. 2, E‐value=6e−8, 21 of 540 annotated genes overlapped with 43 genes in this set; ‘fatty‐acid metabolism’ ranked No. 3, E‐value=2e−5, 18 of 540 annotated genes overlapped with 43 genes in this set). These results confirm our observations in human HCC samples and show that loss of miR‐122 alone can up‐regulate seed‐matched genes and down‐regulate mitochondrially localized genes that are critical for metabolic functions.
Characterization of miR‐122‐regulated orthologous genes
To the extent that miR‐122 targeting is conserved from mouse to human, we can expect mouse genes up‐regulated by anti‐miR‐122 to be orthologous to human genes anti‐correlated with miR‐122 in HCC. Further, if biological pathways regulated by miR‐122 are conserved as well, mouse orthologs of genes positively correlated with miR‐122 should be down‐regulated by anti‐miR‐122. To this end, we compared the mouse signature genes with human genes anti‐correlated and correlated with miR‐122. Restricting the analysis to orthologs represented on the human and the mouse microarrays among both putative direct targets and putative downstream targets, human and mouse genes showed significant overlap for regulations in the same direction (overlap of human genes anti‐correlated with miR‐122 and mouse genes up‐regulated by anti‐miR‐122; hypergeometric P‐value=2.1e−4; overlap of human genes correlated with miR‐122 and mouse genes down‐regulated by anti‐miR‐122; hypergeometric P‐value=1.6e−29), without significant overlap for regulations in the opposite direction (hypergeometric single‐test P‐value>0.5). The overlap between genes negatively correlated with miR‐122 in HCC and up‐regulated in anti‐miR‐122‐treated mice defines a set of putative miR‐122 primary targets (Supplementary Table Ia). The overlap between genes positively correlated with miR‐122 in HCC and down‐regulated in anti‐miR‐122‐treated mice defines putative secondary miR‐122 targets (Supplementary Table Ib).
To further show that orthologous genes are regulated by miR‐122 in the same directions in both systems, we compared their average expression levels and their relationships with miR‐122 (Figure 3B). All significantly regulated transcripts on the mouse microarray that can be mapped to human orthologs were binned by the correlation of the human orthologs to miR‐122 in HCC patient tissue and the average expression ratio (HCC versus non‐tumor in human, antagomir treated versus control in mouse liver) for orthologs in each bin was computed. As expected for the human transcripts, the anti‐correlated genes were more highly expressed in tumor relative to non‐tumor, whereas the correlated genes were more under‐expressed. Similar dependence was observed in mouse liver treated by the anti‐miR‐122, even though the binning was based on correlation in HCC.
miR‐122 regulation of mitochondrial metabolic gene network
The mechanism of repression of mitochondrially localized genes by decreasing miR‐122 is not obvious. We reasoned that a direct link between the degree of induction of seed‐matched genes and the magnitude of repression of mitochondrial genes through loss of miR‐122 could be observed in a group of mice, as time, dose, and individual animal characteristics caused a range of responses. In fact, the mean regulations of seed‐matched and mitochondrial gene sets are well correlated across the varied responses in the treated mice (Figure 4A), supporting a direct link between induction of miR‐122 targets and repression of mitochondrial function.
To test whether the regulation of seed‐matched genes and mitochondrial genes in human liver was directly related across the range of patient conditions, we performed a similar analysis on a separate transcriptome profiling dataset from 180 HCC, 40 cirrhotic, and 6 normal liver tissue samples. For this test, we selected genes negatively correlated with miR‐122, seed matched in both mouse and human 3′UTRs, and mitochondrially localized genes positively associated with miR‐122 in both mouse and human samples. Expression levels of these gene sets were strongly anti‐correlated with each other in individual HCC tumors, adjacent non‐tumor tissue, and in cirrhotic livers (Figure 4B). These data suggest that the relationship observed in mouse is functionally conserved in human beings. To validate the direct correlation between miR‐122 and some of the primary and secondary targets, we determined expression of two putative direct targets, SMARCD1 and MAP3K3 (MEKK3), a target described in the literature, CAT‐1 (SLC7A1), and three putative secondary targets, PPARGC1A (PGC‐1α), succinate dehydrogenase (SDH) subunits A and B, after transfection of miR‐122 mimetic into PLC/PRF/5 HCC cells. The expressions of the seed‐matched genes, SMARCD1, MAP3K3, and CAT‐1 were reduced upon increased expression of miR‐122, and the putative secondary target genes, PPARGC1A, SDHA, and SDHB, showed increased expression (Figure 4C). In contrast, the transfection of anti‐miR‐122 resulted in an increased expression of seed‐matched genes and reduced expression of mitochondrial genes (Supplementary Figure 2).
Function of miR‐122 in mitochondrial energy metabolism
We used functional classification of genes by total ancestry (Yu et al, 2007) to investigate whether common functions might link expression of miR‐122 targets with mitochondrial function. We first defined a set of putative miR‐122 targets: genes negatively correlated with miR‐122 expression in HCC tumor and adjacent tissue with correlation coefficients <−0.4, having mouse orthologs up‐regulated by anti‐miR‐122 with mean log10 expression ratios >0.09 and geometric mean P‐values of up‐regulation <0.05, as calculated by a microarray error model (Weng et al, 2006) over multiple animal treatments. Next, we defined a set of putative miR‐122 secondary targets, applying the opposite criteria to select genes positively correlated with miR‐122 in human tissue and down‐regulated by anti‐miR‐122 in mouse. We then identified all miR‐122 secondary targets connected to miR‐122 targets by functional similarity (Figure 5A). The most connected secondary target, with 27 functional similarities, was PPARGC1A (PGC‐1α). Functional similarity was frequently associated with more direct connections. For example, among the putative primary targets connected to PPARGC1A, MED1 (TRAP220) (Wallberg et al, 2003), and SMARCD1 (BAF60a) (Li et al, 2008) are fellow transcriptional co‐activators; and LCMT1 (PPM1) (Leulliot et al, 2004), PPP1CC, ATF4 (CREB2), MAP3K3 (MEKK3), and MAPKAP2 (MK2) (Wu et al, 2001) may be involved in regulation of PPARGC1A expression. As mentioned previously, transfection of miR‐122 mimetic could increase PPARGC1A expression. To further validate the link of miR‐122 and PGC‐1α protein, we transfected PLC/PRF/5 with miR‐122‐expression vector, and the increase of PGC‐1α protein levels was observed (Figure 5B). Importantly, transfection of both miR‐122 mimetic and miR‐122‐expression vector led to a significant reduction of lactate content in PLC/PRF/5 cells (Figure 5C), whereas the anti‐miR‐122 treatment led to an increase of lactate production in the PLC/PRF/5 cells (Supplementary Figure 3). Together, the data support the function of miR‐122 in mitochondrial metabolic functions.
miR‐122‐regulated pathways contribute to patient survival
As anti‐miR‐122‐treated animals tolerate the loss of miR‐122 well (Jopling et al, 2005; Krützfeldt et al, 2005; Budhu et al, 2008; 2008a, 2008b), the question remained as to the relevance of loss of miR‐122 function to morbidity and mortality for HCC patients. Indeed, miR‐122‐expression levels on their own were not significantly related to patient survival at the P<0.05 level (log‐rank P‐value=0.09 in tumor tissue and 0.17 in adjacent non‐tumor tissue). However, when comparing between surviving and deceased patients, the miR‐122 secondary targets were expressed at significantly higher levels in both tumor (ANOVA P‐value=0.02) and adjacent non‐tumor (ANOVA P‐value=0.006) tissues among the survivors. Patients in the bottom 50% for expression of the miR‐122 secondary targets in either tumor or adjacent non‐tumor tissue showed significantly shorter survival times (Figure 6). As there is no correlation between tumor and adjacent non‐tumor samples from the same patient of miR‐122 levels (r=0.02) or of miR‐122 secondary target levels (r=−0.06), prognostic significance of each tissue appears independent.
Normal liver function includes building biomolecules for export to consumer tissues of the body (Yokoyama et al, 2005). This catabolism requires energy and liver tissue is rich in mitochondria. Loss of mitochondrial function may in turn be associated with loss of liver function. Dysregulation of normal mitochondrial functions may also contribute to cancer metabolism and hepatocarcinogenesis, as the connection between mitochondrial dysfunction and cancer is well known (Brandon et al, 2006). It is possible that loss of miR‐122 expression in HCC may contribute to loss of liver function, a contributor to morbidity and mortality in HBV‐associated HCC. We have shown that loss of expression of miR‐122 positively correlated genes predicts poor survival for HCC patients. In our patient cohort, expression levels of miR‐122‐regulated pathways in both tumor and adjacent non‐tumor tissue appear to be independently functioning as good markers of patient prognosis.
Although miR‐122 negatively correlated genes were enriched for cell‐cycle function in HCC (E‐value=8e−18), the miR‐122 seed‐matched transcripts showed only random overlap with cell‐cycle transcripts (single‐test hypergeometric P‐value=0.78). These observations suggest that there is likely no direct connection between up‐regulation of potential miR‐122 targets and up‐regulation of the proliferative apparatus in tumor tissue. Further, we did not find strong prognostic power for patient survival in miR‐122‐expression levels, although a smaller study has (Coulouarn et al, 2009). We also did not find strong prognostic power in expression levels of proposed miR‐122 direct targets, whereas expression of the putative secondary targets, transcripts positively correlated with miR‐122, is predictive of patient survival. Two possible implications of these observations are that the downstream target pathways are more connected to clinical outcome than the primary targeting mechanisms and that downstream pathways may form a more sensitive measure of miRNA activity than direct measurements of miRNA‐expression levels (Davis et al, 2009).
Genes that were up‐ and down‐regulated in HCC and in anti‐miR‐122‐treated mice (putative primary and secondary miR‐122 targets), appear to function as a biological network. A number of these genes are connected by functional similarity as determined by the total ancestry method of functional classification analysis (Yu et al, 2007). In addition, a number of proposed primary and secondary target genes found in this work have been previously reported to be associated in various functional networks. The networks associated with statin (an HMG‐CoA reductase inhibitor to lower cholesterol levels) treatment, high‐fat feeding or fasting, expression of the obesity causal gene Zfp90, and with cohesive gene expression in normal liver overlap significantly with miR‐122 primary and secondary targets (2005, 2008; Chen et al, 2008).
PPARGC1A is the miR‐122 secondary target most connected by functional similarity to genes up‐regulated by loss of miR‐122 in both human HCC and anti‐miR‐122‐treated mice. PPARGC1A transcription is activated by cAMP‐response element‐binding proteins (CREB) (Wu et al, 2001). Among those genes connected to PPARGC1A in this study, PPP1CC negatively regulates CREB, and LCMT1 (Leulliot et al, 2004) activates PPP2A, which inhibits CaM‐kinase activation of CREB and MEK/ERK signaling in the MAP‐kinase pathway (Wu et al, 2001). The MAP‐kinase pathway represented by MAP3K3 and MAPKAP2 among genes up‐regulated by loss of miR‐122 leads to phosphorylation of PPARGC1A that both activates the protein and enhances SCF/Cdc4‐mediated degradation of the protein (Olson et al, 2008). These published observations suggest that multiple miR‐122 targets may contribute to the down‐regulation of PPARGC1A observed with loss of miR‐122.
PPARGC1A is proposed to be the master regulator of mitochondrial biogenesis (Ventura‐Clapier et al, 2008), suggesting that loss of PPARGC1A expression may contribute to the loss of mitochondrial gene expression correlated with loss of miR‐122 expression. PPARGC1A‐over‐expressing mouse strains show uncontrolled mitochondrial biogenesis (Lehman et al, 2000), whereas PPARGC1A‐knockout mouse strains show decreased expression of mitochondrial genes with strain‐dependent compensatory phenotypes (Benton et al, 2008). PPARGC1A and HNF‐4 are thought to act together to stimulate cholesterol biosynthesis (Rodgers and Puigserver, 2007), so loss of PPARGC1A may contribute to the reduction in plasma cholesterol seen in anti‐miR‐122‐treated animals (Jopling et al, 2005; Esau et al, 2006; 2008a, 2008b). SMARCD1(BAF60a), which stimulates fatty‐acid oxidation in conjunction with PPARGC1A without changing its expression level (Li et al, 2008), is proposed to be a primary target of miR‐122 in this study and in a recent publication (Gatfield et al, 2009), suggesting that increased fatty‐acid oxidation seen with miR‐122 depletion (Esau et al, 2006; Gatfield et al, 2009) may be a direct effect. Another study found that reduction of miR‐122 levels in non‐alcoholic steatosis was associated with increased expression of lipogenic genes (Cheung et al, 2008). In anti‐miR‐122‐treated mice in this study, expression levels for genes encoding cholesterol biosynthesis and lipid metabolism are not tightly anti‐correlated with expression levels of miR‐122 seed‐matched genes (data not shown), suggesting that the observed effects of anti‐miR‐122 on cholesterol and lipid synthesis may be further downstream of miR‐122. Microarray studies of anti‐miR‐122 in high‐fat‐fed mice may be of interest in elucidating the connection between miR‐122 and cholesterol biosynthesis.
Interestingly, down‐regulation of SDH subunits A and B is associated with loss of miR‐122 in both mouse and human tissues in this study. Increased expressions of both the genes were detected in cell culture after treatment with miR‐122 mimetic. Loss‐of‐function mutations in genes encoding subunits B, C, or D of SDH lead to loss of mitochondrial function and to hereditary paraganglioma or in the case of SDHB, also to phaeochromocytoma or renal cell carcinoma (King et al, 2006). A decline in SDH function concomitant with the loss of miR‐122 may thus increase the risk of oncogenesis.
Recently, miR‐122 has been suggested to act as a tumor suppressor in HCC (Tsai et al, 2008; Coulouarn et al, 2009). In the study of Tsai et al, a combination of bioinformatics and tumor profiling was used to identify 45 genes as potential miR‐122 targets. About half of the proposed targets were anti‐correlated with miR‐122 in the tumor and non‐tumor tissues profiled in our study. A total of 11 of the genes were also up‐regulated in the anti‐miR‐122‐treated mouse livers we profiled, suggesting cross‐species conservation of the regulations these authors identified. However, ADAM17, followed up in more depth by Tsai et al, was not significantly anti‐correlated with miR‐122 in our profiles. The study of Coulouarn et al found gene‐expression clusters associated with high and low miR‐122 levels in 32 HCC tissue samples. Genes whose increased expression was associated with lower miR‐122 levels in these samples included predicted miR‐122 targets and genes up‐regulated in anti‐miR‐122‐treated mice, whereas genes whose increased expression was associated with higher miR‐122 levels were enriched for lipid metabolism functions and were more likely to be well expressed in control mice. This study emphasized the function of HNF1A and HNF3, transcription factors mediating hepatocyte differentiation and liver functions, in potential regulation of miR‐122 expression. Our study is unable to support a primary or secondary function for these genes. HNF3 components were uncorrelated with miR‐122 in tumor and non‐tumor tissues profiled in our study and were not consistently regulated in anti‐miR‐122‐treated mouse livers. HNF1A showed no significant relationship to miR‐122 in tumor and non‐tumor profiles; we have no data on mouse expression.
Taken together, our results imply that normal mitochondrial function in liver, including expression of mitochondrion‐associated metabolic pathways, may be maintained in part by miR‐122 expression. Impaired mitochondrial functions are observed in many tumor types, suggesting an alternate possibility that the observed decline in mitochondrial function in HCC may be tumor related rather than miRNA related (Jopling et al, 2005). Our observations that mitochondrial function pathways and miR‐122 levels also decline coordinately in cirrhotic liver and in anti‐miR‐122‐treated mouse livers argue against this explanation.
Other connections between loss of miR‐122 expression and changes in liver function have been proposed. CAT‐1 (SLC7A1) was shown to be a direct target of miR‐122 (Chang et al, 2004; Jopling et al, 2006), and although it is negatively correlated with miR‐122 levels in HCC in this study, it is unregulated in anti‐miR‐122‐treated mouse livers profiled herein. Bcl‐w, recently found to be targeted by miR‐122 (Lin et al, 2008), was negatively correlated with miR‐122 levels in HCC in this study and is up‐regulated in anti‐miR‐122‐treated mouse livers, supporting a pro‐apoptotic function for miR‐122 in HCC and indicating a survival advantage to its down‐regulation in HCC. Expression of miR‐122 precursors is known to be circadian; in a recent study, eight genes were identified as showing circadian accumulation in microarray experiments, showing up‐regulation in mouse livers treated with anti‐miR‐122 and having 3′UTRs down‐regulated by miR‐122 mimetics (Gatfield et al, 2009). In total, 11 other genes were identified as showing up‐regulation by anti‐miR‐122 and having 3′UTRs down‐regulated by miR‐122 mimetics, but without circadian accumulation. In this study, 13 of these 19 genes were up‐regulated by anti‐miR‐122 treatment in mice, including 7 of the 8 genes showing circadian accumulation. However, only one gene, SMARCD1 (BAF60a), also showed expression negatively correlated with miR‐122 expression in our HCC samples, emphasizing the importance of cross‐species analysis. Cyclin G1, found by others (Gramantieri et al, 2007) to be anti‐correlated with miR‐122 in HCC, was not significantly correlated or anti‐correlated with miR‐122 in our samples. Similarly, although N‐myc has been suggested as a target of miR‐122 (Girard et al, 2008), its expression levels are not significantly correlated or anti‐correlated to miR‐122 levels in this study.
Other published studies indicate that miR‐122 is a host factor for HCV replication (2006, 2005; Shan et al, 2007; Chang et al, 2008; Henke et al, 2008; Lupberger et al, 2008) and show that HCV‐infected HCC patients usually do not show a reduction in and may in fact show an increase in miR‐122 expression (Varnholt et al, 2008), although higher miR‐122 levels have also been shown to predict better response of HCV patients to standard therapy (Sheikh et al, 2008). HCV appears to target mitochondria directly causing liver dysfunction (Sarasin‐Filipowicz et al, 2009) by a route not dependent on decreasing miR‐122 expression.
The primary targets of an miRNA may be distributed over a variety of functional categories while resulting in a coordinated secondary response, potentially through synergistic action (Linsley et al, 2007). In our study, we found increased lactate production in tissue culture cells after treatment with anti‐miR‐122. Reduced mitochondrial oxidative phosphorylation is commonly observed in cancer cells, and we postulate that the reduced expression of miR‐122 may contribute to this effect in HCC. In light of the observed connection between miR‐122 expression and mitochondrial function pathways in liver, we speculate that increasing miR‐122 expression may possibly improve mitochondrial function in liver and perhaps in liver tumor tissues. Therefore, it is of great interest to determine the phenotypic changes of HCC after miR‐122 targeting delivery in our established HCC mouse models (Zender et al, 2008; Liu et al, 2009).
Our findings reveal potential new biological functions of miR‐122 in liver physiology. We have observed the decrease of miR‐122, a liver‐specific miRNA, in HBV‐associated HCC, and loss of miR‐122 appears to correlate with the decrease of mitochondrion‐related metabolic pathway gene expression in HCC and in non‐tumor liver tissues, a result that is consistent with the outcome of treatment of mice with anti‐miR‐122 and is of prognostic significance for HCC patients. Further investigation will be conducted to dissect the regulatory function of miR‐122 on mitochondrial metabolism in HCC.
Materials and methods
For human HCC patient samples, resected tumor and adjacent non‐tumor liver tissues were collected from patients who had undergone hepatectomy for curative treatment of HCC at Queen Mary Hospital, Pokfulam, Hong Kong between 1990 and 2007. Informed consents were obtained from patients regarding the use of the liver specimens for research. Demographic and clinicopathologic features are summarized in Supplementary Table II and elsewhere (Yi et al, 2008; Hao et al, 2009; Lee et al, 2009; Xu et al, 2009).
Animal care and treatments
All animal experiments were conducted according to the Institutional American Association for the Accreditation of Laboratory Animal Care Guidelines. Male C57BL/6 mice were obtained from The Jackson Laboratory and were housed four to five animals per cage with a 12‐h light/dark cycle. Oligonucleotides were dissolved in saline and administered to mice based on body weight by intraperitoneal injection. In one treatment regimen, animals were given four doses of 100 mg ASO/kg body weight, administered every other day, and killed 1 day after the last dose. Animals treated for 4 weeks were treated twice weekly with 50 mg ASO/kg body weight, and killed 2 days after the last dose. Mice were killed in the morning, and liver was removed for further analysis. There were no statistically significant elevations in plasma transaminase levels after either treatment regimen and no significant histological findings.
Nucleic‐acid reagents used in this study were as follows. The miR‐122 mimetic consisted of a 5′‐UGGAGUGUGACAAUGGUGUUUG‐3′ guide strand annealed to a [iB]AACACCAUUGUCACACUCGAAU[iB] passenger strand, where iB represents an inverted abasic cap. The Let‐7 mimetic consisted of a 5′‐UGAGGUAGUAGGUUGUAUAGUU‐3′guide strand annealed to a [iB]CUAUACAACCUACUACCUGAAU[iB] passenger strand. For in vitro experiments, the anti‐miR‐122 antagomir consisted of a single strand, 5′‐dCs;dCs;lnaAs;dTs;dTs;lnaGs;dTs;dCs;lnaAs;dCs;dAs;lna‐5methylCs;dTs;dCs;lna‐5methylCs;dAs‐3′. The control antagomir consisted of a single strand, 5′‐lna‐5methylCs;dCs;lnaAs;dTs;dTs;lna‐5methylCs;lnaTs;dCs;dAs;lna‐5methylCs;dAs;lna‐5methylCs;dTs;lnaGs;lna‐5methylCs‐3′. For in vivo experiments, antagomirs were 2′‐O‐methoxyethyl (2′‐MOE) phosphorothioate modified. The anti‐miR‐122 antagomir consisted of a single strand, 5′‐ACAAACACCATTGTCACACTCCA‐3′. The control antagomir consisted of a single strand, 5′‐CCTTCCCTGAAGGTTCCTCCTT‐3′.
Total RNA was purified from mouse in vivo samples using an RNeasy kit (Qiagen, Valencia, CA). RNA from anti‐miR‐122‐treated mice was compared with RNA from mice injected with phosphate‐buffered saline. Three to five mice were analyzed for each treatment or control group. Microarray hybridizations were performed as described (Jackson et al, 2003). Isolation of RNA from human samples was achieved using the following procedures. The milled tissue samples were homogenized in cryopreservation tubes with a vortex mixer after addition of 750–1000 μl of TRIzol reagent. Chloroform was added to the TRIzol/GITC lysate (1:5) to facilitate separation of the organic and aqueous components using the phaselock (Eppendorf) system. The aqueous supernatant was further purified using the Promega SV‐96 total RNA kit, incorporating a DNase treatment during the procedure. Isolated total RNA samples were then assayed for quality (Agilent Bioanalyzer) and yield (Ribogreen) metrics. Hybridization, labeling, and scanning of microarrays were completed following the manufacturer's recommendations (Affymetrix). Human samples were profiled on a custom Affymetrix array, RM‐HU01Aa520485 RSTA Custom Affymetrix 1.0. Mouse samples were profiled on a custom Agilent array, RSTA Mouse 3.0 A1. The miRNA profiling was performed by custom quantitative PCR assays as described (Raymond et al, 2005). Microarray data transformation and analysis was performed as described (Irizarry et al, 2003; Jackson et al, 2003; Eklund et al, 2006). Upon publication of this manuscript, microarray data will be available in GEO (http://www.ncbi.nlm.nih.gov/geo/info/linking.html), under accession number GSE22058 (released on 4 June 2010).
Regulated transcripts were identified in microarray gene‐expression signatures using a P‐value cut‐off (P<0.01 or P<0.05, as specified). No cuts were placed on fold change in expression unless specified. Data were analyzed using Rosetta Resolver™ and MATLAB™ software. Regulated transcripts were tested for enrichment of transcripts belonging to gene sets in the GO Cellular Component (Ashburner et al, 2000) and KEGG Pathways (Kanehisa et al, 2008) annotation sources or of transcripts containing one or more copies of a hexamer sequence in the 3′UTR relative to a background set (i.e. the set of genes represented on the microarray) using the hypergeometric distribution. The P‐value for enrichment of each gene set in an annotation source or of each of 4096 possible hexamers was subjected to Bonferroni's correction.
PLC/PRF/5 and MHCC‐97L cells were transfected with 10 nM of either let‐7 or miR‐122 mimetic, and cell lysates were collected 48 h post‐transfection using TRIzol (Invitrogen, Carlsbad, CA). For the miR‐122‐knockdown assay, PLC/PRF/5 cells were transfected with 80 mM of either miR‐122 antisense or control miR‐122 antisense, and cell lysates were collected 48 h post‐transfection using TRIzol (Invitrogen). Real‐time qPCR assays were performed to evaluate the expression levels of CAT‐1, SMARCD1, MAP3K3, PPARGC1A, SHDA, and SDHB using Power SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA). For detection of miR‐122 expression, TaqMan MicroRNA Assay specific for miR‐122 (Applied Biosystems) was used, and procedures were in accordance with manufacturer's protocol. The numbers of miR‐122 copies in samples were quantified with a standard curve.
The miR‐122‐expression vector was generated by cloning of the genomic region containing mature miR‐122 sequence and its 5′‐ and 3′‐flanking sequence into pcDNA3.1 vector (Invitrogen). PLC/PRF/5 cells were transfected with 2 μg per well of purified miR‐122 plasmids in six‐well plates using Lipofectamine 2000 (Invitrogen) as previously described (Liu et al, 2010). For western blot analysis, cell lysates were prepared in RIPA buffer 48 h post‐transfection and probed with a rabbit monoclonal antibody (clone 3G6) against PGC‐1α (1:1000 dilution; Cell Signaling Inc., Beverly, MA), and the blot was developed using chemiluminescent detection substrate solution (Millipore, Billerica, MA).
Cellular lactate level of PLC/PRF/5 cells after miR‐122 antagomir treatment or ectopic expression of miR‐122 was determined using a commercial lactate assay kit (BioVision, Mountain View, CA) as described by Christofk et al (2008). All procedures were in accordance with the manufacturer's instruction.
The work is partly supported by the small project fund of the Hong Kong University, NMRC Block Vote, and Start‐up fund of the National University of Singapore to JML. We gratefully acknowledge Sheung‐Tat Fan of Queen Mary Hospital for his expert clinical advice in HCC, Douglas Bassett and Alan Sachs of Merck and Co., Inc. for guidance and support of these studies, and C Frank Bennett and Christy Esau of Isis Pharmaceuticals, Inc. for provision of the anti‐miR‐122‐treated mouse liver samples and for helpful discussions. In addition, we thank Ke Hao, Tao Xie, Steven Bartz, Walter Strapps, Lyndon Mitnaul, Luiz Miguel Camargo, Robert Phillips, Peter Shaw, Eric Schadt, Peter Linsley, and Carolyn Buser‐Dopner of Merck and Co., Inc. for scientific input and helpful discussions.
Conflict of Interest
The authors declare that they have no conflict of interest.
Supplementary Table 1 [msb201058-sup-0001.xls]
Supplementary Tables SI–II, Supplementary figures S1–3 [msb201058-sup-0002.pdf]
Source data for figure 3A [msb201058-sup-0001-SourceData-S1.txt]
Source data for figure 5 [msb201058-sup-0002-SourceData-S2.txt]
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.
- Copyright © 2010 EMBO and Macmillan Publishers Limited