Identification of target genes for wild type and truncated HMGA2 in mesenchymal stem-like cells

HMGA2 belongs to a family of nuclear proteins that contain an N-terminal part that recognizes and binds to AT-rich regions in the DNA [1] and an acidic C-terminal tail that probably modulates their interactions with DNA [2], and proteins [3]. HMGA proteins are not transactivators on their own, but modulate the assembly of transcriptional complexes at various levels [4‐8], and may also regulate gene transcription through protein-protein interactions without direct contact with DNA [9‐11] or by altering chromatin structure [12‐16]. Interestingly, a recent study suggested that both HMGA1 and HMGA2 are taking active parts in the formation of senescence-associated heterochromatin foci and maintenance of the growth-arrested state [17, 18].

HMGA2 is expressed during embryonic development while it is almost undetected in normal adult tissues [19, 20]. HMGA2 plays a critical role early in adipogenesis, probably by regulating the proliferation of preadipocytes during differentiation [21], and mice lacking functional HMGA2 protein exhibit a pygmy phenotype with drastic reduction of adipose tissue [22]. On the other hand, over-expression of a truncated HMGA2 protein lacking the acidic C-terminal tail results in a giant phenotype with multiple lipomas [23, 24].

Aberrations in the chromosomal region 12q14-15 that affect HMGA2 are frequent in a variety of tumours. In benign tumours of mesenchymal origin, HMGA2 is often rearranged by translocation, and the resulting chimeric transcripts are formed by fusion of the DNA-binding domains, coded by exons 1-3, to ectopic sequences [25‐27]. In sarcomas, HMGA2 is frequently and selectively amplified and rearranged [28]. We have cloned and sequenced a number of these cancer-associated ectopic sequences from 12q as well as other chromosomes [29]. The only common factor that was found was the loss of the 3'-untranslated region (UTR) as well as the last two exons, resulting in fusion proteins containing as little as one extra amino acid replacing the C-terminal part [29].

Recently it was shown that the HMGA2 3'-UTR contains target sites for the let-7 miRNA, and thus the above mentioned rearrangements lead to increased levels of HMGA2 protein due to loss of miRNA-mediated repression [30, 31]. Thus the attention has shifted from possible oncogenic effects of loss of the acidic domain to effects of increased expression, even of the wild type protein. Furthermore, it turns out that the balance between let-7 and HMGA2 governs the exit of cells from the undifferentiated and self-renewing state, and HMGA2 is now thought to be central in cancer in general [32‐35]. Because HMGA2 most frequently is rearranged in well-differentiated liposarcomas, border-line tumours resembling adipose tissue, most sarcoma cell lines, representing highly malignant cancers with a different tissue type, would not be appropriate to detect a phenotype when the gene is over-expressed. We therefore chose an immature, stem-like mesenchymal cell line with adipogenic potential to investigate whether the wild type and truncated forms of the protein activate the same pathways, by performing gene expression analysis.

The deregulation of expression caused by the absence of the 3'-UTR in our transgenes, and thus loss of let-7--mediated inhibition, resulted in abundant expression of HMGA2 protein (Fig. 1c). The eGFP fusion proteins were shown to localise in the nucleus just like the endogenous protein. Their distribution to the nucleoli (Fig. 1a, arrowheads) was unexpected though HMGA2 has previously been found to interact with the mainly nucleolar nucleophosmin/B23 protein [61]. Furthermore, the localization of endogenous HMGA2 in the nucleoli was clearly restricted to numerous discrete foci (Fig. 1a, arrow). The distribution of these sites was occasionally scattered throughout the nucleolus or more often restricted to the periphery of the nucleolar body were they seemingly co-localized with B23. In this outer part of nucleolus the late processing of rRNAs takes place before they assemble into ribosomal subunits [62], suggesting that HMGA2 might be involved in these processes. However, this could also be a result of B23 acting as a molecular chaperone [63], sequestering HMGA2 to the nucleolus, as described for IRF-1 [64], pRb [65] and TRF2 [66].

Both wild type and truncated HMGA2 abrogated growth inhibition and adipogenesis (Fig. 1d), probably by a common mechanism. It appears that both HMGA2 proteins prevented the growth arrest necessary for adipogenic differentiation, as cellular protein accumulated at the same rate as in un-induced cells. Interestingly, a recent study showed that the level of HMGA2 is highly regulated during adipogenesis by let-7 and knockdown of HMGA2 by this miRNA inhibited both clonal expansion and the transition to terminal differentiation [67]. HMGA2 seems to act in a narrow window early in the adipogenesis where it is involved in expanding the population of preadipocytes and keeps the cells in an undifferentiated state. This would be consistent with our HMGA2 over-expressing keeping the mesenchymal stem-like cells in a proliferative state, preventing differentiation into adipocytes. Transgenic mice overexpressing truncated HMGA2 gene, on the other hand, develop well differentiated lipomas and abundant adipose tissue [24]. This discrepancy is not surprising, as the conditions during culture in vitro and development in vivo are very different, among other things due to the micro¬environment. Furthermore, the absence in hMSC- TERT20 cells of p16, encoded by the INK4A/ARF locus, and possibly other oncogenic alterations in our model system [37] may be expected to contribute to the results, although this blockage of differentiation was confirmed in preliminary experiments using another, non-tumourigenic telomerase-immortalised MSC line (Stabell et al, unpublished).

It has recently been reported that HMGA2 represses transcription of INK4A/ARF [68] and suggested that this might be mediated through JUNB, a transactivator of INK4A/ARF [69]. Consistent with previous findings [3, 68], both HMGA2 forms down-regulated the expression of NF-κB regulated genes in the cells, and in addition to binding sites for NF-κB, many of the down-regulated genes identified here, including JUNB, also contain potential intrinsic binding sites for HMGA2 (Additional file 3 Table S3). These findings suggest that HMGA2 may repress JUNB by interfering with the NF-κB mediated transactivation of this gene. However, since the INK4A/ARF locus is deleted in our cell model, we were not able to confirm an effect on this gene by over-expressing HMGA2. One might speculate that the higher number of genes affected by the truncated protein could be due to a regulatory or moderating function of the c-terminal domain, but this remains to be investigated further.

In spite of abnormal dynamics of growth and differentiation of our model system, we expect the transcription response to HMGA2 overexpression to reflect authentic regulatory functions in an immature mesenchymal context, as these cells show differentiation properties similar to primary mesenchymal stem-like cell cultures.

HMGA2 is known to participate in the formation of heterochromatin in senescent cells [18] and we found that both full-length and truncated HMGA2 proteins co-localized extensively with DAPI-dense regions, probably enriched in heterochromatin (Fig. 1a). The modulation of chromatin by HMGA2 could also be responsible for the coordinated induction of the chromosome 19 KRAB-ZNF genes. Several members of this family modulate cell growth and survival, and they are implicated in malignant disorders [70], although the functions of most of them are largely unknown. Noteworthy, a recently study showed that Apak, encoded by the ZNF420 gene, repress p53-mediated apoptosis [71] and it was suggested that various KRAB-type zinc-finger proteins may modulate p53 activity. Although the level of ZNF420 transcript increased more than two-fold in cells over-expressing HMGA2, no effects on pro-apoptotic genes were observed (data not shown).

Interestingly, our most up-regulated gene, SSX1, encodes a protein that also is involved in chromatin modulation, and is already involved in sarcoma oncogenesis [72]. Members of this gene family are seen expressed in germ cells and mesenchymal stem cells, decrease upon differentiation into adipocytes and osteocytes [73, 74], and are expressed in various tumor types [75]. SSX proteins co-localize with Polycomb group (PcG) proteins in the nucleus and likely act through these chromatin regulators to maintain repression of target genes [76, 77]. This could suggest that they play a key role in maintaining mesenchymal cell stemness [78], and that increased SSX1 expression could reflect a cancer stem cell phenotype. Although the SSX gene family is highly conserved, the variable activation of SSX genes reported in malignant cells [74] suggests gene-specific transcription control. Their restricted expression pattern is mainly governed by epigenetic silencing [79, 80], which could be affected by HMGA2_TRUNC. An AT-rich region 1 kb upstream from transcription start of SSX1 contains a likely binding site for HMGA2 (^-1076 TATTAATAT), although it remains to be proven that it binds specifically to HMGA2_TRUNC. Recent findings suggest that both HMGA2 [81] and SSX proteins [73, 82] are taking part in the epithelial mesenchymal transition, where epithelial cells convert to motile mesenchymal cells [83]. However, the significance of these processes in already mesenchymal cells is unclear. Interestingly, we observed a change in both cell lines to a more restricted but still mesenchymal phenotype by the repression of some epithelial markers, such as keratin 14 (KRT14) and the P selectin ligand CD24, and up-regulation of mesenchymal-specific genes such as CD44, probably shifting from a multi-lineage potential [84] to a more restricted but proliferative state. The down-regulation of the NF-κB pathway might signify an undifferentiated mesenchymal phenotype committed to adipogenic lineage depending on further signalling cues.

The induction of CXCL12 gene expression by HMGA2_TRUNC is probably the cause of down-regulation of HLA-DRA and other major histocompatibility complex class II (MHC-II) proteins on the cell surface, due to this cytokine's ability to repress the class II transactivator [85]. The fact that interferon induced HLA-DRA protein levels to a similar level in parental and HMGA2_TRUNC cells supports this interpretation. Loss of MHC-II expression in lymphomas correlates with poor survival [86], and absence of HLA-DR expression is seen in other cancers [87]. Their lost ability to present tumor antigens on their cell surfaces may contribute to their escape from immunosurveillance and therefore this might also provide a selective advantage to sarcoma cells expressing truncated HMGA2.

Since HMGA2 truncation and amplification is observed initially in the development of well-differentiated liposarcomas, borderline tumours of adipocytic differentiation, and always in combination with amplification of the p53-blocking MDM2, we do not expect a very "malignant" phenotype when HMGA2 is overexpressed on its own. On the other hand, a phenotype where adipogenesis is inhibited would be consistent with a role in adipogenic tumours. When truncated HMGA2 is overexpressed in transgenic mice, they develop adipose hypertrophy, which indicates that adipocytic differentiation is not completely blocked [24]. Like our constructs, theirs lack the 3' UTR, thus preventing down regulation of HMGA2 by let-7 as differentiation progresses, and the most likely explanation for the discrepancy is that the balance of proliferation and differentiation over weeks or months in vivo is sufficiently different for our short-term in vitro conditions, or there are differences in how human and mouse cells are regulated at this level. Although we get the same result in another non-transformed hMSC line, this model is also based on ectopic activation of telomerase, which could affect the result[88, 89].

To investigate these questions, we will determine the activity of truncated HMGA2 in sarcoma cells by siRNA-based knock-down in cells with the natural malignant genetic background.

Our results show that over-expression of truncated HMGA2 induces a more mesenchymal (stem-like) phenotype, characterized by resistance toward differentiation, over-expression of SSX1, lost expression of certain epithelial markers and strengthened expression of mesenchymal markers. We suggest that amplification and truncation of HMGA2 in sarcoma provide the tumors with a more stem-like phenotype.