Background
HMGA2 belongs to the high-mobility group A family of non-histone chromosomal proteins. With its three AT-hooks it can bind to the minor groove of AT-rich regions of DNA, affecting the transcription of target genes by modulating the DNA structure and facilitating or inhibiting the organization of enhanceosomes [
1‐
3]. The HMGA2 protein has been shown to have important functions in cellular growth and differentiation, with a particular impact on the progression of the undifferentiated mesenchyme during fetal development [
4].
Hmga2 knockout mice display reduced body weight with decreased fat levels as well as impaired fibroblast proliferation, whereas transgenic mice expressing activating mutations show somatic overgrowth with gigantism and lipomatosis [
5‐
7]. Similarly, a boy with a constitutional chromosomal rearrangement resulting in deregulation of
HMGA2 displayed a number of features indicative of disturbed mesenchymal differentiation, such as marked somatic overgrowth, skeletal abnormalities and multiple lipomas [
8].
Limited studies on the expression patterns of the
HMGA2 gene have shown that it is active in most tissues during fetal development, and that it is transcriptionally silent, or at least expressed at undetectable levels, in most adult differentiated tissues [
9]. Aberrant deregulation, however, has been identified in a number of neoplasms, notably of mesenchymal origin [
2]. In benign mesenchymal tumors, such as lipoma, the deregulation is often due to a cytogenetically visible rearrangement of the
HMGA2 locus in 12q15, resulting in the expression of a truncated protein lacking the acidic C-terminus encoded by exons 4 and 5, a fusion protein consisting of the AT-hook domains encoded by exons 1–3 and more or less extensive parts from another gene, or, less commonly, expression of a full-length protein; apparently, all three forms – truncated, fusion and full-length – of the HMGA2 protein have transforming potential when expressed in the relevant cell type [
6,
7,
10‐
15].
The loss of the C-terminal domains was long thought to be the cause of neoplastic transformation, but the finding that some tumor-associated translocations disrupt the 3'-untranslated region (3'-UTR) instead of the open reading frame (ORF), resulting in expression of the full-length HMGA2 protein, suggested that separation of repressive sequences in the 3'-UTR from the 5'-part of the gene could be an important mechanism behind transcriptional up-regulation [
12,
15,
16]. Indeed, the 3'-UTR of
HMGA2 is known to contain multiple binding sites for the
let-7 family of microRNAs, small non-coding RNAs that inhibit gene expression at the posttranscriptional level, and it was recently shown that targeted mutation of these binding sites as well as functional inactivation of
let-7 results in up-regulation of
HMGA2 by reducing
HMGA2 mRNA degradation in the cytoplasm [
17,
18].
Approximately 75% of conventional lipomas harbor chromosomal rearrangements of 12q13-15, strongly indicating the involvement of
HMGA2 in these cases [
19]. The remaining lipomas display other aberrations, including recurrent translocations affecting band 6p21, which harbors the
HMGA1 gene, deletions of 13q, and supernumerary ring chromosomes. The importance of
HMGA2 expression in these lipomas and other lipomatous tumors, remains poorly investigated. Over-expression of full-length or truncated
HMGA2 has been demonstrated in some lipomas without 12q-rearrangement, as well as in atypical lipomatous tumors, but no systematic analysis of the status of
HMGA2 in different cytogenetic subsets of lipomas or in other lipomatous tumors has been performed. In the present study, we used fluorescence in situ hybridization (FISH) and quantitative RT-PCR (qRT-PCR) to study the genomic status and expression level of full-length and truncated
HMGA2 in various adipocytic tumors, including conventional lipomas, angiolipomas, spindle cell lipomas, hibernomas, atypical lipomas, well-differentiated liposarcomas, and myxoid liposarcomas. All cases expressing full-length
HMGA2, as well as a few cases expressing truncated
HMGA2 that served as controls, were further analyzed with regard to mutations in the 3'UTR.
Discussion
Among the 35 conventional lipomas that were analyzed here, four expression patterns were observed: (1) strong (>100-fold) expression of the full-length gene (exons 1–2 as well as 4–5) or (2) a truncated gene (only exons 1–2; differential expression), (3) moderate (10–100-fold) expression of the entire gene, and (4) weak/absent expression of any part of the gene. Although we cannot fully exclude that low expression in some case was due to admixture of normal cells, it is of interest to note that the type of expression pattern varied with cytogenetic subgroup. All five lipomas with t(3;12) showed strong (log10 ratio between 2.72 and 3.15), differential expression. The t(3;12) is the most common cytogenetic aberration in conventional lipoma, resulting in an
HMGA2/LPP fusion gene. In cases reported in the literature, the fusion gene always contains exons 1–3 of
HMGA2 and, usually, exons 9–11 of
LPP, which is at odds with the finding of increased (log10 ratios 1.04 and 1.49) expression of exons 4–5 in two of our cases. However, a few reported lipomas with t(3;12) have also expressed the reciprocal
LPP/HMGA2 chimera and some tumors with rearrangements of
HMGA2 express the wild type allele [
14,
28‐
31].
In contrast, only one of five lipomas with a t(5;12), the second most common translocation in conventional lipomas, showed differential expression in the present study; the other four cases all showed aberrant expression of the entire gene. This is in line with previous FISH data on this particular translocation, demonstrating that the genomic breakpoints usually occur outside the
HMGA2 locus [
25]. The only case here with differential expression also had a split
HMGA2 locus at FISH analysis, supporting the view that the molecular outcome of the t(5;12) is heterogeneous. Cytogenetic analyses of conventional lipomas have identified numerous other translocation partners to the
HMGA2 locus in 12q [
19]. Although only a few of these partners have been examined at the molecular level, it is well known from FISH studies that the breakpoints may occur inside as well as outside
HMGA2. Thus, it was not unexpected that all five lipomas with variant 12q13-15 rearrangements in the present study, representing five different recombination partners, showed overexpression of
HMGA2 and that some of them had expression of the full-length gene whereas others had differential expression. Interestingly, strong expression of the full-length gene was seen in one case with a t(1;12)(p32;q15) and a split FISH signal, with the 3'-part of
HMGA2 translocated to chromosome 1. Combined, these findings indicate that this particular translocation led to the fusion of
HMGA2 with a strongly expressed gene in chromosome 1. However, high expression of the other, wild type, allele cannot be excluded.
The molecular pathways in conventional lipomas with rearrangements of 6p or deletions of 13q remain to be clarified. It is commonly believed that rearrangements of the
HMGA1 gene, located in band 6p21, could substitute for
HMGA2 rearrangements, but this has been verified in only a small number of cases, [
19,
32‐
34] and no candidate target for 13q-deletions has been identified. The notion that the molecular pathogenesis of these lipomas is different from that in 12q-positive ones was strengthened by the finding here that only one case in each group showed strong
HMGA2 expression. Further, indirect support for the hypothesis that most lipomas with 6p21 or 13q-rearrangements develop through pathways that do not involve
HMGA2 could be found among the lipomas that showed aberrations that do not belong to any of the known cytogenetic subgroups of conventional lipomas; four of five showed aberrant (log10 ratio 1.80–2.17) expression of the 5'part or of the entire gene. Thus, most of the non-recurrent cytogenetic aberrations observed are probably secondary changes, accrued during tumor growth.
Ring chromosomes and giant markers (hereafter referred to as rings) constitute the cytogenetic hallmark of atypical lipomatous tumors, and it has been shown that these rings consistently contain amplified sequences from chromosome 12 [
35‐
38]. Occasionally, also tumors diagnosed as conventional lipoma display rings, and it has been debated whether these cases represent highly differentiated atypical lipomatous tumors or whether some are true conventional lipomas with a similar genetic profile [
19,
36,
39‐
42]. The former interpretation seems the more likely; not only do tumors diagnosed as conventional lipomas with rings share the clinical characteristics of atypical lipomatous tumors (male predominance, larger size, preferential location in the thigh), but the genomic contents of the rings are the same [
42‐
44]. In the present study, the expression of
HMGA2 was increased (full-length in three and differential in two) in all five cases diagnosed as conventional lipomas with rings. However, the expression levels (median log10 ratio for exons 1–2, 2.80) were more similar to those seen in lipomas with t(3;12) and t(5;12) (median log10 ratios 3.00 and 3.13, respectively) than in tumors diagnosed as atypical lipomas and well-differentiated liposarcomas (median log10 ratios 3.46 and 3.55, respectively), whereas the fraction of lesions with differential expression was similar in all three tumor entities. This raises the interesting possibility that the level of cellular atypia may be influenced by the level of
HMGA2 expression in lipomatous tumors with rings. However, the rings are highly complex chromosome aberrations and it cannot be excluded that the amplification and expression levels of other genes influence the phenotype.
Whereas atypical lipomatous tumors in the extremities seldom recur after surgery and only rarely transform into high-grade lesions, morphologically and cytogenetically identical lesions in the retroperitoneum commonly dedifferentiate, eventually often killing the patient. To emphasize these differences in clinical behaviour, the former are often referred to as atypical lipomas, and the latter as well-differentiated liposarcomas. In the present study, we analyzed ten atypical lipomatous tumors of the extremities and five of the retroperitoneum. They consistently showed high expression of
HMGA2, arguing for the importance of deregulated
HMGA2 expression also in this subtype of lipomatous tumors [
36,
45].
Taken together, these results demonstrate that all 35 lipomatous tumors with cytogenetically visible 12q13-15 rearrangements – translocations, inversions, rings – have aberrant expression of HMGA2, and that some of these rearrangements, like the t(3;12), consistently result in differential expression of the first three exons whereas other rearrangements, like the t(5;12), most often lead to expression of the entire gene. The lack of HMGA2 expression in most cases with rearrangements of 6p21 or deletions of 13q strongly support the view that these lipomas develop through other genetic pathways. The consistently high expression of HMGA2 in conventional lipomas with ring chromosomes provides further support for the idea that these tumors belong to the same biologic entity as atypical lipomas and well-differentiated liposarcomas.
As expected, none of the five myxoid liposarcomas, which all had a t(12;16) resulting in a FUS-DDIT3 fusion gene (data not shown), expressed HMGA2. Similarly, most hibernomas and spindle cell lipomas, which at the chromosome level typically show translocations affecting chromosome band 11q13 and deletions of 13q, respectively, did not show any deregulation of HMGA2. However, one case each demonstrated moderate expression of the full-length gene (hibernoma, log10 ratio exons 1–2, 1.60) or only of the amino-terminal part (spindle cell lipoma, log10 ratio exons 1–2, 1.76). Although both cases were morphologically typical, it is well known that hibernomas and spindle cell lipomas may contain more or less extensive areas resembling conventional lipomas, presumably reflecting differentiation. Possibly, such areas might be associated with deregulated expression of HMGA2.
Angiolipomas often occur as multiple tumors, and an autosomal dominantly inherited predisposition is apparent in some families [
46,
47]. Repeated cytogenetic analyses have, with a single exception, resulted in normal karyotypes [
23], suggesting that they are either non-neoplastic or that they arise through submicroscopic mutations [
46]. In the present study, five cases could be analyzed with regard to expression level of
HMGA2, two of which were also studied by FISH; none of the latter showed any cryptic rearrangement of the
HMGA2 locus. In all cases, the expression levels were highly similar for exons 1–2 and 4–5, and the median expression level (log10 ratio 1.10 for exons 1–2) was higher than that in myxoid liposarcomas, hibernomas, spindle cell lipomas, and lipomas with 13q-deletions or 6p21-rearrangements, but lower than that in conventional lipomas with cytogenetically visible 12q-rearrangements. Thus, it seems as if also angiolipomas, at least to some extent, are associated with aberrant
HMGA2 expression, but not on the basis of gross chromosomal rearrangements.
To study the role of the 3'UTR for deregulated
HMGA2 expression, RT-PCR was performed in 18 lipomatous tumors, including nine with aberrant expression of the full-length gene, four with differential expression, and five with weak or unknown expression (Table
1). The 3'UTR of
HMGA2 is 2996 bp long, according to the sequence reported in 1996 by Ashar et al. [GenBank:NM_003483][
48]. It contains a number of potential termination sites before the one that ends the reported full-length UTR. In addition, there are 10 AUUUA elements, which are thought to be important for deadenylation and subsequent decay of the mRNA [
15]. We are not aware of any previous, systematic
in vivo studies of the expression of the 3'UTR, neither in tumors nor in non-neoplastic tissues, but several experimental studies have suggested that the 3'UTR plays an important role in the posttranscriptional regulation of
HMGA2. In 2001, Borrmann et al. measured the luciferase activity of the full-length and five different deletion constructs of the
HMGA2 3'UTR [
15]. They found that truncation of the distal 593 bp reduced the activity, suggesting the existence of positive regulatory elements at the 3' end of the UTR, but that further deletions significantly increased the luciferase activity. This increase was interpreted to be due to the loss of the AUUUA elements. More recently, it was shown that the 3'UTR of
HMGA2 contains eight, functional or putative, binding sites for the
let-7 family of microRNAs, and that mutational inactivation of these binding sites increased the mRNA levels of
HMGA2 [
17,
18,
49,
50]. In line with these findings, it is known that
let-7 and
HMGA2 expression levels are inversely correlated during embryogenesis. This agrees well with the strong expression of the entire 3'UTR in all three amniocytic cell cultures included here. In contrast, only two of the tumors, a lipoma with t(5;12) and a relocated
HMGA2 locus and an angiolipoma, showed convincing expression of any of the three regions. The remaining tumors, regardless of morphologic subtype and type of
HMGA2 expression, showed only faint expression of no more than two of the three regions investigated, providing further
in vivo support for the notion that loss of regulatory sequences from the 3'UTR is essential for aberrant expression of
HMGA2 in lipomatous tumors. However, as loss or significant reduction of expression of the 3'UTR was a consistent feature also of tumors with expression of the full-length gene and normal
HMGA2 loci at FISH analysis, there must be other mechanisms than chromosomal rearrangements leading to this transcriptional silencing. It was recently shown by Sandberg et al. (2008) that proliferating cells tend to shift from producing longer to shorter mRNA isoforms, and they suggested that a switch to more proximal polyadenylation sites might be a mechanism for these cells to escape regulation by microRNAs [
51]. How this shift from expressing mRNAs with full-length 3'UTR, as seen in the amniocytic cultures, to mRNAs without the 3'UTR, or at least with a 3'UTR too short to be detected by the assays used here, is achieved in lipomatous tumors remains to be clarified. Our and others results indicate that the expression of
HMGA2 [
9,
52] as well as its 3'UTR varies from one tissue to another. Based on the present study, however, the lack of 3'UTR expression seems not to be due to mutations within the 3'UTR itself. Furthermore, nor could we find any evidence that a SNP (rs1042725) within the 3'UTR that has been associated with body height [
53] has any impact on the development of lipomatous tumors.