Renal cell carcinoma primary cultures maintain genomic and phenotypic profile of parental tumor tissues

The clear cell subtype of renal cell carcinoma (ccRCC) accounts for 85% of all RCCs and occurs as familial or, more often, sporadic forms. It is characterized by recurrent genetic anomalies, like copy number alterations (CNAs) and loss of heterozygosity (LOH), that involve specific chromosomes (chrs) and result in deletions with LOH on chrs 3p (often involving the von Hippel Lindau (VHL) locus in 3p26-p25), 6q, 8p, 9p and 14q, and duplications of chrs 5q and 7 [1–3]. Evidences suggest that this peculiar pattern of genomic instability represents a tumor-specific molecular fingerprint useful for diagnostic and prognostic applications [3–5]. However, more work still needs to completely clarify the complex molecular pathogenesis of ccRCC. Although the involvement of the VHL tumor suppressor gene has been demonstrated in all familial and in 80-90% sporadic ccRCCs, the remaining 10-20% harbors wild-type VHL [6, 7], suggesting that, despite their identical histological phenotype, these tumors have an intrinsic molecular heterogeneity that still needs to be unraveled [6, 8]. The recognition of this molecular heterogeneity might improve the selection of patients for targeted therapies and allow the identification of specific oncogenes and tumor suppressor genes to be used as novel clinical markers or therapeutic targets.

The current availability of high-throughput platforms to assess molecular changes at genome-wide level might provide an opportunity to achieve this goal. Presently a comprehensive and detailed genomic profiling of DNA alterations is possible by using the single nucleotide polymorphism (SNP) array technology. Unlike CGH technique, the SNP array platform allows the simultaneous analysis of both chromosomal and allelic imbalances [9] and the distinction between LOH associated with CN changes (such as hemizygous deletions) and CN neutral status (often termed as uniparental disomy) [10]. In addition, this technology is able to provide information about the fraction of normal and tumor cells present in the tumor tissue samples [11]. This allows a more complete definition of the complex genetic rearrangements associated with cancer pathologies. Moreover, since the regions of LOH accompanied by deletion (representing the second hit of Knudson's hypothesis) are of particular interest because they may contain genes involved in tumor etiology, the assessment of deletion/LOH areas represents a useful approach to identify regions potentially harboring novel tumor suppressor genes [12]. Accordingly, in a previous study, we used the SNP array technology to characterize the whole-genome DNA profile of a collection of ccRCC tissue samples to find novel chromosomal regions and genes potentially interesting as candidate tumor markers [13].

However, in ccRCC specimens, as in most solid tumor tissues, the molecular analyses may be affected by tissue heterogeneity due to the presence of necrotic areas and non-tumor cells, such as tumor-infiltrating leukocytes, endothelial cells and fibroblasts [14]. With the purpose to increase the quality of data by minimizing this "background noise", several different technical approaches have been explored [15]. Tissue heterogeneity must be considered also when evaluating which are the most appropriate computational tools to process and analyze array-based CN data [11, 16].

Thus, to overcome the problem of tissue heterogeneity and to prospectively perform in vitro functional studies aimed to better understand ccRCC molecular pathogenesis, it is necessary to have a viable and more homogeneous cell material retaining the phenotypic and genomic profile of original tissue. A possible strategy to face these requirements is to adapt fresh ccRCC tissue specimens to grow in vitro as primary cell cultures, which provide a good quality, homogeneous and well-characterized cellular material, enriched with tumor cell component [17] and retaining at the first passages the phenotypic and proteomic profile of the corresponding tissues from which they derive [15, 18, 19]. Hence, primary cultures represent a better in vitro tumor model than stable cell lines, that can be even very different from the original tissues and thus not at all representative [6]. Anyway, the reliability of data obtained from primary cultures strictly depends on their careful cytological characterization, especially regarding possible cell contaminations that might influence data interpretation with misleading effects.

In this study, we investigated whether ccRCC primary cultures, established from surgical tumor specimens and phenotypically well-characterized, maintain the DNA profile of parental tumor tissues and allow a better discrimination of CNAs and LOH with respect to original tissues. Till now, the comparison of DNA profile between tumor primary cultures and parental tissues has been reported, with discordant results, in melanoma [20], neuroblastoma [21] and glioblastoma [22], using either SNP array or array-CGH techniques, and in RCC [23] only using traditional CGH on metaphase spreads and short-term primary cultures not extensively characterized.

In this study, we investigated whether the in vitro model of ccRCC primary cell cultures, established from surgical tumor specimens, well reproduces the DNA profile of parental tumor tissues, thus allowing a more confident CNAs and LOH discrimination with respect to the original tissues. Tumor cells from ccRCC surgical specimens may be adapted to in vitro growth with high efficiency, independently from any clinico-pathological characteristic of patients, as we previously confirmed also in a wide series of well-characterized samples [18, 19]. The growth and survival rate of our ccRCC cultures were in agreement with those reported by other authors [14] and proved that our primary cultures grow for the first four passages without difficulties [18]. The cellular composition of our cultures was very homogeneous: more than 90% of cultured cells were of proximal tubular origin, with morphological characteristics typical of ccRCC, and more than 60% of cells were positive for CA9, a biomarker present in almost all ccRCC cases and expressed in most, although not all, of the malignant clear cells of each single positive case [26, 28]. When evaluated by Western Blot, CA9 expression was quite strong in all samples, except for 73PG culture where it was undetectable. Notably, about this latter culture, we found an identical normal genomic profile in primary culture and parental tissue in terms of both CNAs and LOH (with the resolution power of the 50K SNP array platform), although the tumor tissue sample, unlike primary culture, showed a CA9 expression by immunohistochemistry (data not shown). This finding is in agreement with another paper reporting a ccRCC case which expressed CA9 in the surgical tissue sample but not in the corresponding primary culture [15]. Moreover, in our experience with a wide series of samples [19], we observed other few ccRCC cases showing a CA9 positivity at tissue level but not in corresponding primary cultures. Excluding tissue sampling mistakes, this behavior might be due to a transitory expression of CA9 during tissue excision. In fact, it is known that surgical conditions, like tissue ischemia after renal artery clamping and subsequent hypoxia, may up-regulate the expression of downstream targets of HIF-1 (hypoxia inducible factor 1), including CA9, that conversely should not be induced under normoxic conditions like those applied for primary cultures [29]. On the other hand, the outgrowth of normal tubular epithelial cells in 73PG culture is unlikely based on the extensive phenotypic characterization we performed on all our cultures [19].

Up to now, karyotypic characterization of RCC primary cultures has been performed only by classical cytogenetic G-banding [30] or CGH technique [23]. In our knowledge, this is the first study applying the Affymetrix SNP array technology to assess at genome-wide level both CNAs and LOH in ccRCC primary cultures. Globally, our cultures confirmed the typical ccRCC genomic signature [13, 23, 30]. These cultures typically showed alterations on at most 4 or 5 chromosomes. Only two samples made exception: 73PG culture did not show alterations in both CN and LOH profiles, while 60CC showed CNAs on all chromosomes. In particular, 60CC primary culture, as well as its corresponding tissue, showed an atypical ccRCC genomic profile more similar to that of chromophobe subtype (characterized by wide losses on chrs 1, 2, 6, 10, 13, 17 and 21), notwithstanding its typical "clear cell" histology.

It should also be noted that while 73PG specimen derived from a small in size tumor, suggestive of an early stage of neoplastic progression, 60CC specimen derived from the largest one, probably associated with a more advance stage of tumor progression characterized by an aberrant and "clear cell" atypical genomic profile [2].

Up today, very few genomic comparisons between short-term primary cultures and parental tumor tissues have been performed. In glioblastoma, for example, the genomic profile of primary cultures, assessed by array-CGH, resulted considerably different from that of parental tumors, with changes progressively occurring already after 2 weeks of culture, resulting in an inconsistent representation of tumor biology [22]. In melanoma [20] and neuroblastoma [21], instead, using SNP array technology, primary cultures showed to encompass the spectrum of significant alterations present in primary tumors, thus providing a genetically appropriate in vitro model for functional genomics characterizations.

Concerning RCC, up till now the comparisons between primary cultures and tumor tissues mainly regarded phenotypic characterizations and proteomic profiling [15, 18, 31]. A comparison of the genomic profile between primary cultures and parental tissues has been reported only by Sanjmyatav et al. by using traditional CGH [23]. Their DNA profiling showed a poor overlap of CNAs between ccRCC primary cultures and parental tumor tissues. In fact, only 3 out of 8 ccRCC cultures exactly showed the same DNA alterations present in the corresponding tissues. In addition, other 3 cultures did not show any of the CNAs found in the corresponding tissue samples and resulted diploid. Probably, the poor overlap of CNAs obtained by Sanjmyatav et al. [23] was due both to the low sensitivity and resolution level of the technique used (i.e. CGH on metaphase chromosome spreads) and to the not extensive cytological characterization of primary cultures. These findings highlight the importance of a careful phenotypic characterization of primary cultures for a correct interpretation of genomic results.

Globally, these few genomic comparison studies between primary cultures and parental tissues point out that the use of tumor primary cultures as in vitro model for genetic analysis or functional studies must be distinctively evaluated tumor by tumor and that in any case it depends on the level of phenotypic characterization of primary cultures.

Looking at the genome-wide CN profiles, our results indicated that 7 out of 9 well-characterized primary cultures exactly reproduced the DNA profile of the corresponding tumor tissues. The other two cultures (59RG and 66SML) maintained all but one the CNAs showed by the original tissues. These findings are in agreement with Lin et al. who evidenced that in melanoma some significant alterations present in original tissues (as a deletion on chr 13q) were almost undetectable in cultured cells, despite the landscape of genomic alterations was strikingly similar [20]. However, it must be pointed out that even if primary cultures might have lost some genomic alterations present in original tumor tissues, they probably correspond to some "passenger mutations", that do not confer selective growth advantage to tumor cells and thus might be lost in cultured cells, and not to "driver mutations" [32].

Although the good overlap of CN profiles between each our primary culture and corresponding tissue, we obtained a CN correlation mean value equal to 0.73 (by Spearman regression method), a value just a little higher than that reported for glioblastoma cultures and tissues (Pearson mean 0.62) [22]. The reason is in the wide range of variance we obtained among our 9 cases (range 0.30-0.99), mainly due to the contribution of one case (66SML) on the mean. Although this variability seems too high for conclusive messages, it must be noted that in our series it essentially depends on a software issue. In fact, we used the CNAG "HMM-CN state" data, that is the SNP copy number status inferred by HMM algorithm and visualized in the HMM-CN state track. However, in the presence of a high sample heterogeneity, as in the case of 66SML tissue (illustrated in Figure 4), the HMM-CN state data missed to detect CN aberrations, even if present, because their CN values did not reach the software threshold for a statistical significance. For this reason, the 66SML sample had the lowest CN correlation coefficient between culture and tissue (Spearman 0.30), a value that does not really reflect a poor overlap but only a software constraint.

Similarly to those found in neuroblastoma [21], the SNP call concordance indexes and the Spearman's correlation coefficients calculated on CNAG SNP allelic calls indicated a strong correlation at SNP genotype level between our cultures and parental tissues, 66SML sample included, thus confirming that each primary culture really derived from its corresponding tissue.

Notably, in six samples, CNAs were more evident and better discriminated in primary cultures than in corresponding tumor tissues. This phenomenon principally occurred in deleted regions and it is due to the different cellular composition of the two samples and it was also observed in neuroblastoma tissues and derived cell lines [21]. Tumor tissues are heterogeneous and comprise a mixture of tumor and normal cells (endothelial cells, leukocytes, fibroblasts). For this reason, the copy number values corresponding to DNA alterations of tumor cells are inevitably "diluted" by the diploid values coming from normal cells. Differently, primary cultures were more homogeneous in terms of tumor component, as shown by their phenotypic characterization, and thus presented better defined CNAs. Tumor tissue heterogeneity reflects also on LOH detection. In fact, the software used in the analysis not only detected the allele retained by the tumor, but also the second allele still present in normal cells; in these cases the LOH call will be missed [33]. In our ccRCC primary cultures, instead, hemizygous deletions were accompanied by LOH calls, confirming the great sample homogeneity that enables to discriminate CNAs and LOH. Thus, the presence of LOH calls in deleted regions could be adopted as a parameter to evaluate the degree of tumor sample purity and the level of normal cell contamination and consequently the origin of cultured cells, highlighting the power of SNP array technology with respect to CGH technique. Such an evidence appears more important taking into account that since today there is not a single universally applied method able to certainly make these evaluations [34].

The high cell homogeneity observed in our ccRCC cultures might explain also the additional CNAs found in 4 of them but not in original tissues. The acquisition of these alterations de novo during in vitro growth is unlikely since cultures were analyzed at first confluence. Moreover, the genomic analysis of some cultures at second and third confluences provided evidence that these cultures did not accumulate further alterations during passages and remained stable for at least 3 weeks. We could therefore conclude that these alterations might fail to be seen in the heterogeneous tumor tissues because present in a very small number of cells (< 20%), thus resulting undetectable for CN analysis algorithms [11, 24].

Actually, the true picture of genomic alterations occurring in a tumor can be obtained only performing a genomic analysis directly on tumor cells isolated, for example, by laser capture microdissection, which allows a > 90% of purity [35] and circumvents the problem of DNA alteration "dilution" due to tissue heterogeneity. However, such a useful technical approach has the not negligible inconvenience of providing cellular material not exploitable for eventual subsequent functional studies.

By performing the genome-wide copy number profiling of a collection of ccRCC primary cultures and corresponding tumor tissues, we demonstrated that these well-characterized primary cultures maintained the genomic alterations of parental tumors. Moreover, their DNA profile remained stable for at least 3 weeks, till to third confluence. Importantly, RCC primary cultures provided greater cell homogeneity and enrichment in tumor component than parental tissues, as proved also by phenotypic characterization, thus enabling a better discrimination of DNA alterations. In this context, SNP array technology demonstrated to be a powerful tool able to confirm the origin of cultured cells and to evaluate sample homogeneity and normal cell contamination. The observation that ccRCC primary cultures retain not only the phenotypic features and the proteomic profile of original tumor tissues but also their genomic profile proves that these short-term cultures are a reliable in vitro model that well represents ccRCC genetics and biology and that prospectively could be used for functional approaches.