Fax: (310) 794-3513
Deletions of chromosomes 3p and 14q molecularly subclassify clear cell renal cell carcinoma
Article first published online: 18 JAN 2013
Copyright © 2013 American Cancer Society
Volume 119, Issue 8, pages 1547–1554, 15 April 2013
How to Cite
Kroeger, N., Klatte, T., Chamie, K., Rao, P. N., Birkhäuser, F. D., Sonn, G. A., Riss, J., Kabbinavar, F. F., Belldegrun, A. S. and Pantuck, A. J. (2013), Deletions of chromosomes 3p and 14q molecularly subclassify clear cell renal cell carcinoma. Cancer, 119: 1547–1554. doi: 10.1002/cncr.27947
- Issue published online: 8 APR 2013
- Article first published online: 18 JAN 2013
- Manuscript Accepted: 10 DEC 2012
- Manuscript Revised: 5 DEC 2012
- Manuscript Received: 10 OCT 2012
- renal cell carcinoma;
- von Hippel-Lindau deletion;
- hypoxia-inducible factor 1α;
- tumor suppressor gene;
- cancer-specific survival;
- recurrence-free survival
The short arm of chromosome 3 (3p) harbors the von Hippel-Lindau (VHL) tumor suppressor gene, and the long arm of chromosome 14 (14q) harbors the hypoxia-inducible factor 1α (HIF-1α) gene. The objective of this study was to evaluate the significance of 3p loss (loss VHL gene) and 14q loss (loss HIF-1α gene) in clear cell renal cell carcinoma (ccRCC).
In total, 288 ccRCC tumors underwent a prospective cytogenetic analysis for alterations in chromosomes 3p and 14q. Tumors were assigned to 1 of 4 possible chromosomal alterations: VHL +3p/+14q (VHL wild type [VHL-WT]), VHL +3p/−14q (VHL-WT plus HIF2α [WT/H2]), −3p/+14q (HIF1α and HIF2α [H1H2]), and −3p/−14q (HIF2α [H2]).
Among patients who had loss of 3p, tumors with −3p/−14q (H2) alterations were larger (P = .002), had higher grade (P = .002) and stage (P = .001), and more often were metastatic (P = .029) than tumors that retained 14q (H1H2). All patients who had tumors with −3p/−14q (H2) had worse cancer-specific survival (P = .014), and patients who had localized disease (P = .012) and primary T1 (pT1) tumors (P = .008) had worse recurrence-free survival. In patients who had pT1 tumors, combined 3p/14q loss was an independent predictor of recurrence-free survival (hazard ratio, 11.19; 95% confidence interval, 1.91-65.63) and cancer-specific survival (hazard ratio, 15.93; 95% confidence interval, 3.09-82.16). The current investigation was limited by its retrospective design, single-center experience, and a lack of confirmatory protein analyses.
Loss of chromosome 3p (the VHL gene) was associated with improved survival in patients with ccRCC, whereas loss of chromosome 14q (the HIF-1α gene) was associated with worse outcomes. The results of the current study support the hypothesis that HIF-1α functions as an important tumor suppressor gene in ccRCC. Cancer 2013. © 2013 American Cancer Society.
It has become apparent that RCC is not a single tumor entity but, rather, a complex family of epithelial malignancies, each with unique underlying genetic alterations.1 The most common histologic subtype of kidney cancer is clear cell renal cell carcinoma (ccRCC), a subtype classically associated with loss of chromosome 3p in >50% of patients. Chromosome 3p25-26 harbors the von Hippel-Lindau (VHL) gene, and the biallelic inactivation of the VHL gene leads to ccRCC formation2; whereas other patients with ccRCC lose VHL protein function through epigenetic alterations, such as gene silencing by methylation. Although the prognostic role of VHL loss in ccRCC has not been consistently defined, most reports suggest that tumors with loss of 3p and VHL tumor suppressor function are associated with less advanced tumor stages and better survival.3, 4
On a molecular level, the VHL gene product acts as part of a ubiquitin ligase complex that targets the α subunits of the hypoxia-inducible factor (HIF) for proteasomal degradation when oxygen is present.5 The loss of the VHL tumor suppressor protein in RCC results in a state of “pseudohypoxia” through the stabilization and activation of HIF signaling. There are at least 3 different α-subunits of HIF, and the subunits HIF-1α and HIF-2α are the most studied. Although preclinical experiments have suggested that HIF-2α acts primarily as a tumor promoter in VHL-deficient ccRCC,6 recent reports suggest that HIF-1α acts instead as a tumor suppressor. Moreover, those reports suggest that ccRCC can be further stratified into distinct subgroups: tumors with wild type (WT) VHL alleles (designated VHL-WT), VHL-deficient tumors that over express both HIF-1α and HIF-2α (designated H1H2), and VHL-deficient tumors that exclusively express HIF-2α (designated H2).7 The latter tumors are characterized by enhanced activity of the proto-oncogene v-myc myelocytomatosis viral oncogene homolog (c-MYC) and the mitogen-activated protein kinase/extracellular signal-regulated kinase (MAPK/ERK) pathway, leading to uncontrolled cell proliferation, escape from immune surveillance, and metastasis.8
Cytogenetic analysis has emerged as a powerful tool to predict the prognosis of patients with ccRCC.4, 9 Recent cytogenetic studies indicate that tumors with deletions of 14q are associated with more advanced tumors and worse survival.4 Because HIF-1α protein expression is prognostically relevant,10 and 14q deletions encompass the HIF-1α gene (ie, located on chromosome 14q 23.2),2 we hypothesized that cytogenetic alterations could serve as surrogates for defining tumor types based on HIF-1α and HIF-2α protein levels. At our institution, we have prospectively performed routine cytogenetic analyses on primary RCC specimens since 2001. By using these data, we sought to subclassify ccRCC according to loss of the VHL gene (3p loss) and/or the HIF-1α gene (14q loss). Therefore, we assigned patients into 4 separate groups based on the occurrence of loss 3p and 14q and compared the groups for differences in their clinicopathologic features, recurrence-free survival (RFS), and cancer-specific survival (CSS).
MATERIALS AND METHODS
Patient Selection and Classification
Three hundred fifty-four patients who received treatment at the University of California-Los Angeles (UCLA) for ccRCC between October 1999 and April 2007 formed the primary study cohort. Clinicopathologic features were gathered from the Institutional Review Board-approved UCLA Kidney Cancer Database. The database included patient age, sex, Eastern Cooperative Group performance status,11 Fuhrman grade,12 tumor size, CSS, and RFS. All investigated tumors were pathologically confirmed ccRCCs according to the 2009 TNM classification.13 Computed tomography or magnetic resonance imaging studies of the chest, abdomen, and pelvis as well as x-rays in a subset of patients were obtained to assess the extent of disease before surgery. Patients were followed according to the risk-adjusted UCLA kidney cancer surveillance protocol, as described previously.14
Tumor samples were collected immediately after surgery and minced into 2-mm-thisk to 3-mm-thick pieces. After tissue dissociation with collagenase II (Worthington Biochemical Corporation, Freehold, NJ), the cells were washed and subsequently cultured in RPMI 1640 supplemented with 20% fetal bovine serum and 1% penicillin/streptomycin. After harvesting, the cells were subjected to hypotonic treatment (potassium chloride, 0.075 M) for 30 minutes at 37°C and fixed in methanol and acetic acid (3:1). Chromosomes were banded using pancreatin solution for G-banding with the Giemsa technique. Twenty metaphases were investigated and analyzed in accordance with International Standing Committee on Human Cytogenetic Nomenclature.15 The detailed protocol of our cytogenetic study has been published previously.9
For this report, we included only patients who had sporadic ccRCC, because familial RCC may obscure the true role of a cytogenetic aberration. In addition, tumors that had a normal karyotype were excluded. For the final cohort, 288 remaining ccRCC tumors were analyzed. The cytogenetic results accrued here were obtained without any a priori knowledge of what the karyotype would exhibit. Thus, each sample that did not have −3 or −14 acted as an internal control.
Patient Assignment and Statistical Analysis
For the purposes of this analysis, loss of chromosome 3p was considered a surrogate for loss of the VHL gene,16 and loss of chromosome 14q was considered a surrogate for loss of the HIF-1α gene.2 Thus, patients were assigned to 4 groups: 1) without loss of chromosome 3p (VHL gene) or loss of chromosome 14q (HIF-1α) (+3p/+14q), 2) without loss of 3p but with loss of 14q (+3p/−14q), 3) with loss of 3p only (−3p/+14q), and 4) with loss of both 3p and 14q (−3p/−14q).
Continuous variables were reported as means ± standard deviation (SD) and medians (interquartile range [IQR]). Normal distribution was tested with the Kolmogorov-Smirnov test. Although age (P = .312) was normally distributed, tumor size (P = .001) was not. On the basis of whether or not the continuous data were normally distributed, we quantified associations with either the Student t test or the Mann-Whitney U test, respectively. The Pearson chi-square and the Fisher exact test were used for comparisons of categorical variables.
The primary endpoint of the study, CSS, was calculated from the date of surgery to the date of cancer-specific death or last contact. A secondary endpoint was RFS, which we defined as freedom from cancer recurrence because of distant or local metastases. Disease was considered a recurrence if metastases or de novo lesions were diagnosed at least 3 months after surgical intervention, in an effort to exclude the possibility of previously misclassifying unrecognized metastases as recurrent disease. Both CSS and RFS were estimated using the Kaplan-Meier method. The association of each variable with either CSS or RFS was assessed with univariate Cox regression analysis. Next, characteristics that demonstrated an association in univariate analysis were evaluated using multivariate Cox proportional-hazards modeling. In multivariate analyses, a backward-stepwise selection process with the likelihood-ratio criterion (cutoffs for inclusion and exclusion criteria were P < .05 and P > .10, respectively) was used. The rank of elimination was established when a variable was removed from the equation, and the hazard ratio (HR), 95% confidence interval (CI), and P value for the removed variable were obtained during the removal step. The Schoenfeld global test was used to test the proportional-hazards assumption in the Cox models. Analyses were made separately for all tumors, localized tumors, and small (primary T1 [pT1]) tumors. The accuracy of Cox regression models was tested with the Harrell concordance (C)-index, and different C-indices were compared with Harrell U statistics.17 All tests performed were 2-sided to assign statistical significance to P values ≤ .05. The IBM SPSS Statistics 19 software package (SPSS Inc., Chicago, IL) was used for all analyses.
Patients and Tumor Characteristics
The patients and their pathologic tumor characteristics are summarized in Table 1. When 3p was not deleted, none of the comparisons differed significantly between patients with +3p/+14q and those with +3p/−14q. Conversely, when 3p was lost, patients with −3/−14q had larger tumors (P = .002), higher Fuhrman grades (P = .002), more advanced pT classifications and TNM stages (P = .001), and more frequently presented with distant metastases (P = .029) compared with those who had the −3p/+14q genotype.
|No. of Patients (%)||No. of Patients (%)|
|Feature||+3p/+14q: n = 93 (32.3%)||+3p/−14q, n = 17 (5.9%)||P||−3p/+14q, n = 114 (39.6%)||−3p/−14q, n = 64 (22.2%)||P|
|Age: Mean±SD, y||63.5±11.9||60.6±12.6||.344||60.2±12.65||59.3±10.2||.635|
|Median tumor size [IQR], cm||6.10 [5.9]||4.70 [5.1]||.425||4.50 [4.2]||6.85 [5.2]||.002|
|Primary tumor classification|
|pT1||40 (43)||11 (64.7)||.285||75 (65.8)||24 (37.5)||.001|
|pT2||8 (8.6)||0 (0)||11 (9.6)||8 (12.5)|
|pT3||42 (45.2)||6 (35.3)||28 (24.6)||32 (50)|
|pT4||3 (3.2)||0 (0)||0 (0)||0 (0)|
|Pathologic lymph node status|
|pN0/pNx||79 (84.9)||12 (70.6)||.351||113 (99.1)||60 (93.8)||.600|
|pN1||6 (6.5)||2 (11.8)||1 (0.9)||1 (1.6)|
|pN2||8 (8.6)||3 (17.6)||0 (0)||3 (4.7)|
|M0||53 (57)||7 (41.2)||.292||98 (86)||46 (71.9)||.029|
|M1||40 (43)||10 (58.8)||16 (14)||18 (28.1)|
|I||32 (34.4)||7 (41.2)||.211||71 (62.3)||21 (32.8)||.001|
|II||5 (5.4)||0 (0)||10 (8.8)||6 (9.4)|
|III||15 (16.1)||0 (0)||17 (14.9)||17 (26.6)|
|IV||41 (44.1)||10 (58.8)||16 (14)||20 (31.3)|
|1||7 (7.5)||2 (11.8)||.190||13 (11.4)||4 (6.3)||.002|
|2||42 (45.2)||4 (23.5)||70 (61.4)||24 (37.5)|
|3||32 (34.4)||10 (58.8)||22 (19.3)||28 (43.8)|
|4||12 (12.9)||1 (5.9)||9 (7.9)||8 (12.5)|
|0||4 (58.1)||11 (64.7)||.761||79 (70.5)||42 (71.2)||.996|
|1||37 (39.8)||6 (35.3)||31 (27.7)||16 (27.1)|
|2||2 (1.8)||0 (0)||2 (1.8)||1 (1.7)|
After a median follow-up of 39.4 months (IQR, 61.8 months), 74 patients (25.7%) had died from ccRCC. Overall, any loss of 3p (VHL) was associated with improved survival (P = .009), whereas loss of chromosome 14q (HIF-1α) was associated with worse survival (P = .042) (Fig. 1A,B). Patients who had the −3p/+14q genotype had the best survival outcomes, whereas patients with the +3p/−14q genotype had the worst survival outcomes. The 1-year and 5-year survival probabilities were 92.1% versus 74.5%, respectively, in those with the in −3p/+14q genotype and 80.5% versus 41.9%, respectively, in those with the +3p/−14q genotype (HR, 4.00; 95% CI, 1.75-9.16). The median survival was 115.2 months (95% CI, 21.9-208.6 months) in the +3p/+14q group, 49.7 months (95% CI, 34.6-64.7 months) in the +3p/−14q group, and 87.4 months (95% CI, 56.9-117.9 months) in the −3p/−14q group; and the median survival was not reached in the −3p/+14q group. When comparing the survival outcomes of patients with +3p/+14q and those with +3p/−14q tumors, we discovered a clear trend toward worse survival outcomes in patients who had loss of 14q. However, the difference was not statistically significant (P = .193) (Fig. 1C). Similarly, compared with patients who had the −3p/+14q genotype, patients who had loss of both 3p and 14q (−3p/−14q) had a >2-fold greater risk of dying from ccRCC (HR, 2.2; P = .014) (Fig. 1D).
Because the investigated deletions were associated significantly with CSS on univariate analysis for all stages (P = .004) and in patients who had pT1NanyMany disease (P = .04), but not for patients who had localized tumors (P = .251), multivariate analyses were performed only for all stages and for those with pT1NanyMany disease (Table 2). On multivariate analysis of all tumors, there was no association between loss of 3p (VHL gene) or 14q (HIF-1α gene) with CSS. However, compared with patients who had the +3p/+14q genotype, patients who had pT1 tumors with the −3p/+14q genotype or the −3p/−14q genotype had a 6.2-fold and 15.9-fold increased risk of cancer-specific death, respectively (P = .041 and P = .001, respectively) (Table 2). Moreover, the C-index increase from 92.4% to 92.9% (statistically significant; P = .010) when cytogenetic data were added to the pT1 model.
|All Patients, n = 288||Small Renal Masses: pT1NanyMany, n = 150|
|Feature||HR||95% CI||P||HR||95% CI||P|
|M-classification||8.10||4.60-14.24||< .001||83.39||17.33-401.19||< .001|
Of 199 patients who had nonmetastatic disease, we identified 28 (14.1%) who developed recurrent disease. Twenty-two of those 28 patients (11.1%) had a recurrence because of distant metastases, and 6 (3%) developed de novo lesions in the same or contralateral kidney. Because of the small number of patients with localized tumors who had the +3/−14q genotype and experienced disease recurrence (7 patients with 1 recurrence), these patients were excluded from further RFS analyses. Among patients with localized tumors, however, those who had −3p/−14q alterations had a significantly worse RFS than those who had −3p/+14q alterations (HR, 2.9; P = .012) (Fig. 2A). Among patients with pT1NanyMany, those who had the −3p/−14q genotype had 5 times the risk of disease recurrence compared with patients who had the −3p/+14q genotype (HR, 5.1; P = .008) (Fig. 2B).
For all patients who had localized disease, genetic data demonstrated a strong trend toward statistical significance (P = .053; −3p/−14q vs −3p/+14q: HR, 2.95; P = .012) in univariate analysis of 3p/14q loss; however, this significance was lost in multivariate regression analysis (P = .127) (Table 3). Conversely, the −3p/−14q genotype was an independent predictor of RFS in patients who had pT1 tumors (HR, 11.2; P = .007) (Table 3); and, when cytogenetic data were added to the multivariate pT1 model, the C-index increased significantly from 72.4% to 80.6% (P = .013).
|All Patients: pT1-pT3N0M0, n = 199||Small Renal Masses: pT1N0M0, n = 150|
|Feature||HR||95% CI||P||HR||95% CI||P|
|Tumor size||1.20||1.10-1.32||< .001||1.88||1.23-2.87||.004|
Chromosome 3p harbors the VHL tumor suppressor gene, whereas chromosome 14q harbors the HIF-1α gene. Renal epithelial tumors with loss of VHL function represent the majority of ccRCCs, and these tumors have been stratified further into those that express both HIF-1α and HIF-2α (H1H2) and those that express only HIF-2α (H2).7 In the current study, we evaluated the influence of individual and concomitant VHL gene and HIF-1α gene deletions on clinicopathologic features, CSS, and RFS in patients with ccRCC. We report 2 principal findings: 1) loss of chromosome 3p (VHL gene) was associated with more favorable pathologic features and with improved RFS and CSS, and 2) loss of chromosome 14q (HIF-1α gene) was associated with worse RFS and CSS. Furthermore, loss of 14q (HIF-1α) was the most important predictor of RFS in patients who had pT1NanyMany disease when we compared tumors with the H2 (−3p/−14q) and H1H2 (−3p/+14q) genotypes.
Patients who had VHL-WT (+3p/+14q) tumors and VHL-WT/H2 (+3p/−14q) tumors did not differ in their clinicopathologic features, whereas patients who had H2 (−3p/−14q) tumors had more advanced tumor stages and Fuhrman grades. The VHL protein is part of a ubiquitin ligase complex that targets the HIF-α subunits.5 Thus, even if the tumor suppressor gene HIF-1α is lost (+3p/−14q), the VHL protein is able to abolish HIF-2α. Conversely, tumors that have loss of the VHL gene may prevent the tumor suppressive function of HIF-1α. This may explain why patients who have H2 (−3/−14q) tumors, compared with those who have H1H2 (−3p/+14q) tumors, have more advanced clinicopathologic features (Table 1).
Our findings are congruent with reports suggesting that ccRCC tumors that arise in the setting of loss of the VHL gene are associated with better survival outcomes.3, 4 One potential explanation for this observation is that VHL loss may result not only in HIF stabilization and the up-regulation of downstream targets, such as vascular endothelial growth factor, but also in the up-regulation of HIF downstream products that reportedly are correlated with improved survival. For example, the tumor-associated membrane antigen carbonic anhydrase IX (CAIX) is regulated by HIF-1α, and high CAIX expression reportedly is associated with improved survival outcomes.3 The contention that HIF-1α primarily functions as a tumor promoter in ccRCC18 is currently being disputed by recent evidence suggesting instead that HIF-1α acts as a tumor suppressor.6, 19 Shen et al more specifically demonstrated only wild-type HIF-1α has tumor suppressive effects, whereas HIF-1α variants resulting from genomic deletions are deleterious to the tumor-suppressive effects of HIF-1α in proliferation assays.2 Chromosome 14q may harbor several other putative tumor suppressor genes, like the Hippo pathway genes SAV1 (salvador homolog 1) and FRMD6 (4.1 ezrin radixin moesin [FERM] domain-containing 6 [14q22]).20 Shen et al also addressed this point in their investigation by screening 16 kidney cancer cell lines for altered transcripts of these genes. In their discussion, those authors concluded that only HIF-1α, but not neighboring genes on chromosome 14q, are changed. However, in their study, Shen et al only investigated alterations caused by focal deletions of 14q. Thus, it is less likely that the results observed in our current study are related to the losses of other tumor suppressor genes and not to the loss of the HIF-1α gene. However, the finding of no focal deletions does not exclude the association of the complete loss of 14q with more aggressive tumor behavior of ccRCC. Therefore, our study cannot exclude the possibility that the losses of chromosomes 3p and 14q involve other genes that are critical in ccRCC tumor biology. Investigations into other tumor suppressor genes on these chromosome parts should be the subject of future examinations.
In our current patient cohort, almost 66% of patients had loss of chromosome 3p; and, among these patients, the additional loss of 14q (HIF-1α) was correlated with more advanced tumor stages and significantly worse survival. H2 ccRCCs, with loss of VHL (3p) and loss of HIF-1α (14q) but with HIF-2α expression,7 have been described as having unique physiologic properties compared with other ccRCCs. For example, a recent study demonstrated that, even under normal oxygen conditions, cell lines that express only HIF-2α (786-O) behave like hypoxic cells and produce α-ketoglutarate from reductive glutamine metabolism for lipogenesis.21 In addition, enhanced tumorigenetic effects of HIF-2α on ccRCC have been associated with an increase in c-Myc activity,7 which, itself, appears to affect prognosis.22 It is noteworthy that c-Myc, in turn, is an important driver of glutamine metabolism. These metabolic alterations have the potential to make HIF-2α-expressing tumors independent of glucose in generating lipogenesis. The preservation of glucose is mandatory for production of ribose and other biosynthesis precursors in fast-growing cells. This mechanism may explain in part the aggressive nature of H2 (−3p/−14q) tumors and VHL-WT/H2 (+3p/−14q) tumors.21 Moreover, in a study by Gordan et al, the effect of HIF-2α on c-Myc was apparent in low tumor stages (pT1 and pT2) but not in advanced tumor stages.7 This may explain why we were only able to demonstrate an independent association of H2 (−3p/−14q) status on CSS with pT1 tumors. Finally, the worst survival outcome was noted in patients who had VHL-WT/H2 (+3p/−14q) tumors. This group represents only a small subset of tumors with loss of the HIF-1α gene but without loss of the VHL gene. The difference in survival between patients with VHL-WT/H2 (+3p/−14q) tumors and those with VHL-WT (+3p/+14q) tumors was not statistically significant, but this lack of an observed difference may be explained by the small number of patients available for analysis.
Loss of 14q was the most important predictor of RFS in patients with pT1 tumors. In recent years, the increase in the incidence of pT1 tumors represents the largest number of new RCC diagnoses,23 and it has been questioned whether surgical treatment is mandatory for all patients with these tumors,24 particularly those who have tumors that fit into the pT1a category. Deciding which of these patients may safely be considered for surveillance rather than immediate treatment is hampered by the lack of existing prognostic biomarkers. Our results suggest that loss of the VHL and HIF-1α genes may be able to serve as an important prognostic marker in assigning patients to active surveillance (from biopsy) or adjuvant treatment after surgical treatment.
The current study has several limitations. First, the chromosomal changes serving as surrogates for presumed VHL, HIF-1α, and HIF-2α protein expression were not directly measured. However, a recent study has demonstrated low HIF-1α and high HIF-2α protein expression in ccRCC tumors that lost chromosome 14q.25 Moreover, although loss of 3p in the current study approximated the reported occurrence of VHL mutations in other studies, there is still the possibility that there are more tumors with VHL loss (ie, because of epigenetic events like promoter hypermethylation). Gerlinger et al recently reported that heterogeneous composition of the tumor can challenge biomarker development and personalized medicine.26 Indeed, tumor diversity is a limitation of the current study. However, our final karyotype results are based on recognition of the presence of major clones as defined in the International System of Human Cytogenetic Nomenclature.15 Despite long-term follow-up, the number of patients with localized tumors who developed recurrent disease remains low but is concordant with the low recurrence rate reported in previous studies.27 Finally, this was a single-center study and, thus, will need to be validated in a prospective fashion across multiple centers.
In conclusion, we observed that loss of chromosome 3p (the VHL gene) is associated with improved survival, whereas loss of chromosome 14q (the HIF-1α gene) is associated with worse RFS and CSS. These findings support the hypothesis that HIF-1α functions as a tumor suppressor gene in ccRCC. Genetic subclassification of ccRCC according to VHL and HIF genotype may assist in identifying high-risk patients for enhanced surveillance after nephrectomy, helping to select appropriate patients with small renal masses for active surveillance and/or personalizing the use of specific targeted therapies in those with metastatic disease.
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURE
The authors made no disclosures.
- 13Urological tumours—kidney. In: Sobin LH, Gospodarowicz MK, Wittekind C, eds. TNM Classification of Malignant Tumours. Oxford, United Kingdom: Wiley-Blackwell; 2009: 255-257., , .
- 152009: An International System for Human Cytogenetic Nomenclature. Farmington: CT: Karger; 2009., , . ISCN