1
Supplementary Fig. 1 – Flowchart of the conducted statistical analyses. ccRCC = clear
cell renal cell carcinoma; CSS = cancer-specific survival.
479 samples in TCGA ccRCC cohort
(https://tcga-data.nci.nih.gov/tcga/) [19] with
• available RNA-Seq data
• complete CSS information
• non-missing data for important
clinicopathological features
8 SAGE libraries from [23] representing
gene expression in human nephron regions
Identification of 97 marker genes
distinguishing nephron regions (Fig. S2)
Interprofile correlation between
tumor samples and nephron regions
Consensus clustering based on
interprofile correlations (Fig. S3A)
Removal of 16 non-ccRCC
samples (Table S2)
463 ccRCC samples
Introduction of nephron-region-scores
(columns of Fig. 1A)
Re-clustering based on interprofile
correlations (Fig. 1A)
Association of CSS
to nephron-region-scores
Kaplan-Meier analysis of
clusters for CSS (Fig. S4)
S3-score constitutes minimal
and sufficient Cox model
Group assignment of ccRCC samples based on S3score to facilitate clinical application (Fig. 1B):
• S3-score cut-off: -0.167
• 322 ccRCC samples in “high” group (above cut-off)
• 141 ccRCC samples in “low” group
2
Supplementary Fig. 2 – Heat map of the selected 97 genes and their expression in
different parts of the normal human nephron (expression data from Cheval et al [23]).
Glomeruli (Glom), initial and terminal part of proximal tubule (S1/S3), medullary and
cortical thick ascending limbs of the loop of Henle (mTAL/cTAL), distal convoluted
tubules (DCT), and cortical as well as outer medullary collecting ducts (CCD/OMCD).
3
Supplementary Fig. 3 – Clustering of renal cell carcinoma (RCC) tumours by means of
gene expression correlation between profiles of tumours and profiles of human
nephron cell types. (a) Interprofile Spearman rank correlation coefficients (IPCs) were
calculated using 97 genes that differentiate nephron regions (glomeruli [Glom], initial
and terminal part of proximal tubule [S1/S3], medullary and cortical thick ascending
limbs of the loop of Henle [mTAL/cTAL], distal convoluted tubules [DCTs], cortical and
outer medullary collecting ducts [CCD/OMCD]). The heat map shows z scores
obtained by centering and standardising the IPC per the Cancer Genome Atlas (TCGA)
sample. Samples were clustered per tumour entity using a resampling-based
consensus clustering approach. Resulting clusters are color coded (see color bar on
the left) (b) Somatic copy number variations of RCC tumours (samples are arranged in
the same order as above). GISTIC2 gene-level copy number estimates represent
homozygous deletion (−2), single copy deletion (−1), diploid normal copy (0), low-level
copy number (1), and high-level copy number amplification (2). Grey color indicates
missing CNV data. ccRCC = clear cell renal cell carcinoma; chRCC = chromophobe
renal cell carcinoma; pRCC = papillary renal cell carcinoma.
4
Supplementary Fig. 4 – Kaplan-Meier curves showing cancer-specific survival of (a)
clear cell renal cell carcinoma clusters and (b) papillary renal cell carcinoma clusters.
Colours of the lines correspond to colours of cluster analyses given in Figure 1a of
the article. The clusters were identified using a resampling-based consensus
clustering approach. CI = confidence interval; HR = hazard ratio.
5
Supplementary Fig. 5 – Distribution of patient characteristics and clinicopathologic
parameters in the high and low groups, as identified by the S3-score (cut-off: −0.167).
Associated p values are shown in Supplementary Table 3.
6
Supplementary Fig. 6 – Cancer-specific survival of clear cell renal cell carcinoma
(ccRCC) tumours with different tumour stages predicted by the S3-score. KaplanMeier curves showing cancer-specific survival of ccRCC patients with stage I (n =
221), stage II/III (n = 164), and stage IV (n = 78), respectively, for high and low groups,
as identified by the S3-score.
7
Supplementary Fig. 7 – Cancer-specific survival of clear cell renal cell carcinoma
(ccRCC) tumours predicted by the S3-score. Kaplan-Meier curves showing cancerspecific survival of ccRCC patients with different stage, size, grade, and necrosis
scores for high and low groups, as identified by the S3-score.
8
Supplementary Fig. 8 – S3-score significantly improves established prediction scores.
Chi-square statistic values depict the improvement of the model likelihood when S3score (red) and ClearCode34 (CC34) signature (blue) are sequentially added to the Cox
model, initially including only the SSIGN score. The chi-square statistic comparing the
initial model with the null model is given as a baseline value (177.73). *p < 0.05, **p <
0.01, ***p < 0.001.
9
Supplementary Fig. 9 – Intratumour heterogeneity of the S3-score across different
tumour regions of 10 clear cell renal cell carcinoma samples. Color codes are defined
by median S3-score per patient, indicating membership to high or low group with
good or poor prognosis, as identified by the S3-score. S3-score cut-off (−0.167) is
represented by the dashed blue line.
10
Supplementary Fig. 10 – Scatter plot of the S3-score and S1-score in the Cancer
Genome Atlas clear cell renal cell carcinoma cohort (n = 463) along with the Spearman
rank correlation coefficient.