RUNX1 Is a Driver of Renal Cell Carcinoma Correlating with Clinical Outcome

Kidney cancer is the seventh commonest cancer in the United Kingdom with around 12,500 diagnoses and 4,500 deaths annually (Cancer Research UK; www.cancerresearchuk.org; accessed July 2019). Around 85% of kidney cancers are classified as renal cell carcinomas (RCC), of which, clear cell renal cell carcinoma (ccRCC) accounts for the vast majority (75%+; refs. 1, 2). Since the mid 1970′s, age-standardized kidney cancer mortality rates have increased by 74% while the incidence rate has increased by 85% in relative terms in the last 25 years (Cancer Research UK). Various environmental risk factors such as smoking, hypertension, and obesity contribute to kidney cancer development, but there is also a strong genetic contribution to the development of the disease (3). Many of these genetic alterations lead to changes in the transcriptional profile of the kidney cancer cells (4, 5).

Currently RCC represents a pressing clinical challenge due to its increasing incidence. Early-stage nonmetastatic RCC can be treated by partial or radical nephrectomy, however, early-stage disease is often asymptomatic, resulting in patients more commonly presenting with advanced disease that has a much poorer prognosis (6). Standard of care for high-risk, advanced metastatic or recurrent RCC involves targeted tyrosine kinase inhibitors (TKI) primarily against the VEGF and mTOR pathways, which have modest improvement over previous cytokine therapies (7, 8). Recently, combinatorial use of TKI with immune checkpoint inhibitors against programmed cell death complex (PD-1 and PD-L1) have shown promising results in stage III clinical trials (9). However, the outlook for high-risk patients remains poor and there is both a need for novel biomarkers of poor prognosis and identifying targetable genetic drivers.

Foremost among the genetic alterations that occur in kidney cancer is the loss of the short arm of chromosome 3 that contains the tumor suppressors VHL and PBRM-1, BAP-1, and SETD2 and occurs in up to 90% of cases of ccRCC (10, 11). VHL protein functions as an E3 ubiquitin ligase targeting the hypoxia inducible factor (HIF) family of transcription factors for proteasomal degradation. Loss of VHL therefore causes a transcription factor–driven change in gene expression, leading to the development of kidney cancer (1). The other tumor suppressor genes commonly deleted (PBRM-1, BAP-1, SETD2) all act directly or indirectly, through epigenetic changes in methylation status, to cause alterations in gene expression of kidney cancer cells (12). While much remains unknown about the transcription factors important for kidney cancer, these genetic alterations highlight the important role transcriptional misregulation plays in kidney cancer and the pressing need to identify the key factors involved.

RUNX1 is a member of an evolutionarily conserved family of RUNX genes that encode transcription factors. Together with its heterotypic binding partner CBFβ, RUNX1 forms a DNA-binding complex required for normal mammalian development (13). RUNX1 also has established roles in various types of cancer (14) where classically, RUNX1 chromosomal translocations and mutations are key drivers of hematopoietic malignancies and leukemia (15). Increasingly, however, RUNX1 has been shown to play “context dependent” roles in solid tumors such as in the breast, where both RUNX1 gain and loss of function has been associated with cancer (16–19). RUNX1 has also been implicated in cancers of the ovary and uterus (20), prostate (21), and skin. To date, very little is known about a functional role for RUNX1 in either normal kidney development or kidney cancer. There is some evidence of increased expression of a RUNX1 chromosomal translocation product in ccRCC patient samples (22) and RUNX1 has been shown to be expressed in mouse models of kidney fibrosis (a feature of chronic kidney disease correlated with RCC), involving RUNX regulation of TGFβ-driven EMT (23).

Here we show that RUNX1 is expressed in human ccRCC and that high protein expression correlates with poorer survival. This is functionally relevant as deletion of RUNX1 in human ccRCC cell lines disrupted tumor cell growth in vitro and in vivo, and enabled identification of a novel set of RUNX1-dependent genes in ccRCC. By utilizing a genetically engineered mouse (GEM) model of kidney cancer, we were able to interrogate the role of RUNX1 in tumor formation and genetically confirm that in vivo deletion of Runx1 slows kidney cancer development. Finally, we reveal that the related transcription factor RUNX2 is expressed in ccRCC, also associating with poorer survival. Our results provide the first evidence that RUNX proteins are novel players in kidney cancer and functionally contribute to disease progression and clinical outcome.

The following antibodies were used: RUNX1 (8529), RUNX2 (8486), GAPDH (3683), horseradish peroxidase-conjugated anti-rabbit secondary antibody (7074; Cell Signaling Technology); SERPINH1 (10875-1-AP), and STATHMIN3 (11311-1-AP; ProteinTech). Primary antibodies used for immunoblotting at 1:1000 dilution: Ki67 SP6 (RM-9106-S; Thermo Fisher Scientific).

Cells were lysed in Pierce RIPA buffer (Thermo Fisher Scientific), protein extracts resolved on 10% NuPAGE Novex Bis-Tris gels (Life Technologies) and transferred to Hybond-ECL nitrocellulose membranes (Amersham). All membranes were stripped and reprobed for GAPDH.

IHC staining for RUNX1/RUNX2 was performed on 4-μm formalin-fixed paraffin-embedded (FFPE) sections previously dry heated at 60°C for 2 hours. IHC performed on Agilent Autostainer link48. Sections manually dewaxed through xylene, graded alcohol, tap water before heat-induced epitope retrieval (HIER) with sections heated to 98°C (25 minutes); rinsed in Tris buffered saline with Tween (TBST), peroxidase blocked (Agilent, UK), washed in TBST before application of antibody at previously optimized dilution (RUNX1 1:75, RUNX2 1:300) for 40 minutes. Sections were washed in TBST before application of rabbit EnVision (Agilent) secondary antibody for 35 minutes and rinsed in TBST before applying Liquid DAB (Agilent, UK) for 10 minutes. Sections were washed in water, counterstained with hematoxylin, and coverslipped using DPX. Ki67 (1:200 dilution) and SERPINH1 (1:80 dilution and high pH antigen retrieval). Digital images were captured on a Leica SCN400f slide-scanner (x20). Quantification of Ki67 performed manually using HALO image analysis software (Indica Labs).

The tissue microarray (TMA) contained cores from 184 patients diagnosed with ccRCC within the Greater Glasgow NHS Trust between 1997 and 2008 and obtained from Greater Glasgow and Clyde NHS Biorepository as described previously (24, 25). Briefly, to address tumor homogeneity, three cores measuring 0.6 mm² from three different tumor-rich areas, as identified by a specialist pathologist, were used to construct the three TMAs. After IHC for RUNX1 or RUNX2 (above) and hematoxylin costaining, the proportion of tumor cells with RUNX nuclear positivity was manually quantified using the weighted histoscore (H-Score) method. This involved calculating a semiquantitative score by multiplying the percentage of cells showing staining by a score ranging from 0 to 3 representing increasing intensity of staining (score 0-no staining, score 1-weak staining, score 2-moderate staining, and score 3, strong staining) providing a score from 0 to 300 (25). Three TMA sections were stained at the same time and average H-Scores obtained. H-Scores were stratified into quartiles (Q1–Q4), the upper quartile Q4 assigned as RUNX-High and remaining quartiles Q1–3 assigned as RUNX-Low. One third of the TMA was independently scored and agreement assessed by interclass correlation coefficient >0.8 (26). Klintrup–Makinen score is a pathologically defined measure of inflammatory infiltration described previously for this TMA (24, 25, 27). Statistical analysis was performed using SPSS Statistics Version 21.0 (SPSS IBM). Associations between categorized H-scores and available data on variables were analyzed using χ² tests. Kaplan–Meier curves were plotted with corresponding log-rank tests to assess the relationship between these markers and survival. Multivariate analysis was performed using backwards Cox regression conditional technique to test for independence (25).

786-O cells (cultured in RPMI medium), Caki-2 cells (cultured in McCoy 5a medium; Sigma), and HEK293 cells (cultured in DMEM) were provided by Professor Eyal Gottlieb (Beatson Institute, Glasgow, Scotland, 2014). All media were supplemented with 10% FCS, 2 mmol/L l-glutamine, penicillin/streptomycin, and 0.5 μg/mL amphotericin B (Sigma). All media reagents were from Gibco unless otherwise stated. Cells were of low passage and cultured for approximately 2 months after recovery from frozen vials. RCC cell lines (786-O and Caki-2) were authenticated using Promega GenePrint 10 system. Short tandem repeat multiplex assay was performed that amplifies 9 tetranucleotide repeat loci and the Amelogenin gender determining marker (December 2016). Cells were routinely tested for Mycoplasma.

RUNX1 MISSION shRNA lentivirus DNA constructs (Sigma) were used for targeting human RUNX1 (sh1: TRCN0000338489, sh5: TRCN0000013660). Lentiviruses were produced by transfecting HEK293 cells with 10 μg of the relevant shRNA expression vector (pLKO) with 7.5 μg PsPax2 and 4 μg pVSVG packaging vectors (Tronolab) using the calcium chloride method; complete medium replacement was done 5 hours after transfection. Forty-eight hours after transfection, viral supernatant was removed, sterile filtered (0.45-μm pores), and used to infect adherent 786-O and Caki-2 cells overnight in the presence of 8 μg/mL polybrene (Sigma). Live-cell visualization of GFP confirmed successful transduction. Cells were maintained in medium containing 2 μg/mL puromycin (Sigma). For CRISPR/CAS9 deletion, guide RNAs (gRNA) targeting human RUNX1 were designed using the Zhang Lab tool (MIT, Boston, MA). The gRNA sequence used was 5′-ATGAGCGAGGCGTTGCCGCT-3′. 786-O cells were transfected using lipofectamine (Thermo Fisher Scientific); 8 μL of Lipofectamine in 250 μL serum-free medium (SFM) and 2 μg DNA in 250 μL of SFM (with GLN), were each left for 10 minutes at room temperature then the Lipofectamine/DNA mix incubated at room temperature for 30 minutes before being added to 2 × 10⁵ 786-O cells (plated overnight) and incubated at 37°C for 5 hours prior to a medium change. Forty-eight hours after transfection, cells were cultured in medium containing 2 μg/mL puromycin for 48 hours. Transfected cells grew back as individual colonies, which were picked, expanded, and screened for RUNX1 deletion by immunoblotting.

A total of 2 × 10⁴ 786-O (pX Ctrl) and 786-O RUNX1 CRISPR clones (CRISPR A1/CRISPR A3) were plated in triplicate in 12-well plates, trypsinized, and counted using the Casyton cell counter 96 hours later. Cell count for pX Ctrl cells was normalized to 1 and CRISPR clones expressed as a proportion; experiments were repeated at least four times. A total of 7 × 10³ Caki-2 cells were plated and counted as above. For xCELLigence assay 3 × 10³ 786-O cells were plated in quadruplicate into wells of an E Plate 16. The impedance applied to an electric field over time caused by cells growing in the plate is proportional to the number of cells in the plate and is represented as a cell index when measured using the XCELLigence Real Time Cell Analysis System (Roche Diagnostics GmbH). Experiments performed in quadruplicate at least three times with separate batches of cells. For MTS cell viability assays, 3 × 10³ 786-O cells or 1 × 10³ Caki-2 cells were plated in quadruplicate in 96-well plates; every 24 hours, a 20% volume of CellTiter96 MTS assay reagent (Promega) was added per well and incubated for 1 hour prior to reading absorbance at A490. Experiments were repeated at least three times with separate batches of cells.

A total of 1 × 10⁵ 786-O cells were left to adhere overnight in complete medium. Medium was removed and replaced with complete medium containing 10 μmol/L EdU and incubated at 37°C for 30 minutes. Cells were washed twice in PBS and sampled immediately or 6 hours later. Cells were costained with 50 μg/mL propidium iodide (Sigma) for 30 minutes with gentle rocking and then analyzed on an Attune NTX flow cytometer. All experimental conditions performed in triplicate three separate times. Flow cytometry data were analyzed using FlowJo.

A total of 3 × 10³ 786-O cells were plated (24-well plate) and allowed to adhere overnight. The next day, medium was changed to complete medium containing 5 μmol/L of Sytox Green. The plate was imaged every hour for 68 hours on an Incucyte FLR imaging system. Confluence and number of Sytox-positive cells per well were calculated using Incucyte software.

786-O cells were plated in a 96-well image lock plate and allowed to adhere overnight in complete medium. At confluence, the plate was scratched using the wound maker (Essen Biosciences) and medium changed. Closure of the wound was imaged every hour over 24 hours and analyzed using Incucyte ZOOM live cell imaging system. All experiments were performed in quadruplicate three times.

All animal experiments performed under UK Home Office Project Licences (60/4181 and 70/8645) with ethical approval from the Beatson Institute and the University of Glasgow under the Animal (Scientific Procedures) Act 1986 and EU directive 2010. Mice were maintained in a purpose-built facility in a 12-hour light/dark cycle with continual access to food and water.

Eight- to 10-week-old female CD1-Foxn1^nu (nude) mice were obtained from Charles River, UK. A total of 5 × 10⁵ 786-O* cells were injected directly into the kidney capsule in 20 μL growth factor–reduced Matrigel. Mice were continually assessed for signs of kidney impairment, and kidney tumor development monitored by ultrasound Imaging. Mice were humanely sacrificed at clinical endpoint or 18-week time-point. Parental 786-O cells were initially passaged once in vivo through the kidney (as described above) and a secondary 786-O cell line (referred to as 786-O*) was established in culture using an adapted version of the method described here (28). Briefly, the kidney was excised and normal tissue removed. The tumor was finely chopped into a paste and incubated with 140 rpm rotation at 37°C for 10 minutes in 10 mL of 1 mg/mL type 2 collagenase (Sigma). The tube was vortexed vigorously for 30 seconds before a second 10-minute incubation. Cells were washed with RPMI and passed through sequential 100-, 70-, and 40-μm filters. RUNX1 was deleted from the 786-O* line by CRISPR/CAS9 as described above. 786-O* vector control and CRISPR cells were confirmed to not express CAS9 prior to engraftment.

A total of 5 × 10⁵ 786-O cells (pX Ctrl, CRISPR A1, and CRISPR A3) were plated and sampled 48 hours later. Whole RNA was extracted using RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. RNA was DNAse treated using RNAse-free DNAse set (Qiagen). RNA quality was tested on an Agilent 2200 Tapestation using RNA screentape, all RNA integrity value ≥9.6. Libraries for cluster generation and DNA sequencing were prepared following an adapted method from Fisher and colleagues (29) using Illumina TruSeq Stranded mRNA LT Kit. Quality and quantity of DNA libraries assessed on an Agilent 2200 Tapestation (D1000 screentape) and Qubit (Thermo Fisher Scientific) respectively. Libraries were run on Illumina Next Seq 500 using High Output 75 cycles kit (2 × 36 cycles, paired end reads, single index). Quality checks on raw RNA-sequencing (RNA-seq) data files was done using fastqc version 0.11.7 and fastq_screen version 0.11.4. RNA-seq paired-end reads were aligned to the GRCh38 (30) version of the mouse genome using tophat2 version 2.1.0 with Bowtie version 2.2.6.0. Expression levels were determined and statistically analyzed by a combination of HTSeq version 0.6.1, the R environment, version 3.5.0, utilizing packages from Bioconductor data analysis suite and differential gene expression analysis is based on the negative binomial distribution using DESeq2. Full datasets produced from this study are publicly available on the Sequence Read Archive database, accession number: PRJNA605312.

Cyp1aCre; Apc^fl/fl; p21^−/− mice (hereafter referred to as CAP) were characterized in the Sansom lab as described previously (31). These mice were crossed with Runx1^fl/fl mice (a kind gift from Prof. Nancy Speck, University of Pennsylvania, Philadelphia, PA; Jax:010673, B6;129-Runx1^tm3.1Spe/J; ref. 32) and/or Runx2^fl/fl mice (33). Tumor mice (equivalent numbers males/females in each cohort) were monitored for signs of tumor development and subsequently checked three times a week for signs of endpoint renal failure (blood in urine, hunching, swollen kidneys; ref. 31). At endpoint kidneys were fixed in 10% neutral buffered formalin and embedded in paraffin for subsequent histologic analyses.

The TMA was analyzed using SPSS. All other statistical analyses were performed using Graphpad Prism. The specific statistical tests used are indicated throughout. All error bars represent ± SEM unless otherwise indicated.

In silico analysis of The Cancer Genome Atlas (TCGA) ccRCC dataset (5) revealed that RUNX1 alterations occur in 6% of ccRCC cases and strikingly, the vast majority of these alterations are mRNA upregulations (96%; Supplementary Fig. S1A). A 6% alteration rate is comparable with the rate at which other established genes involved in ccRCC are altered (34) such as MTOR (11%), PI3KCA (8%), PTEN (8%), TP53 (7%), while interestingly, the pattern of alterations is much more varied for these genes (Supplementary Fig. S1B). Kaplan–Meier survival analysis of the patients with ccRCC with RUNX1 mRNA upregulation shows that they have a statistically significant decrease in survival compared with the unaltered cohort (log-rank P = 0.0008, RUNX1 Unaltered median = 76.98 months, RUNX1 mRNA upregulation median = 36.21 months; Supplementary Fig. S1C). This is in line with a recent report interrogating the TCGA dataset (35). Furthermore, data obtained using the pan-cancer RNA-seq KM plotter (36) tool analyzing clinical survival data from 530 patients with ccRCC also show that high RUNX1 expression correlates with poorer overall survival (P < 0.0001; Supplementary Fig. S1D).

Tissue microarrays (TMA) have previously been used to investigate the protein expression level of RUNX1 in human epithelial tumors of the breast (17, 37) and ovary (38). Therefore, as an independent validation of in silico observations, we immunostained a TMA containing 184 tumor samples from patients with ccRCC. RUNX1 is clearly expressed in cell nuclei in a subset of ccRCC patient samples and is not expressed in the nontumor kidney sample contained within the TMA (Fig. 1A). RUNX1 staining was scored by the weighted average histoscore (H-Score) method to quantify the range of RUNX1 expression (see Materials and Methods). The TMA was stratified into quartiles and the upper quartile (RUNX1-High, H-Score: 30–225, mean = 87.5, n = 46) was compared with the remaining lowest scoring cores (RUNX1-Low, Q1-Q3 H-Score: 0–26.7, mean = 4.1, n = 138; Fig. 1B). Patient survival information was available for 183 patients. Kaplan–Meier survival analysis revealed that RUNX1-high patients had a significantly poorer cumulative survival than RUNX1-low patients (log-rank P = 0.007; Fig. 1C). The survival rate was consistently lower year on year and at 5 years from diagnosis was 68% for RUNX1-high patients compared with 88% for RUNX1-low, Wilcoxon P = 0.005 (Fig. 1C). Assessment of clinicopathologic characteristics showed that there was no significant association with RUNX1 H-Score and age, grade, necrosis, or recurrence (Table 1). However high RUNX1 expression was significantly associated with a high Klintrup–Makinen (KM) score, a pathologically defined measure of inflammation previously described for this TMA (see Materials and Methods). The average RUNX1 H-Score was significantly higher for patients with a high KM score compared with low (34.1 vs. 15.2, t test P = 0.0027; Fig. 1D). Accordingly, RUNX1-high patients were distributed 28% KM low versus 72% KM high, compared with 55% KM low versus 45% KM high for RUNX1-low patients (Fig. 1E). These data reveal for the first time that RUNX1 protein is aberrantly expressed in human ccRCC and that high RUNX1 expression is an independent marker of poor prognosis [P = 0.027HR = 1.58; 95% confidence interval (CI),1.054–2.372] when combined with age, stage, grade, and tumor necrosis. These data also reveal that RUNX1-high patients have an increase in inflammatory infiltration compared with their RUNX1-low counterparts.

Having conclusively shown that RUNX1 expression correlates with poorer survival in ccRCC, we wanted to ascertain a functional role for RUNX1 in this disease setting. To this end, RUNX1 expression was modulated in human ccRCC cell lines. Lentiviral delivery of different short hairpin RNAs (shRNA) was used to knockdown RUNX1 expression in 786-O and Caki-2 cells (sh1 and sh5) compared with a scrambled control shRNA (Scr; Fig. 2A, inset; Supplementary Fig. S2A). shRNA-mediated knockdown of RUNX1 caused a decrease in cell index (proportional to the number of adherent viable cells) over a 125-hour period in culture in the 786-O cell line as assayed using the xCELLigence assay system (Fig. 2A and B). Cell number was also significantly reduced in a second cell line (Caki-2) with RUNX1 knockdown (Fig. 2C). In addition, cell viability after RUNX1 knockdown in 786-O and Caki-2 cells was reduced as assessed by the MTS assay (Supplementary Fig. S2B and S2C). To validate these findings, 786-O cells were transfected with gRNA targeting RUNX1, and CAS9 nuclease. Complete knockout of RUNX1 protein was confirmed in 786-O RUNX1 CRISPR clones (CRISPR A1 and CRISPR A3) by immunoblot (Fig. 2D, inset). CRISPR deletion of RUNX1 also caused a more pronounced decrease in cell index (Fig. 2D and E) and a decrease in cell number (Fig. 2F) in both 786-O CRISPR clones.

To understand the nature of the growth defect observed in the RUNX1 knockout cells, the rate at which they were actively synthesizing DNA by incorporation of the thymidine analogue EdU was assessed. 786-O control and RUNX1-deleted cells were pulsed with EdU by incubation for 30 minutes in medium containing EdU, then sampled by fixation in 4% PFA immediately after EdU incubation (T0) or 6 hours later (T6). The cells were costained for EdU and PI (propidium iodide) and analyzed by flow cytometry as shown for the T6 time-point (Fig. 2G). There was no difference in total EdU incorporation between control and RUNX1-deleted cells at T0 (Supplementary Fig. S2D) and at T6 (Fig. 2H). However, there was a clear reduction in the G1* population representing EdU⁺ cells that have transitioned through S-phase and returned to G₁ in both the RUNX1–deleted cell lines (Fig. 2I and population highlighted in box in Fig. 2G). This suggests that the RUNX1-deleted cells face a delay in transitioning through the S/G₂ stages of the cell cycle. Finally, the number of dead cells was assessed by time-lapse imaging of the control and RUNX1-deleted cells in the presence of SYTOX Green nucleic acid stain. This revealed that the number of SYTOX-positive dead cells per well, as a proportion of percent confluence, was higher in the RUNX1-deleted cells compared with control, especially at earlier time-points (Fig. 2J). Confluence and the number of SYTOX-positive dead cells per well are shown individually in Supplementary Fig. S2E and S2F. Together, these data indicate that knockout of RUNX1 causes a reduction in cell growth in ccRCC cell lines and that RUNX1 CRISPR cells have a subtle delay in progression through the cell cycle and an increase in cell death.

To further investigate the effect of RUNX1 knockout in physiologically relevant assay systems, the effect of deletion on cell migration using in vitro scratch wound assays was assessed. This revealed that RUNX1-deleted cells exhibited decreased wound closure and reduced relative wound density over a 24-hour period (Fig. 3A–C). To establish whether RUNX1 deletion effects ccRCC development in vivo, and to circumvent the low tumorigenicity of the 786-O cell line, we generated a secondary cell line (hereafter referred to as 786-O*) by passaging 786-O cells through the kidney in vivo (see Materials and Methods). RUNX1 was deleted in these 786-O* cells by CRISPR/CAS9 as performed above (Supplementary Fig. S3A) and these 786-O* RUNX1-deleted cells showed a similar growth defect to the parental cells (Supplementary Fig. S3B and S3C). RUNX1-deleted and control 786-O* cells were injected directly into the kidney capsule of CD1-Nu/Nu recipient mice and their tumor growth was monitored by ultrasound over an 18-week period. This revealed that at 10 weeks postsurgery, 7 of 13 mice injected with the control cells had formed tumors compared with 0 of 13 for the RUNX1-deleted 786-O* cells (Fig. 3D). When sacrificed at 18 weeks, 8 of 13 mice injected with RUNX1-proficient cells had grossly observable kidney tumors, whereas just 1 of 13 of the recipients with RUNX1-deleted cells had a small tumor growth (P = 0.011; Fisher exact test; Fig. 3D; Supplementary Fig. S3D). Four of 13 control mice exhibited gross lung metastases while none of the RUNX1-deleted group did. Kidney tumors from the control group and the single tumor arising in the RUNX1-deleted cohort were stained for RUNX1 and its closely related family member RUNX2. RUNX1 was highly expressed in all control tumors tested (n = 4) while it was absent from the RUNX1-deleted tumor as expected (Fig. 3E). However, it was notable that RUNX2 was present in both control and RUNX1-deleted tumors. These data support our findings that RUNX1 is important for growth and survival of human ccRCC cells and that deletion of RUNX1 hampers tumor growth and development in vivo.

As RUNX1 deletion causes a defect in ccRCC cell growth, we wanted to understand the significant downstream players by assessing how deletion of RUNX1 affects the global transcriptional profile in human 786-O ccRCC cells. RNA sequencing was performed on whole RNA extracts from control and RUNX1 CRISPR cells (as used in Fig. 2). Several hundred genes were significantly differentially expressed (P < 0.05, >2-fold up or downregulation) in either RUNX1-deleted cell line (A1 = 1185, A3 = 1296). This revealed a novel RUNX1 regulated signature of 724 genes common to both clones that were significantly differentially expressed, with 710 altered in the same direction in both RUNX1 CRISPR clones compared with the control cells (Fig. 4A). Excluding uncharacterized genes, pseudogenes, and novel transcripts, 661 genes are significantly differentially expressed with 394 upregulated and 267 downregulated on RUNX1 deletion in both clones. Principal component analysis revealed exceptionally high agreement between the datasets with 97% of the variance explained by RUNX1 deletion (Supplementary Fig. S4). Full lists of the regulated genes are available in Supplementary Data File S1 where they are ranked by fold change and significance. Gene ontological analysis using Metacore revealed the main biological pathways that were altered on RUNX1 deletion. This encompasses a range of pathways such as cell adhesion and ECM remodeling, Eph, and Ephrin signaling, angiogenesis and glutathione metabolism (Fig. 4A). The most altered pathway was cell adhesion and ECM remodeling, which included changes in expression of genes such as MMP1, MMP16, SERPINE2, Fibronectin, and Syndecan 2 (Fig. 4B). The average fold change on the x-axis [x = Average log₂(fold change)] was plotted against significance on the y-axis [y = −log₁₀(Max(P_adj)] in a volcano plot to visually depict the most significantly differentially expressed genes (Fig. 4C). Two such genes, STMN3, which encodes a protein that plays a role in microtubule dynamics in the cell cycle (upregulated +46.3x, red circle Fig. 4C) and SERPINH1 (HSP-47), increased expression of which has been shown to be a marker of poor prognosis in ccRCC (downregulated −4.1×, blue circle Fig. 4C) were validated by Western blot analysis, which supported the findings of the RNA-seq data (Fig. 4D). Interestingly, the second most altered gene ontology was Eph and Ephrin signaling, which are downstream targets of the WNT signaling pathway that is itself modulated by RUNX1 activity (Fig. 4E). Finally, CPT1A, which has been shown to be suppressed in ccRCC, was increased on RUNX1 deletion (Fig. 4F). These data have, for the first time, identified a group of genes whose expression is significantly altered as a consequence of the level of RUNX1 in human ccRCC. This shows that deregulation of RUNX1 expression affects a wide range of key pathways, many of which are related to kidney cancer and cancer progression.

To further explore the functional role of RUNX1 in a physiologic setting, we turned to a GEM model where we could intrinsically modulate RUNX1 levels. First, we ascertained the levels of RUNX1 in a GEM model of kidney cancer available in our lab in which Cre recombinase expressed in the kidney epithelium drives deletion of the tumor suppressor Apc on a p21-null background (31). Normal kidneys and kidney tumors from this model (AH-Cre;Apc^fl/fl;p21^−/− referred to as CAP) were stained for RUNX1 to reveal that while RUNX1 is not expressed in normal kidney, it is significantly upregulated in kidney tumors (Fig. 5A). We proceeded to cross this CAP model with a conditional knockout of Runx1 (Runx1^fl/fl) (32). RUNX1 deletion in the tumors of CAP;Runx1^fl/fl mice was confirmed at the protein level by IHC, which showed absence of RUNX1 (Fig. 5B). Cohorts of CAP;Runx1^+/+ and CAP;Runx1^fl/fl mice were aged until clinical endpoint. Kaplan–Meier analysis shows that survival of CAP;Runx1^fl/fl mice was significantly extended (log-rank P = 0.0365) compared with their CAP;Runx1^+/+ counterparts, with a mean survival of 104.6 versus 78.6 days, t test P = 0.0415 (Fig. 5C and D). Tumors were immunostained for the proliferation marker Ki67, which exhibited lower positive staining in tumors from the CAP;Runx1^fl/fl mice compared with CAP;Runx1^+/+ (Fig. 5E). This was confirmed by quantification using the HALO imaging platform, which revealed that tumors from CAP;Runx1^+/+ mice had a higher proportion of Ki67⁺ cells than from CAP;Runx1^fl/fl mice (34% vs. 24.4%, t test, P = 0.0154; Fig. 5F). Finally, tumors immunostained for SERPINH1 (downregulated in RNA-seq, Fig. 4E) revealed SERPINH1 is highly expressed in CAP;Runx1^+/+ tumors compared with normal kidney (Fig. 5G). Deletion of RUNX1 causes a significant decrease in SERPINH1 levels in kidney tumors in line with our RNA-seq data. Taken together, these data from our GEM model of kidney cancer confirm in vivo that deletion of RUNX1 leads to improved survival and less tumor proliferation.

While deletion of Runx1 significantly delayed tumorigenesis in the GEM model, these animals still succumbed to disease. We hypothesized that the related RUNX family protein RUNX2 might be expressed and contribute to disease progression. Indeed, RUNX2 was expressed both in the CAP;Runx1^+/+ and CAP;Runx1^fl/fl tumors (Fig. 6A). Attempts to model deletion of RUNX2 in this model of kidney cancer were hampered by the nonviability of AH-Cre;Runx2^fl/fl mice (suggesting a possible limiting requirement for RUNX2 in embryonic development). Heterozygous deletion of Runx2 did not affect survival either on a Runx1^+/+ or Runx1^fl/fl background in the CAP model. Although, it is important to note that RUNX2 protein expression was still observed in these tumors (Supplementary Fig. S5A–S5D). Interestingly, however, in silico analysis of RUNX2 expression in the TCGA human ccRCC dataset (5) revealed that RUNX2 is altered in 8% of patients with ccRCC (93.5% are mRNA upregulations, Supplementary Fig. S5E). The pan-cancer RNA-seq KM plotter tool (described in Supplementary Fig. S1C) revealed human patients with ccRCC with high RUNX2 expression had poorer survival (log-rank P ≤ 0.0001; Supplementary Fig. S5F). It is noteworthy that RUNX3 is also upregulated in ccRCC; however, unlike RUNX1 and RUNX2, it does not significantly correlate with disease-free survival (Supplementary Fig. S5E and S5F). To directly assess the expression of RUNX2 in human patients with ccRCC, the same TMA in Fig. 1 was used to show that RUNX2 protein is also expressed in human ccRCC (Fig. 6B). While the RUNX2 H-Score was on average lower than RUNX1 (Supplementary Fig. S5G), when stratified into quartiles based on RUNX2 H-Score (Supplementary Fig. S5H), patients with high RUNX2 also had a statistically significant decrease in survival compared with patients with low or no RUNX2 expression (log-rank P = 0.0478; Fig. 6C). At five years post diagnosis, survival for the RUNX2-low quartile was 87% compared with 73% for RUNX2 high (Fig. 6C, inset table). A positive correlation between RUNX1-high and RUNX2-high expression was also observed (Supplementary Fig. S5I). Assessment of clinicopathologic characteristics for RUNX2 expression also showed no correlation with age, grade, necrosis, and recurrence (Supplementary Table S1). However, similar to RUNX1, RUNX2 also correlated with a high KM score (Fig. 6D and E). These data reveal that the related transcription factor RUNX2 is also important in human ccRCC with high expression being indicative of a poorer prognosis.

Taken together, we have identified a novel role for the RUNX family of transcription factors in kidney cancer where both RUNX1 and RUNX2 are expressed and act in an oncogenic fashion that aids the progression of the disease.

This study underscores the importance and functional relevance of the developmentally important transcription factor RUNX1 in kidney cancer. Interrogation of The Cancer Genome Atlas shows RUNX1 to be upregulated in ccRCC (this study and refs. 35, 39), which we can now corroborate using histochemical analysis of an independent cohort. Patients with ccRCC with poor prognosis have high RUNX1 expression, while deletion of RUNX1 reduced kidney cancer cell growth and prevented or delayed tumor development. In addition, we have shown for the first time that RUNX2 is also expressed in patients with poor prognosis. This work opens up a new and unexplored avenue of research into the RUNX genes' enigmatic functions in neoplastic disease and identifies the RUNX genes as novel players in the genetic landscape of kidney cancer.

Initial gene expression observations in silico revealed that alterations in RUNX1 occur at a similar frequency as perturbation of other kidney cancer drivers. Strikingly, almost all RUNX1 alterations were mRNA upregulations, suggesting increased expression is the mode by which RUNX1 contributes to ccRCC. RUNX1 activity appears to be associated with the neoplastic state in renal cells as normal tissues show little evidence of expression. This may reflect the specific ccRCC transcriptome, defined by the recurring molecular changes that typify this disease. Given the importance of hypoxia-induced factor (HIF) activation in the pathogenesis of this disease, it is worth noting that a number of studies have pointed to the interplay between HIFs and RUNX transcription factors including: physical interaction; coregulation of target genes; and in the case of RUNX2, stabilization of the HIF protein (40–42). Moreover, the RUNX genes are themselves regulated by HIFs, raising the possibility that they are both downstream targets, and act to potentiate the oncogenic signal (43). Molecular characterization of RCC subtypes revealed that increased immune cell infiltration gene expression signatures associated with the poorest performing patients specifically in ccRCC (4, 44, 45) and immune checkpoint inhibitors are currently in clinical trial for advanced disease (8, 9). In this regard, it is interesting that our study reveals a positive correlation between RUNX and local inflammatory cell infiltration. Increased systemic inflammation is a known feature of renal cancer (46, 47) and integration of KM score with systemic biomarkers was able to predict poor prognosis in ccRCC (27). Gene ontology profiles suggest that immune and inflammatory processes dominate the expression landscape of ccRCC (10, 11).

RUNX1 deletion in ccRCC cell lines perturbed the cell cycle and reduced cell viability while Runx1 deletion in our GEM model decreased tumor growth and tumor cell proliferation. This genetically confirmed an oncogenic role for RUNX1, which was also highly upregulated in another murine model of ccRCC (48). There is considerable evidence that RUNX1 has an important role in proliferation in organisms as diverse as nematodes (49), sea urchins (50), and mammals; although whether it promotes or restricts cell division depends on the cellular context (51, 52). Similarly RUNX1 is known to regulate cell survival differently in different cell types (53, 54), and downstream mediators of survival have been identified in some tissue systems (55). Our cell line data is given greater physiologic relevance by the observation that RUNX1-null ccRCC cells almost entirely failed to grow in a kidney xenograft model. Furthermore, our data showing a pro-proliferative effect in cell lines and tumors together with enhanced cell survival suggests an exclusively oncogenic role for RUNX1 in the context of renal cancer cells.

Using RNA-seq, we revealed that deletion of RUNX1 induces profound gene expression changes. KEGG analysis of RCC expression profile studies has emphasized that upregulated genes are associated with significant cell adhesion changes and interactions between cytokines and their receptors (56). In this regard, it is worth noting that our gene analysis showed enriched expression of genes involved in cell–ECM interactions and cell–cell interactions such as Eph–Ephrin signaling, suggesting RUNX1 may be contributing to a common oncogenic pathway in renal cancer. One of the most significantly downregulated genes in our human RUNX1-deleted RCC cells and murine tumors was SERPINH1. Importantly, SERPINH1 is a potential negative prognostic marker in ccRCC (57) and is required for collagen folding and secretion (58), therefore its expression may contribute to changes in tissue architecture that promote tumor development. SERPINH1 also associates with enhanced TGFβ signaling and both RUNX1 and RUNX2 have been shown to be involved in TGFβ-induced kidney fibrosis (23, 59). The role of collagen in ccRCC is currently unclear; however, collagen density and alignment have recently been shown to be significantly higher in patients with high-grade tumors compared with low grade (60). The fatty acid metabolism enzyme CPT1A increased markedly on RUNX1 deletion, suggesting a negative correlation. CPT1A is reduced in ccRCC where suppression causes lipid droplet accumulation (a prominent feature of ccRCC) and tumor development (61). Intriguingly, suppression of CPT1A in ccRCC is mediated by the HIF family and is therefore an example of potential RUNX1 interplay with the VHL–HIF signaling axis. We also observed a pronounced increase in STATHMIN3, a microtubule-binding protein important for the formation of mitotic spindles. Overexpression of STATHMIN3 has been associated with delayed cell cycle in leukemia (62). Future studies using chromatin immunoprecipitation analysis would provide valuable insights into which genes are directly modulated by RUNX activity and functionally contribute to the RUNX1-related phenotype in renal cells.

The long-term trend for kidney cancer is one of growing global incidence, and improved treatments for advanced disease remains an unmet clinical need. Human patients with the highest RUNX1 expression in our study had the poorest prognosis and a 20% reduction in survival rate at 5 years post diagnosis (68% vs. 88%). Indeed, RUNX1 associated with poorer survival independent of age, grade, and stage. These data identify RUNX1 as a novel prognostic biomarker and as a potential therapeutic target in human ccRCC. This is encouraging given the active pursuit of therapeutic agents that can block the transcriptional function of the RUNX proteins (63). However, the wider consequences of directly targeting RUNX in kidney cancer would need to be established in the context of the sustained requirement for RUNX function in other tissues.

In general, the relationship between the RUNX genes and other hematopoietic and solid tumors is complex with both a tumor suppressor and a pro-oncogenic role described in leukemia (14, 15), breast (18), and prostate cancer (21). As such ccRCC may be the optimal choice for exploring novel therapeutic agents that block RUNX function. This is further given credence considering that the related family member RUNX2 is also associated with poorer survival and increased inflammation in patients with ccRCC, and was also highly expressed in our GEM model. Indeed RUNX1 and RUNX2 co-occurred in a selection of patient samples as well as in GEM tumors. Although our staining was carried out on serial sections, dual IHC for both RUNX1 and RUNX2 could give valuable insights into the spatial localization and consequence of co-occurrence of the RUNX transcription factors. Although beyond the scope of this study, it will be important to dissect this interplay between the RUNX proteins in ccRCC and how they each contribute to the disease phenotype. Furthermore, while we have not specifically investigated RUNX3 in our system, in silico analysis revealed it is also upregulated in kidney tumors at the mRNA level. Intriguingly, akin to that observed in pancreatic adenocarcinoma (64), transcriptomic upregulation of RUNX3 did not relate to patient survival. Nonetheless, Whittle and colleagues elegantly demonstrated that high RUNX3 in their pancreatic cancer TMA did correlate with poor prognosis and conveyed a prometastatic phenotype. Therefore, it will be interesting to study RUNX3 further in the context of ccRCC to ascertain whether its role recapitulates that seen in pancreatic cancer. Future studies will use compound genetic models and anti-RUNX drugs to investigate the consequence of total ablation of RUNX function in kidney cancer.

O.J. Sansom reports receiving a commercial research grant from Novartis, AstraZeneca, and Cancer Research Technology/BMS. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N. Rooney, O.J. Sansom, E.R. Cameron, K. Blyth

Development of methodology: N. Rooney, S.M. Mason, C. Nixon, O.J. Sansom, E.R. Cameron, K. Blyth

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Rooney, S.M. Mason, L. McDonald, J.H.M. Däbritz, K.J. Campbell, S. Howard, D. Athineos, J Edwards

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Rooney, S.M. Mason, K.J. Campbell, A. Hedley, D. Athineos, J.D.G. Leach, J. Edwards, E.R. Cameron, K. Blyth

Writing, review, and/or revision of the manuscript: N. Rooney, S.M. Mason, K.J. Campbell, W. Clark, J.D.G. Leach, O.J. Sansom, J. Edwards, E.R. Cameron, K. Blyth

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.M. Mason, A. Hedley, C. Nixon, W. Clark, J.D.G. Leach

Study supervision: K. Blyth

Other (histopathology interpretation): J.D.G. Leach

We would like to acknowledge the Core Services and Advanced Technologies at the Cancer Research UK Beatson Institute (C596/A17196), with particular thanks to the Biological Services Unit, Histology and Molecular Technologies. We thank the NHS Greater Glasgow and Clyde Biorepository for supplying the TMA; Tom Hamilton and Sandeep Dhayade for assistance with xenograft surgery; Professor Eyal Gottlieb for providing 786-O and Caki-2 cell lines and David Stevenson for advice on CRISPR strategy. We thank Catherine Winchester for critical reading of the manuscript. This work was funded by CRUK core funding C596/A17196 (KB and OJS labs - CRUK Beatson Institute) and Renal Cancer Research Fund Scotland (to J. Edwards) who supported the TMA production.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.