Identification and validation of NFIA as a novel prognostic marker in renal cell carcinoma

Abstract

Prognostic tools are an essential component of the clinical management of patients with renal cell carcinoma (RCC). Although tumour stage and grade can provide important information, they fail to consider patient‐ and tumour‐specific biology. In this study, we set out to find a novel molecular marker of RCC by using hepatocyte nuclear factor 4A (HNF4A), a transcription factor implicated in RCC progression and malignancy, as a blueprint. Through transcriptomic analyses, we show that the nuclear factor I A (NFIA)‐driven transcription network is active in primary RCC and that higher levels of NFIA confer a survival benefit. We validate our findings using immunohistochemical staining and analysis of a 363‐patient tissue microarray (TMA), showing for the first time that NFIA can independently predict poor cancer‐specific survival in clear cell RCC (ccRCC) patients (hazard ratio = 0.46, 95% CI = 0.24–0.85, p value = 0.014). Furthermore, we confirm the association of HNF4A with higher grades and stages in ccRCC in our TMA cohort. We present novel data that show HNF4A protein expression does not confer favourable prognosis in papillary RCC, confirming our survival analysis with publicly available HNF4A RNA expression data. Further work is required to elucidate the functional role of NFIA in RCC as well as the testing of these markers on patient material from diverse multi‐centre cohorts, to establish their value for the prognostication of RCC.

Introduction

Renal cell carcinoma (RCC) comprises a range of tumour types arising from the epithelial cells of the nephron, the functional unit of the kidney [1]. The nephron is a tubular structure covered with a single epithelial layer, with highly specialised cell types executing discrete and highly specific uptake and excretion of substances in a section‐specific and strictly regulated fashion [2, 3]. According to the WHO classification, there are more than 50 established and provisional distinct subtypes of RCC, each with a defined morphology and genetic characteristic [4]. These characteristics can be partly traced back to the cell type in the nephron from which the tumour originated. With the advent of single‐cell sequencing, the transcriptional wiring in different sections of the nephron is currently being delineated, providing a detailed map of the different cell types along the nephron [5].

The three major RCC types are clear cell RCC (ccRCC), papillary RCC (pRCC), and chromophobe RCC (chRCC), together representing 95% of all RCCs. ccRCC and pRCC represent 75 and 15% of all RCCs, respectively, and it is believed that they stem from the proximal cells of the nephron [6]. Indeed, by comparison with normal proximal tubule cells, ccRCC and pRCC maintain a range of proximal characteristics, including key transcription factors (TFs). In contrast, chRCC shows strong transcriptional similarities to intercalated distal nephron cells [7, 8]. All major RCC subtypes have been extensively characterised with respect to genetic aberrations, and it is likely that the genetic landscape of each subtype is selected for based on the constitution of the transcriptional wiring of the cell of origin. For example, ccRCC is initiated by a functional loss of the von Hippel–Lindau (VHL) tumour suppressor gene leading to a constitutive activation of the hypoxia‐inducible factors (HIF1A and HIF2A) and their respective transcriptional programme [9, 10]. Recent data indeed indicate that the oncogenic effect of HIF activation in proximal tubule cells is dependent on transcriptional cooperation with tissue‐specific transcriptional networks, including PAX8 and HNF1B [11].

Another transcriptional regulator implicated in RCC biology is the hepatocyte nuclear factor 4A (HNF4A), a TF belonging to the nuclear receptor family, primarily expressed in liver, gut, kidney, and pancreatic beta‐cells, where it acts as a master transcriptional regulator [12]. In the kidney, HNF4A is specifically expressed in the proximal tubule cells, regulating key target genes vital to the designated function of this part of the nephron [13, 14, 15]. We have shown that the HNF4A transcriptional activity is maintained and characterises ccRCC and pRCC, which are derived thereof. Furthermore, ccRCC tumours with increased malignancy display lower levels of HNF4A expression and related target genes, indicating that de‐differentiation and loss of lineage fidelity play a major role in RCC tumorigenesis [8]. HNF4A expression has been extensively studied at a transcriptional level, but validation at the protein level in RCC patients is limited to one study with only 30 ccRCC patients [16]. In our current study, we consolidate these findings utilising a larger patient cohort whilst including pRCC, the second most common RCC subtype.

Based on the concept that loss of lineage fidelity is associated with RCC progression, we investigated other TFs that mimic HNF4A expression in non‐tumour kidney tissue (herein referred to as normal kidney) and RCC. One standout gene was nuclear factor 1A (NFIA), expressed in the proximal tubules of the nephron. NFIA transcriptionally defines and regulates the development of multiple organs [17] including liver [18], brain [19], kidney [20], and lung [21]. NFIA has been studied in brain tumours where it acts as a tumour suppressor gene [22, 23] and in oesophageal cancer; NFIA expression provides prognostic information [24]. To the best of our knowledge, neither the protein expression nor the role of NFIA in RCC has been previously studied.

In the current study, we confirm the grade and subtype‐specific expression of HNF4A in RCCs and explore its utility as a prognostic marker. Second, we show that NFIA, a proximal tubule‐associated TF, is a novel prognostic marker for ccRCC and pRCC, with similar association to tumour aggressivity as HNF4A.

Materials and methods

Patient characteristics and tissue collection

The study comprised 376 patients with histopathologically verified RCC. Subtype‐specific patient characteristics are summarised in Table 1. All patients were diagnosed at the Department of Urology, Umeå University Hospital, Sweden between 1990 and 2010. At metastatic disease, most patients were treated with interferon or hormones, whilst a minority had palliative treatment only. All patients were subject to yearly follow‐up, screened in the medical records and screened for being alive in the Swedish National Population Register. The last follow‐up was done in December 2020. Cancer‐specific survival (CSS) time was defined as the time from diagnosis to the date of death from RCC or alive at the end of December 2020. Histopathological classification of RCC type was performed according to the Heidelberg classification [25]. Nuclear grading was performed according to Fuhrman et al [26]. The updated TNM classification 2017 was used for tumour stage grouping [27]. In the stage grouping, patients with Nx were joined with N0, and patients with Mx joined with M0. The distribution of Fuhrman grades, tumour, node, and metastases status across grouped stages in ccRCC and pRCC cohorts is presented in supplementary material, Figures S1 and S2. Tumour size, defined as the largest tumour diameter, was measured primarily on the computed tomography (CT) or magnetic resonance imaging (MRI) scans. The median (range) tumour size was 7 (1–25) cm. The venous invasion was defined as tumour invasion in major renal veins, verified microscopically in the renal hilum. All patients were followed with clinical and radiological examinations.

Table 1.

TMA patient cohort characteristics

Characteristics	ccRCC (n = 313)	pRCC (n = 50)
Age range	30–90	20–90
Median age	67	67
Sex
Female	138	16
Male	175	34
Grouped staging
I	112	19
II	48	13
III	71	9
IV	81	8
Grade
G1	30	7
G2	95	20
G3	130	16
G4	58	5
Median follow‐up (years)	4.8	6.8
Dead by disease	153	22

Tissue microarray construction

For tissue microarray (TMA) construction, tumour sections were screened by a pathologist and representative tumour cores (0.6 mm in diameter) were arranged in recipient TMA paraffin TMA blocks, as described previously [28]. Most tumours were represented by 2–4 valid tissue cores. Eight ccRCC and two pRCC patients were excluded from analysis due to loss of material during processing. All samples from included patients were histopathologically re‐evaluated and TMAs, containing duplicate 1.00 mm cores, were constructed from both non‐tumour and tumour kidney tissue using a tissue array machine (Beecher Instruments, Microarray Technology, MD, USA).

Immunohistochemistry staining

The TMA blocks were sliced into 4 μm sections and treated according to standard procedures including de‐paraffinisation and rehydration. Immunostaining was performed using Dako's Autostainerplus with the EnVisionFlex High pH‐kit (Dako, CA, USA) and the following antibodies: HNF4A (HPA004712, Atlas Antibodies, Bromma, Sweden) and NFIA (HPA006111, Atlas Antibodies, Bromma, Sweden) both at 1:200 dilution.

Image acquisition and data analysis

Digital images of slides were extracted using a Zeiss Axion scanner (Jenna, Germany) at ×200 magnification. Image analysis was processed in QuPath v.0.3.2 (Queen's University, Belfast, Northern Ireland). TMA slides were de‐arrayed and pre‐processed as previously described [29]. Cells were detected using cell detection and cell classifiers were created and trained for each tumour type to separate tumour cells from stromal and immune cells. To evaluate the staining intensity, 3,3'‐Diaminobenzidine (DAB) staining intensity classification was performed, and the Allred score was calculated by QuPath for each core. For further statistical analysis, the average of the Allred score of all cores from each patient was calculated.

Ethical approval

The patients had provided informed consent, orally before year 2000, and informed and written consent from year 2000. The study was reviewed and approved by the Ethical Review Board (Dnr: 2015‐146‐31M and Dnr: 2018‐296‐32M) and the Ethical Board of Sweden (Dnr: 2019‐02579). All procedures regarding patients were in accordance with the Helsinki Declaration. The data used were anonymised and throughout the project, all data were treated under the regulations of the General Data Protection Regulation Act.

TCGA analysis

The Cancer Genome Atlas (TCGA) data were downloaded from the GDC data portal (https://portal.gdc.cancer.gov/) using the R package TCGAbiolinks [30]. Additional clinical data regarding grade of the KIRC cohort were downloaded separately from firebrowse.org. For the data analysis, transcripts per million unstranded data of tumour samples were extracted for each RCC type, followed by a log2 transformation after adding an offset of 1. TF network activity was analysed using the R package single‐cell regulatory network inference and clustering (SCENIC) according to Aibar et al [31]. In the first step, potential TF targets were identified by GENIE3 followed by a TF‐motif enrichment analysis and identification of direct targets (regulons) using RcisTarget. The last step consisted of scoring the activity of regulons performed by AUCell. Finally, the network activity was binarised.

Single‐cell RNA‐sequencing analysis

Single‐cell RNA‐seq data of ccRCC patients were obtained from Obradovic et al [32]. Raw data were downloaded from https://data.mendeley.com/datasets/nc9bc8dn4m/1. For the re‐analysis, the data were subset into normal adjacent tissue and further split into a CD45 positive and a CD45 negative subset. For the analysis in this paper, only the subset of CD45 negative of adjacent normal tissue was used. Re‐processing of the data was performed in R using Seurat (v.4.1.1) [33] and harmony (v.0.1.0) [34] was used for data integration. Initial Quality control comprised the removal of cells with fewer than 200 and more than 5,000 features detected as well as a mitochondrial content >25%. Furthermore, features needed to be expressed in a minimum of 50 cells to be included in the data.

Before data integration, the data were normalised using Seurat's function NormalizeData followed by a feature selection to find the 2,000 most variable genes. In addition, the data were scaled using ScaleData and a principal component analysis was performed. Finally, RunHarmony was used for data integration to adjust for patient‐to‐patient variation. For visualisation, uniform manifold approximation and projection was performed on the first 20 dimensions of the harmony reductions. The Louvain algorithm implemented in Seurat was used for cluster analysis with a resolution parameter of 0.5. Seurat's function FindAllMarkers was applied for differential gene expression analysis to annotate clusters. After initial annotation, the data were again subset to include only cells of the nephron. This subset was re‐processed once again using the same workflow as stated above.

Statistical analysis

Data were compiled, analysed, and visualised using R (v.4.2.1). Survival analysis was performed by using the log‐rank test and the Kaplan–Meier method, where patients who died from non‐RCC causes and/or had poor or missing cores were censored. As a threshold for the Kaplan–Meier analysis, the median was used for the TCGA data and an Allred score of 4 for the TMA analysis. Pearson correlation was used for the correlation analysis between the HNF4A and NFIA staining. Multivariable testing was performed on all clinical parameters that were found to be significant with univariate analysis using the Cox regression model. P < 0.05 was considered statistically significant.

Results

Analyses of TFs in RCC cohorts

Our previous work investigating transcriptional programs in the normal nephron and RCC revealed HNF4A as a tumour‐defining transcriptional marker in ccRCC and pRCC [8]. Experimental studies have corroborated the role and significance of HNF4A in RCC tumorigenesis and lineage fidelity [13, 14, 35]. In our current study, we set out to identify other novel TFs, apart from HNF4A, that may be operating within RCC tumours to build upon the molecular definitions of the disease as well as understand clinical implications. We identified expression signatures of the histological RCC subgroups that can be intimately associated with specific TF networks. We used the SCENIC package that is designed to infer TFs and gene regulatory networks or ‘regulons’. Using this approach, we could corroborate the signifying HNF4A activity in ccRCC (KIRC) and pRCC (KIRP) within the TCGA data sets (Figure 1A). In contrast, and as previously reported, the FOXI1 signature is highly enriched in chRCC [8]. Thus, our SCENIC analysis defined key TFs seem to be transcriptionally active and defined the different molecular subtypes of RCC. Using our analysis, we identified NFIA, a TF not previously studied within the context of RCC. The expression of NFIA and its target genes are highly enriched in the ccRCC and pRCC cohorts, whilst the same is undetected or extremely low in chRCC which is known to arise from the distal segments of the nephron. The network enrichment signature of NFIA in ccRCC and pRCC subtype cohorts mimics the enrichment signature to that of HNF4A. These observations were further evidenced through our correlation analysis of HNF4A and NFIA RNA in the TCGA cohort, where a significantly positive correlation was seen in ccRCC (R = 0.5) and pRCC (0.45) tumours (Figure 1B) (validation in supplementary material, Figure S3). In line with this, our analysis of HNF4A and NFIA in normal kidney single‐cell data (Figure 1C) show an enrichment of these TFs in the proximal tubular cells S2 and S3 (Figure 1D).

Figure 1.

Analysis of TCGA RCC cohorts and kidney single‐cell RNA‐seq data. (A) Regulon activity of target genes by selected transcription factors within RCC subtypes in the TCGA cohort, with number of target genes indicated in brackets. (B) Correlation of NFIA and HNF4 RNA expression in ccRCC (n = 540) and pRCC (n = 290) samples within the TCGA patient cohort. R = Pearson's correlation coefficient. Line is line of best fit; shaded area represents 95% CI. (C) Re‐analysis dependent clustering of kidney single‐cell RNA‐seq data. (D) RNA expression of NFIA and HNF4A in normal kidney cell types based on re‐analysis.

Pan‐TCGA HNF4A / NFIA expression

We next sought to dissect the expression pattern of NFIA in TCGA‐based RCC cohorts, relate this expression to HNF4A and finally validate the observations from our SCENIC analysis. To broadly understand the expression of NFIA in cancer and more specifically in RCC, we analysed its expression at the transcriptional level in the TCGA cohort which included 33 cancer types (Figure 2A). In line with previous reports on the role of NFIA in development and cancer, it is broadly expressed across many tumour types [22]. This includes ccRCC (KIRC) and pRCC (KIRP) whilst having relatively low expression in chRCC which is in line with our SCENIC analysis. Given that NFIA expression is primarily enriched within the proximal tubular segments of the nephron and to an extent is conserved in ccRCC and pRCC tumours, this data further validate our understanding of the proximal origin of these RCC subtypes [36, 37]. Based on this reasoning, it is unsurprising that chRCCs relatively lack NFIA expression as they are thought to arise from the intercalated cells of the collecting duct. We also explored the expression of HNF4a within the same tumour panel and as expected, there was a greater disparity in expression level across tumours, reflecting the highly tissue‐specific expression of this TF. As previously reported, expression of HNF4A within the RCC cohorts is largely restricted to ccRCC and pRCC whilst remaining relatively low in chRCC reflecting the cell of origin characteristics of the respective tumour subtypes [8] (Figure 2B). As a result of the low levels of HNF4A and NFIA expression in the distal segments of the nephron and chRCC, we excluded this subtype from subsequent analyses.

Figure 2.

Pan‐TCGA expression of NFIA and HNF4A. Relative RNA expression (transcripts per million) of (A) NFIA and (B) HNF4A in 33 cancer types from TCGA data including ccRCC (KIRC, n = 540), pRCC (KIRP, n = 290), and chRCC (KICH, n = 65) RCC subtypes.

HNF4A and NFIA expression correlates with favourable survival

To better understand the role HNF4A and NFIA play in RCC, we analysed TCGA‐based RNA expression of NFIA and HNF4A in relation to grade (Figure 3A) and stage in ccRCC and only stage in pRCC (TCGA grade data is not available for pRCC) (Figure 3B,D). Higher grades displayed lower levels of NFIA or HNF4A expression in ccRCC (p < 0.0001 and p < 0.01, respectively). Furthermore, we dichotomised tumours based on expression level for HNF4A or NFIA and examined patient overall survival between these two assigned groups (Figure 3C,E). Within the TCGA ccRCC and pRCC tumours, there is a tendency for higher stage tumours to display relatively lower levels of NFIA expression in comparison to low‐stage tumours. In line with these observations, when RCC tumours were grouped by subtype, Kaplan–Meier survival plots show that high NFIA expression confers better CSS probability in ccRCC (p < 0.0001) and pRCC (p < 0.041). High HNF4A expression correlated with favourable CSS probability in ccRCC (p > 0.001) but not in pRCC.

Figure 3.

HNF4A and NFIA expression confers favourable survival in TCGA RCC cohorts. (A) Expression of NFIA and HNF4A in ccRCC patients within TCGA, split by Fuhrman grade. n = 14 (G1), 236 (G2), 206 (G3), and 76 (G4). (B) Expression of NFIA in ccRCC and pRCC tumours within TCGA, split by stage. ccRCC, n = 272 (stage I), 59 (stage II), 123 (stage III), and 83 (stage IV). pRCC, n = 172 (stage I), 21 (stage II), 52 (stage III), and 15 (stage IV). (C) Kaplan–Meier survival analysis of ccRCC and pRCC patients within TCGA based on high (>median) or low (<median) expression of NFIA. ccRCC, n = 270 (high), 270 (low). pRCC, n = 145 (high), n = 145 (low). (D) HNF4A expression in ccRCC and pRCC tumours within TCGA, split by stage. (E) Kaplan–Meier survival analysis of ccRCC and pRCC patients within TCGA based on high (>median) or low (<median) expression of HNF4A.

HNF4A and NFIA expression confers better CSS probability in the validation TMA cohort

To validate our TCGA data‐derived findings, we opted to immunohistochemically stain a TMA of RCC tumours consisting of ccRCC (Figure 4A) and pRCC (Figure 4B) for HNF4A and NFIA proteins. We related the Allred score (a combination indicator of positively stained cell percentage and intensity) of these TMA cores to tumour grade, stage, and patient CSS (Figures 5 and 6). Lower levels of NFIA were observed with higher ccRCC tumour stages (p < 0.05) and grades (p < 0.0001) (Figure 5A). A similar association was observed for HNF4A in relation to tumour grade (p < 0.001) but not stage (Figure 5C). These results corroborate the mRNA data, showing that the loss of NFIA and HNF4A protein is associated with increased tumour malignancy in ccRCC. In concordance, we were able to confirm that both high NFIA (p < 0.0001) and HNF4A (p < 0.0001) protein expression confers favourable survival in ccRCC patients (Figure 5B,D).

Figure 4.

HNF4A and NFIA immunohistochemistry (IHC) staining on TMA tumours. (A) Representative IHC staining of HNF4A and NFIA on ccRCC tumours within the TMA, with Allred score range indicated. Scale bar represents 50 μm. (B) Representative IHC staining of HNF4A and NFIA on pRCC tumours within the TMA, with Allred score indicated. Scale bar indicates 50 μm.

Figure 5.

HNF4A and NFIA expression confers favourable CSS in the ccRCC patient TMA. (A) Allred score of ccRCC tumours stained with NFIA, split by stage and grade. Stages I–IV, n = 108, 47, 69, and 80. Grades I–IV, n = 26, 94, 130, and 55. (B) Kaplan–Meier survival analysis of ccRCC tumours based on high (≥4, n = 219) and low (<4, n = 83) NFIA Allred score. (C) Allred score of ccRCC tumours stained with HNF4A, split by stage and grade. Sample numbers same (A). (D) Kaplan–Meier survival analysis of ccRCC tumours split by high (≥4, n = 158) and low (<4, n = 145) HNF4A Allred score. Post hoc log‐rank test *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Figure 6.

NFIA but not HNF4A expression confers longer CSS in pRCC TMA. (A) Allred score of pRCC tumours stained with NFIA, split by stage and grade. Stages I–IV, n = 19, 13, 9, and 7. Grades I–IV, n = 7, 20, 16, and 5. (B) Kaplan–Meier survival analysis of pRCC tumours based on high (≥4, n = 32) and low (<4, n = 16) NFIA Allred score. (C) Allred score of pRCC tumours stained with HNF4a, split by stage and grade. Same sample numbers as (A). (D) Kaplan–Meier survival analysis of pRCC tumours split by high (≥4, n = 35) and low (<4, n = 12) HNF4a Allred score. Post hoc log‐rank test *p < 0.05.

Similarly to the TCGA data analysis in pRCC, lower NFIA levels were only observed in stage III compared to stage I (p < 0.05) and no trend with relation to grade was observed (Figure 6A). Furthermore, a stage or grade‐dependent HNF4A expression trend was not observed (Figure 6C). In line with the TCGA data, high HNF4A protein expression did not result in a survival benefit for patients with pRCC tumours (Figure 6D). For pRCC patients, high NFIA expression conferred a survival benefit (p = 0.02) (Figure 6B). As the pRCC cohort encompasses a limited number of patients (n = 50), analysis on a larger pRCC cohort is warranted to further clarify the role of HNF4A and NFIA in this RCC subtype.

Prognostic significance of NFIA and HNF4A in ccRCC

To investigate whether NFIA expression was an indicator of CSS in ccRCC patients in the TMA cohort, we performed univariable Cox regression analyses. High NFIA expression in ccRCC patients indicated a favourable effect on CSS (hazard ratio [HR] = 0.44, 95% CI = 0.32–0.62, p value < 0.001) (Table 2). Other clinicopathological parameters were also assessed via univariable Cox regression analysis to identify parameters that would influence CSS in ccRCC patients. When univariably tested, six out of nine parameters indicated a statistically significant effect on CSS. Subsequently, multivariable Cox regression analysis was performed to ascertain whether NFIA expression level could serve as a prognostic factor independently of these six clinicopathological features. We found that high NFIA expression could independently indicate better CSS in ccRCC patients (HR = 0.46, 95% CI = 0.24–0.85, p value = 0.014). A similar univariable and multivariable analysis was performed for HNF4A and although HNF4A univariably had a significant effect on CSS, we failed to detect such an effect on CSS when adjusted for other clinicopathological features (supplementary material, Table S1), suggesting that HNF4A protein expression is not an independent prognostic marker in ccRCC.

Table 2.

Evaluation of the prognostic role of NFIA on CSS in univariable and multivariable analyses

		Univariable			Multivariable
Covariate	Units	Hazard ratio	95% CI	P value	Hazard ratio	95% CI	P value
NFIA	<4 Allred Score	ref.
	≥4 Allred Score	0.44	0.32–0.62	<0.001	0.46	0.24–0.85	0.014
Grade	III–IV	ref.
	I–II	0.36	0.25–0.53	<0.001	0.30	0.15–0.61	0.001
Stage	III–IV	ref.
	I–II	0.14	0.10–0.21	<0.001	0.14	0.06–0.33	<0.001
Age		0.99	0.98–1	0.16
Sex		0.91	0.66–1.3	0.59
C‐reactive protein		1	1–1	<0.001	1.01	1.01–1.02	<0.001
Weight		0.99	0.97–1	0.037	1.01	0.99–1.03	0.951
Venous invasion	No venous growth	ref.
	Sinus vein	2.01	1.04–3.89	0.038	2.45	0.85–7.03	0.096
	Renal vein	2.37	1.57–3.59	<0.001	0.95	0.40–2.26	0.916
	Vena cava	2.61	1.68–4.06	<0.001	0.68	0.31–1.46	0.312
Tumour diameter		1.01	1.01–1.02	<0.001	1.00	0.99–1.01	0.842
Bilateralness	No
	Primary	1.19	0.49–2.9	0.703
	Secondary	1.07	0.34–3.36	0.906

ref., reference category.

Discussion

Although the clinical outlook for RCC patients has improved in the last decade, mainly due to the use of molecular therapies such as kinase and checkpoint inhibitors, RCC remains a difficult to treat disease entity. Up to 30% of RCC patients develop tumour recurrence after being considered disease‐free at primary diagnosis [38, 39, 40]. This has highlighted the importance of not only gaining further information about RCC biology but also identifying patients that may have poorer prognoses. TNM classification and histological grade are currently the standard for determining RCC patient prognosis; however, its accuracy remains suboptimal as it is likely to disregard patient‐related factors [41]. In the current study, we have identified a novel molecular marker, NFIA, that can independently prognosticate RCC patients. We have validated our findings through comparisons to the well‐established RCC and proximal tubule TF, HNF4A, and presented novel data on its expression in pRCC with relation to clinical parameters.

HNF4A is one of the most extensively studied tissue‐specific TF regulatory networks [42]. In line with this, we based our search for a novel TF that is operating within RCC in a similar manner to HNF4A. Our SCENIC analysis identified a gene coding for TF NFIA, which operates within the same RCC subtypes as HNF4A. On this basis, we founded our subsequent analyses of NFIA in comparison to HNF4A, using the former as a blueprint for how an influential TF within RCC may be uncovered. Our analysis of HNF4A and NFIA in the single‐cell kidney transcriptome and pan‐TCGA cohorts further likened the expression of NFIA to HNF4A. Both TFs are primarily expressed in the proximal tubule of the kidney, more specifically the S2 and S3 populations. HNF4A is an important lineage defining TF in proximal tubule development [13, 14] and maintenance [43]. In fact, studies also show that NFIA is a key factor in development where haploinsufficiency leads to a range of perinatal abnormalities including renal defects [20]. Given the key role of these two TFs in proximal tubule development and homeostasis, it is no surprise that there is a large overlap in expression within the subpopulations of proximal tubular cells. Our pan‐TCGA expression analysis of the two TFs shows that they are widely expressed across multiple tumour types, but more importantly, within RCC, the tumour subtype‐specific expression pattern is conserved between the two TFs. This likely indicates the retention of some cell of origin gene expression characteristics even after transformation which warrants further investigation into the potential role of NFIA in RCC.

We also looked at the correlation of these two TFs in RCC subtype cohorts within the TCGA, which was subsequently validated by correlation analysis using the TMA cohort (supplementary material, Figure S3). In both analyses, NFIA expression positively correlated significantly with HNF4A expression. This could indicate that NFIA may be under the regulation of HNF4A or they are regulated by overlapping signalling pathways. Alternatively, they operate independently but display similar associations to disease progression in ccRCC. Further work is required to elucidate the mechanisms pertaining to their expression, e.g. through the analysis of binding sites on both genes. Regardless of the regulatory mechanisms in place, the positive correlation provides further relevance to NFIA in RCC by associating it with HNF4A, a well‐described TF behaving as a tumour suppressor gene in ccRCC [16].

To gain further clinically relevant information pertaining to NFIA, we compared its expression to clinical variables such as tumour grade, stage, and CSS, and compared this data to that of HNF4A. We performed these analyses on the TCGA RCC patient cohort and were able to validate our findings with analyses of the TMA. In both TCGA and TMA cohorts, we observed that the expression of NFIA significantly decreased with increasing tumour malignancy. Interestingly, this trend was not replicated with HNF4A, where our TMA data showed a significant decrease in HNF4A expression with increasing grade and stage, observations that were not replicated in the TCGA data set. Additionally, high tumour expression of NFIA or HNF4A conferred longer CSS compared to low‐expressing ccRCC tumours in both TCGA and TMA cohorts. Given the discrepancy in HNF4A expression in relation to tumour stage and grade between the TCGA and TMA cohorts, we believe that our TMA data portray the biologically relevant setting, as it is based on protein data. Post‐translational regulation is likely to play a role in HNF4A abundance in tumours, which could serve as an explanation for why differences in HNF4A abundance between stages and grades (in ccRCC) or stages (in pRCC) are not detected in RNA data from the TCGA. Furthermore, our TMA data are able to validate previously published TMA data from ccRCC by Gao et al [16]. We are the first to validate that HNF4A does not correlate with stage or grade and its expression level does not confer any CSS advantage in pRCC. Despite pRCCs and ccRCCs both arising from proximal tubule segments of the nephron, pRCCs are genetically and transcriptionally less uniform than ccRCC tumours [44, 45], where the latter can be defined by a near‐universal loss of VHL function [46]. This could explain the inability of HNF4A expression to segregate patients based on stage, grade or CSS in pRCC. However, our TMA‐based NFIA data suggest that the expression of this TF in pRCC can segregate patients based on stage and CSS but not grade, despite the high biological variability displayed by pRCC tumours.

Founded on our results that NFIA and HNF4A can segregate patients based on stage, grade and CSS in the ccRCC setting, we explored the potential use of NFIA and HNF4A as independent prognostic marker in ccRCC. HNF4A was unable to independently predict CSS, but we found that NFIA is able to predict poor CSS in ccRCC patients independently of clinicopathological parameters such as tumour grade and stage. Although HNF4A is previously described in RCC literature, we show that NFIA may be a better prognostication marker. Such evaluations are critical to refine the pool of potential prognostication markers within the management of RCC. With studies of this nature, there are challenges with regard to the applicability of immunohistochemistry‐based methods pertaining to reagents and quantification. Additionally, observations need to be validated with other independent cohorts. Nevertheless, robust prognostication tools are essential for the effective clinical management of RCC patients and the initiation and subsequent translation of these biomarker investigations are a key step in improving patient survival.

Author contributions statement

HA and RdA conceived the study. RdA, MI and CM performed the experiments and collected data. RdA, MI and SS carried out data analyses. SS adapted software and performed bioinformatic analyses. RdA and SS prepared figures. BL collected clinical samples and provided clinical data. HA obtained funding. RdA drafted the manuscript. All authors revised the manuscript and approved the submitted version.

Ethics approval and consent to participate

The study was reviewed and approved by the Ethical Review Board (Dnr: 2015‐146‐31M and Dnr: 2018‐296‐32M) and the Ethical Board of Sweden (Dnr: 2019‐02579). All procedures regarding patients were in accordance with the Helsinki Declaration. The patients had informed consent, orally before year 2000, and informed and written consent from year 2000.

Supporting information

Figure S1. Correlation analysis of HNF4A and NFIA expression in the TMA cohort

Figure S2. Distribution of Fuhrman grades across T‐grouped stages within ccRCC and pRCC patient cohorts

Figure S3. Distribution of T, N, and M status' within T‐grouped stages in ccRCC and pRCC patients

Table S1. Evaluation of the prognostic value of HNF4A on CSS in univariable and multivariable analyses

Data availability statement

Data are available upon request to the corresponding author.