A total of 16 primary RCC patient tumour samples (Table I) and 5 primary RCC cell were kindly provided by Dr Helén Nilsson, Department of Translational Medicine, Lund University for validation of selected biomarkers. Patient tumour samples were retrieved from the material collection established after Lund University ethical committee approval (approval no. LU680-08), where informed written consent was obtained before archival. Human renal tissue dissociation and culturing was performed as described in Appendix S1 and previously described in Hansson et al (23). Tumour tissues obtained consisted of ccRCC, p1RCC, p2RCC and chRCC tissues and cells that were predetermined by a trained pathologist. ccRCC cell lines SNU-349 (Korean Cell Line Bank) and 786-O (ATCC) were cultured in RPMI-1640 (Corning, Manassas, USA) and DMEM (Corning, Manassas, USA) respectively, both medium supplemented with 10% foetal bovine serum (Gibco, MA, USA). All cell lines were incubated in a humidified chamber in 5% CO₂ at 37˚C, were STR authenticated, had a passage number of not more than 20 and were mycoplasma free. Healthy blood for controls were obtained from the Blood donation centre in Lund with informed written consent from patients and ethical permission obtained by the regional health care provider (Region Skåne, Lund, Sweden; approval no. 2018:19).

CA9 and SLC6A3 were selected as ccRCC subtype specific markers based on a literature search and previously published data from Hansson et al (23). The subtype specific expression of these markers was confirmed in the TCGA data set as described below. The same approach was also used to identify subtype specific markers for the other two subtypes, pRCC and chRCC. Level 3 RNA-sequencing data processed using the ‘UNC V2 RNA-Seq’ workflow was downloaded from The Cancer Genome Atlas (TCGA) data portal (https://portal.gdc.cancer.gov/), as described in Lindgren et al (12). Statistical analyses were performed using the R software (http://www.r-project.org). Differential gene expression between clear cell, papillary and chromophobe RCC (KIRC, KIRP and KICH, respectively) tumours was determined using the Limma Bioconductor package (24). For the analysis of taxonomy groups, samples from all 3 TCGA kidney cancer projects [chromophobe RCC (KICH), clear cell (KIRC), papillary kidney carcinoma (KIRP)] were merged into one single dataset (25). Molecular RCC taxonomy groups as well as clinical and histopathologic parameters were used as presented in the article by Chen et al (26). The selection criteria for sub-type specific expression were minimal expression in human blood (based on the blood datasets found in the Human Protein Atlas, available from https://www.proteinatlas.org) and distinct RCC sub-type specific expression. For the two types of papillary RCC defined by Chen et al (26) (p.e1 and p.e2) optimal markers for the more common p.e1 subtype (but also displaying elevated expression in p.e2 tumours) were selected (Table SI). These selection criteria resulted in a panel of 7 genes (Table II).

cDNA synthesis for cell lines and primary tumours was performed using the High-Capacity RNA-to-cDNA kit. Quantitative PCR (qPCR) was performed on a QuantStudio 7 Flex Real-Time PCR system using TaqMan Gene Expression Master Mix and TaqMan Probes (Table III) (all from Applied Biosystems, Vilnius, Lithuania). Relative gene expression from qPCR data was calculated using the double delta Cq method and normalised to β-actin levels (27).

Cell diameter of RCC cell lines and primary cell lines were measured on a Nucleocounter NC-3000 (ChemoMetec, Allerod, Denmark) using NC-Slide A8 with the Cell Viability and Cell Count Assay, according to manufacturer's instructions.

Whole blood (7.5 ml) was processed for each run on the ClearCell FX system (Biolidics, Singapore). Firstly, red blood cells (RBCs) were lysed by a 10-minute incubation with a RBC lysis buffer (G-Bioscience, St. Louis, MO, USA) and discarded via centrifugation and removal of supernatant. The resulting cell pellet containing nucleated cells was resuspended in 4 ml of ClearCell FX re-suspension buffer prior to being loaded onto the ClearCell FX system and running ‘Protocol 1’. Renal cancer cell line SNU-349 and primary RCC cells were used to establish the recovery efficiency of the ClearCell FX tumour cell isolation system for RCC cells. Cells were labelled by incubating in serum free media (Corning, Manassas, USA) with a final concentration of 25 µM CellTracker Green CMFDA (Invitrogen, Carlsbad, USA) dye for 45 min. Cells were then washed with DPBS and diluted to varying numbers before being counted and spiked into 7.5 ml of healthy donor blood. The spiked blood sample was then processed and run through the ClearCell FX system according to manufacturer's instructions. The isolated output cells were then pelleted, resuspended in a lower volume and aliquoted into a 96-well plate. Labelled and isolated cells were counted using an inverted immunofluorescence microscope to calculate the recovery efficiency.

RNA extraction for ClearCell FX output samples were performed according to manufacturer's instructions using the Norgen Biotek Single Cell RNA Purification kit (Norgen Biotek Corp., Thorold, Canada) and RNA was eluted in 14 µl of PCR clean water. Global pre-amplification of RNA and cDNA clean-up was performed on RNA extracts from ClearCell FX system outputs using the SMART-Seq v4 Ultra Low Input RNA kit (Takara Bio Inc., Shiga, Japan) according to manufacturer instructions. Pre-amplified cDNA was purified using AMPure XP magnetic bead solution (Beckman Coulter, USA) and eluted in 20 µl TE buffer. RNA concentration, RNA integrity number and pre-amplified cDNA concentration from ClearCell FX system outputs was assessed on the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, USA). Primary patient tumour tissue (<30 mg) was processed by disruption in a TissueLyser using Trizol (Qiagen, Hilden, Germany). Homogenization was achieved by flushing the disrupted sample through a QIAshredder column. Subsequent RNA extraction was performed using the Qiagen RNeasy Mini (Qiagen, Hilden, Germany) kit according to manufacturer's instructions. RNA concentration and quality from cultured cells and primary tumour material was assessed on a Nanodrop 2000 (Thermo Fisher Scientific, Waltham, USA).

Global pre-amplification was monitored in real time to obtain the optimal cycle number by adding 0.1X SYBR Green (Sigma Aldrich, Darmstadt, Germany) and allowing the reaction to run for 40 cycles (28).

Mann-Whitney U, mean and SEM calculations were performed on GraphPad Prism software (Graph Pad Software, San Diego, USA). Kruskal-Wallis with Dunn's comparison calculations were performed on R-software (Vienna, Austria) and GraphPad Prism.

In this study we set out to develop a method for efficient isolation of CTCs from RCC patient blood. Previous studies have indicated that label-free, size-based enrichment is particularly well suited for RCC, which led us to employ the size-based ClearCell FX system for our study (16,29).

In order to confirm that RCC tumour cells were large enough to be successfully enriched in a size-dependent system, we measured the cell diameters of two ccRCC cell lines along with 5 primary cell lines (ccRCC and p1RCC). This data showed that cultured RCC tumour cells SNU-349 and 786-O are approximately 18 µm in diameter whilst cultured primary tumour cells are larger and measure between 20-30 µm (Fig. 1). These values are greater than the lower ClearCell FX isolation size threshold of 14 µm suggesting that our choice of enrichment method is likely suitable for isolation of CTCs from RCC patients. A small fraction of leucocytes from whole blood are larger than 14 µm, these are also presumably enriched and isolated together with the tumour cells. Furthermore, we measured the cell diameter of four non-RCC tumour cell lines (MCF-7, T47D, Mel JuSo, PC3) to confirm that tumour cells in general, and RCC cells in particular, are larger than most leucocytes.

To establish the recovery efficiency of the ClearCell FX system with RCC cells, we performed cell spike-in experiments. For these experiments we employed the ccRCC cell line SNU-349 and primary RCC cell lines. The cells were labelled with a fluorescent tracker and the number of fluorescent cells were counted (range of cells spiked-in 10-232) before mixing them with 7.5 ml of healthy donor blood. After lysis of the reticulocytes, the blood sample was subjected to size-based isolation using the ClearCell FX system. The output, containing a background of approximately 10,000 leukocytes and the enriched labelled SNU-349 cells were thereafter analysed using a fluorescent microscope. These experiments show that on average, the system is able to recover 50% of SNU-349 cells and 66% of primary RCC cells that are spiked into whole blood (Fig. 2A). Spiked-in and recovered cell numbers are reported in Table SII. We also performed immunofluorescence staining on ClearCell FX enriched samples originating from whole blood spiked with SNU-349 cells. To distinguish RCC cells from leucocytes we stained cells for CD45 (leukocytes) and CA9 (a well-established marker for ccRCC cells). We were able to clearly distinguish CA9 positive, CD45 negative SNU-349 cells on a background of only CD45 positive leucocytes (Fig. 2B)

Next, we wanted to establish and validate a method to detect RCC-specific RNA transcripts from the ClearCell FX enriched CTC fraction. First, RNA was extracted and subjected to quality and quantity assessment (Fig. S1). Once the quality was deemed acceptable (RIN >5), we performed global pre-amplification of reverse transcribed cDNA to yield sufficient sample material and to increase the relatively low transcript copy numbers from the enriched SNU-349 cells present on a background of leukocytes in the sample. Prior to global pre-amplification of cDNA from the CTC enriched samples, the pre-amplification process was monitored with quantitative PCR using SNU-349 cDNA, in order to determine the optimal number of cycles required (Fig. S1B). A successful pre-amplification reaction should yield an increase in fragments in the 400-10,000 base-pair range, which was observed in our samples via fragment analysis. (Fig. S1C).

In order to identify ccRCC specific marker genes suitable for assessing the presence of ccRCC cells in our spiked-in samples, we analysed data from the publicly available TCGA (The Cancer Genome Atlas) database and the subsequent sub-type classification by Chen et al (26), as defined when pooling the three RCC datasets. Using Limma analyses, we extracted a gene list with the most differentially expressed genes, when comparing ccRCC to pRCC and chRCC (Table SI). CA9 and SLC6A3 were amongst the most differentially expressed genes in ccRCC and were selected for further exploration (Fig. 3A). CA9 is a well-documented hypoxia-driven gene and SLC6A3 displays a highly ccRCC specific expression pattern, as described by us and others (23,30). We tested the reproducibility of the digital PCR assays of these marker genes on pre-amplified cDNA (Fig. 3B). Increasing the input cDNA yielded higher copies/µl in a linear fashion. Next, we assessed the limit of detection of our assays within the context of enriched tumour cell samples. We performed a titration of tumour cells (SNU-349) spiked into 7.5 ml of whole blood resulting in samples with defined numbers of tumour cells after enrichment. We reliably detected transcripts of CA9 at the one cell level with more transcripts being detected with higher cell numbers. Similarly, SLC6A3 transcripts could also be detected from enriched samples containing ≥1 to 121 cells (Fig. 3C). However, transcripts were less reliably detected at the one cell level with SLC6A3. Enriched samples from healthy blood controls (4 samples per marker) were also tested, consistently giving readouts of less than 1 copy/µl.

With the aim of broadening the applicability of our method to include pRCC and chRCC, we identified genes that are differentially and specifically expressed within these subtypes (Table SI). When selecting markers for p1RCC and p2RCC, markers were chosen so they indicated a papillary subtype and not based on a papillary type 1 or 2 subtype. The selection criteria for a marker included minimal or absent expression in human blood, based on analyses of blood datasets within the Human Protein Atlas (31,32) and high RCC subtype specific expression based on Limma analyses of TCGA expression data, which is based on the Chen et al (26) classification of RCC tumours. These selection criteria resulted in a panel of 5 genes in addition to the ccRCC markers CA9 and SLC6A3 (Fig. 4).

To validate the sub-type specific expression of the selected markers we analysed RNA extracted from primary tumour samples (Fig. 5). Only RCC subtypes that were confirmed to be one of the three primary subtypes were included in this patient cohort (Table II). Our expression data was in-line with the pattern seen in the TCGA data set. The two markers for the ccRCC subtype (SLC6A3, CA9) clearly distinguished this subtype from the rest in the tumour samples tested. Using a joint set of markers to distinguish the papillary RCC subtype also allowed for clear separation of this subtype from the other two. Finally, the markers for the chRCC subtype distinguished these tumours extremely well, showing elevated RNA expression in a subtype specific manner. We additionally tested all markers on 6 healthy-volunteer blood samples and found them to be negative for our RCC specific markers (Fig. S2).