Liquid biopsy in mice bearing colorectal carcinoma xenografts: gateways regulating the levels of circulating tumor DNA (ctDNA) and miRNA (ctmiRNA)

Human CRC cells (HT-29, LoVo and LS174T) were purchased from the American Type Culture Collection (ATCC). HT-29 cells were derived from a KRAS-WT, differentiated colorectal adenocarcinoma. LoVo cells (originally described as Dukes’ type C, grade IV) and LS174T cells (Dukes’ type B) harbor a heterozygous KRAS c.38G > A mutation (G13D) and a heterozygous KRAS c.35G > A mutation (G12D), respectively. Somatic mutations and their Variant Allelic Frequencies (VAFs) were confirmed by interrogating the Cancer Cell Line Encyclopedia ( https://portals.broadinstitute.org/ccle ). Cell identity was confirmed by HLA genotyping, as described [8]. Cells were grown in RPMI 1640, 10% fetal bovine serum, 1% l-glutamine and 1% penicillin/streptomycin (all from Euroclone, Milan, Italy) in a humidified atmosphere containing 5% CO₂. To permit the accumulation of DNA and miRNAs in culture supernatants, cells were grown for 72 h to 70% confluency and manually counted by the Trypan Blue exclusion method. Cells were harvested and either pelleted and snap-frozen (for miRNA extraction, see below), or Formalin Fixed and Paraffin Embedded (FFPE), as routinely done for histological specimens (for DNA extraction, see below). Spent medium was then collected, spun twice at 4 °C (2000×g for 20 min and 16,000×g for 10 min), and immediately stored at − 80 °C in aliquots.

Animal studies were performed according to Directive 2010/63/EU and Italian Decree Law 26/2014. They were approved by the EU Research Executive Agency (REA), the Intramural Regina Elena Board for Animal Welfare, and the Italian Ministry of Health (prot n. 700–2015-PR, dated July 17th, 2015). Tumor xenotransplants were established by inoculating 3 × 10⁶ cells from the HT-29, LS174T and LoVo cell lines in the flank of 4-month old Nu/CD1 mice (Charles River Laboratories, Italy). Xenotransplants were allowed to grow to two different sizes (300 and 1000 mm³, 6 mice per group for each of the three cell lines). Additional independent experiments were carried out on smaller cohorts of mice and individual animals. Tumors were taken at sacrifice along with blood, and were divided into two fragments: one half was snap-frozen in liquid nitrogen, the other half was FFPE-processed exactly as routinely done for human diagnostic histopathology. Frozen tissues were used as the source of miRNAs, FFPE samples were used as the source of genomic DNA, for morphological evaluation (hematoxylin/eosin stain), and for immunohistochemistry (see Additional file 1). Blood was collected in 6 mL BD Vacutainer K₂E tubes (BD, #368857) and centrifuged within 1 h at 2,000×g for 20 min at 4 °C. Plasma was recovered and further centrifuged at 16,000×g for 10 min at 4 °C to remove cell debris, and stored at − 80 °C until extraction.

DNA was obtained according to the manufacturer’s (Qiagen, Hilden, Germany) instructions from four different sources. Genomic DNA was extracted by the QIAamp® DNA FFPE Tissue Kit from FFPE inclusion slices of either (a) cell pellets of cultured cells or (b) tumor xenotransplants. ctDNA was extracted with the QIAamp® CNA kit from either (c) 2 mL of culture supernatants or from (d) 2 mL of mouse plasma. DNAs from (c) and (d) were eluted in a final volume of 30 μL, and stored at − 20 °C. All DNAs were quantified with the Qubit dsDNA assay kit (LifeTechnologies, Carlsbad, CA, USA) prior to dPCR analysis.

RNA was extracted from frozen cell (5 × 10⁵) pellets by the TRI-Reagent (Sigma-Aldrich, St.Louis, MO, USA) and from cell culture supernatants according to Turchinovich et al. [9] with minor modifications. Briefly, 400 μL of culture supernatants were lysed in 1.2 ml of TRIzol LS (Thermo Fisher Scientific, Waltham, MA, USA) in the presence of 12 μg of carrier glycogen (Thermo Fisher Scientific) and 400 attomoles of C. elegans miR-39-3p (cel-miR-39-3p) (Thermo Fisher Scientific) to control miRNA recovery efficiency. Following phase-partitioning by nuclease-free chloroform (Sigma-Aldrich), miRNAs were purified using miRNeasy Serum/Plasma columns (Qiagen). RNA was eluted in 18 μL of nuclease-free water (Sigma-Aldrich). The snap-frozen tumor fragment was homogenized using an IKA T10 Basic Ultraturrax (IKA Werke, Staufen, DE) in TRI-Reagent at maximum speed for 1 min and then processed as above. For the analysis of ctmiRNA, blood plasma (150 μL) was treated to disrupt exosomes and denaturate miRNA-binding proteins in 5 volumes of QIAzol Lysis Reagent (Qiagen). After the addition of 400 amoles of cel-miR-39-3p, total RNA was purified with the miRNeasy Serum/Plasma Kit (Qiagen) in a final volume of 18 μL. All RNAs were stored at − 80 °C until use.

DNA samples were dispensed in the chip-based QuantStudio™ 3D Digital PCR System (Thermo Fisher Scientific), and amplified using selected TaqMan™ SNP Genotyping Assays for human KRAS mutations (Table 1). dPCR reactions were set up in a final volume of 15 μL, containing 7.5 μL of 2× QuantStudio™ 3D Digital PCR Master Mix, 0.75 μL of TaqMan® SNP Genotyping assays (Life Technologies), and different DNA templates, each adjusted in 6.75 μL water, as follows. Genomic DNAs from cells and tumor tissues were 50 ng; DNAs from culture supernatants were 6.75 μL equivalent to 450 μl of supernatant; ctDNAs were 6.75 μL equivalent to 45 μl plasma. Thermal cycling was as follows: 10 min at 96.0 °C, 39 cycles at 56.0 °C for 2 min, 30 s at 98.0 °C, and a final elongation step of 2 min at 60 °C. Copy numbers were calculated assuming that a human haploid genome DNA corresponds to 3.3 pg DNA.

Levels of miR-141, miR-221 and miR-222 were assessed by Reverse Transcription droplet digital PCR (RT-ddPCR). 300 ng of total RNA (from cells and tissues) or 3 μL of purified miRNAs (corresponding to 65 μL of supernatants and 25 μL of plasma) were reverse transcribed using the TaqMan™ miRNA Reverse Transcription Kit (Thermo Fisher Scientific) according to the manufacturer’s manual, in a final reaction volume of 20 μl. 1.3 μL of cDNA, undiluted (for supernatants and plasma) or diluted 1:100 (for cells and tissues), was amplified in a final volume of 20 μL, in the presence of ddPCR Supermix for Probes (no dUTP) 2X (Bio-Rad) and TaqMan™ miRNA 20X assays (Thermo Fisher Scientific) (Table 1). 40 μL of droplets emulsion was automatically generated using Automated Droplet Generator (AutoDG) (Bio-Rad) and amplified in a thermal cycler. The following thermal cycler conditions were used: 95 °C for 10 min, 40 cycles of 95 °C for 15 s and 60 °C for 1 min and a final step of 98 °C for 10 min. Droplets were analyzed using the QX200 Droplet Reader, and data analysis was performed with QuantaSoft version 1.7.4 (Bio-Rad).

To assess ctDNAs and miRNAs, a model study was designed that includes cells, culture supernatants, tumor xenografts, and blood samples, as outlined in Fig. 1. Three different CRC cell lines (HT-29, LoVo and LS174T) were selected as proxies of clinically evident cancers and sources of soluble analytes. Because RAS mutations are assessed in the routine diagnosis of colorectal carcinoma to assign epidermal growth factor receptor (EGFR) blockade therapy [10], we selected cells that are either wild-type for KRAS (HT-29), or carry the G12D and G13D mutations (LS174T and LoVo, respectively). DNAs and miRNAs were extracted from both cells and culture supernatants, as outlined in Fig. 1 (left). The same cells were also modeled as xenotransplants in nude mice (Fig. 1, right). Matched tumor tissue and blood samples were taken at sacrifice. All biological specimens were then subjected to dPCR, ddPCR and RT-qPCR analysis as outlined in Materials and Methods.

DNAs were obtained from: (1) cell pellets of HT-29, LS174T and LoVo cells grown in culture flasks to approximately (±5%) the same numbers; (2) equal amounts of supernatants from the above flasks; (3) genomic DNAs from tumor xenotransplants grown to 300 mm³ (±15%) in size (all tumors); and (4) equal amounts of pooled blood plasma obtained from the above animals (n = 6). DNAs and ctDNAs were subjected to dPCR for KRAS G12D and G13D, and for control PIK3CA H1047R and FBXW7 R505C mutations, to test DNA fragments originating from the same (KRAS) as well as distinct (PIK3CA and FBXW7) genomic loci in two different (LS174T and LoVo) CRC cell lines. Mutated RAS sequences were detected in LoVo and LS174T (Fig. 2, panels I-X), but not in KRAS-WT HT-29 cells (E-H), as expected. VAFs in cells and tissues were approximately 50% in all cases (I, M, K, O, U, V), as expected, with the exception of KRAS in LoVo (Q and S, approximately 75%), confirming copy number variation of the KRAS MUT allele in the absence of chromosome 12 duplication (see whole genome sequencing data in the Cancer Cell Line Encyclopedia at https://portals.broadinstitute.org/ccle ). However, and interestingly, when free soluble DNAs were considered, a strikingly higher variability was readily observed in both VAFs and absolute copy numbers per mL. Specifically, KRAS G12D DNA from LS174T was almost 60 times less abundant than G13D DNA from LoVo (958 vs 56,359 copies/mL; compare J and R) in culture supernatants and, viceversa, it was approximately 8 times more abundant (compare L and T; 139 vs 17 c/mL) in blood, demonstrating a 500-fold difference in the ability of the two cell lines to generate accumulation of RAS ctDNA in cell supernatants vs blood of tumor-xenotransplanted mice. Likewise, comparison of PI3KCA and FBXW7 (compare N to V, and P to X) demonstrated greater (25 times) and lesser (7 times) relative abundances of the two mutations in culture supernatants and blood, resulting in a 175-fold difference that is in full agreement with KRAS ctDNA blood accumulation preferences (LS174T > LoVo). Altogether, ctDNA levels appear to reflect a cell line-dependent property only marginally affected by the genomic regions of origin of the soluble analytes. Preferential ctDNA accumulation in the blood of LS174T xenografts was reproduced in: (a) two separate cohorts of 6 mice carrying larger (1000 mm³) tumors, (b) two additional, distinct plasma samples obtained by pooling blood from two cohorts of three animals each, and (c) two plasma samples from two individual animals (data not shown). Altogether, our results suggest the existence of a minimum of two gateways. The first gateway predominantly operates in LS174T cells by controlling DNA scavenging towards the microenvironment, since this cell line releases in vitro much less DNA than LoVo. Conversely, the latter gateway becomes evident in LoVo cells when they are grown as tumor xenografts, and prevents the accumulation of mutated ctDNA in blood.

At least 3 major cellular/histopathological features are known to account for DNA accumulation in the extracellular milieu and/or in blood: cell proliferation, tissue necrosis and tumor micro-vasculature [4]. A simple explanation of ctDNA accumulation based on cell proliferation is unlikely, since LS174T proliferate much faster than LoVo cells both in culture and as xenografts (doubling times of 24 h vs 96 h in culture, and 4 days vs 17 days in the mouse, respectively), but the preferential accumulation of LS174T DNA is evident only in blood. It may then be hypothesized that this preferential ctDNA release might depend on the tissue architecture, e.g. necrosis or micro-vasculature. However, morphological evaluation revealed higher percentages of necrotic areas in low-ctDNA-releasing LoVo as compared to LS174T xenografts (30% vs 45%). Moreover, staining with antibodies to the endothelial marker CD31 demonstrated very similar numbers (about 14/microscopic field) of blood micro-vessels in LS174T and LoVo xenografts (Additional file 1: Figures S1 and S2). These results suggest that ctDNA gateways operate independently of tissue architecture, e.g. they are likely dependent on constraints on ctDNA release that operate both at single-cell and whole-body levels, but independent from housekeeping functions associated with cell proliferation, tumor necrosis and tumor vascularization.

Next, we wondered whether similar gateways regulate circulating miRNAs. Three miRNAs (miR-221, miR-222 and miR-141) were selected since they were previously described to affect colorectal cancer biology and clinical behavior [11, 12]. RNAs from dry frozen pellets and the corresponding culture supernatants were reverse transcribed, and amplified by ddPCR. 2D plots of representative miR-222-3p experiments and the complete dataset for the three miRNAs (cell lines and supernatants) are shown in Fig. 3a (top and bottom panels respectively). In parallel, in Fig. 3b, are reported the copy number/μl of RNA of all the three miRs analized in cells and supernatants (dots and diamonds respectively). From Fig. 3 it may be seen that miR-222 is more abundant in the supernatants from the cells in which it is expressed most (for instance 5.45⁴ and 7.38³ miR-222 molecules/μl of RNA are present in the supernatant of LoVo and HT-29 cells, respectively; similarly, the intracellular accumulation of miR-222 was 9.01⁶ and 3.32⁶ molecules/μl of RNA). Data reported in Fig. 3b indicate that miR-221 and miR-222, that belong to the same miRNA cluster, are proportionately expressed in cells and supernatants, whereas miR-141 levels do not correlate between cells and supernatants, since HT-29 cells express high levels of miR-141 (1.06⁶ molecules/μl of RNA, compared the lower 6.04⁵ amounts of LoVo cells) but do not efficiently release it into the medium (8.12² molecules/μl of RNA, compared the higher 4.96³ amounts of LoVo supernatants). Additional ddPCR results are displayed in Additional file 1: Figure S3 and were confirmed by RT-qPCR (Additional file 1: Figure S4), although this latter technique systematically overestimated miRNA concentrations. To exclude that bovine miRNAs (that are contained in FBS and are highly homologous to human miRNAs) might contribute significant noise to PCR assays aimed at tumor miRNAs, we reverse transcribed from complete culture medium and from medium conditioned by cells grown in culture to 70% confluency. As shown in Additional file 1: Figure S5, the contribution of bovine miRNA was negligible. Altogether, these results rule out trivial technical artifacts and demonstrate a specific gateway preventing miR-141 release into HT-29 supernatants. When the results shown in Fig. 3 are taken together with those shown in Fig. 2, it can be concluded that, similar to ctDNAs, miR-141 was released into HT-29 cell supernatants 10 and 6.5 times less abundantly with respect to LoVo and LS147T cells, respectively; on the contrary, release of miR-141 in blood of HT-29 xenografted mice was found similar to that observed in the blood of LoVo and LS147T xenografted mice.

Next, we analyzed tumor xenotransplants and their corresponding plasma specimens using the same approach (Fig. 4). A representative example of miR-222 content is shown in Fig. 4a and all the quantitative data for miR-141, miR-221 and miR-222 are given in Fig. 4b. This resulted in a graphical outline very similar to that of Fig. 3. Comparison of the two figures (representative 2D ddPCR plots and the complete dataset from all xenotransplants in the top and bottom panels, respectively) demonstrates at glance that miR-141, poorly released in the culture supernatant (Fig. 3), is instead clearly detectable in plasma (Fig. 4) as the other microRNA analyzed, i.e. miR-221 and miR-222. Also in this case, miRNA levels were independently assessed by RT-qPCR, that confirmed ddPCR results (Additional file 1: Figures S6 and S7). Finally, we investigated whether cross-species miRNA homology might influence our in vivo results. To this end, we quantitated baseline, endogenous miR-221, miR-222 and miR-141 levels in tumor-free, healthy nude mice. Both RT-ddPCR and RT-qPCR demonstrated that circulating miRNAs were detectable even in the absence of tumor growth (Fig. 5). Thus, whereas bovine miRNAs do not contribute significant background in vitro (see above), mouse miRNAs do, and this affects the interpretation of the in vivo results. The differences between tumor-bearing and tumor-free mice were clearly appreciable for all the three miRNAs analyzed on HT-29 xenografts. In particular, it should be underlined that, miR-141 was significantly more abundant in the plasma of mice bearing HT-29 tumor xenografts as compared to their tumor-free littermates. This result is compatible with the hypothesis suggesting that tumor masses induce systemic miR-141 accumulation in spite of the intrinsic poor ability of HT-29 cells grown in vitro to release this miRNA in the culture supernatant.