All patients with invasive ductal carcinoma of the breast were preliminarily selected for this prospective study at the Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan from March 2019 to December 2019. Totally 43 patients were selected according to the inclusion criteria and subdivided. The mean age of the patients was 49.87 years (range, 39–57 years). The patients were suspected to have newly diagnosed or postoperative recurrent liver metastasis as assessed by fluorine-18-fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) imaging. The breast tumors were reviewed and liver metastases were confirmed by two expert pathologists (ZeL and LG). The patients were excluded if the liver nodules were benign lesions, primary liver tumors, or metastases of other organs besides the breast. Next, the demographic information and detailed history of the included patients were documented according to an interviewer-administered questionnaire. Laboratory tests including blood routine, liver function tests [aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP), prothrombin time (PT), albumin, and bilirubin] and a ¹⁸F-FDG PET/CT were performed. All participants received different systematic and local therapies at the Department of Oncology, The Affiliated Hospital of Southwest Medical University (Luzhou, Sichuan).

Fig. 1 shows the present study model in detail. Firstly, a suspected malignant liver nodule was identified by ¹⁸F-FDG PET/CT scan in all patients, which was then confirmed by pathology within a week. Next, CTCs from each selected BCLM patient were captured with the CellSearch platform using antibody enrichment, and were further isolated under a fluorescence microscope with 94% specificity. Then, genomic DNA (gDNA) of single a CTC was amplified using the multiple annealing and looping based amplification cycle (MALBAC) method. WGS was planned for CTCs as well as newly diagnosed or recurrent liver metastases samples from each CTC positive patient. Furthermore, we analyzed genome-wide CNVs in these cases to introduce the major genomic variations that are specific to and reproducible in BCLM. Finally, noninvasive CTC-based BCLM diagnostics were compared with ¹⁸F-FDG PET/CT and histological findings.

All tissue samples were fixed and embedded in Tissue Tek II OCT (Miles Scientific), frozen 15 min in isopentane, precooled in liquid nitrogen, and then stored at −80°C for future pathology assays. Next, the best frozen sample was oriented and cut as 5-µm-thick cryostat sections for hematoxylin and eosin (H&E) analysis and histological conformation. Microscopic analysis of all slides was performed using light microscopy (Olympus Corp.) linked to a computerized imaging system (Image-Pro Plus V6.0; Media Cybernetics, Inc.). The cases were coded and measurements were made in a blinded manner by two expert pathologists (LiZ and SI). Subsequently, from each patient, 7.5 ml fresh blood was collected in a Cell Save blood collection tube (Immunicon Inc.) and stored at 15–20°C for CTC enumeration within 3 days after collection and 10 ml was collected for gDNA extraction. The gDNA was extracted from fresh peripheral blood leukocytes, using the Qiagen DNA extraction kit (Qiagen). Primary tumor and liver metastasis tissues were obtained by standard core needle methods.

¹⁸F-FDG PET/CT imaging was provided by the Department of Nuclear Medicine, The Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, China. Before scanning, all patients had fasted for at least 6 h and their blood glucose levels were measured to be within the normal range prior to intravenous injection of 370 MBq of ¹⁸F-FDG. The PET/CT parameters were 120 KV, 80 mAs, 3 min/bed, 0.813 pitch, and a 3.8 mm reconstruction of layer thickness. Immediately after CT scanning, a PET emission scan that covered the identical transverse field of view was acquired in 3-min acquisition time per bed position. Data were reconstructed using an iterative reconstruction technique and attenuation correction derived from the CT data. The CT, PET and fused PET/CT images were transmitted to an Extended Brilliance Philips workstation version 4.1 (Philips Healthcare). The PET images were evaluated qualitatively for regions of focally increased metabolism. An appreciably increased uptake level in comparison to surrounding tissue was considered as malignant lesions. Semi-quantitative analysis using the standardized uptake value (SUV) and the mean ± SD of maximum-pixel SUV (SUVmax) of the lesions was calculated. All the PET/CT scans were evaluated by the unit of nuclear radiologists and two nuclear radiologists (SI and QW).

CTCs from 7.5 ml blood sample were captured with the CellSearch Circulating Tumor Cell Kit (cat. no. K062013; Veridex, LLC) for 1 h at room temperature. This kit is a high-throughput isolation kit that is mostly used for enumeration and isolation of CTCs using magnetic beads conjugated to anti-epithelial cell adhesion molecule (EpCAM) antibodies (26,27). EpCAM-positive cells were immediately isolated using a magnetic field. To distinguish cancer cells from leukocytes, 1 ml PBS and isolated CTCs were stained with 10 µl DAPI for 15 min in the dark, 200 µl anti-CK-8 (8)-phycoerythrin antibodies for 30 min and 20 µl anti-CD45-allophycocyanin antibodies for 60 min. The cancer cells stained DAPI⁺, anti-CD45⁻, and CK-8⁺, while leukocytes stained DAPI⁺, anti-CD45⁺, and anti-CK-8⁻. Following immunostaining, the enriched CTCs were re-suspended with PBS and were then manually counted under a fluorescence microscope (10× lens, Axio Imager A2; Carl Zeiss). We obtained full coverage of all the stained CTCs using micropipetting within 5 min. For high-ratio CTC enumeration, UV-exposed water was used to wash the isolated single CTCs repeatedly to minimize DNA contamination. The isolated single CTCs were used for further genome amplification. The identification of all CTCs was performed by an expert pathologist who is specialized in the pathological diagnosis of metastasis (LiZ, SI and QW).

Whole genome amplification (WGA) of an isolated single CTC was performed using MALBAC Single Cell Whole genome amplification Kit (cat. no. YK001A; Yikon Genomics, Inc.). The amplified DNA was purified by the DNA clean-up kit (cat. no. CW2301; CWbio) and the fragment size generated by WGA was between 300–2,000 bp, as determined by gel electrophoresis. Quantitative PCR (qPCR) was performed on seven randomly selected loci in the genome to check the integrity of amplification. All seven loci were amplified with reasonable Ct number and a deep average ‘unique mapped of raw’. The 10× WGS analyses were performed in CTC specimens of newly diagnosed BCLM patients with a high CTCs and also in CTC specimens of recurrent BCLM patients with low CTCs. CTCs were regularly amplified with an average amplified gDNA of 900 ng/cell. Then, we applied 30× WGS sequencing of liver metastases to both newly diagnosed and recurrent BCLM.

The design of whole genome library and sequencing were detected using Illumina paired-end libraries (Illumina, Inc.) methods as previously described (28). Briefly, 900 ng to 1 µg of extracted DNA from the amplification products of CTCs by MALBAC or from tumor tissues was sheared into fragments of approximately 300 bp using a Covaris E220 system (Covaris Inc.). The adaptor-ligated DNA was prepared, and enrichment q;PCR was performed on an aliquot of adaptor-ligated DNA to complete the adaptor for Illumina PE sequencing. A total of 600 ng of pooled-DNA from four barcoded libraries was used for hybridization and post-hybridization amplification following the manufacturer's protocol (SureSelectXT Target Enrichment System for Illumina Paired-End Sequencing Library; Illumina, Inc.). Then, the result of the PCR amplification of single CTCs were quality checked and WGS was performed with the Illumina HiSeq 2000 (Illumina, Inc.). to achieve an average of ~10× and ~30× coverage depth, respectively.

Here, we used the standard probe to determine genomic CNVs in single CTCs that was established by Zong et al (29). In brief, paired-end sequencing reads of each CTC and tumor sample were aligned with the human hg19 reference genome using Burrows-Wheeler Aligner v0.6.1 (30) and the available public online University of Santa Cruz (UCSC) database (http://genome.ucsc.edu/) (30). The Firehose pipeline (level 4) was used to manage input and output files and submit analyses for execution (31). Genome-wide detection of single-nucleotide and CNVs of a single human cell was performed using ControlFreeC (32). A binary array, which indicates whether a single cancer cell has higher coverage than normal leukocytes, was taken as output in Hidden Markov Models-based calling algorithms (HMMs) (33,34). The copy number analysis was performed by applying data on the Ginkgo dataset (http://qb.cshl.edu/ginkgo) and two R packages (HMM copy and DNA copy), with hg19 as the reference genome. Enrichment tests were conducted at the arm level to identify significantly gained and lost chromosome arms. In addition, Gene Set Enrichment Analysis (GSEA) was used for a functional assessment of the recognized disease pathways among different CTC-shared CNVs (35,36). Accordingly, we used pathway analyses to find the potential biological functional assessment of CTC-shared CNVs via R software (v3.3.1) (37,38).

According to the CellSearch machine-default, patients with at least five CTCs/7.5 ml were considered CTC-positive. In this study, comparison of group differences was carried out with a one-way analysis of variance (ANOVA) test and then Turkey multiple comparison post-hoc analysis. All statistical analyses were performed using SPSS software v21.0 (IBM, Corp.). All tests were repeated three times or more. Data are presented as means ± standard deviation (SD) or median (range). A linear regression analysis was carried out to determine independent factors for the diagnosis of CTCs. For data not distributed normally, comparisons between three groups were made using a Kruskal-Wallis one-way analysis of variance, followed by a post-hoc Dunn's test. For all tests, two-sided P-values and adjusted P-values of <0.05 were considered statistically significant. All charts were designed using GraphPad Prism v5.0 (GraphPad Software, Inc.).

The demographics and clinicopathological characteristics of the 43 selected patients are detailed in Table I. After considering all exclusion/inclusion criteria, 43 BCLM patients were included in this study. As of now, there is no established cut-off value for a prognostic number of CTCs in BCLM. During this study, we divided our patients by their CTC counts as either lower than, equal to, or higher than the median number of CTCs (5–15 CTCs/7.5 ml blood) to determine any possible correlation with clinicopathological features. Using these criteria, 60% (26 of 43 patients) of patients were categorized as positive-CTCs (≥5 CTCs/7.5 ml blood) and 40% (17 of 43 patients) of patients were categorized as negative-CTCs (<5 CTCs/7.5 ml blood). Meanwhile, almost 46% (12 of 26 patients) of patients were categorized as high-CTCs (>15 CTCs/7.5 ml blood) and 54% (14 of 26 patients) of patients were categorized as low-CTCs (≥15 CTCs/7.5 ml blood). Comparison of patients with positive/negative CTCs and those with high/low CTCs showed similar age, body mass index (BMI), disease duration, and SUVmax of PET/CT scan. However, there were more patients with low-CTCs than those with high-CTCs in both the newly diagnosed and the recurrent metastasis groups (P=0.04). Histologically, more than 65% of all samples were hormone receptor positive, either ER- or PR-positive (30 of 43 patients), and HER2-negative (26 of 43 patients). Metastases to the bone, lung, and brain were sufficiently frequent for statistical analysis. Sites of involvement in the 43 patients were: Liver + bone (12 patients), liver + lung (21 patients), liver + lung + bone (3 patients), and liver + lung + brain (2 patients). The calculated detection efficiency of CTCs was >61% when 6–24 tumor cells were present per 7.5 ml peripheral blood (26 of 43 patients), and there was a 100% success rate in the detection of histopathological variables from captured tumor cells. In this study, the median CTCs was 32.35 per 7.5 ml (interquartile range 0–248). Interestingly, newly diagnosed metastatic patients had a higher number of CTCs than recurrent metastatic patients (Fig. 2A, mean CTC count 38.30 vs. 25.50; P=0.032).

We performed a linear logistic regression analysis to predict the accuracy of CTC detection and characteristic of BCLM. Fig. 2B shows that the average recovery was calculated to be 81% (R²=0.91). We assessed possible sources of cell loss and found that nearly 10% of cells were in the waste from cell isolation and sampling, and another 9% of cells died in the magnetic tubing. Therefore, counting of CTCs may not identify patients with a high risk of postoperative recurrence but could be a more precise indication for the diagnosis for BCLM.

To confirm liver metastases of newly diagnosed and recurrent MBC, ¹⁸F-FDG PET/CT scan was conducted. Serial biopsy sections from paraffin blocks of different regions of resected or punctured tumor samples were examined by H&E staining. Pathologically, all breast cancer samples were identified as invasive ducal carcinoma samples. Of the 43 patients, 23 newly diagnosed and 20 postoperative recurrent liver metastases were found by ¹⁸F-FDG PET/CT. The SUVmax of PET/CT between the the negative-CTC group was not significantly different than that of the positive-CTC group (6.12±3.44 vs. 5.97±2.82; P=0.39). Similarly, no significant difference was seen in the SUVmax between the low-CTC and high-CTC groups (4.59±3.27 vs. 4.96±2.73; P=0.76) (Table I). Representative H&E staining and ¹⁸F-FDG PET/CT images of typical newly diagnosed MBC patient no. 13 (BCLM13H, Table SI) showed this patient to be in the high-CTC group while recurrent MBC patient no. 3 (BCLM03L, Table SI) belonged to the low-CTC group. There was high proliferation of bile duct cells and infiltrating inflammatory cells in newly diagnosed liver metastases (Fig. 3A). Large necrotic, dense lymphocytic infiltrate and desmoplastic rim are the main structural depolarization in recurrent liver metastases (Fig. 3B). A significant increase in the number of hepatocytes and significant diffuse infiltration by poorly differentiated carcinoma cells were evident in connective tissues from both newly diagnosed and recurrent liver metastatic groups. The ¹⁸F-FDG PET/CT images clearly show that newly diagnosed liver metastases (Fig. 3C) and the recurrent liver metastases (Fig. 3D) have similar imaging characteristics and SUVmax value. A meaningful increase in necrotic area in metastatic hepatic lesions on the right lobe of the liver was a prominent feature in recurrent liver metastases.

The 7.5 ml blood samples from BCLM patients were used for CTC isolation using high efficiency Cell Search technology. We evaluated the quality of the blood samples before CTC isolation and none of the samples showed hemolysis. The stored CTC suspension was placed under a fluorescence microscope to select individual CTCs. Physical and three-color immunofluorescent characterization of CTCs isolated from patients with BCLM are shown in Fig. 4. There was a large number of cell debris under fluorescence microscope (>75%) and few cells with intact morphology (>11%). The basis for affinity-binding systems used for CTC enrichment and identification is the selection of specific tissue-type and cancer-specific markers, such as leukocyte (WBC)-expressed cell membrane CD45 (a tyrosine phosphatase) and epithelial cell (EPC)-expressed cytoplasmic CK-8, as well as DAPI nuclear staining. Then, we further distinguished the EPCs from CTCs by size, where the cutoff was a cellular diameter <8 µm for CTCs and a diameter >8 µm for EPCs (27,39,40). Morphologically, CTCs from BCLM patients were recognizable with typical deformations of a neoplastic cell: Hyperchromatic nuclei by fluorescence microscopy with an irregular shape, high nuclear-to-cytoplasmic ratio, as described by Krebs et al (41). Enriched CTCs can be isolated based on their distinct physical (size or deformability) or fluorescence properties from EPCs and WBCs (Fig. 4). The CTC sample appears as CD45⁻/Nucleus⁺/CK8⁺, while WBCs appear as CD45⁺/Nucleus⁺/CK8⁻. Cells that stain CD45⁻/Nucleus⁺/CK8⁺ with larger diameter <8 µm were identified as EPCs. Furthermore, the dead cell mass sample shows cytoplasmic CK8⁺. Obviously, the triple-immunostaining method is slightly limited by the specificity and availability of antibodies. Of a total of 2,081 captured cells from all samples, 1,359 (65.3%) of the tumor cells were identified as CTCs, 430 (20.7%) as WBCs, and 292 (14.1%) as EPCs. The details of the CTC count from each BCLM patient are shown in the Table S1.

By comparing the negative- and positive-CTC groups, the initial disease stage and HER2 status were strongly correlated with the presence of CTCs (R=0.75, P=0.04 and R=0.72, P=0.032, respectively). These findings suggest that the CTC count may be a prognostic factor for survival of BCLM patients. Strikingly, by comparing the low- and high-CTC groups, it was found that the CTC number was significantly correlated with the size of metastasis and metastatic tumor number (R=0.81, P=0.03 and R=0.77, P=0.03, respectively). These data indicate that CTC counting could be a viable method by which to monitor distant metastasis. Additionally, we did not find any significant correlation between CTC presence or number and hormone receptor in the multivariate analysis model (Table II). In general, high CTC frequencies were correlated with disease severity and metastatic progression, which also suggests an effective value for CTCs in predicting the prognosis of BCLM.

Visualization of 10× genome-wide gene copy numbers was performed in CTC specimens of the newly diagnosed MBC patient no. 13 (no. BCLM13H, Table S1) and CNVs of recurrent MBC patient no 3 (no. BCLM03L, Table S1). We then applied 30× WGS sequencing of liver metastases from these two patients. Our primary finding showed that the average ‘unique mapped of raw’ of the single cell genome sequencing products was 79.58% (78.74-80.17%), which further implies that uniform amplification of a single cell genome can be used for genomic copy number analysis. As shown in Fig. 5, the CNV pattern of CTCs and liver metastases in newly diagnosed and recurrent MBC patients were significantly different. A few CNVs were found in either the newly diagnosed liver metastases (Fig. 5A) or the recurrent liver metastases (Fig. 5C), in which gain and loss regions accounted for ~7.4 and ~0.26% of their entire genomes, respectively. Remarkably, the region of 0–22,800 kbp of chromosome 13 and 0–22,750 kbp of chromosome 14 were the most significant CNVs in newly diagnosed and recurrent liver metastases, respectively. In comparison with CTCs of recurrent BCLM, there are nearly ~45% gain and ~4.7% loss regions in the CTCs isolated from the newly diagnosed metastases group (Fig. 5B and D).

In the CTCs of recurrent BCLM, the copy number slightly increases in most CNV regions with 1 copy as the lower limit and 3 copies as the upper limit (Fig. 5D). In addition, we also found a few regions with copy number between 3–5 copies located on chromosomes 9,8,11,13,14,15, and 22 (Fig. 5D). Other regions with copy numbers between 5–9 on these chromosomes accounted for ~2.4 and ~2.1% of the entire genome (Fig. 5C). Nearly all CNVs that arise in newly diagnosed liver metastases (Fig. 5A) were found in CTCs of newly diagnosed BCLM (Fig. 5B), but there are many more CNVs in newly diagnosed liver metastases. Furthermore, in CTCs of newly diagnosed liver metastases, chromosomes 3, 9 and Y had significant CNVs (Fig. 5B). In contrast, the CNV pattern of isolated CTCs of recurrent BCLM patient (Fig. 5C) was similar to recurrent liver metastases (Fig. 5D) (nearly 82% of the gain/loss regions). The CNVs on the remaining chromosomes showed some similarity between CTCs of recurrent BCLM and recurrent liver metastases.

GSEA analysis was performed for a functional assessment of the genes involved in specific CTC-shared CNVs. The significant biological pathways are sorted in Table III and Fig. 6. We found four common target pathways: β-defensins (hBDs), defensins, antimicrobial peptides, and Ub-specific processing proteases between newly diagnosed liver metastases (Fig. 6A), recurrent liver metastases (Fig. 6C), CTC-shared CNVs of newly diagnosed BCLM (Fig. 6B), and CTC-shared CNVs of recurrent BCLM (Fig. 6D). High expression of genes involved in defensin and hBD functions were enriched in all four groups. Moreover, we found chemokine receptors bind chemokines, GPCR ligand binding, G alpha (i) signaling events, and class C/3 (metabotropic glutamate/pheromone receptors) pathways with their response genes were higher in the CNVs of newly diagnosed liver metastases compared with CNVs of recurrent liver metastases; which may partly account for the poor prognosis in newly diagnosed BCLM patients (Fig. 6).

A pathway analysis and protein-protein interaction (PPI) networks of commonly dysregulated genes are presented in Figs. 7 and 8, receptively. Our findings show that GPCR ligand binding, chemokine receptor binds chemokine, and G alpha signaling events were enriched in patients with liver metastatic phenotype, both newly diagnosed and recurrent (Fig. 7A and C). Whereas Ub-specific processing protease, deubiquitinating, anti-microbial peptides, defensins and hBDs pathway/biological function were enriched in CTC-positive patients (Fig. 7B and D). The emapplot (Fig. 8) and cnetplot of PPI analysis (Fig. S1) indicate that all these enriched pathways and genes may play similar and critical roles in the development and progression of liver metastasis. In addition, the enriched genes have previously been reported to play critical roles in anti-angiogenesis, immunomodulation, and cell growth. These findings show that these genes are involved in: i) antitumor immunity with non-immunogenic tumor antigens (anti-microbial peptide pathway enriched in the KEGG pathway analysis); ii) suppression of cell growth, cell migration via cell cycle arrest in G1/S checkpoint (hBDs and defensins are enriched in the KEGG pathway analysis); iii) intravasation competency through ECM remodeling and collagen catabolism (defensins and GPCR ligand binding are enriched in the KEGG pathway analysis).