Patients

All patients and healthy controls were enrolled in the protocol approved by Siriraj Institutional Review Board (SIRB), a certificate of approval number Si348/2019 and Si105/2021. Informed consent was obtained from all subjects. Both groups of subjects were diagnosed by surgeons, colonoscope, and tissue biopsy pathological report. Patients with CRC stage IV who treated consecutively at the Faculty of Medicine Siriraj Hospital were enrolled, whether the patient had other diseases that can cause leukocytosis. All patients were collected peripheral blood before receiving any treatments. After surgical resection, patients were diagnosed TNM stage by using pathological (American Joint Committee on Cancer, AJCC, 7th edition) or clinical classification. Healthy control subjects who were completely done colonoscopy or diagnosed by surgeons at the Faculty of Medicine Siriraj Hospital were enrolled.

Patient samples

All human samples, fixed formalin embedded paraffin (FFPE) tissues, and peripheral blood samples were collected and analyzed with approved protocols in accordance with the ethical requirements and regulations of the Siriraj Institutional Review Board (SIRB), a certificate of approval number Si348/2019 and Si105/2021. Peripheral blood was obtained from stage IV CRC patients (N = 10) before surgical resection and healthy controls (N = 10). For each subject, 20 ml of blood sample in a 10 ml Vacuette® K2EDTA tube, EDTA (Greiner Bio-One, Frickenhausen, Germany) and a 10 ml BD Vacutainer® sodium heparin N tube, heparin (BD, NJ, USA) was collected. A total of six primary FFPE colorectal adenocarcinoma tissues and two matched liver metastases of FFPE were collected from three CRC patients analyzed with their peripheral blood in the Siriraj Cancer Center laboratory, Faculty of Medicine Siriraj Hospital.

Detection and quantification of CTCs and CHCs from human peripheral blood using the IsoFlux liquid biopsy system

Initially, 10 ml of peripheral whole blood samples in EDTA tube from stage IV CRC patients (N = 10) and healthy controls (N = 10) were processed using the CTCs enrichment kit from the IsoFlux liquid biopsy system (Fluxion Biosciences, CA, USA), which is based on EpCAM expression [31]. Briefly, Leucosep® tubes (Greiner, Kremsmünster, Austria) and ficoll-paque (GE Healthcare, IL, USA) were used, according to the manufacturer's instructions, to obtain the peripheral blood mononuclear cell (PBMC) fraction. The EpCAM coated magnetic bead and the IsoFlux system were used for EpCAM-positive cell selection [32]. Cells were stained with the circulating tumor cell enumeration kit (Fluxion Bio-Sciences, CA, USA), according to the manufacturer’s instructions. Briefly, immunofluorescence staining was performed using anti-cytokeratin (CK), anti-CD45 (CD45), and Hoechst 33,342 (nucleus) (IsoFlux CTC Enumeration Kit; Fluxion Bio-Sciences, CA, USA). The stained cells were mounted in multiwell plates with sensoplate glass bottom (Greiner Bio-One, Frickenhausen, Germany) for imaging. Imaging was performed using an inverted epifluorescence motorized microscope (Cytation1 imaging reader; Biotex, VT, USA). Automate quantification was performed for whole sample regions by imaging software: Gen5 3.08 (Biotex, VT, USA) with individuals blinded to the clinical status of the patients or healthy controls. CTCs were identified as those with an intact nucleated cell showing CK+/CD45− [32]. CHCs were identified as those with an intact nucleated cell showing CK+/CD45+.

Tumor-associated circulating endothelial cells (rare cells) isolation and analysis

For rare cell detection, 10 ml of blood samples in a heparin tube from patients with stage 4 CRC (N = 10) and healthy controls (N = 10) were sent to X-Zell (X-Zell, Bangkok, Thailand) within 12 h. The samples were processed according to the X-Zell procedures [33]. Briefly, each sample was subjected to red blood cell lysis and CD45-based high flow magnetic white blood cell depletion. The remaining cells were subjected to multiplexed cryo-immunostaining with antibodies directed against CD31 (mouse IgG1 with Alexa Fluor594, WM59; Biolegend, CA, USA), CD34 (mouse IgG1 with Brilliant Violet421, 581; BD Bioscience, CA, USA), CD45 (mouse IgG1 with Pacific Orange, HI30; EXBIO, Vestec, Czech Republic), Vimentin (rabbit IgG1 with Alexa Fluor488, EPR3776; Abcam, Cambridge, UK), pan-Cytokeratin (mouse IgG1 with Pacific Blue, C-11; Abcam, Cambridge, UK), and EpCAM (mouse IgG1 with Pacific Blue, VU-1D9 (CD326); EXBIO, Vestec, Czech Republic). The company analyzed and classified atypical cells.

The nCounter® analysis and molecular classificationRNA extraction and the nCounter® analysis

Total RNA was extracted from two sectioned FFPE tissues (5 μm thickness) using a high purity FFPE RNA isolation kit (Roche Diagnostics, IN, USA), strictly according to the manufacturer’s instructions. The nCounter® analysis system was used to perform the assay (Nanostring Technologies, WA, USA). A pan-cancer progression panel kit was used to measure the expression of 770 genes. The raw counts of each target gene were normalized by the geometric mean counts of 11 housekeeping genes (HRNP1, RPL27, RPL9, RPL6, RPL30, OAZ1, PTMA, RPS29, UBC, RPS12 and RPS16) and spiked controls. A threshold count value equal to 20 was used for background thresholding and normalizing the samples for differences in hybridization.

Classification of subtypes of CRC based on deep learning

A gene expression data set from Siriraj Hospital's CRC cohort was logarithm transformed and converted from genetic information to functional spectra associated with biological pathway activities. Subsequently, a DeepCC model (DeepCC R package version 0.1.1), containing a trained artificial neural network, was performed to extract advantageous features and classify the Siriraj hospital gene expression data into four CMS classes, CMS1, CMS2, CMS3, and CMS4 [34]

In situ detection of CTCs and CHC from human peripheral blood using spatial proteomic analysis

PBMC from two stage IV CRC patients and two healthy controls were isolated from whole blood samples in a heparin tube using ficoll-paque (GE Healthcare, IL, USA) following the manufacturer's instructions. Briefly, whole blood was first diluted with Dulbecco's phosphate-buffered saline (DPBS; Thermo Fisher Scientific, MA, USA), overlaid on ficoll-paque, and then centrifuged at 800 × g for 20 min without brake. After centrifugation, the mononuclear cells at the interface were transferred to a new tube, diluted with DPBS, and pelleted at 800 × g for 10 min. The cells were resuspended with FACS buffer (Thermo Fisher Scientific, MA, USA). Cells then adhered to poly-d-lysine-coated slides (Thermo Fisher Scientific, MA, USA) through incubation at 37 °C for 15 min, permeabilized with Triton-X (Sigma-Aldrich, MO, USA), and fixed with 4% paraformaldehyde (PFA; Thermo Fisher Scientific, MA, USA) [4]. Before staining, slides were made with 1X citrate buffer pH 6.0 (Sigma-Aldrich, MO, USA) at high pressure and temperature for 15 min. Morphological markers included Syto13 for nuclei, CD45 for immune cells, Pan-CK (CK) for adenocarcinoma cells (GeoMx® morphology markers; NanoString Technologies, WA, USA) and EpCAM (mouse IgG2 with Alexa Fluor647, 9C4; Biolegend, CA, USA). for the epithelial cell adhesion molecule. The slides were scanned and imaged using a GeoMx® instrument (NanoString Technologies, WA, USA). CHCs were identified as those with an intact nucleated cell showing CK+/CD45+.

Exploring THC in patient FFPE tissues using spatial proteomic analysis

5-μm thick FFPE sections from three CRC patients (six FFPE primary colorectal adenocarcinoma tissues and two matched FFPE liver metastases) were strictly prepared for DSP using manual instruction from the GeoMx instrument and the GeoMx immune cell profiling panel kit with 24 proteins (NanoString Technologies, WA, USA). Briefly, FFPEs were deparaffinized by incubating slides in (R)-(+)-Limonene (Sigma-Aldrich, MO, USA), and rehydrated with various concentrations of ethanol. Antigen recovery was performed with 1X citrate buffer pH 6.0 (Sigma-Aldrich, MO, USA) at high pressure and temperature for 15 min. Morphological markers included Syto13, CD45, CK (GeoMx® morphology markers; NanoString Technologies, WA, USA) and EpCAM(mouse IgG2 with Alexa Fluor647, 9C4; Biolegend, CA, USA).

The stained slides were loaded onto a GeoMx® instrument (NanoString Technologies, WA, USA) and scanned before the regions of interest (ROI) (approximately 20 nuclei/ ROI) were selected. The four colors of the morphological marker demarcate the regions of THC, tumor-immune cells, epithelial cells, and stromal cells. UV illumination was performed, and oligonucleotides were released. The photocleaved oligonucleotides released were collected using a microcapillary tube inspiration robotic system and transferred to a 96-microwell plate. The barcodes were counted in the nCounter® analysis system using standard procedures (NanoString Technologies, WA, USA). Normalized counts were calculated using three housekeeping proteins: GAPDH, HISTONE H3 and S6.

Spatial transcriptomic analysis of THC in FFPE tissues at the gene expression level (Visium CytAssist)

For spatial transcriptomic construction and sequencing, FFPE sections were obtained from two CRC patients (two FFPE primary colorectal adenocarcinoma tissues and two matched FFPE liver metastases). FFPE samples that passed the RNA quality control (DV200 > 50%) were strictly prepared according to the Visium CytAssist spatial gene expression for the FFPE tissue preparation guide (CG000518, 10X Genomics, CA, USA). The library preparation was strictly performed according to the Demonstrated Protocol (CG000520, 10X Genomics, CA, USA) and proceeded with the Visium CytAssist Spatial Gene Expression for FFPE-Tissue Preparation Guide (CG000495, 10X Genomics, CA, USA) by our laboratory at Siriraj Cancer Center, Siriraj Hospital Faculty of Medicine, Mahidol University. Libraries were sequenced by Novogene Co., Ltd. (Singapore) using a NovaSeq 6000 platform (Illumina, CA, USA). For data analysis, Space Ranger 2.1.0 May 2023 (10X Genomics, CA, USA) and GRCh38-2020-A reference were used to process FASTQ files. Downstream analyzes were performed using Loupe Browser 7.0 (10X genomics, CA, USA). The p value reported here has been given by the Wilcoxon’s test and adjusted for multiple tests via the Benjamini–Hochberg procedure.

Pathway analysis

Ingenuity Pathway Analysis Software (IPA 41280214 and 41,280,202, Ingenuity® Systems, https://digitalinsights.qiagen.com/, accessed on February 19, 2024) was used to examine the biological pathways. The IPA software (IPA 412480214 and 41,280,202) uses a manually curated database that contains information from several reputable sources, including published journal articles and gene annotation databases. Fisher’s exact test was used to calculate the probabilities between the input gene set and the pathway.

Statistical analysis

The analyzes were performed using Prism 9 software (GraphPad Software, Inc., CA, USA). The Mann–Whitney U test or the T test was used to compare whether there was a difference in the dependent variable between the two independent groups. The T test or Chi-squared was performed to test the statistical significance of clinical characteristics. Statistical significance was established at p value < 0.05.