Abstract

   Tumor dissemination is a key event in tumor progression. During this
   event, a main role is played by circulating tumor cells (CTCs), immune
   cells, and their interaction. How the immune system supports the
   survival and proliferation of CTCs is not fully elucidated. In this
   study we established an in-vitro co-culture system consisting of immune
   cells and CTCs from the same patient, which increased the success rate
   in the establishment of CTC-derived long-term cell cultures. In this
   system, we characterized the immune cells of successful co-cultures and
   the signals they exchange with cancer cells, including cytokines and
   extracellular vesicle (EV) content. Using this protocol, we stabilized
   four CTC-derived cell lines from patients with metastatic
   gastroesophageal cancer, which were cultured for over a year and
   characterized from a genetic and molecular point of view. The four cell
   lines harbor shared chromosomal aberrations including the amplification
   at 8q24.21 containing MYC and deletion 9p21.3 containing CDKN2A/B and
   the IFN type I cluster. The transcriptomic profile of CTC cell lines is
   distinct from primary tumors, and we detected the activation of E2F,
   G2M and MYC pathways and the downregulation of interferon response
   pathway. Each cell line shows a degree of invasiveness in zebrafish
   in-vivo, and the most invasive ones share the same mutation in RAB14
   gene. In addition, the four cell lines secrete cell-line specific EVs
   containing microRNAs that target YAP, BRG1-AKT1, TCF8-HDAC pathways.
   Overall, we highlight how the immune system plays a key role in the
   proliferation of CTCs through EV signaling, and how CTC cell line
   genomic and transcriptomic alterations make these cells less visible
   from the immune system and likely responsible for the survival
   advantage in sites distant from the microenvironment of origin.

   Subject terms: Cancer models, Oesophageal cancer, Tumour biomarkers

Introduction

   Gastric, gastroesophageal junction (GEJ), and esophageal cancers often
   present as a metastatic disease. Their incidence has been rising in the
   past decades and contribute to 6% of all cancer related deaths [[56]1].
   Although many treatment options are available, the prognosis of
   advanced gastroesophageal cancer (GEC) remains poor [[57]1]. In recent
   years, the molecular and genomic characteristics of this tumor have
   been extensively described [[58]2–[59]4], although providing little
   information about the biological mechanisms of GEC aggressiveness.

   Circulating tumor cells (CTCs) are considered key elements for disease
   progression, considering their pivotal role in the metastatic spread.
   CTCs are the fraction of a solid tumor that detaches from the primary
   site and travel within the bloodstream allowing the cancer to spread to
   distant organs [[60]5, [61]6]. While numerous CTCs and CTC clusters are
   shed during the primary tumor growth [[62]7–[63]9] the dissemination
   process remains highly inefficient, with only a small fraction
   (approximately 0.02%) of these cells successfully completing the
   metastatic cascade and forming secondary tumors [[64]10–[65]12].

   The number of CTC in the circulation is a prognostic biomarker in many
   tumor types, and CTC counting in GEC has been recognized to have a
   prognostic value [[66]13, [67]14]. In this context, we previously
   demonstrated that esophageal cancers shed not only epithelial CTCs, but
   also mesenchymal and hybrid “unconventional” CTCs [[68]15] during
   treatment, thus confirming the relevance of epithelial to mesenchymal
   transition (EMT) program activation in cancer evolution [[69]16]. We
   also described that esophageal CTCs harbor extensive genomic
   alterations, especially in relapsed patients, enriched for regions
   associated to the innate immune system pathways [[70]16], but how these
   aberrations impact on metastasis formation is still an open question.

   While CTC count is reaching its full potential in the clinic [[71]17],
   we still have a limited knowledge about CTC biological role in cancer
   metastasization. Growing evidence suggests the existence of an active
   crosstalk between CTCs and non-neoplastic cells [[72]18] including
   platelets [[73]19–[74]21], cancer-associated fibroblasts CAFs
   [[75]22–[76]24] and immune cells [[77]25, [78]26]. Indeed, the
   cooperation between CTCs and different cell types in the bloodstream
   influences various aspects of the metastatic cascade [[79]27]. CTC
   immune escaping is related to their interaction with “educated”
   platelets [[80]28, [81]29] and natural killer cells (NK) [[82]30];
   while extravasation is promoted by neutrophil assistance [[83]26].
   Seminal studies have shown that myeloid-derived suppressor cells
   (MDSCs) [[84]25] and CAFs [[85]23] are also involved in the metastatic
   cascade of CTCs outside the primary site by promoting their
   dissemination and facilitating their metastatic growth.

   Understanding the crosstalk between CTCs and non-tumor cells that
   sustain a favorable metastatic niche and establishing in vitro models
   of pre-metastatic niche formation is therefore crucial.

   In this study, we developed a custom co-culture system for the
   autologous co-culture of patients derived CTCs and peripheral blood
   mononuclear cells (PBMCs) to increase the success rate in long-term CTC
   cell culture establishment and identify the interactions between CTCs
   and immune cells that are responsible for CTC growth and survival. We
   found that fibroblasts and monocyte-derived macrophages lead CTC
   proliferation through a crosstalk consisting of cytokines and
   extracellular vesicle (EVs) delivered microRNAs (miRNAs).

   We propagated four CTC cell lines growing as tumoroids, which
   recapitulate the pathological characteristics of the original tumor,
   and investigated their invasive potential in a zebrafish model. We
   further characterized the CTC specific molecular and genetic features
   and the content of CTC-derived EVs.

   Our study provides convincing evidence that the proliferation of CTCs
   is supported by immune system cells, possibly through EV content and
   cytokine crosstalk. As a result of this study, we immortalized four
   CTC-derived cell lines allowing a step forward to the knowledge of the
   metastatic cascade of gastroesophageal cancer.

Results

Successfully establishment of CTC-derived cell lines from patients with
gastroesophageal cancer

   After an initial failure in CTC culturing from 18 metastatic GEC
   patients using approaches based on mono-cell culture (enriched CTCs)
   [[86]31], we decided to test the hypothesis that CTCs could acquire
   their full growth and metastatic potential from the interaction with
   other cells, and specifically immune cells. We settled a co-culture
   system for CTC propagation using both enriched CTCs and peripheral
   blood cells from 6 patients with metastatic GEC.

   The co-culture system consisted of a well insert comprising Alvetex
   scaffold technology -a porous polystyrene 3D scaffold [[87]32], in an
   ultra-low attachment well plate.

   The 3D scaffold is seeded with PBMC from patient and the bottom of the
   well was seeded with CTCs enriched from the same patient. The
   co-culture was then maintained under serum-free growth conditions with
   4% O2.

   After 20–30 days, cell aggregates were visible only in the bioreactors.
   After 25–30 days, growing CTCs were present in four out of six
   co-cultures and were further propagated outside the bioreactor. All
   bioreactor components (cells, supernatant, EVs) were collected for
   subsequent analyses.

   Long-term continuous CTC cell lines were established from four patients
   (LLCTC, NM9CTC, RGCTC, CACTC) and successfully propagated beyond the
   20th passage. This result indicates that the bioreactor was successful
   in 66% of cases, which compared to the 0% success rate of the mono-cell
   cultured mode is impressive. The clinical and pathological
   characteristics of the source patients are listed in Table [88]1. LLCTC
   and NM9CTC were both derived from esophageal adenocarcinoma. RGCTC and
   CACTC cell lines were derived from squamous esophageal and gastric
   cancer, respectively.

Table 1.

   Clinical and pathological characteristics of the source patients of
   CTC-derived cell lines.
   Cell Lines ID/Patient Primary tumor histology Grade Mestatic at
   diagnosis Resection of primary tumor Distant metastasis
   LLCTC/4 Esophageal Adenocarcinoma 3 Yes No Lymph nodes, Liver,
   Carcinosis, Adrenal gland
   CACTC/14 Diffuse Gastric Cancer 3 Yes No Lymph nodes, Adrenal gland
   RGCTC/16 Squamous Esophageal Carcinoma 3 Yes No Lymph nodes, Carcinosis
   NM9CTC/17 Esophageal Adenocarcinoma ND No Yes Liver, Carcinomatosis,
   Andrenal gland
   [89]Open in a new tab

   Interestingly, the four cell lines display distinctive growth behavior
   in-vitro. LLCTC cells initially form cell chains which fold in a
   sheet-like monolayer and then stratify in a more complex structure.
   NM9CTC cells create sealed layers and form small cyst-resembling
   structures, very similar to an organoid of esophagus, with the apical
   side of the polarized epithelium facing the lumen filled of tumor
   cells. They both form weakly aggregated structures, and many single
   cells grow in suspension (Fig. [90]1a).

Fig. 1. Morphological and phenotypic characteristics of cultured CTCs.

   [91]Fig. 1
   [92]Open in a new tab

   a Representative images of the four CTC-derived cell lines. Tumoroid
   structures can be noticed in all cell lines (black arrows). 5x (upper)
   and 10x (lower) magnification, phase contrast, scalebar 100 µm. b FACS
   analysis of EpCAM-positive cell populations (red curve) and negative
   control (cyan curve) The percentage of EpCAM+ cells is indicated. c
   Histological characterization of CTC-derived cell lines. Panels show
   cellular staining with hematoxylin/eosin (H&E), cytokeratin-7 (K7),
   cytokeratin-20 (K20), caudal type homeobox 2 (CDX2) and tumor protein
   p63.

   The epithelial status of the proliferating cells was verified testing
   EpCAM as epithelial-specific marker using FACS analysis. All four cell
   lines were strongly EpCAM-positive: LLCTC 98.9%, NM9CTC 96.4%, CACTC
   99.5%, RGCTC 96.8% (Fig. [93]1b).

   CTC cell lines recapitulate the histological features of the original
   tumors (Fig. [94]1c); as confirmed in-blind by the pathologist. The two
   cell lines derived from adenocarcinomas (LLCTC and NM9CTC) show
   epithelial-like and mucinous-like structure. Signet ring-like cells are
   present in LLCTC structures, all cells have strong positivity for CK7,
   CK20 signal is patchy, and CDX2 is absent. NM9CTC cells, on the other
   hand, show features distinct from the other CTC cell lines: a cuboidal
   layer lacking mucinous or signet-ring cells; all cells are negative for
   both CK7 and CK20, and highly positive for nuclear CDX2. RGCTC and
   CACTC cell line morphology was confirmed to be esophageal squamous cell
   carcinoma and diffuse gastric carcinoma, respectively. RGCTC cells are
   negative for CK20 and p63, while some cells are strongly positive for
   CK7. In contrast, CACTC cell line appears negative for CK7, patchy for
   CK20 and strongly positive for CDX2.

CTC-derived cell lines behavior in a zebrafish model

   Zebrafish emerged as a useful vertebrate model system to study the
   metastatic cascade in vivo [[95]33–[96]35] we tested the invasiveness
   of CTC-derived cell lines in this in-vivo model. In a
   endothelium-EGFP-labelled zebrafish embryo (Tg(Fli1a:EGFP)) we injected
   fluorescently labelled CTC-derived cells in perivitelline space. Tumor
   cell invasion was followed 24- and 48 h post-injection by confocal
   microscopy.

   Three out of four cell lines (LLCTC, RGCTC and NM9CTC) showed invasive
   ability in zebrafish. The invasive competence of the cells was
   calculated as the fraction of fishes with tumor cells in the
   circulatory system. After 24 h from injection, we detected 12/21
   positive zebrafish embryos in LLCTC, 10/18 in RGCTC, and 2/9 in NM9. At
   48 h post-injection the number of positive zebrafish embryos increased
   to 15/21 ( + 3) in LLCTC, 11/18 ( + 1) in RGCTC and 4/9 in NM9 ( + 2)
   (Fig. [97]2a, b).

Fig. 2. Invasive competence of CTC-derived cell lines in in-vivo zebrafish
model.

   [98]Fig. 2
   [99]Open in a new tab

   a Temporal monitoring of the percentage of embryos with CTC
   intravasation. b Representative images of engrafted zebrafish with
   tumor cells shed in the circulation 24 h post-injection (c)
   Representative images of tumor cell contacts with the circulatory
   system near the site of injection. At 24 h, sprouting vessel formations
   are visible, in addition to circulating CTCs. d Representative image of
   metastases formation 48 h post-injection. Green: zebrafish endothelial
   cells, red: CTC-derived tumor cells.

   Timelapse experiments (24–39 h post-injection) showed that tumor cells
   moved within the bloodstream both passively and actively via
   interactions/adhesion with the endothelial layer (movie [100]1).

   We noticed sprouting vessel formation occurred in the subintestinal
   vessels (SIV) area 24 h after injection (Fig. [101]2c). The
   characteristic feature of these vessels was their directly contact with
   the tumor cells, suggesting the intravasation of CTC-derived cell lines
   through vessel formation. At 48 h post-injection micrometastases
   localized mainly in the tail were also visible (Fig. [102]2d).

Genomic and transcriptomic analyses of CTC-derived cell lines

   To characterize the CTC-derived cancer cell lines, we performed a
   genomic analysis on the four CTC-derived cell lines revealing a wide
   spectrum of copy number aberrations (CNAs) (Fig. [103]3a). GISTIC
   analysis highlighted recurrent chromosomal aberration shared among most
   cell lines (Fig. [104]3b). In particular gain of chromosome 8q24.21
   containing MYC gene [[105]36] and loss of chromosomes 1p36.33, 3q11.2,
   4q28.3, 13q22.3, 22q11.22, 1p36.33 and 9p21.3 which contain, among
   others, the tumor-suppressor genes CDKN2A/B and a type I IFN gene
   cluster [[106]37], was also shared among most cell lines.

Fig. 3. Genomic landscape of the four CTC-derived cell lines.

   [107]Fig. 3
   [108]Open in a new tab

   a Graphical representation of CTC cell line copy number aberration
   (CNA) distribution across all chromosomes. Red dots represent
   amplifications, blue dots represent deletions (b) Significant copy
   number gains (red) and losses (blue) detected by GISTIC 2.0 (shown as
   peaks) shared between the cell lines.

   To investigate whether the CTC cell lines had a transcriptional profile
   distinct from primary tumors, we obtained their transcriptomic profile
   and compared the CTC-derived cell lines transcriptome with that of
   non-pretreated resected esophageal (ESCA) and stomach (STAD) cancer
   samples available in TCGA database. Principal component analysis (PCA)
   revealed that CTC samples have a distinct gene expression profile and
   in particular, three CTC-derived cell lines clustered together (RGCTC,
   CACTC and NM9CTC) and LLCTC cell line diverged the most (Fig. [109]4a).

Fig. 4. Gene expression analysis of the four CTC-derived cell lines.

   [110]Fig. 4
   [111]Open in a new tab

   a Unsupervised principal component analysis (PCA) of the gene
   expression profiles of 4 CTC cell lines (each one in triplicate) and
   TGCA esophageal (ESCA) and stomach (STAD) cancer samples. b Hallmark
   GSEA comparison of CTC-derived cell lines versus TGCA dataset of GEC
   primary tumors showing normalized enrichment scores for the represented
   genesets (FDR < 25%, p-value < 0.01). Red: genesets enriched in CTC
   cell lines; blue: genesets enriched in primary GECs c Hallmark GSEA
   comparison of in-vivo non-invasive CTC-derived cell line (CACTC) versus
   in-vivo invasive CTC-derived cell lines (RGCTC, LLCTC and NM9CTC)
   showing normalized enrichment scores for the represented genesets
   (FDR < 25%, p-value < 0.01). Red: genesets enriched in non-invasive CTC
   cell line; blue genesets enriched in invasive CTC cell lines (d)
   Histogram representing somatic mutations detected in at least two cell
   lines. The different colors indicate different groups of cell lines.

   This observation led us to investigate whether there were distinct
   transcriptional programs in CTC and primary tumors from the same origin
   (Fig. [112]4b). Expression profiles obtained from ESCA/STAD-TCGA
   dataset and from CTC-derived cell lines were analyzed with GSEA using
   “hallmark gene sets” (h.all.v2023.1.Hs.symbols.gmt). GSEA analysis
   highlighted 11/50 gene sets enriched in CTC-derived cell lines (10 gene
   sets are significant at False Discovery Rate, FDR < 25%) and 39/50 gene
   sets enriched in ESCA/STAD dataset (35 gene sets are significantly
   enriched at FDR < 25%). Especially compelling are the gene sets
   enriched in CTC cell lines, which include HALLMARK_E2F_TARGETS,
   HALLMARK_G2M_CHECKPOINT, HALLMARK_MYC_TARGETS_V2 all with FDR 0.0 and
   p-value < 0.01. Equally meaningful are the top-three gene sets enriched
   in ESCA/STAD, including HALLMARK_INTERFERON_GAMMA_RESPONSE,
   HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION and
   HALLMARK_ALLOGRAFT_REJECTION all with a FDR 0.0 and p-value < 0.01.

   Five of the top 50 in genes differentially expressed CTCs compared to
   ESCA/STAD samples (Fig. [113]S1) resulted highly expressed in all the
   four CTC cell lines: PCF11-AS1 (RNA gene, affiliated with the lncRNA
   class), MIR7161, SFT2D3 (involved in fusion of retrograde transport
   vesicles derived from an endocytic compartment with the Golgi complex),
   DND1 (encoding a protein that binds microRNA-targeting sequences of
   mRNA, inhibiting microRNA mediated repression) and H3C2 (a histonic
   protein).

   The distinct in-vivo invasive properties of the CTC-derived cell lines
   prompted us to investigate their specific transcriptional programs.
   GSEA gene sets positively enriched in non-invasive CACTC cell line
   resulted: HALLMARK_WNT_BETA_CATENIN_SIGNALING, HALLMARK_COAGULATION,
   HALLMARK_KRAS_SIGNALING_UP. While GSEA gene sets enriched in invasive
   CTC cell lines were 11 and the top three were: HALLMARK_G2M_CHECKPOINT,
   HALLMARK_E2F_TARGETS, HALLMARK_MITOTIC_SPINDLE (Fig. [114]4c).

   Using RNA sequencing data, we identified somatic mutations in all CTC
   cell lines. The number of non-synonymous or frameshift mutations ranged
   from 26 in CACTC to 46 in LLCTC, 70 in RGCTC and 106 in NM9CTC cell
   line; all the exonic variants are listed in Supplemented Table [115]S1.

   We found mutations in known esophageal and gastric cancer genes
   including TP53, ARID1A and PIK3CA [[116]4]. We also detected mutations
   in genes involved in DNA repair (PARP1, BCL-10, CCNB, MBD4, HMGB1),
   histone deacetylases (HDAC9, HDAC7) and HIF1A.

   Specific mutations shared between two or three cell lines are reported
   in Fig. [117]4d. We found no common mutations in all four lines, nor
   distinctive for tumor type, but interestingly the p.I60fs RAB14
   mutation is present in the three invasive CTC cell lines (LLCTC, RGCTC
   and NM9CTC).

Characterization of extracellular vesicles secreted by CTC-derived cell lines

   We hypothesized that CTC-secreted extracellular vesicles (EVs) could be
   relevant to the process of metastasis.

   We investigated this hypothesis by comparing the characteristics of EVs
   secreted by the four CTC cell lines with the EVs released by two
   commercial tumor cell lines. Specifically, OE-33 and NCI-N87 derived
   from esophageal [[118]38, [119]39] and gastric cancer [[120]40],
   respectively.

   To maximize the EV production, cells were cultured in the 3D scaffold
   Alvetex. EVs were enriched through 70 nm qEV10 Size Exclusion Columns.
   The estimated number/mL of EVs was determines by nanoparticle tracking
   analysis (NTA). We did not observe any difference in the number of EVs
   between the six samples, all ranging from 10^9 to 10^10 particles/mL
   (Fig. [121]S2). We evaluated the EV phenotypic traits by using FACS
   analysis and MACSPlex exosomes kit. We observed that all the EV samples
   were strongly CD9/CD81/CD63 positive and retained the CD326 marker of
   epithelial origin. Although we did not detect a phenotypic marker
   present only on CTC-derived EVs, we noticed that HLA-BC, CD41b, CD24,
   CD44 and CD29 were detected at higher levels in LLCTC and RGCTC
   CTC-derived cell lines (Fig. [122]5a).

Fig. 5. Characterization of extracellular vesicles secreted by cell lines.

   [123]Fig. 5
   [124]Open in a new tab

   a Phenotypic characterization of extracellular vesicles isolated from
   supernatant of exhausted culture media of cell lines obtained by FACS
   analysis. Histograms represent the mean fluorescence intensity (MFI) of
   the detected membrane proteins with values > 500 in at least one of the
   samples. MFI values are reported in the side table. b Cluster analysis
   and heatmap representation of CTC and commercial GEC cell lines based
   on the 14 miRNAs contained into EV and differentially expressed between
   the two groups.

   We investigated the EV cargo by comparing the miRNA profile of
   CTC-derived and tumor-derived EVs. We found 18 miRNAs differentially
   expressed in the two EV types; interestingly, 14 miRNAs are carried
   specifically in EVs secreted by CTC cell lines (Fig. [125]5b), on the
   contrary miR-555-5p, miR-934, miR-100-5p and miR-4796-5p were detected
   only in tumor cell lines (Fig. [126]5b). The target enrichment analysis
   of these miRNAs conducted with Metacore showed an enrichment in miRNA
   validated target genes belonging to YAP1, BRG1/AKT1 and TCF8/HDAC2
   networks (Fig. [127]S3).

Characterization of cellular components in co-cultures (bioreactors)

   Since four CTC cell lines were successfully established only in the
   co-culture setting, while 2 out of 6 failed to grow, we decided to
   investigate the features of the PBMC populations in “successful” vs.
   “failing” co-cultures. We recognized differentiated leukocytes both in
   the 2D and 3D compartment of the bioreactor (Fig. [128]6a).

Fig. 6. Characterization of immune cell differentiation in co-cultures.

   [129]Fig. 6
   [130]Open in a new tab

   a Schematic representation of the co-culture (bioreactor): Isolated
   PBMCs were seeded in the 3D compartment, while enriched CTCs are seeded
   in-suspension (2D) with residual PBMCs. After 30 days, proliferating
   CTCs are visible (brightfield images) and differentiated monocytes are
   evident both in the 3D scaffold (cells are immune-stained with DAPI,
   blue) and at the bottom of the ultra-low attachment well (cells are
   stained with phalloidin, red, and counterstained with DAPI, blue). b
   Leukocyte differentiation at the bottom of the well is dependent on the
   presence of differentiated leukocytes in the 3D scaffold. When seeded
   without the 3D compartment, leukocytes do not differentiate (adhere at
   the bottom of the ultra-low attachment well) (2D PBMC w/o 3D PBMC), but
   rather persist in-suspension and then die in about 30 days. Leukocytes
   in 2D compartment adhere and differentiate only in the presence of
   differentiated leukocytes in the 3D compartment (2D PBMC w/ 3D PBMC). c
   Representative immunofluorescence staining of differentiated leukocytes
   in the 2D-compartment indicating differentiated monocytes with a mixed
   population of macrophages M2-like profile (CD206 + /CD163 + ) and
   M1-like profile (CD11c + ), nuclei are in blue. d Representative images
   confirming the differentiation of monocytes into macrophages in patient
   co-cultures, Cells multinucleation were verified through the cell-cell
   fusion marker CD13 (red), CD146 (yellow) as a monocyte-derived subtype
   of macrophages and CD3 (green) to discriminate any residual
   T-lymphocyte. e Expression of cell type or lineage-specific genes
   including ACTA2, PDGFRb (fibroblasts/myofibroblasts), CD163, CD206
   (macrophages) and CD14, CD45 as monocyte lineage, using Digital PCR.
   Expression analysis revealed a mixed population of CD163 + /CD206+
   macrophages and fibroblast-like cells ACTA2/PDGFRb+ on 3D compartments
   seeded with patient’s PBMCs (with, plain colours, or without
   proliferating CTCs checkered pattern).

   At first, we investigated whether the leukocyte differentiation was
   compartment dependent. Using the same in-vitro conditions, we seeded
   healthy donor PBMCs in separated compartments. Specifically, we seeded
   a controlled number of leukocytes in the 2D compartment in the absence
   of cells in the 3D and vice versa, and a complete co-culture condition
   where leukocytes were present in both compartments (Fig. [131]6b).

   After 20 days of culturing, no adhering leukocytes were detectable in
   the single 2D culture condition. We identified adherent leukocytes in
   the single 3D culture and in the complete co-culture conditions, using
   confocal microscopy by nuclei staining. Only in the complete co-culture
   condition, adherent leukocytes in the 2D compartment were present.
   These cells did not proliferate but rather showed a tight adhesiveness
   on an immobilization-inhibiting surface (ultra-low attachment)
   suggesting a differentiation program activation leading to matrix
   production.

   Adherent leukocytes in the 2D compartment of the co-culture showed a
   mixed morphology: there were large cells with an egg-like shape with
   numerous folds and projections, elongated bipolar cells, cells
   exhibiting lamellopoidia, smaller rounded cells with cytoplasmic
   processes with a star-shaped morphology (Fig. [132]6a).

   The morphological appearance and ability to adhere on an ultra-low
   plastic led us to hypothesize that they derived from a monocyte
   population differentiated into macrophages. Indeed, confocal microscopy
   analysis showed that the cells in the 2D compartment are
   myeloid-derived (CD14 + ) with a macrophages M2-like profile
   (CD206 + /CD163 + ) positive for HLA-DR and macrophages M1-like profile
   (CD11c + ), while negative for dendritic markers CD1a, CD80, CD83. The
   CD11c/CD206/CD163 signature was detected also in the patient-derived
   leukocytes cultured the 2D compartment (Fig. [133]6c, Fig. [134]S4).

   As cells showed multinucleation, we verified their positivity for CD13,
   a cell-cell fusion marker, CD146 as a monocyte-derived subtype of
   macrophages and CD3 to discriminate any residual T-lymphocyte.
   Patients’ leucocytes in the 2D compartment were actual multinucleated
   cells positive for cell-cell fusion marker CD13, negative for CD146,
   and negative for CD3 (Fig. [135]6d).

   Due to the high autofluorescence of the scaffold, fluorescence
   microscopy could not be used for the identification of leucocyte
   differentiation in the 3D compartment, therefore we extracted the RNA
   from these cells and employed a Digital PCR testing to assess the
   expression of cell type or lineage-specific genes including that of
   fibroblasts/myofibroblasts (ACTA2, PDGFRb), macrophages (CD163, CD206)
   and monocyte (CD14, CD45) lineages (Fig. [136]6e). We tested the
   expression of these genes in the co-culture in the presence or absence
   of CTCs. The results showed that 3D leukocytes completely loose the
   lineage markers (CD45 and CD14) and differentiate into a mixed
   population of macrophages (CD163 + /CD206 + ) and fibroblasts
   (ACTA2 + /PDGFRb + ).

Secretome composition of co-cultures

   To gain insight into cell-to-cell crosstalk within the tumor
   microenvironment, we investigated the secretome (cytokines and EVs) of
   the co-cultures.

   We investigated the cytokine composition secreted by the leukocyte
   populations grown in separate compartments and in the co-culture system
   by performing an 80-cytokine array on supernatants. This analysis
   showed that the two distinct compartments secrete specific cytokines,
   and that there is a specific cytokine-based crosstalk within the
   complete co-culture (Fig. [137]S5). Since leukocytes in the 3D
   compartment drive the differentiation of those in the 2D compartment,
   we identified ENA-78 as a cytokine secreted exclusively in the isolated
   3D compartment and in the complete co-culture. Nine cytokines were
   detected only in the complete co-culture: GRO, GRO alpha, IL-6, I-309,
   MCP-1, MCP-2, MCP-3, MCP4 (Fig. [138]S5a). Then, we analyzed the same
   cytokine panel in the bioreactor with CTCs; we did not identify any
   additional cytokine discriminating between the presence and absence of
   proliferating CTCs (Fig. [139]S5b).

   To evaluate the phenotype of EVs produced in the co-cultures, we
   performed the multiplex assay MACSPlex which allows the detection of 37
   EV membrane proteins including EV markers CD9, CD63 and CD81.

   We first compared the MACSPLEX profile of GEC patient-derived PBMC
   co-cultures with that of a healthy donor PBMC co-culture considering a
   100 median fluorescence intensity (MFI) as a robust positive signal
   [[140]41]. All the samples were positive for exosome markers CD9, CD63
   and CD81. Phenotypic profiles are very different between patients and
   healthy donors, as EV markers of the latter are typical of cells with
   immunomodulatory activity, specifically: HLA-DR, CD40, CD11c, CD29
   (above the threshold of 500 MFI). CD41b was found to be unique EV
   marker of PBMC co-cultures of gastroesophageal cancer patients, and no
   signal was detected in healthy donor EVs (Fig. [141]S6a).

   Comparing the phenotypic profile of EVs secreted in co-cultures of EGC
   PBMCs in the presence or absence of proliferating CTCs (Fig. [142]S6b),
   we confirmed the positivity for the exosomal markers CD9, CD63, CD81
   and the presence of epithelial marker EpCAM on EVs in co-cultures with
   proliferating CTCs. In contrast, CD11c (dendritic cells) and MSCP1 were
   detectable at high ( > 100 MFI) levels in EVs in co-cultures without
   proliferating CTCs.

   Since microRNAs are one of the main components of EVs, we compared the
   miRNA content of EVs in successful CTC co-cultures and failing CTC
   co-cultures. Specifically, we compared the miRNA profiles of EVs
   secreted in: healthy donor co-culture (N = 3), CTC-failure co-culture
   (N = 2), CTC-success co-culture, with growing CTCs (N = 2) and without
   CTCs (N = 2), and CTC-derived cell lines expanded after the co-culture
   (N = 4). Comparing the miRNA profile of all EV samples (PCA analysis
   (Fig. [143]S7)) we observed that the EVs of successful co-cultures have
   a microRNA content that differs from failing/healthy co-cultures or
   co-cultures without CTCs and become more similar to EVs released by
   tumor cell lines. This content is maintained only when the CTC cell
   line grows with the support of the co-culture system.

   We identified five miRNAs delivered in EVs of CTC-success co-culture
   with CTCs, as shown in the supervised heatmap where this class of
   samples clustered independently (Fig. [144]S8). METACORE analysis
   showed that these miRNAs target genes that belong to NANOG, p53 and
   STAT1 pathways (Fig. [145]S9).

Discussion

   In this study, we exploited the interaction between cells of the immune
   system and CTCs to sustain CTC proliferation. Extending Virchow’s
   hypothesis by Balkwill and Mantovani stating that certain types of
   inflammation are the “fuel that feeds the flames” [[146]42], we
   established in-vitro co-cultures between the host immune cells and CTCs
   from the same patient. In this context, the co-culture generates a
   microenvironment which sustain CTC indefinite proliferation. Using an
   optimized co-culture system (bioreactor) we obtained 4 CTC cell lines
   starting from EGC CTC enriched from the peripheral blood; which were
   not able to grow when cultured using the same conditions, but without
   the immune microenvironment.

   The 4 CTC-derived cell lines replicate the structure and morphology of
   the tumor of origin, as well as EpCAM positivity, the marker of choice
   for CTCs of epithelial origin. Despite the technical difficulties
   associated with injecting compact tumoroids in zebrafish embryos,
   in-vivo zebrafish model intravasation and extravasation experiments
   showed that three out of 4 cell lines have invasive characteristics.
   Indeed, they enter the bloodstream within 24 h (through sprouting
   vessel formation) and originate micro-metastases within 48 h after
   injection. This unexpected result further consolidates the consensus
   regarding the heterogeneity of CTCs [[147]10, [148]43, [149]44] not
   only in phenotypic aspects [[150]16], but also in behavioral patterns
   within the metastatic cascade.

   By investigating the genetic features of the CTC-derived cell lines, we
   verified that they bear amplification of chromosome 8q24.21 which is
   known to contain the oncogenic gene MYC [[151]2, [152]4]; of particular
   interest are also the chromosomal deletion of 9p21.3 containing the
   tumor suppressor gene CDKN2A/B and the IFN type I gene cluster. These
   amplification/deletion patterns were previously described in colon stem
   cells generating CTCs in mouse models [[153]36]. In addition, in a
   recent study, Iriondo et al. described that suppression of type I IFN
   signaling by hypoxic memory increases the metastatic capacity of tumor
   cells, including cell lines derived from breast cancer CTCs [[154]45].
   These recurring genetic findings strongly suggest that these traits
   represent a CTC selective survival advantage in the bloodstream through
   amplification of oncogenic genes and deletion of mediators of immune
   evasion genes. These genetic characteristics are well reflected in the
   transcriptomic profiles of the CTC-derived lines, as they diverge from
   transcriptomic profiles of the gastroesophageal cohort of the Cancer
   Genome Atlas (TCGA). In particular, gene sets related to E2F, G2M and
   MYC targets emerge as upregulated, while gene sets related to antibody
   recognition such as gamma-interferon responses, EMT, allograft
   rejection emerge as downregulated, suggesting that CTCs are
   immune-evading tumor cells.

   By exploring the differences in the transcriptional profile of
   non-invasive and invasive CTC cell lines, we observed an enrichment in
   gene sets of Wnt/beta-catenin, coagulation and K-RAS signaling in
   non-invasive cell lines. The four cell lines have a significant number
   of mutations typical of gastroesophageal cancer (TP53, ARID1A, PI3KCA)
   [[155]46], but also in genes involved in DNA repair [[156]2] and
   histone deacetylase activity in cancer [[157]47]. Despite the high
   mutational rate, we did not identify common mutations; although, the
   frameshift mutation pI60fs in RAB14 was shared among all three invasive
   CTC cell lines. This mutation has not been described before, but recent
   findings reported RAB14 as an oncogene that is involved in the
   trafficking of tumor derived EVs by enhancing the angiogenic potential
   of endothelial cells [[158]48].

   This result led us to investigate the EVs secreted by CTCs. CTC-derived
   EVs retain a strong expression of EpCAM epithelial marker and while no
   phenotypic difference was detectable from EVs from commercial EGC cell
   lines, we observed a change in the pattern miRNAs delivered as CTC cell
   line cargo. In particular, 14 miRNAs are distinctive of CTC derived EVs
   and the pathway enrichment analysis showed that these miRNAs are
   involved in the regulation of YAP, BRG1-AKT1, TCF8-HADAC2 networks.
   MiR-3606-5p, the miRNA most enriched in CTC-EVs is known to regulate
   TGF-beta receptor type II [[159]49]. Interestingly, miR-379-3p and
   miR-199b-5p are reported to act as proliferation inhibitors in tumors
   and cell lines [[160]50] while miR-199-5p cargo of CAF-derived EVs
   promoted gastric tumorigenesis in vivo through the pathway of AKT1 and
   mTORC1 [[161]51]. The miR-410-3p and miR-4299 were implicated in
   promoting EMT/radioresistance via PTEN/PI3K/mTOR axis [[162]52]
   proliferation and metastasis [[163]53].

   Since CTCs grew only in the bioreactor, in co-culture with immune
   cells, we wanted to investigate the composition in terms of cell type
   and secretome of the bioreactor. Within the co-cultures, we observed
   the formation of a microenvironment mainly composed by two
   monocyte-derived cell types: macrophages and fibroblasts. Our
   co-cultures revealed that their differentiation occurs in a
   hierarchical order: differentiated monocytes on the 3D compartment
   prompts monocytes differentiation on the 2D compartment.

   In particular, the 3D compartment is composed of a mixed population of
   CD206 + /CD163+ macrophages and fibroblasts expressing ACTA2/PDGFBRbeta
   which are known as distinctive for cancer associated fibroblasts
   [[164]54, [165]55]. The presence of this mixed population in the 3D
   compartment, induces the differentiation of monocytes in the underlying
   co-culture (2D compartment) into M2 (CD206 + /CD163 + ) and M1
   (HLA-DR + /CD11c + ) macrophages; furthermore, some of these cells were
   polynucleated probably as a result of cell-cell fusion processes, as
   suggested by CD13 positivity [[166]56].

   The microenvironment generated by these cell populations is what
   induced CTC proliferation, as supported by the inability of CTCs to
   grow outside the bioreactor.

   This effect is in line with the dormancy-awakening mechanism described
   in previous works [[167]57]. Since the dormancy phenomenon of
   “temporary mitotic and growth arrest“ [[168]58] is reversible
   [[169]59], dormant disseminated CTCs [[170]60, [171]61] may awaken as a
   consequence of their interaction with factors produced by activated
   endothelial cells [[172]62], stromal cells [[173]63], osteoclasts
   [[174]64] neutrophils through neutrophil-derived extracellular fibers
   (NET) [[175]65], and the most studied myeloid cells, like
   monocyte-derived macrophages [[176]66, [177]67].

   The CTC proliferation latency (20-30 days) suggests its dependency from
   monocytes differentiation and the presence of soluble factors produced
   by these cells.

   The cytokines secreted in the bioreactor (MCP-1/CCL2 [[178]68,
   [179]69], ENA-78/CXCL5 [[180]68, [181]70]) have been implicated in the
   cross-talk between cancer associated fibroblasts (CAFs) and tumor
   associated macrophages (TAMs) [182]71, [183]72 Within the co-cultures,
   we find additional cytokines that contribute to the establishment of an
   inflammatory tumor microenvironment, such as SPP1 [[184]73], MCP-3
   [[185]68], RANTES [[186]68, [187]74], IL-8 [[188]69, [189]75].

   Of note, we did not identify distinctive cytokines released only in the
   presence of proliferating CTCs, thus suggesting that these
   microenvironment components are driven specifically by the immune cell
   compartment.

   Interestingly, we found differences in the phenotypic profile of EVs
   derived from patients’ co-cultures: EVs in GEC patient-derived
   bioreactors express CD41b while EVs in bioreactors with PBMC from
   healthy donors display an immuno-modulatory phenotype
   (HLA-DR/CD40/CD11c/CD29).

   We also identified 5 unique miRNAs in EVs of bioreactors with
   proliferating CTCs, but not present in EVs from failing/healthy
   co-cultures or bioreactors without CTCs, thus suggesting that the
   crosstalk between fibroblasts, macrophages and CTCs occurs mostly
   through the EV exchange. In addition, this EV content is maintained
   only when the CTC cell line grows with the support of the co-culture
   system, thus suggesting the relevance of the tumor-immune cell
   crosstalk in the very early phases of the metastatic process.

   Metacore analyses showed that these miRNAs are involved in the
   regulation of NANOG, p53 and STAT1 pathways. Among these miRNAs,
   mir-483 is located in the intronic sequence of the IGF2 gene, has been
   observed to induce in-vitro the proliferation of hepatocellular
   carcinoma cells [[190]76, [191]77]

   Our data indicate that an in-vitro inflammatory environment [[192]42]
   generated in co-cultures with autologous immune system cells is able to
   induce CTC proliferation, leading to the successful establishment of
   CTC-derived cell lines. The inflammatory environment is determined by
   the progressive monocyte differentiation toward CAF-like fibroblasts
   and M2/M1 macrophages, as the result of the secretion of specific
   cytokines. These cells communicate with CTCs through the release of
   EVs, whose miRNA cargo changes as a consequence of the crosstalk.
   Specifically, 5 miRNAs contained in EVs are most likely involved in the
   initial in vitro proliferation and survival of CTCs. The CTC-derived
   cell lines are able to independently form 3D structures that resemble
   the tumor of origin, retain positivity to EpCAM, and genetically
   exhibit 8q24.21 amplification containing the MYC oncogene. From a
   transcriptional point of view, these CTC cells lines show
   immune-evasion characteristics that likely drive their survival
   advantage in the bloodstream.

Methods

CTC and PBMC isolation and cell line establishment

   Patients with plurimetastatic gastrointestinal cancer were enrolled at
   the IRCCS Istituto Romagnolo per lo Studio dei Tumori (IRST) “Dino
   Amadori” (Italy) between 2015 and 2019 (N = 24). From each patient we
   collected about 20 mL of peripheral blood. The study was conducted
   under Romagna Ethics Committee (CEROM) of Meldola approval (ref. B063)
   according to the recommendations of the declaration of Helsinki. All
   patients provided informed consent prior to blood acquisition.

   About 20 mL of peripheral blood samples was drawn into a EDTA
   vacutainers (BD Diagnostics, Franklin Lakes, NJ, USA) and processed
   within 3 h.

   The workflow for cells co-culturing is outlined in Fig. [193]S10.

PBMC preparation for culture

   PBMCs were isolated from 3 ml of peripheral bloodusing density gradient
   centrifugation via Ficoll-Paque(GE Healthcare, Pittsburgh, PA, USA).
   Peripheral blood was diluted (1:2) with sterile DPBS 1X (GIBCO) and
   layered gently over Ficoll-Paque. After centrifugation (1200 g, 30 min
   without break) PBMC buffy coat was enriched at the interphase between
   plasma and Ficoll-Paque. Buffy coat was carefully aspirated without
   disturbing red blood cells layer, isolated PBMCs were washed with DPBS
   X twice and counted for seeding.

CTC preparation for culture from 18-20 ml of peripheral blood

   CTCs were enriched using negative depletion reagent with a minimal
   handling protocol in order to minimize CTC loss. Peripheral blood was
   firstly incubated with RosetteSep™ CTC Enrichment Cocktail (StemCell
   Technologies, Vancouver, Canada) (using 40 ul of RosetteSep™ solution
   per ml of blood instead of 50 ul/ml) for 20 min at room temperature.
   During this step the tetrameric antibody complexes targeting CD2, CD16,
   CD19, CD36, CD38, CD45, CD66b and glycophorin A on erythrocytes are
   formed.

   Following incubation, blood was diluted (1:2) with sterile DPBS1X and
   layered gently on the top of the insert barrier of a SepMate™ tube
   (StemCell Technologies, Vancouver, Canada) previously loaded with 15 ml
   of Ficoll-Paque (above the insert) and centrifuged at 1200 g for 10 min
   brake on. During centrifugation tetrameric antibody complexes of
   unwanted cells together with erythrocytes (serving as anchor) are
   pelleted at the bottom of the tube. SepMate top layer, containing
   enriched CTCs, was poured off into a new 50 ml tube. Enriched CTCs were
   washed with DPBS1X and resuspended with serum-free culture medium for
   co-culture.

   Serum-free culture medium comprised DMEM/F12 Glutamax (GIBCO, Life
   Technologies, Paisley, UK) supplemented with 15 mM HEPES (Life
   Technologies), 25% IntestiCult™ Organoid Growth Medium (Stem Cell
   Technologies), Heparin Solution (Stem Cell Technologies), 1%
   penicillin-streptomycin (GIBCO), 20 ng/mL human recombinant Epidermal
   Growth Factor (EGF; Peprotech, London, United Kingdom) and 20 ng/mL
   human recombinant Fibroblast Growth Factor (FGF; Peprotech).

Assembly of the co-culture

   The co-culture system consists of 3 compartments: a well-insert with
   Alvetex® scaffold technology (Reprocell; MD, USA) - a 200 ul thick
   polystyrene scaffold, the bottom area of the well composed by an
   ultra-low attachment plastic surface (2D) and in the middle the
   in-suspension area.

   The assembly of the co-culture involved the in-suspension seeding of
   enriched CTCs, following a 1 h PBMCs seeding on 3D well-insert in an
   ultra-low attachment 6-well plate.

   Prior to cell seeding, Alvetex® scaffold well insert was rendered
   hydrophilic by washing it with a sterile solution of 70% ethanol,
   followed by a wash step with DPBS1X.

   To prevent the 3D scaffold from drying out, the co-culture system was
   filled with culture medium to make the 3D scaffold wet. 106 PBMCs (50ul
   volume) were seeded carefully directly onto the center of the Alvetex®
   scaffold. Enriched CTCs were in-suspension seeded below the well
   insert. Co-cultures were maintained in a humified incubator at 37 °C
   with 4% O2 and 5%CO2.

Immunohistochemical and immunophenotypical analysis

   CTC-derived cell lines structures were encapsulated in a 1% Agar (Life
   Technologies) domes, to preserve their 3D-organization. All the domes
   were fixed in 10% formalin and paraffin embedded, with 4 µm sections
   cut for staining. Antigen retrieval and antibodies staining were
   performed on the fully automated slide stainer Ventana (Ventana Medical
   Systems Inc, Roche AG) according to manufacturer’s instructions using
   antibodies directed toward: Cytokeratin-7 (CONFIRM anti-Cytokeratin 7
   (SP52) Rabbit Monoclonal Primary Antibody, Roche Ventana),
   Cytokeratin-20 (CONFIRM anti-Cytokeratin 20 (SP33) Rabbit Monoclonal
   Primary Antibody, Roche Ventana), CDX-2 (CDX-2 (EPR2764Y) Rabbit
   Monoclonal Antibody, Roche Ventana), p63 (VENTANA anti-p63 (4A4) Mouse
   Monoclonal Primary Antibody, Roche Ventana) and counterstained with
   Hematoxylin II.

   Flow cytometric analysis was performed using a FACS-Canto flow
   cytometer (Becton Dickinson, Franklin Lakes, NJ, USA). A total of 10^6
   events per sample were acquired using FACSDiva software (Becton
   Dickinson), events were acquired at the medium rate (approximately 200
   events/seconds) background fluorescence was estimated by substituting
   primary antibody with specific isotype control. The following primary
   antibodies were employed for 15 min in the dark at room temperature
   according to manufacturer’s instructions: CD326-FITC (EpCAM) Antibody,
   anti-human, REAfinity™ (Mitenyi Biotec GmbH, Germany) and FITC
   conjugated REA Control Antibody, human IgG1, REAfinity™ (Mitenyi
   Biotec).

   Fluorescence Microscopy were performed on fixed cells with 4%
   paraformaldehyde, samples were blocked with 2% Bovine Serum Albumin
   (BSA, St. Louis, MO, USA), stained with 1 µg/mL
   4′,6-diamidino-2-phenylindole (DAPI, Life Technologies) and incubated
   overnight at 4 °C with conjugated primary antibodies. List of employed
   antibodies:
   CD206  FITC Anti-Mannose Receptor antibody [15-2] ab270647
   CD163  PE Anti-CD163 antibody [GHI/61]            ab95613
   CD83

   BD Pharmingen™ APC Mouse Anti-Human CD83

   Clone HB15e (RUO)
   551073
   CD3    BD Pharmingen™ FITC Mouse Anti-Rat CD3     554832
   CD13   CD13 Antibody, anti-human, REAfinity™      130-120-238
   CD11c

   BD Pharmingen™ APC Mouse Anti-Human CD11C

   Clone B-ly6 (RUO)
   559877
   CD14   CD14 Antibody, anti-human, REAfinity™      130-110-518
   CD1a

   BD™ CD1a PE

   Clone SK9
   333167
   HLA-DR HLA-DR Antibody, anti-human, REAfinity™    130-111-789
   CD80   CD80 Antibody, anti-mouse, REAfinity™      130-116-461
   [194]Open in a new tab

   Cell images were captured by a Nikon Eclipse Ti2 confocal microscope
   (Nikon Corporation, Tokyo, Japan) using NIS Elements software (Nikon
   Corporation).

CTC monitoring in Zebrafish embryo

   Adult and embryo of Tg(fli1a:EGFP) transgenic zebrafish strain
   [[195]78] were maintained and handled according to and European
   Directive 2010/63/EU and Italian law (D.Lgs 26/2014).

   Fertilized eggs were maintained in fish water supplemented with 0.1% of
   Methylene Blue at 28 °C. Before handling, embryos were anesthetized
   with 0.02% tricaine solution (Ethyl 3-aminobenzoatemethanesulfonate
   salt, Sigma-Aldrich® Merck KGaA). Prior to injection, CTC cells were
   labeled with a red fluorescent viable dye (CellTracker™ Deep Red Dye,
   Invitrogen, Carlsbad, CA, USA) and resuspended with PBS at the
   concentration of 2.5 × 10^5/µL. The cell suspensions were grafted into
   the perivitelline space close to the sub-intestinal vein (SIV) plexus
   of 48 hpf Tg(fli1a:EGFP) embryos using a pulled micropipette. 3 h post
   implantation, the embryos were screened to select the correctly grafted
   larvae. Injected embryos were maintained at 32 °C before image
   acquisition. Circulating cells were monitored in vivo for 24 h and 48 h
   by means of a confocal microscopy (Nikon Corporation, Tokyo, Japan) and
   images analyzed with the NIS Elements software (Nikon Corporation,
   Tokyo, Japan).

Whole genome copy number analysis

   For whole genome copy number aberration analysis (CNA), a cellular
   pellet containing approximately 1000 cells from each cell line was
   subjected to whole genome amplification using the Ampli1 WGA kit
   (Menarini Silicon Biosystems, Castel Maggiore, Italy), following the
   protocol provided by the manufacturer. The quality of each amplified
   DNA using the Ampli1 QC Kit (Menarini Silicon Biosystems). The
   resulting PCR products were run on a 2,5% agarose gel following and QC
   bands were visualized using the Chemidoc XRS System (Bio-Rad, Hercules,
   CA, USA). WGA products were conserved at −20 °C until further
   downstream analysis.

   Libraries were prepared using the Ampli1 LowPass kit for Ion Torrent
   (Menarini Silicon Biosystems), following the protocol provided by the
   manufacturer. For determination of library concentration and size,
   Qubit™ 1X dsDNA High Sensitivity (HS) (Thermo Fisher Scientific,
   Waltham, MA, USA) and Bioanalyzer High Sensitivity DNA Analysis
   (Agilent Technologies, Santa Clara, CA, USA) were used, respectively.
   For sequencing, Ion chips were prepared using the Ion Chef platform and
   sequenced using the Ion S5 sequencing system (Thermo Fisher Sc.).

   CNA calling was performed as described in [[196]79]. Briefly, BAM files
   were processed by Control-FREEC [[197]80] Mann–Whitney and
   Kolmogorov–Smirnov statistical tests were used to assess the
   statistical significance of each CNAs. We applied the Genomic
   Identification of Significant Targets in Cancer (GISTIC) tool [[198]81]
   to identify aberrant chromosomal regions shared between the four cell
   lines.

RNA extraction

   RNA extraction from CTC cell lines was performed using RNeasy Plus Mini
   Kit (QIAGEN, Hilden, Germany). RNA extraction from differentiated PBMC
   was performed using TRIzol reagent (Invitrogen) according to the
   manufacturer’s instructions and purified using the RNeasy MinElute
   CleanUp kit with DNAse treatment (Qiagen). RNA concentration was
   evaluated using the Spectrophotometer Nanodrop-ND-1000 system (Celbio,
   Milan, Italy). RNA extraction from EVs was performed using the
   Plasma/Serum RNA Purification Mini Kit (Cat. 56100, Norgen Biotek, ON,
   Canada), following the manufacturer’s protocol. All samples were stored
   at −80 °C until further analyses.

RNA sequencing and gene expression analysis

   We conducted a short-read, stranded RNA sequencing (RNA-seq) on the
   four CTC-derived cell lines. RNA-seq experiments were performed using 3
   biological replicates for each cell line. RNASeq libraries were
   generated for the four CTC-derived cell lines using Qiagen QIAseq Fast
   Select RNA Removal kit (Qiagen) and Qiagen QIAseq stranded total RNA
   library kit (Qiagen). Specifically, for library generation 500 ng of
   total RNA (RIN > 8) was used as a starting material. Reverse
   transcription, second strand synthesis, end-repair, A-addition and
   adapters ligation were performed as reported in the manufacturer’s
   instructions. Quality Control (QC) of the libraries was performed using
   Agilent High Sensitivity DNA kit 1000 (Agilent). RNA-Seq was performed
   using the Illumina NextSeq 500 Instrument (Illumina) and the NextSeq
   High Output kit v.2.5 paired-end flowcell (Illumina).

   The RNA-seq data preprocessing and analysis workflow, specifically
   tailored for differential expression gene (DEG) analysis between
   primary tumors and circulating tumor cells, was drafted using Snakemake
   v.7.18.1 [[199]82], a robust workflow management system performed in
   Anaconda environment v.23.1.0. Raw sequencing reads were quality
   checked using FastQC v.0.11.8 and filtered to remove low-quality reads
   and adapter sequences using Trimmomatic v.0.39. Filtered reads were
   aligned to the reference genome from Gencode release 41 (GRCh38.p13)
   using STAR v.2.7.10b with default parameters from Release 32 of the GDC
   mRNA quantification analysis pipeline. Gene expression levels were
   quantified and normalized by using DESeq2 R package [[200]83], and
   differential expression analysis was performed to identify genes with
   significant expression changes between conditions. Gene set enrichment
   analysis was conducted using GSEA software [[201]84].

Small RNA sequencing

   Libraries were prepared using the Qiaseq miRNA library kit (Qiagen)
   starting from 5 ul of total RNA and following the protocol for low
   input samples. Libraries were quantified using the Qubit dsDNA HS kit
   (Thermo Fisher), and quality was assessed using the Bioanalyzer DNA HS
   kit (Agilent). Sequencing was performed on the NextSeq550 Instrument
   (Illumina). Library pools were diluted to 1.5 pM and sequenced using
   NextSeq 500/550 High Output Kit v2.5 75 cycles flow cell (Cat. N°
   20024906, Illumina, San Diego, CA, USA) on NextSeq 500 platform
   (Illumina). Sequencing data analysis: raw data (FASTQ) were analyzed
   according to the QIAseq miRNA Primary Quantification pipeline using the
   GeneGlobe Data Analysis Center (available at
   [202]https://geneglobe.qiagen.com/us/analyze Qiagen, Hilden Germany).
   All miRNA UMI (Unique Molecular Identifiers) raw counts were normalized
   using the bioconductor R package DESeq2 (Love 2014, PMID: 25516281).
   Differentially expressed miRNAs were identified using GeneSpring GX
   software (Agilent Technologies). Top scored networks of miRNAs cargo
   EVs were performed using MetaCore^TM software (Clarivate Analytics, UK,
   [203]https://clarivate.com/products/metacore/).

Gene expression analysis with Digital PCR

   Reverse transcription reactions were performed in 20 µL of nuclease
   free water containing 80 ng of total RNA using iScript cDNA Synthesis
   kit (Bio-Rad Laboratories, Hercules, CA, USA).

   The resulting cDNA was stored at −20 °C. For dPCR, the chip-based
   QuantStudio 3D Digital PCR system (Applied Biosystems, Foster City, CA,
   USA) was used to determine the expression of ATCTA2, CD163, CD206,
   CD14, CD45, PDGFRB with B2-microglobulin as an internal control. All
   dPCR experiments were carried out using the chip-based QuantStudioTM 3D
   Digital PCR system (Applied Biosystems, Foster City, CA, USA), in
   accordance with the “Minimum Information for Publication of
   Quantitative Real-Time dPCR Experiments” (dMIQE) guidelines. All
   multiplex reactions mix were composed by 2X QuantStudio 3D™ Digital PCR
   Master Mix v2 (Applied Biosystems), 20X gene specific assay FAM and 20X
   gene B2 microglobulin assay VIC. In brief, chips were run in GeneAmp
   PCR System 9700 (Applied Biosystems) by applying the following
   conditions: hold at 96 °C for 10 min; 45 cycles of 60 °C for 2 min and
   98 °C for 30 s; hold at 60 °C for 2 min. At the end of the reaction,
   chips were processed with QuantStudio™ 3D Digital PCR system (Applied
   Biosystems) and data analysis was performed using QuantStudio™ 3D
   Analysis Suite™ software (version 3.0.3).

Variant calling and annotation from RNA seq data

   Reads quality assessment was performed before the alignment step using
   FastQC v0.11.5
   [[204]http://www.bioinformatics.babraham.ac.uk/projects/fastqc/].
   Sequencing reads were aligned to the GRCh38 genome assembly (with
   GENCODE v22 reference annotation) using a two-pass method with the STAR
   algorithm v2.7.3a [[205]85]. Following the methods used by the
   International Cancer Genome Consortium (ICGC)
   [[206]https://github.com/akahles/icgc_rnaseq_align], the two-pass
   method includes a splice junction detection step. Aligned reads in BAM
   format were sorted, and duplicate reads were flagged using Picard
   MarkDuplicates v2.9.0-1
   [[207]https://broadinstitute.github.io/picard/]. Systematic biases
   affecting the assignment of base quality scores by the sequencer were
   corrected using a BQSR procedure in a two-stage process performed by
   GATK tools (v4.1.4.1) BaseRecalibrator and ApplyBQSR [[208]86]. Somatic
   short mutations (single-nucleotide variants and insertions/deletions)
   were called using GATK Mutect2 in tumor-only mode. Variants with
   sequencing depth >10 were retained and functionally annotated using
   ANNOVAR (with refGene, avsnp150, ljb26_all, cosmic90, clinvar_20190305,
   intervar_20180118, gnomad211_exome databases) [[209]87].

Cytokines assays

   80-Cytokines analyses were conducted using a membrane-based antibody
   array, Cytokine Array C5 (RayBiotech Life, Inc., GA, USA) according to
   manufacturer’s protocol. Briefly, membranes were incubated with
   blocking buffer and then overnight incubated with 1 mL of neat medium
   under gentle rotation at 4 °C. Membranes were washed and cytokines were
   detected though HRP-Streptavidin/Biotin chemiluminescence detection
   using ChemiDoc MP Imaging System (Bio-Rad). Spot intensity was
   quantified using QuatityOne software (Bio-Rad) and data were then
   analyzed with the RayBio Antibody Array Analysis Tool (RayBiotech),
   which subtracted the background and the normalized data to the positive
   control signals.

Extracellular vesicles isolation and phenotypical analysis

   About 30 mL of exhausted media from co-cultures were intended to EV
   isolation. Media were initially cleared of any cell debris by
   centrifugation of 1000 g for 10 min then concentrated using through
   Centricon Plus-70 centrifugal filter devices (Merck Millipore). EV
   isolation was performed by size exclusion chromatography using qEV10
   Size Exclusion Columns (70 nm, Izon Science) according to the
   manufacturer’s recommendations. After column equilibration and flushing
   with PBS 1X, we collected EV enriched 7–9 fractions using 0.5 mL per
   fraction and performed particle concentration of individual fractions
   by Nanoparticle Tracking Analysis (NTA). NTA analyses were performed
   using a NanoSight NS300 (Malvern Instruments, UK), equipped with NTA
   2.3 analytical software laser, measurements were settled according to
   the manufacturer’s instructions diluting samples with 0.1 um filtered
   PBS1X. Since the EV quantifications indicated that the highest
   concentration was within fractions 7–8, we combined those fractions for
   subsequent experiments.

   EVs phenotypical analyses were performed using MACSPlex Exosome kit
   human (Miltenyi Biotech) following the manufacturers protocol opting
   for overnight capture, suggested for the culture medium. This kit
   enables the detection of 37 markers (CD1c, CD2, CD3, CD4, CD8, CD9,
   CD11c, CD14, CD19, CD20, CD24, CD25, CD29, CD31, CD40, CD41b, CD42a,
   CD44, CD45, CD49e, CD56, CD62p, CD63, CD69, CD81, CD86, CD105, CD133.1,
   CD142, CD146, CD209, CD326, HLA- ABC, HLA-DR DP DQ, MCSP, ROR1 and
   SSEA-4) simultaneously, and two isotype controls (mIgG1 and REA
   control) corresponding to the antibodies. Samples staining was
   performed as in [[210]88]. The samples were washed with MACSPlex buffer
   and analyzed on a BD FACS Canto equipped with two lasers, 488 nm, and
   630 nm (Becton Dickinson), running a BD FACSDiva software (Becton
   Dickinson) recording a minimum of 50 events for each population of
   specific beads. 39 bead populations (37 exosomal surface epitopes + 2
   isotype controls) were distinguished by different fluorescence
   intensities detected in the FITC, PE, and APC channels. Collected data
   were analysed by BD FACSDiva software (Becton Dickinson), median
   fluorescence intensity (MFI) was “background corrected” by blank
   samples and isotypes controls subtractions.

Supplementary information

   [211]Normalized_Counts^ (7.1MB, xlsx)
   [212]Normalized_miRNA^ (687.6KB, xlsx)
   [213]SUPPLEMENTARY FIGURES^ (8.6MB, docx)
   [214]Movie 1^ (701.9KB, mp4)
   [215]Supplemental Table S1^ (2.7MB, xlsx)

Acknowledgements