Metastasis is the primary cause of cancer-related deaths, accounting for an estimated 66% to 90% of fatalities [1]. It is a multistep process involving the dissemination of circulating tumor cells (CTCs) and their colonization of distant organs [2, 3]. A higher number of detected CTCs in cancer patients is associated with shorter survival [4].

We analyzed 544 single-cell RNA sequencing (scRNA-seq) profiles of bona fide CTCs, identified as keratin-positive and aneuploid, from over 3,000 putative CTC profiles available in public databases, as detailed in Supplementary Table S1. Most of the CTCs originated from patients with breast cancer (n = 502, 92.3%), while a smaller number were derived from patients with prostate cancer (n = 42). All experimental methods are described in the Supplementary Materials and Methods.

All bona fide CTCs were positive for KRT18 and negative for PTPRC (CD45), as expected. Three main CTC subgroups were identified (Supplementary Figure S1). We labeled the two epithelial (EPCAM⁺) subgroups as epithelial A (epiA) and epithelial B (epiB), while the third subgroup was mesenchymal (VIM⁺/EPCAM⁻). CAV1 and AXL showed the highest specificity for mesenchymal CTCs, whereas LY6E was the most distinctive gene for epiB CTCs (Supplementary Tables S2–S4). Further analysis revealed that mesenchymal and epiB, but not epiA CTCs, were actively engaged in the cell cycle, as inferred using the R package Tricycle (Supplementary Figure S1). The biological implications of these three CTC subgroups are highly relevant. Mesenchymal CTCs expressed significantly lower levels of KRT18 and other keratins, such as KRT19 and KRT7, compared to epithelial CTCs. Conversely, vimentin, another class of intermediate filaments, was highly expressed in mesenchymal CTCs. The shift from keratins to vimentin is a hallmark molecular event in epithelial-to-mesenchymal transition (EMT). EMT regulators ZEB1, ZEB2, and SNAI2 were upregulated in mesenchymal CTCs, indicating that EMT was responsible for their origin. These findings highlight the need to prioritize the detection and targeting of epiB and mesenchymal CTCs. PD-L1 (CD274), an important target for immunotherapy in clinical practice, was expressed in only a small fraction of mesenchymal CTCs and even less in epithelial CTCs (Supplementary Figure S2). In contrast, two other immune checkpoint genes, CD276 (B7-H3) and PVR (CD155), were highly expressed in CTCs, comparable to their expression in trophoblasts. This suggests an immuno-evasive phenotype common to most CTCs, driven by the expression of CD276 and PVR.

Is there a functional relationship between CTCs and trophoblast cells, as suggested by the co-expression of genes such as CD276, SP6, and LY6E (Supplementary Figure S3)? To address this question, we examined potential links between CTCs and the placenta or early embryo by integrating scRNA-seq profiles of CTCs with those from normal and cancerous breast tissue, early embryos and first- and second-trimester human placenta (Figure 1A, Supplemental Figures S4,S5). The UMAP plot positioned CTCs within a region enclosed by metastatic breast cancer cells, early embryonic cells, and trophoblast cells. We further explored these interrelations using divisive hierarchical spectral clustering (Figure 1B), which confirmed that CTCs, trophoblast cells, and embryonic cells share similar RNA profiles (Supplementary Figure S6 and Supplementary Table S5). To validate these findings, we mapped the CTCs onto the transcriptional landscape of embryo developmental [5] (Figure 1C). A subset of CTCs (n = 72 out of 544, p-value < 0.001) aligned with the trophectoderm (TE), the blastocyst cells that give rise to the trophoblast, facilitating embryo attachment and subsequent invasion to form the placenta. Notably, these TE-like CTCs were predominantly from the epiB subgroup (64 out of 72, Fisher Test p-value < 0.001) and were frequently in the S phase of the cell cycle (Supplementary Figure S7). To further investigate the similarity between epiB CTCs and TE, we performed a transcriptomic correlation analysis across all cell types. In the resulting correlation plot (Figure 1D), epiA CTCs clustered among breast cancer subtypes, while epiB and mesenchymal CTCs clustered with early embryonic stages. In particular, epiB CTCs showed strong similarity to TE and its precursors, pre-lineage cells. This analysis confirmed the strong relationship between epiB CTCs and the TE lineage previously observed.

We hypothesized that the similarity between epiB CTCs and early embryonic stages arises from functional convergence. Specifically, we proposed that key traits essential for the functionality of CTCs (invasiveness and immune-evasion) are encoded in the human genome as part of the trophectodermal program, which leads to the extravillous trophoblast and ultimately to the placenta. To test this hypothesis, we investigated transcription factors that were upregulated and had active gene regulatory networks (GRNs) in both epiB CTCs and TE or its precursors, pre-lineages (Figure 1E). While epiB CTCs exhibited several active GRNs, only CEBPA and ILF2 were shared with TE. Furthermore, while ILF2 was ubiquitously expressed across the dataset, the upregulation of CEBPA mRNA was predominantly restricted to epiB CTCs and TE (Supplementary Figure S8). Additionally, we identified several GRNs that appeared to be specific to either epiB or mesenchymal CTCs (Figure 1E).

Is it possible to define a cellular path for cancer establishment and progression in breast cancer, given the diversity of the scRNA-seq profiles in the dataset we assembled? To reconstruct the lineages leading from normal breast tissue, through various breast cancer subtypes, to metastatic lymph nodes and eventually to CTCs, we inferred pseudotime using Slingshot [6]. Although the analysis was unsupervised, the normal breast clusters were accurately identified as the starting points, culminating in the epithelial A/B CTCs via intermediate cell clusters from ER⁺ and metastatic cancers (Figure 1F). The progression lineages for ER⁺ and HER2⁺ breast cancer shared common evolutionary segments, ultimately leading to the emergence of CTCs (Supplementary Table S6).

We finally identified RNA modules that could be relevant to metastatic evolution. The genes upregulated in CTCs, metastatic lymph nodes, and their respective primary tumors are shown in Supplementary Figure S9 and listed in Supplementary Table S7. Two genes associated with metastasis (ALDOA and PSMA6) were also upregulated in TE. The expression levels of RNA modules implicated in progression are displayed, superimposed on the UMAP plot, in Supplementary Figures S10–S12.

In this study, we aimed to characterize CTCs within the context of the cancer environment using scRNA-seq. We identified a mesenchymal CTC (VIM⁺/AXL1⁺) subpopulation, distinct from the larger epithelial CTCs (EPCAM⁺) population. Importantly, we further divided the epithelial CTCs into two divergent subgroups: epiA, characterized by high CD24/CDH1 expression, and epiB, marked by elevated levels of the stem cell master regulators SOX2/CEBPA. Notably, epiB and mesenchymal CTCs, but not epiA CTCs, exhibited mitotic activity. Of clinical significance, CD276 and PVR, but not PD-L1, were the primary immune checkpoint genes expressed in CTCs. CD276, like PD-L1, is an immune checkpoint that suppresses tumor antigen-specific immune responses and is a target of anticancer agents such as enoblituzumab [7], and CAR T cells [8]. We propose that CD276 and PVR could serve as targets for novel immunotherapeutic strategies to eliminate CTCs.

In conclusion, we identified a novel CTC subtype, epiB, along the lineages of breast cancer progression, characterized by high levels of the stem cell master regulator CEBPA and significant mitotic activity. For the first time, we also demonstrated a link between this CTC subgroup, epiB, and the embryonic trophectoderm. EpiB CTCs may utilize elements of the TE genetic program to invade the vasculature, achieve metastasis, and implement fetal-like immune tolerance. The RNA modules involved in cancer progression that we identified, particularly those of mesenchymal and epithelial B CTCs, could have clinical applications in detecting minimal residual disease [9] and in identifying novel molecular targets in metastasis.

AUTHOR CONTRIBUTIONS

Stefano Volinia conceived and designed the study, collected the data, and performed the analysis. Stefano Volinia, Krystian Jazdzewski, Anna Terrazzan, Jeff Palatini, Tomasz S Kaminski, Eva Reali, and Nicoletta Bianchi discussed and revised the methods and results. Stefano Volinia, Krystian Jazdzewski, Anna Terrazzan, Jeff Palatini, Tomasz S. Kaminski, Eva Reali, and Nicoletta Bianchi drafted the manuscript. All authors read, revised, and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

FUNDING INFORMATION

Italy's MUR PNRR National Center for HPC, big data and quantum computing (CN00000013 CN1) and Poland's National Science Centre project OPUS 24 (2022/47/B/NZ7/03418) to Stefano Volinia. Stefano Volinia was also recipient of a Polish NAWA Ulam Scholarship (BPN/ULM/2021/1/00232) and of an University of Ferrara FAR 2024 grant. Krystian Jazdzewski was supported by Foundation for Polish Science (POIR.04.04.00-00-1DD9/16-00).

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

We confirm that all methods were carried out in accordance with relevant guidelines and regulations. Data were obtained from public databases.