Red Blood Cells Internalize Extracellular DNA via Apoptotic Bodies with Clinical Relevance to Cancer Patients
Zihang Zeng, Zongbi Yi, Jing Hu, Jiali Li, Yu Xu, Xiuli Guo, Qian Ji, Kaixiang Feng, Ying Zhang, Sirui Bai, Yushuang Tan, Yufei Yan, Linzhi Han, Jing Jiang, Tengfei Wang, Xiang Wang, Ziqing Zhan, Ruiying Huang, Jinfang Zhang, Conghua Xie, Binghe Xu
TL;DR
Red blood cells can absorb DNA from the environment, which may help track cancer progression and treatment response.
Contribution
Discovery that red blood cells internalize DNA via apoptotic bodies, linking this process to tumor burden and clinical outcomes.
Findings
RBCs internalize extracellular DNA fragments similar to cell-free DNA.
Apoptotic bodies mediate DNA uptake, causing RBC deformation and oxidative stress.
rbcDNA levels correlate with tumor burden and treatment response in cancer patients.
Abstract
Mature red blood cells (RBCs), the most abundant anucleate cells in humans, have long been overlooked as DNA carriers. Recent evidence implicates RBC‐derived DNA (rbcDNA) as a potential biomarker for cancer diagnostics, yet its origin and how RBCs acquire tumor DNA remain poorly understood. Here, we find that mature RBCs harbor short DNA fragments distinct from genomic DNA. Both in vivo and in vitro experiments confirm that RBCs can internalize extracellular DNA and reflect tumor burden. Oxford Nanopore sequencing of rbcDNA reveals that short rbcDNA fragments are homologous to extracellular cell‐free DNA (cfDNA). We identify apoptotic bodies (apoBDs) as key mediators of extracellular DNA uptake by RBCs, triggering RBC deformation, Heinz body formation, oxidative stress, and vesiculation. Tumor apoBD‐treated RBCs are rapidly cleared in vivo via a partly macrophage‐dependent effect,…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
FIGURE 1
FIGURE 2
FIGURE 3
FIGURE 4
FIGURE 5
FIGURE 6| High‐depth targeted sequencing (n=19) | Measuring DNA concentration (n=71) | |
|---|---|---|
|
| ||
|
| 11 (57.9%) | 18 (25.4%) |
|
| 8 (42.1%) | 53 (74.6%) |
|
| 61 (55.5–69) | 66 (60–70) |
|
| ||
|
| 18 (94.7%) | 34 (48.6%) |
|
| 0 (0%) | 17 (24.3%) |
|
| 0 (0%) | 7 (10%) |
|
| 0 (0%) | 9 (12.9%) |
|
| 1 (5.3%) | 0 (0%) |
|
| 0 (0%) | 1 (1.4%) |
|
| 0 (0%) | 2 (2.9%) |
|
| ||
|
| 0 (0%) | 0 (0%) |
|
| 2 (10.5%) | 14 (20.3%) |
|
| 7 (36.8%) | 19 (27.5%) |
|
| 3 (15.8%) | 15 (21.7%) |
|
| 6 (31.6%) | 20 (29%) |
|
| 1 (5.3%) | 1 (1.4%) |
|
| ||
|
| 4 (21.1%) | 16 (23.2%) |
|
| 3 (15.8%) | 6 (8.7%) |
|
| 3 (15.8%) | 22 (31.9%) |
|
| 8 (42.1%) | 23 (33.3%) |
|
| 1 (5.3%) | 2 (2.9%) |
|
| ||
|
| 6 (31.6%) | 34 (49.3%) |
|
| 13 (68.4%) | 35 (50.7%) |
|
| ||
|
| 1 (5.6%) | 12 (17.6%) |
|
| 1 (5.6%) | 7 (10.3%) |
|
| 3 (16.7%) | 14 (20.6%) |
|
| 13 (72.2%) | 35 (51.5%) |
|
| ||
|
| 3 (18.8%) | 10 (19.6%) |
|
| 8 (50%) | 35 (68.6%) |
|
| 5 (31.3%) | 6 (11.8%) |
|
| ||
|
| 12 (63.2%) | 59 (85.5%) |
|
| 15 (78.9%) | 33 (47.8%) |
|
| 13 (68.4%) | 16 (23.2%) |
|
| 9 (47.4%) | 45 (65.2%) |
|
| 4 (21.1%) | 21 (30.4%) |
- —National Natural Science Foundation of China10.13039/501100001809
- —Major Program of the National Natural Science Foundation of China
- —National Science and Technology Major Project10.13039/501100018537
- —CAMS Innovation Fund for Medical Sciences
- —Fundamental Research Funds for the Central Universities10.13039/501100012226
- —National Major Science and Technology Projects of China10.13039/501100013076
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsErythrocyte Function and Pathophysiology · Cancer Genomics and Diagnostics · Cancer Cells and Metastasis
Introduction
1
Human red blood cells (RBCs) are the most abundant cell type in peripheral circulation, accounting for 40%–50% of the blood volume [1]. During maturation, RBCs expel the nucleus and organelles, such as the Golgi apparatus, endoplasmic reticulum, mitochondria, and ribosomes, to form biconcave discocytes [2]. The unique shape is characterized by a high membrane area‐to‐volume ratio, which facilitates the deformation of mature RBCs to pass through narrow capillaries and enables them to perform their classical function as carriers for oxygen and carbon dioxide [3]. Nevertheless, recent studies have increasingly reported that mature RBCs were involved in various biological processes beyond gas transport, including immune regulation, phagocytosis enhancement, and cardiovascular homeostasis [4, 5, 6]. Pereira‐Veiga et al. demonstrated that RBCs can enhance tumor cell invasiveness through intercellular communication [7]. The current findings emphasize the under‐–explained role of RBCs, the most common cells in the blood, which are like the elephant in the room.
Despite the absence of nuclei, DNA has recently been identified in human mature RBCs. Lam et al. reported that in infectious diseases, RBCs can act as “immune sentinels” by taking up pathogen‐derived CpG DNA via surface TLR9 proteins, thereby activating macrophage‐mediated innate immunity [8]. Liang et al. discovered that mature RBCs contain long DNA fragments and can acquire tumor‐specific mutations through direct contact with cancer cells [9], a phenomenon later confirmed by Thompson et al. [10]. Therefore, these findings suggested that at least part of rbcDNA originated from the uptake of exogenous material. However, the biological source of rbcDNA has not been elucidated, and acquisition through cell‐cell contact alone appears inefficient. Given that tumor cells shed circulating tumor DNA (ctDNA) into plasma as part of cell‐free DNA (cfDNA), it remains unclear whether mature RBCs can take up cfDNA and its biological consequences.
cfDNA, a widely used liquid‐biopsy biomarker, comprises DNA isolated from plasma and mainly derives from apoptosis, necrosis, or active secretion of malignant and non‐malignant cells [11, 12]. However, the low abundance, short half‐life, and high degree of fragmentation present challenges for cfDNA applications [13]. Due to the protective effect of cell membranes, rbcDNA detection holds promise as a novel liquid‑biopsy marker. Sun et al. identified region‑specific differences in rbcDNA reads between colorectal cancer patients and healthy controls, enabling early diagnosis [14]. However, the clinical implications of rbcDNA remain insufficiently characterized. Direct comparisons between rbcDNA and cfDNA, as well as the potential of RBCs as an efficient tumor DNA reservoir, have not been validated.
In this study, we found that rbcDNA contained short fragments distinct from genomic DNA. Using third‐generation long‐read sequencing, we traced the biological origins of human rbcDNA. Short fragments were primarily originated from immune cells and shared homology with cfDNA, while long fragments were enriched for sequences from erythroid progenitor. RBCs can internalize extracellular DNA, as demonstrated in both in vivo and in vitro models. This uptake was mediated by apoptotic bodies (apoBDs), which induced oxidative stress, accelerated in vivo clearance of RBCs, and suppressed local splenic immune responses. Finally, we investigated the clinical potential of rbcDNA and conducted high‐depth targeted sequencing of rbcDNA with matched cfDNA for head‐to‐head comparisons.
Results
2
Mature RBCs Contain Short DNA Fragments, Distinct From Genomic DNA
2.1
To determine whether human mature RBCs carry DNA, we isolated RBCs from peripheral blood samples of our internal lung cancer cohort using density‐gradient centrifugation. Flow cytometry confirmed >99% purity of CD235a+ CD45– RBCs (Figure S1). Using Hoechst 33342, a living cell‐permeable dye that binds double‐stranded DNA (dsDNA), we detected DNA uniformly distributed within mature RBCs, lacking the heterochromatin pattern of nucleated cells (Figure 1A), consistent with the enrichment of rbcDNA in transcriptionally active regions [14]. The DNA mean fluorescence intensity (MFI) in RBCs measured just 0.057% of that in nucleated cells (lung cancer PC9 and A549 cell lines; Figure 1B).
*Mature RBCs contain unique short DNA fragments. (A) Confocal images showed Hoechst 33342 (blue) and CD235a (yellow) in mature RBCs from lung cancer patients and PC9 lung cancer cell line. The DNA in RBCs is uniformly distributed within the cells, which contrasts with the DNA pattern in the nuclei of PC9 cells. Scale bars 20 µm. (B) Flow cytometry analysis of Hoechst 33342 fluorescence intensity in RBCs, PC9, and A549 cells. The DNA contents of nucleated cells PC9 and A549 are conserved. (C) PAGE with silver stain of rbcDNA from lung cancer patients (n = 5) and healthy donors (n = 3). (D) AFM images of rbcDNA from cancer patients. Sky blue highlights potential short DNA fragments. (E) Schematic diagram of multi‐omics analysis of human rbcDNA using ONT and high‐depth targeted sequencing. Created with BioRender.com. (F) Distribution of DNA fragment sizes (<1, 1–5, 5–8, and >8 kb) in rbcDNA (n = 8) and WBC DNA (n=1). Values for rbcDNA represent the combined results of 8 samples. (G) Representative density plots of fragment distribution for rbcDNA and WBC DNA. To prevent overly long fragments (exceeding 100 kb) from obscuring visualization, all DNA fragments longer than 20,000 bp were normalized to 20,000 bp. H) Coefficient of variation of rbcDNA fragment length in cancer patients (n = 6) and healthy donor (n = 2). t‐test. (I) Contributions of 6 end‐motif profiles in different ONT samples (n = 9). End‐motif profiles are identified by NMF algorithm. (J) Relative frequencies of 4‐mer motifs (left) and A/T/C/G bases (top right) in the end‐motif profile I. On the bottom right is the sequence logo plot for profile I. Profile I exhibits a preference for A bases at the 5′ end. PAGE, polyacrylamide gel electrophoresis. AFM, atomic force microscope. CV, coefficient of variation. p‐values annotated as: # ≤ 0.1, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ***p ≤ 0.0001, or not significant (ns).
The fragment distributions of rbcDNA remain unclear. We analyzed rbcDNA from cancer patients and healthy donors using native polyacrylamide gel electrophoresis (PAGE) followed by silver staining. The results showed that rbcDNA was predominantly composed of long fragments (5–15 kb), although the DNA smear below 1 kb was detectable. In several patients, rbcDNA displayed a ladder‐like pattern (bands at <100, 100–200, 300–400, and >500 bp), a distribution absent in healthy individuals (Figure 1C). Furthermore, atomic force microscopy (AFM) confirmed the presence of both long and short (<1 kb) rbcDNA fragments in lung cancer samples (Figure 1D).
To precisely measure rbcDNA fragment lengths in an unbiased, high‐throughput approach, we extracted rbcDNA from 8 samples (6 lung cancer patients and 2 healthy donors). Peripheral blood mononuclear cell (PBMC, subsets of white blood cells, WBCs) DNA served as a genomic DNA control. Full‐length sequencing was performed using Oxford Nanopore Technology (ONT) (Figure 1E). The rbcDNA presented a distinct fragmentation pattern compared to genomic DNA. Most rbcDNA were medium‐sized (1–8 kb, 51.6%) or short (<1 kb, 31.4%), with a median length of 2708.5 bp (interquartile range, IQR: 1867–3811; Figure 1F). Conversely, WBC DNA was mainly long (>8 kb, 47.8%), with few short fragments (11.2%) and a median length of 7491 bp. Representative fragment size distributions confirmed a preference for short fragments in rbcDNA (Figure 1G). Moreover, the coefficient of variation (CV) of rbcDNA fragment lengths was significantly higher in lung cancer patients than in healthy rbcDNA (Figure 1H). Since DNA fragmentation is mediated by nuclease, we next analyzed 5′‐end nucleotide sequences to elucidate nuclease cleavage preferences of rbcDNA. Here, 6 motif profiles were unsupervisedly identified using the non‐negative matrix factorization (NMF) algorithm based on the frequencies of all 256 possible 5′‐end 4‐mer motifs in rbcDNA (Figure 1I; Figure S2). The frequency of profile I was higher in patients CR2, CR6, CR3, and CR5. This profile was characterized by an A at the first position and resembled the cleavage pattern of DNA fragmentation factor subunit beta (DFFB) [15].
RBCs are Capable of Internalizing Extracellular DNA
2.2
We hypothesized that the short rbcDNA fragments did not originate from residual genomic DNA, but rather from the uptake of exogenous DNA. A cell‐derived xenograft (CDX) model was generated by subcutaneous implantation of human PC9 tumor cells into NCG mice (Figure 2A). The rbcDNA from 3 mice with separate tumor volumes (NG1, NG2, and NG4) was subjected to whole‐genome sequencing (WGS; Figure 2B; Figure S3A). The rbcDNA aligned to the human reference genome represented exogenous, tumor‐derived DNA. The proportion of human DNA in mouse RBCs closely tracked tumor volume (Figure 2C; Figure S3B), reaching 0.42% in NG1 (the largest tumor with rib infiltration) and 0.18% in NG2 (the smallest tumor). Moreover, ichorCNA analysis indicated that rbcDNA from NG1 had the highest tumor purity (Figure S3C). These observations confirmed that circulating RBCs can capture tumor‐shed DNA and reflect tumor burden.
Uptake of tumor DNA by RBCs in vivo or in vitro and the origins of rbcDNA. (A) Schematic diagram of establishment of the mouse CDX model using human PC9 cell lines. Created with BioRender.com. Human DNA in mouse RBCs is considered to be of tumor origin. (B) Tumor size of CDX model after 1 month of growth. NG1 tumor has invaded the adjacent rib. The tumor volumes of NG1, NG4, and NG2 decreased in sequence. (C) WGS coverage of human‐derived DNA in mouse RBCs across the genome. (D) The DNA MFI (Hoechst) of RBCs after direct co‐culture with PC9 (left) and A549 (right) cells for 12 h via flow cytometry (n = 3). Paired t‐test. (E) The DNA MFI (Hoechst) of RBCs after indirect co‐culture with A549 and PC9 cells for 24 h via flow cytometry (n = 4). One‐way ANOVA. (F) Representative dot plots showing the fluorescence (Hoechst channel) in RBCs after direct (top) or indirect (bottom) co‐culture with PC9 cells for 24 h in DNA transfer assay. Autofluorescence refers to the Hoechst channel signal detected from RBCs in the co‐culture system without dye treatment. Set the threshold for positive RBCs as greater than autofluorescence. (G) The DNA MFI of RBCs cultured for 24 h with supernatants from PC9 (top) and A549 (bottom) cell lines via flow cytometry (n = 5). Paired t‐test. (H) Sanger sequencing of EGFR amplicons from RBCs co‐cultured with H1975 cells harboring the EGFR L858R mutation for 24 h (n = 3). The red box represents the EGFR L858R gene locus. Red T denotes the wild‐type, black G is the mutant type. (I) The DNA MFI (Hoechst) of untreated RBCs from healthy donors and lung cancer patients via flow cytometry (n = 3). t‐test. (J) The DNA uptake of RBCs from healthy donors and lung cancer patients after direct (top) or indirect (bottom) co‐culture with PC9 cells for different time‐points via flow cytometry (n = 3). The normalized value for DNA uptake in RBCs is based on the difference between the Hoechst MFI of the treated group at each time point and that of untreated RBCs. Two‐way ANOVA. (K) Qubit assay of rbcDNA and paired cfDNA concentrations from lung cancer patients (n = 21). Paired t‐test. (L) Cellular origin contributions of rbcDNA (n = 16) and WBC DNA (n = 1). Values represent the proportions of DNA from different cell types as identified by the deconvolution algorithm. The reference methylation profile is sourced from GSE87196. Samples with “_” in their names belong to the batch 2 of ONT sequencing. (M) The proportions of rbcDNA fragments derived from CD8+ T cells (left) and MEPs (right) in healthy donors (n = 5) and lung cancer patients (n = 11). t‐test. Cell origins are determined by deconvoluting methylation profiles from ONT sequencing. N) Pearson's correlation of TOO of short (<1 kb; left), medium (1–8 kb; middle), and long (>8 kb; right) rbcDNA fragments with that of cfDNA. The cfDNA origin data were obtained from the public cfDNA database reported by Mattox et al. CDX, cell line‐derived xenograft. WGS, whole genome sequencing. MFI, mean fluorescence intensity. MEP, megakaryocyte‐erythrocyte progenitors. GMP, granulocyte‐macrophage progenitors. ONT, Oxford nanopore technologies. TOO, tissue‐of‐origin. Paired tests are used for samples from the same biological source. For multiple tests, p values are corrected using the Benjamini‐Hochberg method. p values annotated as: # ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, or not significant (ns).
Due to the absence of blood circulation, we used the concept of extracellular DNA rather than cfDNA to represent “free DNA” for in vitro experiments. To assess the ability of RBCs to capture tumor DNA, particularly extracellular DNA, human RBCs were either directly contacted with lung cancer cells (PC9 and A549) for 12 h or indirectly co‐cultured through 0.4 µm pore‐size inserts for 24 h. The duration of incubation was constrained by the accumulation of cellular debris (Figure S4). Direct or indirect exposure to tumor cells substantially increased Hoechst DNA fluorescence in RBCs (Figure 2D,E). Next, we designed a DNA transfer assay to examine whether labeled DNA was transferred from PC9 and A549 cells to RBCs (Figure S5). Specifically, tumor cells were Hoechst‐stained, thoroughly washed to remove unbound dye, and subsequently cultured with RBCs directly (12 h) or indirectly (24 h). Unstained co‐cultures served as autofluorescence controls, and untreated RBCs incubated with stained tumor cell supernatant (collected at 24 h) for 2 h to exclude false positive signals from Hoechst leakage. The MFI of rbcDNA was significantly elevated after co‐culture, suggesting the transfer of tumor‐derived DNA into RBCs (Figure S6A–C). Representative samples showed 8.07% and 7.68% of RBCs were Hoechst‐positive after direct contact with PC9 and A549, respectively, versus 2.38% and 4.12% after indirect co‐culture (Figure 2F).
To overcome the pore‐size limitation of indirect co‐culture inserts, supernatants from PC9 and A549 cells were collected by centrifugation at 500 g for 10 min twice. The 24 h incubation with tumor supernatants led to a significant up‐regulation in Hoechst MFI of RBCs (Figure 2G), which was further verified by DNA agarose electrophoresis and PAGE (Figure S7A). Using EGFR as a representative gene, quantitative real‐time PCR (qRT‐PCR) revealed that the relative abundance of EGFR DNA in rbcDNA was significantly higher compared with controls following direct co‐culture (12 h) or exposure to tumor‐cell supernatants (24 h; Figure S7B). Next, RBCs from 3 EGFR wild‐type (WT) lung cancer patients were directly contacted with H1975 lung cancer cells carrying the EGFR L858R (T>G) mutation (Figure S8A–C) or cultured with their supernatants. RBCs were then isolated by density‐gradient centrifugation. Sanger sequencing identified that the RBCs acquired the EGFR L858R mutation from H1975 cells (Figure 2H). These findings substantiated the capacity of RBCs to take up extracellular DNA.
Finally, RBCs from cancer patients displayed a modestly higher baseline DNA MFI than healthy controls (p = 0.071; Figure 2I), while the increase in rbcDNA after exposure to PC9 cells or their supernatants was similar between groups (Figure 2J).
Origin of rbcDNA and its Homology With cfDNA
2.3
We hypothesized that circulating RBCs in humans already harbored internalized cfDNA. To elucidate the origin of rbcDNA, we profiled ONT sequencing to resolve genome‐wide methylation landscapes. Two batches of rbcDNA samples were extracted: batch 1 included 6 lung cancer patients and 2 healthy individuals; batch 2 comprised 5 patients and 3 healthy controls, with the more stringent RBC purification steps (see Methods). One patient provided paired WBC DNA. The average rbcDNA concentration measured by Qubit in 21 lung cancer patients was 283 ng/mL, 16 times higher than the paired cfDNA (Figure 2K). Therefore, the total rbcDNA obtained from 3–4 mL of whole blood is sufficient for reliable ONT sequencing. For tissue‐of‐origin (TOO) inference, we applied a deconvolution algorithm using the cell‐type‐specific methylation reference matrix from the GSE87196 dataset [16], which included 8 cell types. The TOO profiles were highly correlated between batches but distinct from WBC DNA (Figure S9A), indicating conserved rbcDNA origin and negligible leukocyte contamination. The main contributors to rbcDNA were granulocyte‐monocyte progenitors (GMPs, 28.0%) and megakaryocyte‐erythroid progenitors (MEPs, 23.1%), while epithelial cells and fibroblasts contributed only 1.11% and 0.80%, respectively (n = 16; Figure 2L). This suggested that at least 20% of rbcDNA originated from residual nuclear components retained during erythroid precursor differentiation. WBC DNA was mainly derived from lymphocytes (CD4+ T cells: 32.7%; CD8+ T cells: 21.8%; NK cells: 12.5%; B cells: 9.7%), confirming the validity of the deconvolution algorithm. Importantly, the CD8+ T cell contributor was significantly more abundant in rbcDNA from healthy donors than cancer patients (9.8% vs. 4.0%, p = 0.020, n = 16; Figure 2M), whereas the MEP‐derived fraction was lower (18.6% vs. 25.2%, p = 0.039, n = 16).
To achieve deconvolution at a higher cellular resolution, we employed the methylation atlas reported by Loyfer et al. [17] as the reference matrix, which covers 40 tissue and cell types. The results showed that granulocytes were the largest contributors of rbcDNA (39.5%), followed by T cells (8.5%), monocyte–macrophages (8.2%), and erythroid progenitor cells (3.4%). Among epithelial cells, alveolar cells contributed the most (1.6%), with notable contributions also from hepatocytes (1.6%) and pancreatic acinar cells (1.4%). Although the results varied depending on the reference matrix, both analyses demonstrated that rbcDNA primarily originated from granulocytes/GMPs and erythroid progenitors/MEPs.
Given the presence of short DNA fragments in mature RBCs, we conducted multi‐omics integration of methylation and fragmentation features to clarify the TOO of size‐stratified rbcDNA. Due to degradation in batch 2 rbcDNA, we performed analysis only on the 8 rbcDNA samples from batch 1. In short rbcDNA fragments (<1 kb), immune cells accounted for 36.4% of the contributions (top 5 constituents: granulocytes, 16.69%; monocyte‐macrophage lines, 6.52%; T cells, 5.68%; erythroid progenitors, 4.2%; NK cells, 3.95%). This proportion significantly decreased to 20.5% (p < 0.0001) in medium fragments (1 kb‐8 kb) and further declined to 16.6% (p < 0.0001) in long fragments (>8 kb; Figure S9B). Furthermore, when comparing erythroid progenitors, cellular contribution in short DNA fragments (4.2%) was significantly lower than in medium (4.8%, p = 0.045) and long fragments (4.9%, p = 0.015; Figure S9C).
Since TOO characteristics of short rbcDNA fragments were similar to the typical TOO of cfDNA reported by Mattox et al. [18] (granulocytes, 35.87%; megakaryocytes, 17.7%; monocyte‐macrophage lines, 14.79%; hepatocytes, 7.78%; endothelial cells, 3.96%), we next investigated the correlations between the TOO of rbcDNA and that of cfDNA. Remarkably, the TOO of short rbcDNA fragments showed high similarity (r > 0.6) to the TOO profiles of cfDNA, while the medium (1–8 kb) and long (>8 kb) fragments showed no significant resemblance to cfDNA (Figure 2N). The TOO patterns of WBC DNA exhibited marked divergence from both rbcDNA and cfDNA. Furthermore, the TOO of medium and long DNA fragments was conserved across different samples, while short fragments showed sample‐specific patterns (mean coefficient of variation, CV: short fragments 114.3%; medium fragments 69.6%, p < 0.0001; long fragments 62.1%, p < 0.0001; Figure S9D). Overall, these findings indicated that short rbcDNA fragments likely shared a similar biological origin with cfDNA, while longer fragments were more derived from erythroid precursors.
ApoBDs Serve as the Carriers for Extracellular DNA Captured by RBCs
2.4
To investigate the uptake selectivity of RBCs for extracellular DNA, we synthesized FAM‐labeled single‐stranded DNA (ssDNA) and dsDNA fragments of 50, 200, and 1000 bp, and incubated them with RBCs for 24 h. Flow cytometry indicated dose‐dependent absorption of 50 bp ssDNA and dsDNA by RBCs (100 ng–10 µg; Figure 3A). However, uptake of 200 and 1000 bp DNA fragments was not obvious (Figure S10A). Given that circulating extracellular vesicles (EVs) sequester DNA as a component of cfDNA, lipid nanoparticles (LNPs) were formulated using SM‐102, DSPC, cholesterol, and DMG‐PEG 2000, and loaded with the 598 bp EGFR dsDNA fragment amplified by PCR to mimic DNA‐loaded EVs. The LNPs ranged in diameter from the nanoscale to ∼1 µm (Figure S10B). When exposed to equivalent amounts of DNA for 24 h, RBCs exhibited significantly higher Hoechst MFI following LNP DNA treatment compared with naked EGFR dsDNA (Figure 3B). Propidium iodide (PI) cannot penetrate the membranes of viable cells. Flow cytometry of PI staining showed stronger signals in RBCs treated with naked DNA than with LNP DNA for 24 h (Figure 3C). This suggested that LNP DNA was internalized into RBCs more efficiently, rather than adhering to the cell membrane.
ApoBDs mediated absorption of tumor DNA by RBCs. (A) The FAM fluorescence signals of RBCs treated with different concentrations (100 ng, 1 µg, 2 µg, and 10 µg) of FAM modified 50 bp ssDNA and dsDNA for 24 h via flow cytometry (n = 8; left). Two‐way ANOVA. Confocal images displayed FAM modified short dsDNA in RBCs (right). Scale bars 5 µm. (B) The DNA MFI (Hoechst) of RBCs treated with different concentrations (100 ng, 1 µg, 2 µg, and 10 µg) of dsDNA and LNP‐DNA for 24 h via flow cytometry (n = 8). Two‐way ANOVA. C) The PI MFI of RBCs treated with 10 µg dsDNA and LNP‐DNA for 24 h via flow cytometry (n = 5). t‐test. (D) TEM images of apoptotic cells (left) and extracted apoBDs (right) from PC9 and A549 cell lines, after 48 h of 1 µM docetaxel treatment. Scale bars, left, 1 µm; right, 2 µm. (E) Confocal images of A549‐derived apoBDs stained for Hoechst (blue) and CFSE (green). A549 cells are labeled with CFSE, and apoBDs are collected 24 h after docetaxel treatment. CFSE positivity indicates tumor origin. Scale bars 10 µm. (F) The DNA MFI (Hoechst) of RBCs co‐cultured with apoBDs or apoBD‐free tumor supernatant for 24 h via flow cytometry (n = 3; left). The DNA MFI of RBCs co‐cultured with apoBDs, with or without DNase treatment, for 24 h via flow cytometry (n = 4; right). One‐way ANOVA. The apoBDs in the right panel undergo fixation and permeabilization with methanol at −20°C. (G) The DNA MFI (Hoechst) of RBCs in direct or indirect co‐culture with PC9 (left) and A549 (right) cells for 12 h, and with or without docetaxel treatment via flow cytometry (n = 5). One‐way ANOVA. (H) The DNA MFI (Hoechst) of RBCs in direct or indirect co‐culture with PC9 (top) and A549 (bottom) cells for 12 h, and with or without GW4869 treatment via flow cytometry (n = 4). One‐way ANOVA. (I) Naive‐PAGE images of DNA from plasma‐derived apoBDs of lung cancer patients (left) and apoBDs released by A549 and PC9 cells (right). Apoptosis in A549 and PC9 cells was induced by 1 µM docetaxel for 48 h. (J) TEM images of untreated RBCs and RBCs co‐cultured with tumor apoBDs for 24 h. The red circle highlights a low electron density region containing the potential vesicular‐like structures, accompanied by blurred cell membranes. (K) Confocal images of RBCs co‐cultured with A549‐derived apoBDs stained for DiD (red). The FITC channel (green) displayed characteristic autofluorescence. DiD exclusively labeled the apoBDs, and DiD positivity in RBCs confirms transfer of apoBD components to RBCs. Only the large red dot in the center is the true DiD signal. The other small red dots are the autofluorescence of the Heinz bodies. Scale bars 10 µm. LNP, lipid nanoparticles. PI, propidium iodide. dsDNA, double‐stranded DNA. ssDNA, single‐stranded DNA. TEM, transmission electron microscopy. MFI, mean fluorescence intensity. CFSE, carboxyfluorescein succinimidyl Ester. PAGE, polyacrylamide gel electrophoresis. DiD, 1,1′‐dioctadecyl‐3,3,3′,3′‐tetramethylindodicarbocyanine perchlorate. For multiple tests, P‐values are corrected using the Benjamini‐Hochberg method. p values annotated as: # ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, or not significant (ns).
EVs, ranging from apoBDs to microvesicles and exosomes, differ in size, with larger EVs carrying more DNA [19]. To determine the uptake of apoBD‐DNA by RBCs, apoBDs were isolated from tumor cell supernatants by differential centrifugation (Figure S11A). Flow cytometry, performed according to published protocols [20], confirmed that the purity of the collected apoBDs exceeded 80% (Figure S11B,C). Transmission electron microscopy (TEM) demonstrated that PC9 and A549 cells underwent apoptosis and released apoBDs after 48 h of 1 µM docetaxel treatment (Figure 3D). PC9 cells were labeled with CFSE (carboxyfluorescein succinimidyl ester) before apoBD isolation, and confocal microscopy showed abundant DNA within tumor‐derived apoBDs (Figure 3E). ApoBDs from PC9 and A549 cells, as well as the supernatant depleted of apoBDs (containing membrane‐free DNA and smaller EVs such as exosomes), were co‐incubated with RBCs for 24 h. RBCs exposed to apoBDs manifested higher Hoechst MFI than supernatant‐treated or untreated cells (Figure 3F). Next, apoBDs were fixed and permeabilized with methanol at −20°C, followed by treatment with DNase I (10 U 100 µL^−1^). Compared with apoBDs treated by methanol alone, co‐incubation of RBCs with DNase I‐treated apoBDs significantly reduced the Hoechst signal (Figure 3F). Furthermore, docetaxel (1 µM) increased DNA MFI in RBCs after direct (12 h) or indirect (24 h) contact with PC9 or A549 cells (Figure 3G), whereas the exosome inhibitor GW4869 (20 µM) had no consistent effect on rbcDNA fluorescence (Figure 3H).
We reasoned that if cfDNA‐homologous fragments in rbcDNA originated from apoBDs, short DNA fragments should be detectable in the apoBDs. We collected apoBDs from the supernatants of PC9 and A549 cells treated with 1 µM docetaxel (48 h), as well as from plasma samples of lung cancer patients during the chemotherapy interval. Native‐PAGE revealed that plasma‐derived apoBD DNA mainly consisted of fragments shorter than 200 bp (Figure 3I). The apoBDs from cell lines contained not only short fragments displaying a ladder‐like pattern but also long fragments >5 kb, likely resulting from incomplete chromatin fragmentation during apoptosis. Flow cytometric analysis further showed that the average concentration of apoBDs in the supernatant of PC9 cells was 3.2 × 10⁴ mL^−1^, compared with 2.4 × 10⁴ mL^−1^ for A549 cells (Figure S12A,B). In plasma apoBDs, the mean concentration reached 5.4 × 10⁴ mL^−1^. Moreover, plasma apoBDs showed smaller FSC and SSC than in vitro apoBDs (Figure S12C), possibly reflecting the effects of hemodynamic shear stress, degradation, and phagocytic clearance in vivo.
Further, RBCs treated with apoBDs from PC9 and A549 cells indicated low‐density swollen regions containing vesicular‐like structures, blurred membranes, and disrupted phospholipid bilayers, as observed by TEM (Figure 3J). To trace the transfer of apoBD components into RBCs, we labeled the EV membranes of PC9‐apoBDs with DiD (1,1‐dioctadecyl‐3,3,3,3‐tetramethylindotricarbocyanine iodide) and incubated them with unlabeled RBCs for 24 h. Confocal microscopy demonstrated positive DiD signals in RBCs (Figure 3K), suggesting acquisition of labeled lipids. In addition, treated RBCs developed eccentric protrusions with strong autofluorescence in the FITC (green) channel and weaker signals in the DiD (red) channel. Under light microscopy, eccentric protrusions of RBCs appeared refractile (Figure S13A).
Tumor apoBDs Induce Oxidative Stress of Mature RBCs
2.5
To investigate the potential causes of eccentric protrusion formation and autofluorescence, as well as the biological effects of apoBDs on RBCs, we examined RBCs by scanning electron microscopy (SEM) after 24 h of apoBD treatment (Figure 4A). The apoBDs derived from PC9 and A549 cells induced significantly more morphological alterations in RBCs (PC9: 58.4%; A549: 44.0%) than the NC (16.0%), including surface protrusions, cell size heterogeneity, loss of central pallor, and irregular shape (Figure 4B; Figure S14A). TEM images further found multiple electron‐dense regions of variable size, up to 800 nm in diameter, adjacent to the membranes, accompanied by local membrane discontinuities in irregular RBCs (Figure 4C). The ultrastructural features of these dense granules were consistent with Heinz bodies, which represent aggregates of oxidized and denatured hemoglobin deposited on the inner RBC membrane. DCFH‐DA probe demonstrated abundant reactive oxygen species (ROS) in tumor cell‐derived apoBDs (Figure 4D; Figure S13B), and flow cytometry confirmed stronger ROS signals in RBCs after 24 h of co‐culture with PC9‐ or A549‐derived apoBDs (Figure 4E).
RBCs treated with tumor apoBDs undergo oxidative stress and secrete vesicles. (A) SEM revealed protrusions on RBCs treated with A549‐ (top) or PC9‐derived (bottom) apoBDs. These protrusions represent Heinz bodies. (B) SEM images of RBCs treated with A549‐ or PC9‐derived apoBDs (left). Red arrows indicate irregular deformation, green arrows represent acanthocytes, and blue arrows are centrally flattened or spherical RBCs. Scale bars 10 µm. Statistics of RBC alterations (right). One‐way ANOVA. (C) TEM images of RBCs co‐cultured with A549‐ (right) or PC9‐derived (left) apoBDs for 24 h. The red circle indicates disruption of the RBC membrane boundary. (D) Confocal images of A549‐derived apoBDs stained for ROS probe (DCFH‐DA). Scale bars 10 µm. (E) The ROS (DCFH‐DA) of RBCs treated with tumor apoBDs for 24 h via flow cytometry (n = 5). One‐way ANOVA. (F) Confocal images of RBCs co‐cultured with A549 apoBDs. All channels display autofluorescence from unstained samples. The FITC/PE channel exhibits the strongest autofluorescence. Scale bars 30 µm. (G) The autofluorescences (FITC channel) of RBCs co‐cultured with tumor apoBDs or H2O2 at various concentrations (200 µM, 500 µM, and 1 mM) for 24 h via flow cytometry (n = 4). One‐way ANOVA. (H) Confocal images of RBCs co‐cultured with PC9 cells stained for CD235a (yellow) and Hoechst (blue). The yellow particles are RBC debris. Scale bars 20 µm. (I) Boxplot of the numbers of RBCs after 24 h of co‐culture with apoBDs or supernatant without apoBDs from PC9 cells (left) and A549 cells (right) (n = 3). One‐way ANOVA. (J) Confocal images of RBCs treated with A549‐derived apoBDs, with (right) or without (left) vitamin treatment (80 µM vitamin C and 40 µM vitamin E). The image shows the autofluorescence of FITC/PE channels in unstained RBCs. Scale bars 30 µm. (K) RBC counts based on the confocal images in panel J (n = 9). (L) The phosphatidylserine fluorescence of RBCs co‐cultured with tumor apoBDs or apoBD‐free supernatant for 24 h via flow cytometry (n = 4). One‐way ANOVA. (M) TEM images showed RBCs secreting vesicles after 24 h of co‐culture with PC9 apoBDs. Red boxes denote representative areas. Scale bars 1 µm. (N) Representative images of flow cytometry showing phosphatidylserine fluorescence of RBC‐derived vesicles (CD235a+) and other vesicles (CD235a−) from RBCs co‐cultured with PC9‐apoBDs for 24 h. ROS, reactive oxygen species. TEM, transmission electron microscopy. SEM, scanning electron microscopy. For multiple tests, p values are corrected using the Benjamini‐Hochberg method. p Values annotated as: # ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, or not significant (ns).
Oxidized hemoglobin has been reported to have high autofluorescence [21]. Using confocal microscopy, we found intense FITC/PE autofluorescence in RBCs exposed to apoBDs for 24 h, with most signals localized to the membrane and Heinz body regions (Figure 4F). However, autofluorescence of APC (red) and DAPI (Hoechst) channel signals were weak (Figure S13C). RBCs incubated with tumor supernatants depleted of apoBDs for 24 h showed no detectable autofluorescence (Figure S13D). To further test the link between oxidation and autofluorescence, RBCs were treated with gradient levels of hydrogen peroxide (H_2_O_2_; 0 µM, 200 µM, 500 µM, 1 mM) for 6 h, and concentration‐dependent autofluorescence was observed at 24 h (FITC channel, Figure 4G; other channels, Figure S14B). Compared with H_2_O_2_, apoBDs induced higher autofluorescence in RBCs.
Oxidative stress is known to cause RBC injury [22]. Direct contact of RBCs with PC9 or A549 tumor cells for 12 h led to increased RBC debris and reduced cell counts (Figure 4H; Figure S15A). Co‐culture with tumor supernatants for 24 h also resulted in RBC damage (Figure 4I). Similarly, RBCs underwent hemolysis after 24 h exposure to PC9‐ or A549‐derived apoBDs (Figure S15B). The addition of 80 µM vitamin C and 40 µM vitamin E to the RBC‐apoBDs system did not prevent Heinz body formation but resulted in a higher RBC recovery, suggesting that apoBD‐induced RBC damage was partially mediated by ROS (Figure 4J,K; Figure S15C).
Finally, given the close association between oxidative stress and RBC senescence [23], we investigated the exposure of phosphatidylserine on the RBC surface, a classic “eat‐me” signal of aged RBCs. Flow cytometry showed that 24 h treatment with PC9‐apoBDs significantly increased RBC surface phosphatidylserine compared with supernatants depleted of apoBDs, whereas A549‐apoBDs reduced RBC phosphatidylserine levels (Figure 4L). TEM revealed that apoBDs induced vesicle shedding from RBCs (Figure 4M; Figure S15D). We hypothesized that RBC‐derived vesicles carried phosphatidylserine, leading to decreased phosphatidylserine levels on the parent RBCs after vesicle release. Flow cytometry confirmed the presence of phosphatidylserine on the RBC‐derived vesicles (Figure 4N). Compared with the control and supernatants depleted of apoBDs, RBCs subjected to apoBDs for 24 h produced more vesicles (Figure S16A). In particular, A549‐apoBDs induced significantly higher vesicle generation than PC9‐apoBDs (Figure S16B), consistent with the lowest RBC surface phosphatidylserine observed in the A549‐apoBD‐treated RBCs.
ApoBDs Mediate the Rapid Clearance of RBCs In Vivo
2.6
To investigate the in vivo fate of RBCs treated with tumor apoBDs, we enrolled 15 female Sprague‐Dawley (SD) rats and randomly assigned them into 3 groups (n = 5 per group): NC group, apoBD‐treated group, and H_2_O_2_‐treated group (positive control), as illustrated in Figure 5A. Blood was collected from the jugular vein of rats, then RBCs were isolated, and each group was given the corresponding treatment. For the apoBD‐treated group, apoBDs were extracted from the rat breast cancer SHZ‐88 cell line and incubated with RBCs for 24 h. In the positive control group, RBCs were treated with 1 mM H_2_O_2_ for 6 h, followed by medium replacement until 24 h. After treatment, RBCs were labeled with CFSE and reinfused via the tail vein into the original rats. At 6 h after reinfusion, the rats were sacrificed, and both peripheral blood and spleen samples were obtained. Flow cytometry revealed that the percentage of CFSE+ peripheral blood cells was significantly higher in the NC group than in the H_2_O_2_‐treated group, whereas the apoBD‐treated group showed the lowest signal (<1%; Figure 5B,C), demonstrating that RBCs exposed to apoBDs were rapidly cleared from circulation within 6 h.
Tumor apoBDs enhance in vivo clearance of RBCs and attenuate local immune responses. (A) Schematic diagram of rat experiment. Created with BioRender.com. (B) The CFSE positivity rates of peripheral blood cells in the apoBD‐treated group, the H2O2‐treated group, and the NC group via flow cytometry (n = 5). One‐way ANOVA. (C) Histogram of CFSE positivity rates in panel B. (D) Prussian blue staining of the rat spleens in the apoBD‐treated group, the H2O2‐treated group, and the NC group. Left scale bars 1 mm; right scale bars 100 µm. (E) The positive rate of Prussian blue staining in panel D (n = 5). One‐way ANOVA. (F) Confocal images of rat spleen cells in apoBD‐treated group stained for CD68 (red). CFSE (green) stained the infused treated RBCs. Scale bars 5 µm. (G) Hematoxylin & eosin staining of the rat spleens in the apoBD‐treated group, the H2O2‐treated group, and the NC group. On the right are the cell types identified using QuPath software: green, stromal cell; blue, lymphocyte; yellow, macrophage; red, RBC; dark brown, hemosiderin‐laden macrophages, HLM; brown, erythroid precursors. Scale bars 100 µm. (H) Quantification of panel G (n = 5). One‐way ANOVA. (I) IL6 levels of rat serum in the apoBD‐treated group, the H2O2‐treated group, and the NC group via ELISA (n = 5). One‐way ANOVA. (J) IL6 (left) and IFN‐γ (right) levels of rat spleens via RNA‐seq (n = 5). One‐way ANOVA. (K) GSEA analysis of RNA‐seq data from rat spleen. ELISA, enzyme linked immunosorbent assay. TPM, transcripts per million. GSEA, gene set enrichment analysis. NC, negative control. Multiple testing is corrected using the Benjamini‐Hochberg method. p‐values annotated as: # ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, or not significant (ns).
Since senescent or damaged RBCs are typically cleared in the spleen, we performed Prussian blue staining to assess the formation of ferritin or hemosiderin in splenic macrophages following RBC phagocytosis. Though without statistical significance, the apoBDs group showed the highest Prussian blue‐positive rates, followed by the H_2_O_2_‐treated group and the NC group (Figure 5D,E; Figure S17). Confocal microscopy showed CD68+ cells (monocyte‐macrophage lineages) containing CFSE signals from infused RBCs (Figure 5F). However, in some cases, the CFSE and CD68 signals were spatially separated (Figure S18). These findings implied that RBCs subjected to apoBDs were rapidly cleared in vivo, but this process was not solely dependent on splenic phagocytes.
ApoBD‐Treated RBCs Suppress the Localized Immune Response in the Spleen
2.7
Next, we investigated whether RBCs that treated tumor‐derived apoBDs altered immune responses. Hematoxylin and eosin (H&E) staining was performed on rat spleens from above SD rats, and distinct cell types within the red pulp were classified using QuPath‐based digital pathology (Figure 5G). The apoBD‐treated group manifested the highest proportion of infiltrating RBCs in the spleen (Figure 5H; Figure S19), suggesting more severe splenic congestion. In contrast, the proportion of lymphocytes within the red pulp was significantly higher in the NC group than in either the apoBD‐treated or H_2_O_2_‐treated groups. Subsequently, we measured serum cytokine levels by ELISA. Interferon‐γ (IFN‐γ) was undetectable in all samples (<15.6 pg mL^−1^), and serum IL‐6 levels showed no statistically significant differences among the 3 groups (Figure 5I).
To elucidate the local splenic immune response elicited by RBCs, we performed RNA sequencing on rat spleen tissues. IL6 (Ensembl ID: ENSRNOG00000010278) transcripts showed no significant differences among groups (Figure 5J). IFN‐γ (Ensembl ID: ENSRNOG00000007468) RNAs in the NC group were significantly higher than in the apoBD‐treated group (where no reads were detected). Gene set enrichment analysis (GSEA; Figure S20) demonstrated down‐regulation of inflammation‐ and innate immunity–related pathways in the apoBD group compared with NC, while hypoxia and C‐type lectin receptor signaling pathways were suppressed in the H_2_O_2_‐treated group (Figure 5K). Relative to the H_2_O_2_ group, the apoBD‐treated group showed down‐regulation of antigen presentation, cytokine production, and innate immune pathways, along with up‐regulation of the C‐type lectin receptor pathway. C‐type lectins have been reported to be involved in pathogen recognition, cell adhesion, and endocytosis [24]. These findings indicated that tumor apoBD‐induced RBCs attenuated local immune responses in the spleen.
Head‐To‐Head Comparison of rbcDNA and cfDNA Using High‐Depth Targeted Sequencing
2.8
To evaluate the clinical translational potential of rbcDNA, we first collected RBCs from 19 lung cancer patients (94.7% diagnosed with adenocarcinoma), of whom 17 had paired plasma and 4 had paired WBCs, and extracted DNA for high‐depth targeted sequencing. Blood samples were collected during the interval between chemotherapy cycles. The baseline characteristics of the included patients are presented in Table 1. All patients underwent genetic testing of the primary lung lesions, with the majority of driver mutations being EGFR L858R (4 of 19). The DNA library from plasma showed a main peak at 167 bp and a 10 bp periodicity within the <150 bp range, confirming the reliability of the cfDNA we extracted (Figure S21). Targeted sequencing of rbcDNA detected the EGFR L858R mutation with a sensitivity of 50% and a specificity of 73.3%, while cfDNA detection showed 75% sensitivity and 100% specificity with high variant allele frequencies (VAFs) (Figure 6A; Figure S22A). We next performed droplet digital PCR (ddPCR) on rbcDNA and cfDNA from 1 patient harboring a histologically confirmed EGFR L858R mutation and 1 patient with WT EGFR. In the mutant case, cfDNA contained 202 copies positive for EGFR WT and 48 copies positive for mutation (Figure 6B). However, rbcDNA exhibited exclusively EGFR WT signals (652 copies), with no mutation copies detected. Both cfDNA and rbcDNA from the WT patient contained only EGFR WT copies (Figure S23).
Investigation of the clinical implications of rbcDNA in tumor patients. (A) Sensitivity and specificity of detecting EGFR L858R in rbcDNA (left) and paired cfDNA (right) from the 19 lung cancer patients. (B) ddPCR detection of L858R in cfDNA and rbcDNA from an EGFR L858R mutation patient. (C) Scatter plot of the correlation between rbcDNA VAF and cfDNA VAF in somatic variations. The black line represents the diagonal, while the blue line represents the fitted line for VAF. Pearson's correlation. (D) VAF distribution of rbcDNA (left), cfDNA (middle), and WBC DNA (right) in sample ID57. TP53 mutations were highlighted. (E) VAF distribution plot of cfDNA (red)‐based simulated rbcDNA (other colors) and real rbcDNA (blue). The simulated proportion represents the estimated fraction of rbcDNA originated from cfDNA. (F) The length distribution characteristics of somatic mutation‐carrying fragments in rbcDNA of ID29 sample. (G) Length distribution of human‐ or mouse‐derived DNA in RBCs from xenogeneic model. This model involves incubating human RBCs with mouse LLC lung cancer cell culture supernatant for 24 h, followed by WGS analysis of the rbcDNA. (H) MDS plot of CNV profiles for rbcDNA, cfDNA, and WBC DNA. Each point represents a sample, with spatial proximity reflecting similarity. CD, rbcDNA. PD, plasma DNA. WB, WBC DNA. (I) Boxplot of rbcDNA levels among patients with different clinical characteristics. t‐test. (J) The scatter plot showing the relationship between rbcDNA levels and RDW as well as the natural logarithm of CA125. Pearson's correlation and Spearman's correlation. (K) The waterfall plot illustrating the relationship between rbcDNA levels and efficacy evaluation. The cut‐off was 15 ng µL−1. The rbcDNA concentrations in I–K) correspond to the DNA extracted from 200 µL of RBCs per sample and eluted in 80 µL of TE buffer, quantified by Nanodrop. VAF, variant allele frequency. RDW, red blood cell distribution width. LLC, Lewis lung carcinoma. WGS, whole‐genome sequencing. MDS, multidimensional scaling. CNV, copy number variation. PD, progressive disease; PR, partial response; SD, stable disease. p‐values annotated as: # ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, or not significant (ns).
In terms of mutation profile, we utilized the variants detected in paired WBC DNA as germline references and trained a support vector machine (SVM) model to predict somatic or germline mutations using VAFs (see methods). The VAFs of somatic mutations in rbcDNA were significantly correlated with those in cfDNA (r = 0.85, p < 0.0001; Figure 6C), with the regression line skewed toward cfDNA, reflecting lower VAFs in rbcDNA. We next investigated the VAF distributions of rbcDNA, cfDNA, and WBC DNA. Remarkably, the VAF pattern of rbcDNA showed a conserved feature across different samples (Figure S22B). The mutant‐allele tumor heterogeneity (MATH) score was highest in cfDNA, followed by rbcDNA, and lowest in WBC DNA (Figure S22C). The VAFs of rbcDNA were distributed around 3 peaks at 0–0.2, ∼0.5, and ∼1.0, with almost no variants in between (Figure 6D), whereas cfDNA showed a broad VAF distribution below 0.4. Mutations with VAFs close to 0.5 and 1.0 were considered heterozygous or homozygous germline variants. We hypothesized that the low uptake efficiency of exogenous DNA by RBCs compressed the VAFs of somatic mutations into the 0‐0.2 peak region in rbcDNA, while germline variants remain at ∼0.5 or ∼1.0. Based on this rationale, we modeled the rbcDNA VAF distribution using cfDNA to infer the DNA uptake efficiency (see methods). The optimal fitting parameters were determined by the Kolmogorov‐Smirnov test, yielding cfDNA proportion estimated ranging from 0.04 to 0.27 (median, 0.22; Figure 6E; Figure S24) in rbcDNA. Analysis of high‐depth targeted sequencing data using THetA2 further confirmed that cfDNA showed higher tumor purity than rbcDNA (p = 0.098; Figure S25).
In addition, in samples with paired WBC sequencing (ID23, ID25, ID50, and ID57), we detected an average of over 1,000 SNVs in rbcDNA, significantly higher than those in cfDNA and WBC DNA (Figure S26). However, the average target depth of rbcDNA (11,967×) was greater than that of cfDNA (6,389×). To correct for the potential influence of sequencing depth, the rbcDNA BAM files were randomly downsampled, resulting in an adjusted average target depth (6,390×) comparable to that of cfDNA. Although down‐sampling reduced the number of SNVs in rbcDNA (n = 1,232 vs. n = 988), it still exceeded that in cfDNA (n = 668). We then applied stricter down‐sampling to equalize average depth on the target without duplicated reads (remove PCR duplicates; adjusted depth: rbcDNA 1,553× vs. cfDNA 1,607×), where SNVs in rbcDNA (n = 456) became fewer than in cfDNA. Thus, the greater number of SNVs identified in rbcDNA can be attributed to its increased non‐duplicated sequencing depth compared with cfDNA, which likely resulted from the elevated library complexity due to its high DNA concentration.
To characterize the fragment length preference of tumor‐derived DNA in RBCs, we analyzed the sizes of rbcDNA carrying somatic variants. The results showed that somatic mutations were enriched in 2 length ranges, 90–150 and 340–400 bp, while germline variants lacked this pattern (Figure 6F; Figure S27). Based on TLEN (an approximation of the fragment size), rbcDNA fragments of ≤150, 150–340, 340–400, and 400–1000 bp were separately analyzed for variant calling. Fragments ≤150 bp were enriched for low‐VAF (0–0.2; suggestive of somatic origin) variants, but few variants near VAF ∼0.5 or 1.0 (suggestive of germline origin; Figure S28). On the other hand, the 340–400 and 400–1000 bp fragments contained fewer low‐VAF mutations. Then, we co‐incubated human RBCs with supernatants from mouse LLC lung cancer cells for 24 h and performed WGS on rbcDNA (Figure S29A). The proportion of mouse‐derived (i.e., tumor‐derived) DNA in total rbcDNA ranged from 0.26% to 5.99% (Figure S29B). Tumor‐derived DNA fragments were shorter than those of normal origin (Figure 6G; Figure S29C) and were preferentially enriched near transcription start sites (Figure S29D). Consistently, tumor DNA in mouse RBCs from the NCG CDX model (mentioned in Section 2.2) demonstrated the highest mutations in 50‐150 bp fragments (Figure S29E).
For copy number variation (CNV) profiles, the similarity between cfDNA and rbcDNA was greater than that with WBC DNA (mean r 0.706 vs. 0.588). Multidimensional scaling (MDS) analysis based on cosine distance also demonstrated that the CNV of cfDNA was closer to that of rbcDNA than to WBC DNA (Figure 6H). For instance, on the TP53 gene, both rbcDNA and cfDNA showed no significant CNV, while WBC DNA detected many amplifications and deletions (Figure S30). The abnormality of TP53 in WBC DNA may be due to clonal hematopoiesis [13]. Collectively, rbcDNA and cfDNA had similar genomic characteristics; however, the tumor fraction in rbcDNA was relatively low, suggesting that mutation detection may not represent the optimal clinical application for rbcDNA.
rbcDNA Levels Correlate With Tumor Burden and Treatment Response in Lung Cancer Patients
2.9
We collected 71 peripheral blood specimens from patients with lung diseases, including 58 samples from NSCLC, 9 from small cell lung cancer (SCLC), 2 from infectious lung disease, 1 from pulmonary lymphoma, and 1 from unspecified lung cancer. To ensure sample diversity and reflect real‐world conditions, no restrictions were placed on patient enrollment. Among the samples, 69.0% were from stage III–IV lung cancer, and 13.2% were from treatment‐naive patients. Baseline characteristics of the cohort are summarized in Table 1. DNA was extracted from 200 µL of RBCs from each sample and eluted using 80 µL of TE buffer. The eluted DNA was quantified using Nanodrop. The rbcDNA concentration showed no association with gender or histological type (Figure S31A). However, patients over 65 years showed higher rbcDNA levels (p = 0.066; Figure 6I), which may be explained by impaired macrophage‐mediated clearance of DNA‐laden RBCs in aged individuals [25]. We next examined rbcDNA abundances in samples from early‐ versus advanced‐stage patients. No significant differences were found between T1–2 and T3–4, or between stage I–II and stage III–IV patients. Differently, rbcDNA levels were significantly higher in N2–3 and M1 stages compared with N0–1 and M0 samples, respectively. Moreover, patients with a maximal tumor diameter >30 mm indicated elevated rbcDNA levels, indicating a positive association with tumor burden.
Next, we performed a comprehensive analysis of the relationship between rbcDNA levels and hematological tests. The rbcDNA showed no marked correlation with RBC count, WBC count, mean corpuscular volume, D‐dimer, total bilirubin, glucose, alanine aminotransferase, aspartate aminotransferase, creatinine, or platelet count (Figure S31B). In contrast, rbcDNA levels were significantly associated with RBC distribution width (RDW) and the tumor marker CA125 by both Pearson's and Spearman's correlation analyses (Figure 6J). As an indicator of RBC volume heterogeneity, RDW was associated with unfavorable tumor prognosis [7].
For the most recent efficacy evaluation before blood collection, we found higher disease control rates (DCR) of patients with low rbcDNA levels (93.2% vs. 57.1%, p = 0.028; Figure 6K). We analyzed all 8 samples (from 4 patients) with sequential blood collection and complete tumor marker information. In 1 patient, rbcDNA increased from 3.1 to 13.02 ng µL^−1^, accompanied by a rise in CA125 from 26.2 U mL^−1^ to 36.9 U mL^−1^, whereas other tumor markers remained within the normal range. In the 2 patients whose rbcDNA levels decreased, corresponding reductions in tumor marker levels were also observed. One patient with a small change in rbcDNA also had a small variation in tumor marker levels (Figure S31C).
Discussion
3
Previous researchers did not observe the sequestration of cfDNA by mature RBCs [9]. Here, using a large number of RBCs (up to 10^7^), with co‐culture times of up to 12 h with tumor cells and 24 h with tumor supernatant or apoBDs, we observed the uptake of extracellular DNA by RBCs. The flow cytometry, DNA gel electrophoresis, PCR, and Sanger sequencing were utilized to confirm the above findings. Moreover, our study identified apoBDs as the main sources of tumor DNA absorbed by RBCs. ApoBDs have a wide range of sizes between 100 nm and 5 µm [26], while the pore size of the co‐culture inserts is 400 nm, which prevents some apoBDs from passing through. Therefore, we not only performed indirect co‐culture but also used tumor supernatants to treat RBCs. Overall, this study depicted a new scenario in which RBCs took up DNA released by tumors in the peripheral circulation, without the need for direct contact with tumor cells.
The fragment size distribution of rbcDNA identified in this study differed from previous reports. Using agarose gel electrophoresis and AFM, Liang et al. [9] found that rbcDNA mainly consists of long (>10 kb) and medium‐length fragments, while Sun et al. [14] identified that ∼84.1% of rbcDNA fragments were shorter than 1 kb. Using ONT sequencing, we reported that rbcDNA was dominated by short fragments (<1 kb; 31.4%) and medium‐length fragments (1 kb‐8 kb; 51.6%). The discrepancies in fragment sizes likely reflected methodological variations and technical bias. Using native‐PAGE and silver staining, which was more sensitive to short fragments, we detected bands <1 kb and ladder‐like patterns in rbcDNA. Moreover, our AFM imaging revealed both medium‐long (>1000 nm) and short (<100 nm) fragments. However, AFM was constrained by subjective interpretation and the DNA‐mica preparation process, which can generate cationic crystal artifacts indistinguishable from short DNA molecules. By contrast, ONT sequencing provides unbiased, high‐throughput, single‐molecule measurement of DNA fragment lengths, serving as a practical “silver standard” for rbcDNA analysis.
Previous studies have shown that RBCs can acquire DNA from lung cancer cell lines through direct contact in vitro [9]. Consistently, we also found that RBCs captured DNA from lung cancer cells via a contact‐mediated process (Figure 2D,F). However, flow cytometry and confocal imaging revealed that such contact caused RBC injury and fragmentation. Pereira‐Veiga et al. reported that tumor cells contacted with RBCs became more invasive [7]. Although the effects on RBCs were not explicitly investigated, the microscopy images in that study also suggested RBC fragmentation. Furthermore, we observed that tumor‐derived apoBDs were enriched in ROS. Exposure of RBCs to apoBDs increased intracellular ROS, promoted Heinz body formation, induced membrane deformation, and triggered rapid RBC clearance in vivo. These findings indicated a constraint on rbcDNA accumulation: RBCs that take up large amounts of tumor DNA may undergo destruction or rapid clearance, resulting in low tumor‐derived DNA content in circulating RBCs.
In the study by Liang et al. [9], amplicon‐based next‐generation sequencing (NGS) of rbcDNA achieved 100% sensitivity and 86.7% specificity for detecting the EGFR L858R mutation (n = 26). In our cohort, high‐depth targeted sequencing yielded a sensitivity of 50% and specificity of 73.3% (n = 19), while cfDNA achieved 75% and 100%, respectively. Differences in patient baseline characteristics likely accounted for the discrepancy. Liang et al. primarily enrolled early‐stage (I) and treatment‐naïve lung cancer patients, whereas our study focused on stage III–IV cases (69%), most of whom had received systemic therapy (chemotherapy, radiotherapy, targeted therapy, or immunotherapy; Table 1). Such treatments may enhance the release of tumor apoBDs and accelerate the clearance of RBCs enriched in apoBD‐DNA. Future clinical studies are needed to define the utility of rbcDNA for mutation detection and longitudinal tumor monitoring across diverse patient populations.
This study inscribed a fundamental portrait of rbcDNA as a distinct DNA type from both cfDNA and WBC DNA. The rbcDNA was highly heterogeneous, with multiple biological origins and fragment length sources. Through methylation deconvolution, we found that short fragments in rbcDNA were homologous to cfDNA. As a popular liquid biopsy biomarker, cfDNA has the advantage of detecting tumor driver gene mutations through high‐depth targeted sequencing [13]. However, when performing the same sequencing on rbcDNA, its detection sensitivity and specificity were inferior to those of cfDNA. The ddPCR analysis showed that the absolute copy number of the EGFR L858R mutation in rbcDNA was lower than that in cfDNA. This reduced performance likely reflects the higher proportion of non‐tumor DNA in RBCs, which dilutes the malignant DNA signal. Statistical modeling of targeted sequencing data estimated that cfDNA‐derived sequences accounted for approximately 22% (4%–27%) of total rbcDNA. Given that tumor DNA typically constitutes 0.1% to >10% of cfDNA, the corresponding fraction in rbcDNA was estimated at ∼0.004% to >2.7%. Consistently, in our CDX mouse models, tumor‐derived DNA represented only 0.18–0.42% of total rbcDNA, indicating that RBC uptake of tumor DNA was inefficient. As shown in Section 2.2, fewer than 5% of RBCs acquired tumor DNA after 24 h of indirect co‐culture. Consequently, rbcDNA currently isn't a substitute for cfDNA in driver mutation detection. On the other hand, since our data indicated that short rbcDNA fragments harbored the higher tumor DNA fraction, enriching short rbcDNA, using gel electrophoresis or magnetic bead‐based size selection, may enhance the efficiency of tumor DNA detection.
Methylation deconvolution revealed that erythroid progenitor contributions were highest in long and medium rbcDNA, suggesting a greater origin from nuclear remnants. Since necrotic cells generate long DNA fragments bound to cell surfaces [11, 27], we hypothesize that tumor necrosis‐produced long DNA fragments adsorb onto RBC surfaces (i.e., cell surface‐bound DNA, csbDNA), thereby serving as an additional source of long rbcDNA fragments. Supporting this hypothesis, previous studies have reported the detection of RARβ2 methylation only in long fragments of csbDNA from cancer patients [28]. Given that DNA remnants from RBC maturation are theoretically intracellular, specifically extracting RBC‐csbDNA rather than total DNA might reduce non‐tumor DNA contamination. Furthermore, Tamkovich et al., using the Agilent 2100 Bioanalyzer, found that most breast cancer cfDNA was composed of long fragments [28]. Therefore, long rbcDNA fragments could potentially contain tumor‐associated signals.
EVs can be categorized into 3 main types, ranging in size from large to small: apoBDs, microvesicles, and exosomes [29]. ApoBDs contain remnants of apoptotic cells, including DNA, RNA, organelles, and cytoplasmic contents. Phosphatidylserine on apoBDs acts as an “eat‐me” signal, recognized by macrophages, leading to phagocytosis [30]. We found that after co‐culture with apoBDs, the surface phosphatidylserine of RBCs did not upregulate; instead, they released phosphatidylserine‐containing microvesicles. These microvesicles may be taken up by other cells, potentially mediating unknown biological effects. Moreover, apoBDs contained a rich cargo, and their role in facilitating intercellular communication was gaining increasing attention. Dong et al. discovered that tendon cell‐derived apoBDs can promote the proliferation and migration of tendon and stromal cells [31]. It was also found that apoBDs delivered DNA cargo [32]. In 1999, Holmgren et al. observed that apoBDs from EBV‐carrying lymphoma cells could be phagocytosed by surrounding fibroblasts, and these fibroblasts were able to express EBV‐encoded genes in vitro [33]. In this study, we reported that RBCs took up tumor‐derived ApoBDs and developed oxidative stress. However, the mechanisms of apoBDs in the uptake by RBCs were not fully clarified. Whether other cargo, aside from nucleic acids, within apoBDs can also enter RBCs needs further investigation.
Unlike RBCs that trigger immune responses by absorbing pathogen DNA through surface TLR9 [8], we observed in rat experiments that apoBD‐treated RBCs did not elicit systemic immune reactions but instead suppressed local immune responses in the spleen. Beyond pathogen DNA, tumor cell death can release damage‐associated molecular patterns (DAMPs), including DNA. Schaubaecher et al. discovered that extracellular DNA, widely present in the tumor microenvironment [34]. These DAMPs recruited platelets to the tumor tissues via endosomal toll‐like receptors, and exerted pro‐tumorigenic effects through inhibitory immune checkpoint molecules on the platelet surface. Besides the spleen, whether RBCs can interact with DAMPs locally in tumor microenvironment to modulate immune responses remained to be investigated.
Cancer‐related anemia (CRA) is a common comorbidity in cancer patients [35]. The traditional view suggested that the mechanism of CRA was the defect in iron handling associated with low‐grade chronic inflammation [36]. CRA is an important clinical indicator and is correlated with cancer patient prognosis [37]. We found that after co‐culture with apoBDs, RBCs underwent deformation, destruction, and hemolysis, and were cleared in vivo. Since radiotherapy and chemotherapy led to tumor cell apoptosis, we speculated that the release of apoBDs during treatment may contribute to anemia. This could represent a novel mechanism by which CRA was directly induced by treatment, aside from the impact on hematopoietic stem cells. However, the extent, temporal dynamics, and clinical significance of apoBDs‐induced CRA remain unclear.
Under pathological conditions, mature RBCs can carry mitochondrial DNA (mtDNA). Caielli et al. reported that in patients with systemic lupus erythematosus, dysregulation of the ubiquitin‐proteasome system led to an accumulation of mitochondria‐retaining RBCs, which activated macrophage cGAS‐type I interferon signaling [38]. In infectious diseases, RBCs can bind extracellular mtDNA directly through surface TLR9 [8], facilitating mtDNA clearance and mitigating lung injury [39]. In the field of cancer liquid biopsy, mtDNA has emerged as a promising biomarker [13]. Circulating mtDNA exhibited a higher mutation rate than nuclear DNA and displayed tumor specificity; however, its clinical relevance was controversial. Moreover, the impact of differences between tumor‐derived and pathogen‐derived mtDNA on RBC uptake efficiency are not well understood. Taken together, further studies are needed to determine whether RBCs could selectively enrich tumor mtDNA and act as reservoirs of tumor nucleic acids.
This study had several limitations. The exploration of the relationship between rbcDNA and clinical features was retrospective and limited by confounding factors, necessitating prospective studies to confirm the clinical significance. Due to the limited sample size in clinical rbcDNA samples, some results should be interpreted with caution. The specific molecular mechanism by which RBCs internalize tumor apoBDs remained unknown. Whether the clearance of tumor DNA by RBCs affected the tumor microenvironment and the efficacy of immunotherapy requires further investigation. Additionally, different RBC purification and DNA extraction methods may introduce biases or cause potential contamination in rbcDNA. Therefore, further studies are needed to benchmark various rbcDNA extraction strategies and establish best practices.
Conclusion
4
RBCs contained short DNA fragments and possessed the ability to absorb extracellular free DNA. Short rbcDNA fragments in peripheral blood shared homology with cfDNA. The DNA uptake was mediated by apoBDs. ApoBDs influenced the in vivo and in vitro fate of RBCs by inducing oxidative stress. Finally, the clinical applications of rbcDNA were investigated. Although it underperformed cfDNA for detecting driver‐gene mutations, rbcDNA abundance correlated significantly with tumor burden and treatment response. Our findings provided a comprehensive reference for understanding rbcDNA and opened novel insights for liquid biopsy.
Experimental Section
5
Experimental Animals, Cell Lines, and Human Samples
5.1
All animal studies were approved by the Experimental Animal Welfare Ethics Committee of Zhongnan Hospital of Wuhan University and Institutional Animal Care and Use Committee of Wuhan University Center for Animal Experiment (No. ZN2023238 & IACUC‐20241018). Male NCG (NOD/ShiLtJGpt‐Prkdc ^em26Cd52^ *Il2rg^em26Cd22^/*Gpt) mice (5‐6 weeks) were purchased from the Vital River Laboratories Animal Technology (Beijing, China). Female 8–10 week SD rats were purchased from Hubei BIONT Biological Technology Co., Ltd (Wuhan, China). Mice and rats were housed in individually ventilated cages in a specific pathogen‐free (SPF) environment at room temperature and relative humidity of 50%–60%. Lung cancer cell lines, including human NSCLC cell line PC9 (RRID: CVCL_B260), H1975 (RRID: CVCL_1511), A549 (RRID: CVCL_0023), and mouse LLC cells (RRID: CVCL_4358), were purchased from the Type Culture Center of the Chinese Academy of Sciences (Shanghai, China). SHZ‐88 was a DMBA‐induced breast cancer cell line derived from SD rats, established in Nanjing, China in 1991, and currently has no RRID. SHZ‐88 cell lines were provided by Procell Life Science & Technology Co., Ltd. (Wuhan, China; cat #CL‐0209). Despite the lack of the RRID, SHZ‐88 was used to avoid species differences between tumor apoBDs and SD rat RBCs, which did not affect the experimental conclusions. All cell lines were authenticated by short tandem repeat (STR) profiling before use and were not contaminated. A549, H1975, PC9, and SHZ‐88 cell lines were cultured in RPMI‐1640 medium, and LLC cell lines were cultured in DMEM medium containing 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. The cells were grown in a standard incubator at 37°C with 95% humidity and 5% CO_2_.
Blood samples and clinical data were collected from the internal lung cancer cohort at Zhongnan Hospital of Wuhan University. Genetic testing information for theprimary lesions was collected from previous clinical records. Blood samples from the healthy donors were collected from the author and his colleagues. This study received approval from the Medical Ethics Committee of Zhongnan Hospital of Wuhan University (No. 2022032K), and followed the guidelines of the Declaration of Helsinki. All individuals provided informed consent. To ensure diversity and reflect real‐world conditions, no filtering criteria were applied to the patient samples. Efficacy evaluation for lung cancer patients was based on the RECIST 1.1 criteria.
For CDX model, we expanded the PC9 cell line in vitro, digested and centrifuged to obtain the cells, washed it 3 times with PBS, resuspended it in an appropriate amount of serum‐free RPMI‐1640 medium. Tumor cells were implanted subcutaneously in the armpit at a density of 3 × 10^6^ cells 100 µL^−1^ per mouse. For the rat model, first, 500 µL of whole blood was drawn from the jugular vein, and rat RBCs were separated using density gradient centrifugation. In the experimental group, the isolated RBCs were co‐cultured with apoBDs extracted from SHZ‐88 cells for 24 h. Then, the RBCs were again isolated using density gradient centrifugation, labeled with CFSE, and finally, 200 µL of RBCs were injected into the original rat via the tail vein. In the NC group, there was no co‐culture step. In the H_2_O_2_‐treated group, apoBDs was replaced by H_2_O_2_ treatment. Specifically, treat RBCs with 1 mM H_2_O_2_ for 4 h, then replace with fresh medium and culture until 24 h.
Flow Cytometry of RBCs
5.2
Take 10^6^ extracted RBCs, incubate with 1 mL Hoechst 33342 (abbreviation ‘Hoechst’, 10 µg mL^−1^) at 37°C for 15 min, centrifuge at 400 g for 5 min and wash twice with 1 mL 3% BSA/PBS. Label the cells using CD235a antibody (BD Pharmingen PE Mouse Anti‐Human CD235a) and CD45 antibody (BD Pharmingen FITC Mouse Anti‐Human CD45), incubating at 4°C for 1 h. Wash 3 times with 3% BSA/PBS, ensuring all procedures were performed in the dark. Finally, re‐suspend the cells in 1 mL PBS. Flow cytometry was performed using CytoFLEX (Beckman, USA). To correct for autofluorescence, each sample was split into 2 aliquots (stained/unstained) before acquisition, with both groups derived from the same biological source and processed. Samples were run in the same experiment using identical flow cytometer settings. For single‐channel signals, the corrected MFI was defined as the MFI of the stained group minus the MFI of the unstained control. All multicolor experiments were compensated using single‐stained controls and a standard compensation matrix.
ONT Sequencing
5.3
Using the ONT ligation sequencing kit (Oxford Nanopore Technologies), long‐read DNA libraries were prepared according to the manufacturer's instructions. The DNA libraries were loaded onto the R9.4.1 flow cell and sequenced for 48–72 h on the PromethION sequencer (Oxford Nanopore Technologies). Sequencing support was provided by Wuhan Benagen Technology Co., Ltd. We used the Dorado software with default parameters for basecalling, adapter trimming, alignment, and read error correction on the POD5 files. The average depth of ONT sequencing was 7.89× in this study.
Fragmentomics Analysis
5.4
For NGS samples, extract the absolute value of the TLEN column from the deduplicated BAM file as the fragment length. Use the geom_histogram and geom_density functions from the ggplot2 package in R to visualize the fragment length distribution. Use the R ChIPseeker package [40] annotate the fragments of rbcDNA. Extract the first 4 bases from the 5′‐end of each DNA fragment as the 4‐mer motif from ONT data. There were 256 combinations of the bases A, T, C, and G. Calculate the frequency of each of the 256 motifs for each sample, normalizing them to sum to 100. Use the R package NMF [41] on the motif frequencies to identify 6 profiles. Count the occurrences of A, T, C, and G bases at positions in each profile.
Extraction of RBCs, Plasma, and PBMCs Using Density Gradient Centrifugation
5.5
Collect 3–4 mL of whole blood from patients or healthy donors using EDTA‐containing collection tubes. Plasma, RBCs, and PBMCs were separated within 2 h after blood sampling by density gradient centrifugation, beginning with a 10 min centrifugation for whole blood at 1600 g. The supernatant was extracted and centrifuged again at 1600 g for 10 min (4°C), resulting in the plasma sample. Next, 5 mL of PBS was added to the blood cell precipitation, mixed, and then added to 3 mL of human lymphocyte separation medium. Next, centrifuge at 400 g for 40 min at 18°C, with both acceleration and deceleration set to 1. Use a pasteur pipette to aspirate the middle buffy coat, add 5 mL of PBS, centrifuge at 400 g for 10 min to retrieve the PBMC precipitate, and repeat twice. Wash the RBC pellet in the nuclease‐free tube 3 times with 5 mL of PBS, centrifuging at 400 g for 5 min after each wash. For the ONT samples in batch 2 described in Section 2.3, we employed a more stringent rbcDNA extraction protocol. Specifically, after density‐gradient centrifugation, 5 mL of PBS was added to the RBC pellet, followed by filtration twice through a 10 µm cell strainer to remove potential leukocyte contamination [14]. The RBCs were then washed 3 times with 50 mL PBS at 400 g for 5 min each. Nevertheless, fragmentomic characterization revealed degradation in the batch 2 samples; therefore, they were used exclusively for methylation analysis.
rbcDNA, cfDNA, and WBC DNA extraction
5.6
For rbcDNA extraction, collect 1 mL of RBCs, add 2 mL of RBC lysis buffer (Biosharp), mix well. Wait 5 min until the solution was clear. Centrifuge at 400 g for 5 mins, collect the supernatant (potential contaminating leukocytes were not lysed and remained in the pellet). Add 4 µL 100 mg mL^−1^ RNase and incubate for 5 min to remove RNA. The remaining steps followed the instructions provided in the TIANamp Blood/Cell/Tissue Genomic DNA Extraction Kit (TIANGEN). For cfDNA extraction, collect 1‐1.5 mL of plasma and extract the DNA using the MagMAX Cell‐Free DNA Isolation Kit (Applied Biosystems). Extraction steps were carried out according to the kit's manual. Additionally, the PBMCs collected from 3–4 mL of whole blood were used to extract WBC DNA through the QIAamp DNA Mini Kit (50) (QIAGEN). The DNA concentration was measured using NanoPhotometer N60 (Implen, Germany) and Qubit 4 (Thermo Fisher Scientific, USA).
The Confocal Microscope and Fluorescence Microscope for Imaging RBCs
5.7
Coat the coverslips with 0.1 mg mL^−1^ poly‐L‐lysine for 2 h. Wash the coverslips with sterile ultrapure water. After allowing them to air dry, expose the coverslips to ultraviolet light for 1 h. Add 10^6^ RBCs to the coverslips. The next day, incubate the cells with pre‐chilled methanol at −20°C for 10 min. Wash 3 times with PBS (4°C). Then, stain the cells according to the procedure. Perform confocal imaging and immunofluorescence imaging using the Leica STELLARIS 5 SR (Leica, Germany) and Leica DMi8 (Leica, Germany), respectively. Some experiments did not go through fixed steps but used 35 mm confocal dishes to image living cells.
Isolation and Purity Identification of Tumor‐Derived apoBDs
5.8
When the cell confluence was 70%–80%, replace the normal medium with a medium containing 1 µM docetaxel (MedChemExpress). After 48 h, the supernatant was collected and centrifuged twice at 600 g for 5 min to fully precipitate cells and cell debris. The supernatant was collected, leaving at least 1 mL in the 15 mL tube to avoid disturbing the cell pellet. Centrifuge at 2400 g for 45 min, and the resulting pellet was the apoBDs.
The apoBD pellets obtained from differential centrifugation was re‐suspended in 500 µL of 1× Annexin V binding buffer (Bestbio), followed by the addition of 5 µL of Annexin V‐PE staining solution (Bestbio) and 2 drops of TO‐PRO3 Ready Flow (Thermo Fisher Scientific). The mixture was incubated with apoBDs at room temperature for 15 min. All operations were performed in the dark. We applied the flow cytometry strategy reported by Phan et al. [20] to identify apoBD purity.
Co‐Culture Experiment
5.9
In this study, 5 co‐culture systems were established: direct co‐culture, indirect co‐culture, supernatant co‐culture, apoBD co‐culture, and apoBD‐free supernatant co‐culture. Unless otherwise stated, we used 10^7^ RBCs in all experiments. For direct co‐culture, add the RBCs directly into the tumor cell cultures at 70%–80% confluence, and incubate for 12 h. Indirect co‐culture was performed using a culture insert (24 mm Diameter, pore size 0.4 µm), with RBCs placed in the upper insert and tumor cells cultured in the lower layer, followed by 12 h incubation. For the supernatant co‐culture, medium was collected from tumor cells after 24 h, centrifuged at 500 g for 10 min, and subsequently incubated with RBCs for 24 h. For the apoBD co‐culture, induce apoptosis in tumor cells by docetaxel treatment. After 48 h, collect apoBDs from the supernatant, re‐suspend them in the fresh medium, and incubate them with RBCs for 24 h. For the apoBD‐free supernatant co‐culture, extract the supernatant after removing apoBDs, then co‐culture it with RBCs for 24 h. After co‐culture, use density gradient centrifugation to isolate RBCs. To determine the effect of exosomes on RBCs absorbing tumor DNA, treat tumor cells with 20 µM exosome inhibitor GW4869 (MedChemExpress) for 12 h.
DNA Gel Electrophoresis
5.10
Prepare a 1.5% agarose gel using 1× TAE buffer. Heat it for 1.5 min until fully melted. Once cooled to 60°C, add Gel‐Red stain (10000×, Beyotime) at a ratio of 1 µL per 10 mL. Mix the DNA sample with 6× loading buffer (Biosharp). DNA marker was 1 kb Plus DNA Ladder (TIANGEN). Maintain the voltage at 100 V, and stop the electrophoresis when the bands have migrated to approximately 2 cm from the front of the gel. Image the DNA using the UVP ChemStudio (analytikjena, Germany). For DNA PAGE, we used the TBE PAGE gel preparation kit (Beyotime) to prepare 10% separating gels and 4% stacking gels. DNA samples were mixed with 6× loading buffer (Biosharp). DL 15000 DNA Marker (Vazyme) and 100 bp DNA Ladder (TIANGEN) were used as DNA marker. Electrophoresis was performed at 60 V. After completion, gels were silver‐stained following the instructions of the Rapid PAGE Silver Staining Kit (Solarbio).
ddPCR
5.11
The presence of the EGFR L858R (c.2573T>G) mutation was quantified using the QX200 Droplet Digital PCR system (Bio‐Rad) following the manufacturer's instructions. Briefly, rbcDNA was mixed with the ddPCR Supermix for Probes (No dUTP, Bio‐Rad). Mutation‐specific primers and probes targeted EGFR L858R and WT alleles. The forward primer sequence for EGFR L858R was: 5′‐GCAGCATGTCAAGATCACAGATT‐3′. The reverse primer sequence was: 5′‐CCTCCTTCTGCATGGTATTCTTTCT‐3. The TaqMan probe sequence for the mutation allele was 5′‐AGTTTGGCCCGCCCAA‐3′, labeled with 5′ 6‐FAM and 3′ MGB. The TaqMan probe sequence for the WT allele was 5′‐AGTTTGGCCAGCCCAA‐3′, labeled with 5′ VIC and 3′ MGB.
PCR, FAM‐Modified Probe and Sanger Sequencing
5.12
The EGFR gene primers were designed in this study and synthesized by Beijing Tsingke Biotech Co., Ltd. Purify the primers using the HPLC method. The forward primer sequence was: 5′‐TTCGCCAGCCATAAGTCCTC‐3′. The reverse primer sequence for EGFR was: 5′‐GTGACCTTTCCCAATTGCGC‐3′. PCR was performed using ChamQ SYBR qPCR Master Mix (Vazyme) according to the manufacturer's instructions, on the MA‐6000 Real‐Time Quantitative Thermal Cycler (Molarray, China). The EGFR amplicon sequence was 598 bp and included the nucleotide at the EGFR L858R mutation site. Sanger sequencing was also performed by Beijing Tsingke Biotech Co., Ltd., which also synthesized random sequences of ssDNA and dsDNA sequences (50 bp, 200 bp, and 1 kb) labeled with FAM fluorescence at the 5′‐end.
CNV and Tumor Purity Identification
5.13
We used the CNVkit software [42] to identify the CNV profiles. For tumor purity estimation of WGS or ONT, we utilized the ichorCNA software [43]. In high‐depth targeted sequencing, we used the THetA2 software to calculate tumor purity [44]. We compared rbcDNA/cfDNA with WBC DNA using Verdict [45] to generate a VCF file containing somatic mutation information. Subsequently, we used “cnvkit.py export theta” to output the files required for THetA2, and finally ran the model using “RunTHetA”.
WBS of Human‐Mouse Hybrid Genomes
5.14
In this study, the 2 human‐mouse genome mixing experiments were conducted. In the CDX mouse model, we extracted the rbcDNA and performed WGS. The sequencing depths for the 3 samples were 32×, 35×, and 36×, respectively. In vitro co‐culture experiment, we also performed WGS on the rbcDNA after co‐culture with the supernatant of LLC cells (sequencing depth of 22×, 28×, and 42×, respectively). Using the cellranger mkref tool (10x Genomics), we constructed a reference genome combining human hg38 and mouse mm39. We used bwa mem to align reads to the hybrid genome and applied picard MarkDuplicates to remove duplicates. High‐quality alignments were selected using the “samtools ‐bF 12” and “samtools ‐q 30”. We used GATK “RealignerTargetCreator” and “IndelRealigner” to realign the BAM files and applied GATK “BaseRecalibrator” and “ApplyBQSR” for base quality score recalibration. We used GATK “HaplotypeCaller” to call mutations across the whole genome.
CFSE Labeling of RBCs
5.15
We used CFSE to label RBCs in both flow cytometry and rat experiments. Prepare a 20 µM CFSE (Beyotime) working solution using dimethyl sulfoxide, and incubate it with RBCs at 37°C for 15 min. Add 5 times the volume of FBS‐containing medium and incubate for 10 min to remove remaining CFSE from the solution. Centrifuge at 400 g for 5 min, discard the supernatant, and re‐suspend the cells in pre‐warmed (37°C) medium. Incubate at 37°C for at least 15 min to ensure full hydrolysis of CFSE.
TEM of Apoptotic Cells and apoBDs
5.16
Both the apoBDs and the apoptotic cells were fixed with 2.5% glutaraldehyde at 4°C for 15 min and were washed 3 times with 0.1 M phosphate buffer. Then fix at room temperature for 2 h with 1% osmium tetroxide. Afterward, wash 3 times with 0.1 M phosphate buffer. The samples were dehydrated using a graded ethanol series. Subsequently, infiltrate with a penetration agent, fill with embedding epoxy resin, and polymerize in a 60°C incubator for 48 h. After sectioning, stain with a 2% uranyl acetate aqueous solution and lead citrate at room temperature for 15 min. The slides were then air‐dried overnight at room temperature before observation under an electron microscope (Tecnai G 20 TWIN, FEI, USA).
High‐Depth Targeted Sequencing
5.17
For rbcDNA and paired cfDNA, we performed high‐depth targeted capture sequencing. A 1021‐gene panel (Beijing GenePlus Technology Co., Ltd, China) was used for 16 patients, and an 88‐gene panel for 3 patients. Due to technical privacy, the BED files for the panels cannot be disclosed. Library preparation, hybrid capture, and DNA sequencing were performed as previously described [46, 47]. The average sequencing depth for rbcDNA was 3,974×, while the average depth for cfDNA was 1,645×.
AFM Imaging
5.18
The preparation of mica containing rbcDNA was performed as described in previous studies [9]. AFM imaging was performed in air using a Dimension ICON system (Bruker, USA).
Mutation Data Analysis
5.19
For targeted capture sequencing, the reference genome used was hs37d5. Adapters were removed using the Cutadapt software [48], and the Fastq files were aligned to the reference genome using bwa‐mem2. BAM files were sorted using samtools sort. Using sentieon software's “Dedup”, “Realigner”, and “ReadWriter” modules [49] to perform PCR duplicate removal, realignment, and BAM file output, respectively. Somatic mutations were identified using Mutect2 software [50]. For the identification of EGFR L858R, due to the low VAF of mutations in rbcDNA, we directly extracted the base at position 55259515 on chromosome 7 (for hg19), with filtering criteria set as QUAL > = 20 and MAPQ > = 30. To visualize and further analyze mutation data, we used the annovarToMaf function from the R maftools package [51] to convert the output of annotate_variation into MAF format. Using the inferHeterogeneity function from the maftools package, we calculated the MATH scores.
Simulation of the Proportion of cfDNA Absorption by RBCs
5.20
Mutations were classified as germline or nongermline based on whether they were detected in paired WBC DNA. The independent variable of the SVM model was the VAF value of mutations detected in cfDNA or rbcDNA, and the dependent variable was whether the mutation was germline (detected in WBC DNA). The R package e1071 [52] was used to train the SVM model with the best.svm function to optimize parameters. A radial kernel was employed. We used equations to simulate the VAF distribution of rbcDNA: VAF_RBC_ = VAF_cfDNA_ × α+ VAF_WBC_ × (1‐α) × β. The value of β (0 or 1), inferred as the probability of a germline mutation, is obtained by the SVM model. The α represents the proportion of rbcDNA derived from cfDNA. When β = 1, *VAF_cfDNA_
- = *VAF_WBC_
- (germline mutation). By iterating through α values from 0.01 to 1 (in intervals of 0.01), a simulated VAF distribution for rbcDNA can be obtained. The Kolmogorov‐Smirnov test was then used to compare the simulated distribution with the actual distribution, allowing the most suitable α value to be inferred.
Methylation Data Analysis and Deconvolution of Cellular Contributions
5.21
In ONT data, we matched fragment length and the CpG methylations through read ID alignment. In this study, we used the methylation data from 51 tissue and cell types reported by Loyfer et al. [17] as the reference matrix. Using positional information, we matched the rbcDNA fragments to the reference matrix [17], and then used the nnls package in R to perform deconvolution to determine the cellular origin of rbcDNA. We compared the cellular origin of rbcDNA with cfDNA inferred by Mattox et al. [18], with 37 overlapping cell types. Additionally, we analyzed the overall CpG site methylation of rbcDNA and mapped it to the methylation sites on the H450K chip. Using the hepidish function from the R EpiDISH package [53], we performed deconvolution of the CpG methylation data. For “ref2.m”, we utilized reference matrix from the GSE87196 dataset [16]. In the GSE87196, we selected B cells, CD4+ T cells, CD8+ T cells, GMPs, MEPs, and NK cells, averaged the methylation data of the same cell type from multiple samples, and constructed the reference matrix by choosing the top 50 sites that showed the largest differences between each cell type and others.
Prussian Blue Staining and H&E Staining of Rat Spleen
5.22
For Prussian blue staining, tissue of appropriate size was dehydrated and embedded. The paraffin block was sectioned, and the slides were baked. Subsequently, after deparaffinization and Prussian blue solution staining, the slides were mounted with neutral balsam. For H&E staining, the slides were sequentially deparaffinized, stained with hematoxylin, stained with eosin, dehydrated, and then mounted with neutral balsam. Scan the whole slides using the PANNORAMIC scanner (3DHISTECH, Hungary). We used “Analyze” and “Classify” modules of QuPath software [54] to analyze the pathological slides of rat spleens. For H&E staining, we selected characteristic red pulp as the region of interest (ROI), with the area containing at least 9,000 cells. By training and applying an object classifier for each sample individually, the cells within the ROI were classified into lymphocytes, macrophages, stromal cells, hemosiderin‐laden macrophages, erythroid precursors, and RBCs.
ELISA Detection of Cytokines
5.23
At the end of the animal experiment, 1 mL of whole blood was drawn from the rat's jugular vein and allowed to sit for 1 h to clot. The sample was then centrifuged at 4,000 rpm for 15 min, and the supernatant was collected as the serum sample. According to the manufacturer's instructions, serum levels of IFN‐γ (the rat IFN‐γ ELISA kit, LABLEAD, Beijing) and IL‐6 (the rat IL‐6 ELISA kit, LABLEAD, Beijing) were measured.
RNA‐Seq Analysis of Rat Spleen
5.24
A library was prepared using 2 µg of rat spleen RNA. Following the manufacturer's recommendations, sequencing libraries were generated using the NEBNext Ultra RNA Library Prep Kit for Illumina (NEB, USA). The libraries were sequenced on the Illumina platform. Sequencing support was provided by Wuhan Benagen Technology Co., Ltd. The reads were mapped to the rat mRatBN7.2 genome using STAR. Gene expression levels for each sample were quantified using RSEM. Differentially expressed genes were identified for read counts by the R DESeq2 package. GSEA were performed using the R packages clusterProfiler [55]. The OrgDb was org.Rn.eg.db v3.19.1.
Statistical Analysis
5.25
Statistical analyses in this study were performed using R (v4.3.2) and GraphPad Prism (v8.3.0). Comparisons between 2 independent samples were performed using t‐tests, while comparisons between multiple groups were conducted using one‐way ANOVA or two‐way ANOVA. Categorical variables were compared using the Chi‐Squared test or Fisher's exact test. Survival analyses were carried out using the R survival package. Data were presented as mean ± SD. In enrichment analysis, the significance threshold for the false discovery rate was set at 0.05. Multiple testing correction was performed using the Benjamini & Hochberg method. A p‐value less than 0.05 was considered statistically significant, and all p‐values were two‐sided.
Author Contributions
B.X., C.X., Z.Y., and J.Z. supervised the research. B.X., C.X., Z.Y., and ZH.Z. designed experimental design. B.X., C.X., and Z.Y. acquired the funds. J.H., Y.X., K.F., X.G., Q.J., S.B., Y.T., and Y.Y. collected clinical data. ZH.Z., Z.Y., J.L., and T.W. analyzed multi‐omics data. ZH.Z., J.L., K.F., X.G., Q.J., Y.Z., S.B., Y.Y., X.W., and R.H. performed the experiments. ZH.Z., Z.Y., J.H., Y.T., L.H., ZQ.Z., and J.J. analyzed clinical data. ZH.Z., Z.Y., and J.H. wrote the original draft. All authors reviewed and approved the manuscript.
Conflicts of Interest
The authors declare no conflict of interest.
Supporting information
Supporting File: advs73490‐sup‐0001‐SuppMat.docx.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Y. Liu , X. Nie , X. Yao , et al., “Developing an Erythrocyte‒MHC‐I Conjugate for Cancer Treatment,” Cell Discovery 10 (2024): 99.39349449 10.1038/s 41421-024-00713-9PMC 11443136 · doi ↗ · pubmed ↗
- 2L. Mesarec , W. Góźdź , A. Iglič , V. Kralj‐Iglič , E. G. Virga , and S. Kralj , “Normal Red Blood Cells' Shape Stabilized By Membrane's In‐Plane Ordering,” Scientific Reports 9 (2019): 19742.31875042 10.1038/s 41598-019-56128-0PMC 6930264 · doi ↗ · pubmed ↗
- 3M. Risinger and T. A. Kalfa , “Red Cell Membrane Disorders: Structure Meets Function,” Blood 136 (2020): 1250–1261.32702754 10.1182/blood.2019000946 PMC 7483429 · doi ↗ · pubmed ↗
- 4J. Dobkin and N. S. Mangalmurti , “Immunomodulatory Roles Of Red Blood Cells,” Current Opinion in Hematology 29 (2022): 306–309.35916547 10.1097/MOH.0000000000000734 PMC 9547827 · doi ↗ · pubmed ↗
- 5F. Leo , T. Suvorava , S. K. Heuser , et al., “Red Blood Cell and Endothelial e NOS Independently Regulate Circulating Nitric Oxide Metabolites and Blood Pressure,” Circulation 144 (2021): 870–889.34229449 10.1161/CIRCULATIONAHA.120.049606 PMC 8529898 · doi ↗ · pubmed ↗
- 6C. Xu , J. He , H. Wang , et al., “Single‐Cell Transcriptomic Analysis Identifies An Immune‐Prone Population In Erythroid Precursors During Human Ontogenesis,” Nature Immunology 23 (2022): 1109–1120.35761081 10.1038/s 41590-022-01245-8 · doi ↗ · pubmed ↗
- 7T. Pereira‐Veiga , C. Yáñez‐Gómez , A. Pekkarinen , et al., “Red Blood Cell‐Tumor Cell Interactions Promote Tumor Cell Progression,” Journal of Experimental & Clinical Cancer Research 44 (2025): 134.40301999 10.1186/s 13046-025-03376-w PMC 12042424 · doi ↗ · pubmed ↗
- 8L. K. M. Lam , S. Murphy , D. Kokkinaki , et al., “DNA Binding to TLR 9 Expressed By Red Blood Cells Promotes Innate Immune Activation And Anemia,” Science Translational Medicine 13: abj 1008.10.1126/scitranslmed.abj 1008 PMC 906592434669439 · doi ↗ · pubmed ↗