Platelets as immune sensors: monitoring immune dynamics and diagnosing disease states across multiple disorders

Summary

Background

Beyond their classical roles in haemostasis and thrombosis, platelets have been recognised as active regulators of immune responses. However, whether platelets can function as immune sensors capable of monitoring immune status remains largely unexplored.

Methods

We first analysed platelet transcriptomes from patients at different stages following haematopoietic stem cell transplantation (HSCT) to assess their ability to capture immune reconstitution dynamics. We then employed immune cell–platelet co-incubation experiments to elucidate the mechanistic role of RNA transfer in reshaping platelet immune molecular profiles. Subsequently, platelet transcriptomes were examined in immune-related conditions, including HSCT-associated acute graft-versus-host disease (aGVHD), cytomegalovirus (CMV) DNAemia, and immune-inflammatory diseases such as systemic lupus erythematosus (SLE) and ulcerative colitis (UC), to evaluate their capacity to identify disease-specific immune dysregulation. Finally, machine learning models were developed based on platelet immune gene signatures to assess diagnostic performance.

Findings

Platelet transcriptomes accurately reflected the dynamics of immune reconstitution following HSCT. Mechanistically, immune cells directly reprogrammed platelet immune-related molecular features through RNA transfer. Moreover, platelets sensitively detected immune alterations associated with transplant complications, including aGVHD and CMV DNAemia, and effectively monitored immune activity and inflammatory states in SLE and UC. Machine learning models based on platelet immune gene profiles further improved diagnostic accuracy across disease settings.

Interpretation

This study establishes platelets as precise, readily accessible, and noninvasive immune sensors, extending their functional repertoire from immune regulation to immune surveillance and diagnosis in immune-related diseases.

Funding

National Natural Science Foundation of China; CAMS Innovation Fund for Medical Sciences; Science and Technology Projects of Xizang Autonomous Region, China; National Key Research and Development Program of China; Tianjin Municipal Science and Technology Commission Grant; and Natural Science Foundation of Tianjin.

Research in context.

Evidence before this study

Platelets, the smallest anucleate blood cells, play a central role in haemostasis and thrombosis. Beyond these classical functions, accumulating evidence demonstrates that platelets regulate both innate and adaptive immune responses through multiple mechanisms, thereby broadly contributing to immune homoeostasis and to the initiation and progression of infections, autoimmune diseases, and cancer. Although the immunomodulatory roles of platelets have been extensively documented, their capacity to sense and surveil the immune status of the host remains insufficiently characterised. If such an immune-sensing and monitoring function does exist, its potential clinical relevance for disease diagnosis and management has yet to be defined.

Added value of this study

In this study, by analysing platelet transcriptomic data across different stages following haematopoietic stem cell transplantation (HSCT), we demonstrate that the platelet transcriptome accurately reflects post-HSCT immune reconstitution. In particular, it clearly delineates the transitional trajectory from innate to adaptive immunity, providing compelling evidence that platelets can function as immune sensors for monitoring the host immune status. Moreover, in HSCT-related complications, including acute graft-versus-host disease (aGVHD) and cytomegalovirus (CMV) infection, as well as in complex immune-mediated disorders such as systemic lupus erythematosus (SLE) and ulcerative colitis (UC), platelets effectively capture dynamic changes in immune states. Machine learning models built on platelet immune gene profiles further enhance diagnostic accuracy and enable the identification of potential previously undefined biomarkers. Mechanistically, our data suggest that immune cells can directly remodel the immune-related transcriptomic features of platelets via RNA transfer.

Implications of all the available evidence

Our findings establish platelets as precise, readily accessible, and noninvasive immune sensors, extending their recognised roles from immunomodulation to immune surveillance and substantially deepening our understanding of platelet functions within the immune system. Furthermore, this study highlights the potential of platelets as effective tools for monitoring disease-associated immune dysregulation, opening new avenues for the diagnosis, treatment, and prognostic assessment of a broad spectrum of immune-mediated diseases, as well as for their clinical translation.

Introduction

Platelets, the smallest anucleate blood cells, are best known for their essential role in haemostasis through clot formation.¹ However, their functions extend far beyond coagulation, encompassing immune regulation and lymphatic vascular development.2, 3, 4 Platelets express surface glycoproteins and pattern recognition receptors (PRRs), such as GPIIb/IIIa and Toll-like receptors (TLRs), enabling them to recognise and bind pathogens and exert direct antimicrobial activity. Upon activation, platelets secrete CD40L and various cytokines to recruit leukocytes and guide them to sites of inflammation. In addition, platelets interact with immune cells to modulate both innate and adaptive immune responses.^2,5, 6, 7, 8 For example, platelets engage neutrophils via P-selectin and TLR4 to promote neutrophil extracellular trap (NET) formation; regulate T-cell activity through MHC-I; and influence B-cell–mediated antitumour immunity through metabolic modulation,9, 10, 11, 12, 13, 14 underscoring their broad involvement in immune homoeostasis. By influencing the phenotype and activity of immune cells, platelets contribute to the pathogenesis and the progression of infection, autoimmunity, and tumour,15, 16, 17, 18, 19 functioning as both immune mediators and participants in pathology.

Platelets originate from megakaryocytes, which package RNA into platelets before releasing them into circulation.^20,21 Beyond inheriting megakaryocyte-derived RNA, platelets actively exchange materials with other cells,22, 23, 24, 25, 26 acquiring exogenous RNA from their surroundings.^22,23 Thus, the platelet transcriptome comprises both endogenous and interaction-derived RNA. Indeed, platelets can acquire RNA from endothelial cells (e.g., CXADR, COL18A1, SELE) and monocytes (IL1B, S100A9, TLR2),²² and can also transfer RNA to other cells.27, 28, 29 These properties position the platelet transcriptome as a sensitive indicator of physiological and pathological states. Given these features, platelet transcriptomic alterations are increasingly linked to diverse diseases, including cancer and hematologic disorders. Platelet RNA profiles can distinguish patients with cancer from healthy individuals, identify tumour origins, and predict disease progression.30, 31, 32 Our recent work further shows that platelet transcriptomes can predict therapeutic response and long-term platelet recovery in aplastic anaemia,³³ underscoring their utility as sensitive biomarkers of systemic health. Nevertheless, the active role of platelets in sensing and monitoring immune states remains largely unexplored, despite extensive studies on platelet immune regulation.

Given the close interaction between platelets and immune cells and their capacity for RNA exchange, we hypothesised that platelets can capture immune cell-specific transcripts, enabling their transcriptome to reflect immune activity. Supporting this concept, prior studies have confirmed leucocyte-to-platelet RNA transfer, and our work identified antiviral gene upregulation in platelets from patients with SARS-CoV-2 Omicron,³⁴ suggesting that platelets can sense immune states during infection.

To test this hypothesis, we analysed platelet transcriptomes from patients undergoing haematopoietic stem cell transplantation (HSCT), a process involving a transition from innate to adaptive immune reconstitution. Platelet transcriptomes accurately mirrored this transition and sensitively reflected immune alterations associated with complications such as acute graft-versus-host disease (aGVHD) and cytomegalovirus (CMV) infection. Similarly, platelet transcriptomes captured immune changes in systemic lupus erythematosus (SLE) and ulcerative colitis (UC). Machine learning models based on platelet immune gene profiles further improved diagnostic accuracy and revealed previously undefined biomarkers. Mechanistically, our data reveal that immune cells directly reprogramme platelet immune transcriptomes via RNA transfer. Collectively, these findings establish platelets as precise, accessible, and noninvasive immune sensors capable of monitoring immune dynamics through transcriptomic changes.

Methods

Ethics

The research adhered to Declaration of Helsinki and was approved by the Ethics Committee of the Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and the Tianjin Medical University General Hospital (CIFMS2021012-EC-2 and IRB2022-YX-215-01), with written informed consent obtained from all participants.

Clinical sample collection and inclusion criteria

Peripheral blood was collected from healthy donors (HD) and patients with normal short-term (NS) or long-term (NL) reconstitution, aGVHD, or CMV DNAemia post-HSCT. All participants were of Chinese ethnicity, and data on biological sex were collected via self-report by study participants.

Platelet and neutrophil engraftment typically occurs within 2–4 weeks after transplantation, monocytes usually return to normal levels at approximately 1 month, and lymphocyte reconstitution requires substantially more time—ranging from 6 months to over 2 years.³⁵ In addition, day 60 post-transplant is a key clinical time point for identifying delayed platelet recovery.36, 37, 38 Therefore, we defined days 20–60 post-transplant as the short-term reconstitution window and ≥ 180 days as the long-term reconstitution window. Accordingly, for patients with NS, peripheral blood samples were collected 20–60 days after transplantation, following neutrophil and platelet engraftment. Neutrophil engraftment was defined as an absolute neutrophil count > 0.5/μL for three consecutive days, and platelet engraftment as a platelet count > 20,000/μL for three consecutive days without transfusion support.³⁹ For patients with NL, samples were obtained at least 180 days post-transplant, with platelet counts restored to normal levels (100–300 × 10⁹/L). At the time of sampling, patients with neither NS nor NL exhibited evidence of GVHD or bacterial, fungal, or viral infections.

Furthermore, inclusion criteria for patients with aGVHD required a severity grade of II or higher according to the MAGIC criteria, with no concurrent bacterial, fungal, or viral infections at the time of sampling. Patients with CMV DNAemia were included if their venous blood CMV viral load exceeded 1000 copies/mL, and they were free of bacterial, fungal, and other viral infections. Samples from patients with aGVHD and CMV DNAemia were both collected within one week of diagnosis, with sampling conducted 20–60 days post-transplant. Complete response (CR) of aGVHD was defined as the complete resolution of aGVHD manifestations in all organs. Unstable partial response (unstable PR) was characterised by an improvement in aGVHD symptoms without achieving CR, along with disease progression in affected organs on the day of sample collection. No platelet transfusions were administered within seven days prior to sampling for either healthy donors or patients with HSCT.

To address potential selection bias, we performed a comprehensive statistical comparison of baseline clinical characteristics across all groups. For the HD, NS, and NL groups, no significant differences were observed in baseline variables such as age, sex, disease, ABO compatibility, graft type, or GVHD prophylaxis. As expected, platelet, neutrophil, monocyte, and lymphocyte counts—as well as time after HSCT—differed across groups (Table S1). For the NS versus aGVHD comparison, all baseline characteristics (age, sex, disease, ABO compatibility, donor type, graft type, and GVHD prophylaxis) were comparable except for neutrophil counts (Table S2). For the NS versus CMV DNAemia comparison, baseline characteristics were similar, with the exception of lymphocyte counts (Table S3). Thus, the transcriptomic differences identified in our study primarily reflect group-specific biology rather than selection bias.

Platelet isolation

Peripheral blood samples were collected in 5 mL sodium citrate-coated blue-capped vacutainers and processed to isolate platelets using gradient centrifugation. Whole blood was centrifuged at 1200 rpm for 10 min at room temperature to collect platelet-rich plasma, which was then centrifuged again at 300 g for 5 min to remove leukocytes and erythrocytes. The supernatant was subsequently centrifuged at 3500 rpm for 10 min to obtain platelets.

To ensure platelet purity, isolated platelets were stained with anti-human CD41a (BD Biosciences Cat# 559777, RRID:AB_398671), anti-human CD45 (BD Biosciences Cat# 555483, RRID:AB_395875), anti-human CD235a (BD Biosciences Cat# 559943, RRID:AB_397386), and Hoechst 33342 (Solarbio Cat# C0031), followed by flow cytometry analysis using a BD Canto II flow cytometer. Data were processed using FlowJo software (version 10.0.7). Samples with CD45⁺Hoechst⁺ cell contamination < 0.1% were resuspended in 1 mL TRIzol (Invitrogen Cat# 15596018CN) and stored at −80 °C for further use.

Platelet RNA sequencing

RNA sequencing was performed on total RNA extracted from platelets isolated from patients with HSCT and healthy donors. Purified platelets were lysed in TRIzol for total RNA extraction. RNA integrity was assessed using the Agilent 5400 system, and only samples with an RNA integrity number (RIN) ≥ 6 were used for RNA sequencing. Total RNA was reverse-transcribed and amplified into cDNA using the NEBNext® Ultra™ RNA Library Prep Kit for Illumina® according to the manufacturer's instructions. Final cDNA libraries were evaluated using the Agilent 5400 system, pooled, and sequenced on the Illumina NovaSeq 6000 platform.

Transcriptome data preprocessing

After sequencing, raw reads in FASTQ format were processed using Fastp (version 0.23.1) to remove adaptor sequences, poly-N, and low-quality reads, resulting in clean data. These clean reads were then mapped to the human reference genome (hg38) using STAR (version 2.7.1a). The aligned reads were quantified with featureCounts (version 2.0.1), using Gencode v35 GRCh38 for gene annotation. All downstream analyses were conducted in R (version 4.2.1).

Bioinformatics analysis

DESeq2 (version 1.36.0) package in R was used to construct a DESeqDataSet object, removing transcripts with an average count < 10. Principal component analysis (PCA) was performed using built-in plotPCA function in DESeq2. Differential gene analysis was performed by estimating size factors, dispersion and fitting negative binomial generalised linear model. Genes with a p value < 0.05 and an absolute log₂FoldChange > 1 were defined as differentially expressed genes (DEGs). Count data were normalised using size factors, resulting in a gene expression matrix. Gene Ontology: Biological Process (GO: BP) enrichment analysis of the DEGs was conducted using g:Profiler (https://biit.cs.ut.ee/gprofiler/gost). Gene set scoring was calculated using the formula: $Score=\frac{\sum _{i=1}^n{\log }_2\left(x_i+1\right)}{n}$, where x_i represents the expression value of genes in the gene set, and n is the number of genes in the set. Pearson's correlation was calculated using the rcorr function from the Hmisc (version 4.7–1) package in R. Heatmaps were generated using the pheatmap (version 1.0.12) package in R

Functional gene sets used in this study were downloaded from http://geneontology.org/. Immune cell marker genes were downloaded from CellMarker 2.0 (see Table S4). The gene lists related to human neutrophils, monocytes, T cells and B cells were downloaded from https://www.ncbi.nlm.nih.gov/gene (see Table S5). Single-cell transcriptome data for human neutrophils, monocytes, T cells and B cells at different stages after HSCT were sourced from GSE224714. Human platelet transcriptome data for systemic lupus erythematosus (SLE) were sourced from GSE226147. The platelet transcriptome data of patients with ulcerative colitis (UC) were sourced from PRJNA737596.

Isolation of neutrophils and monocytes

Peripheral blood samples were collected in 5 mL vacuum tubes containing sodium citrate as an anticoagulant. Samples were centrifuged at 1200 rpm for 10 min at room temperature. The upper platelet-rich plasma layer was aspirated, and the remaining fraction was used for neutrophil or monocyte isolation.

Neutrophils were isolated using the Human Peripheral Blood Neutrophil Isolation Solution Kit (Solarbio Cat# P9040). Samples were centrifuged at 800–1000 g for 30–40 min at room temperature, and the neutrophil layer was carefully collected and washed with phosphate-buffered saline (PBS; Beyotime Cat# C0221A). After centrifugation at 250 g for 10 min, red blood cells were lysed with Red Blood Cell Lysis Buffer (Solarbio Cat# R1010) for 5 min, followed by centrifugation at 400 g for 5 min. The resulting neutrophil pellet was resuspended in RPMI 1640 medium (Gibco Cat# C11875500BT) supplemented with 10% heat-inactivated fetal bovine serum (FBS; Gibco Cat# A5256701).

For monocyte isolation, peripheral blood mononuclear cells (PBMCs) were first obtained by density gradient centrifugation using Separation Medium (TBD Cat# LTS1077) at 400 g for 30 min at room temperature. Monocytes were then isolated from PBMCs using the EasySep™ Human CD14 Positive Selection Kit II (STEMCELL Cat# 17858) according to the manufacturer's instructions.

To verify the purity of neutrophils and monocytes, cells were stained with anti-human CD66b (BioLegend Cat# 305104, RRID:AB_314496), anti-human CD11b (BD Biosciences Cat# 555388, RRID:AB_395789), anti-human CD14 (BioLegend Cat# 325618, RRID:AB_830691), anti-human CD3 (BioLegend Cat# 300312, RRID:AB_314048), and anti-human CD19 (BioLegend Cat# 392504, RRID:AB_2728416), followed by flow cytometry. The purity of isolated neutrophils (CD66b⁺) and monocytes (CD14⁺) exceeded 90%.

RNA transfer assay

Neutrophils or monocytes were labelled with SYTO® RNASelect™ Green Fluorescent Stain (Life Technologies Cat# S32703) at a 1:1000 dilution for 30 min according to the manufacturer's instructions, followed by washing with PBS. Labelled cells were co-incubated with platelets at a 1:50 ratio for 3 h. After co-incubation, cells and platelets were separated by centrifugation at 400–500 g for 5 min. For flow cytometry, platelets were stained with anti-human CD41 (BioLegend Cat# 303716, RRID:AB_10897646), anti-human CD45 (BD Biosciences Cat# 555485, RRID:AB_398600), and Hoechst 33342 to evaluate the transfer of neutrophil- or monocyte-derived RNA into platelets. For immunofluorescence analysis, platelets were stained with anti-human CD41 (BioLegend Cat# 303706, RRID:AB_314376) to visualise intracellular RNA transfer.

Platelets were labelled with SYTO® RNASelect™ fluorescent green stain for 30 min and washed with PBS. Neutrophils were then co-incubated with platelets at a 1:50 ratio for 3 h, followed by separation via centrifugation at 400 g for 5 min. To assess the transfer of platelet-derived RNA into neutrophils by flow cytometry, cells were stained with anti-human CD66b (BioLegend Cat# 305116, RRID:AB_2566605) and DAPI (Solarbio Cat# C0060). For immunofluorescence analysis of RNA transfer, cells were stained with anti-human CD41 (BioLegend Cat# 303706, RRID:AB_314376), anti-human CD66b (BioLegend Cat# 305118, RRID:AB_2566607), and Hoechst 33342.

RNA sequencing of platelets co-incubated with neutrophils

Neutrophils and platelets were co-incubated at a 1:50 ratio for 3 h, followed by separation by centrifugation at 400 g for 5 min. Platelets were stained with anti-human CD41 (BioLegend Cat# 303716, RRID:AB_10897646), anti-human CD45 (BD Biosciences Cat# 555485, RRID:AB_398600), and Hoechst 33342 to confirm purity. Samples containing less than 0.1% CD45⁺ Hoechst⁺ cells were considered pure and resuspended in 1 mL TRIzol reagent for RNA sequencing.

RT-qPCR

Platelet RNA was extracted using TRIzol, and cDNA was synthesised with the HiScript IV All-in-One Ultra RT SuperMix for qPCR (Vazyme Cat# R433). Quantitative real-time PCR was then performed using Hieff UNICON® Universal Blue qPCR SYBR Green Master Mix (YEASEN Cat# 11184ES08). The reaction mixture consisted of SYBR Green reagent, nuclease-free water, forward primer, reverse primer, and cDNA at a ratio of 5:2:1:1:1. All qPCR assays were conducted on the QuantStudio 6 Flex System. Primer sequences are as follows:

DEFA3-F: 5′CATGGGACGAAAGCTTGGCT.

DEFA3-R: 5′TGCAGGTTCCATAGCGACGTT.

CAMP-F: 5′GGGGCTCCTTTGACATCAGT.

CAMP-R: 5′GGTAGGGCACACACTAGGAC.

IL-7R-F: 5′AAATATGTGGGGCCCTCGTG.

IL-7R-R: 5′AAGTCATTGGCTCCTTCCCG.

CD79A-F: 5′TCTTCCTCCTCTTCCTGCTGTCTG.

CD79A-R: 5′CGTTGGCGTTGTTGCTGCTATTG.

S100A8-F: 5′AAGGGGAATTTCCATGCCGT.

S100A8-R: 5′CGTCTGCACCCTTTTTCCTG.

S100A9-F: 5′CCTCGGCTTTGACAGAGTG.

S100A9-R: 5′CACCAGCTCTTTGAATTCCCC.

S100A12-F: 5′AGCATCTGGAGGGAATTGTCA.

S100A12-R: 5′GCAATGGCTACCAGGGATATGAA.

MMP9-F: 5′TTTGAGTCCGGTGGACGATG.

MMP9-R: 5′GCTCCTCAAAGACCGAGTCC.

GAPDH-F: 5′TGTTGCCATCAATGACCCCTT.

GAPDH-R: 5′CTCCACGACGTACTCAGCG.

Gene expression was normalised to GAPDH, and relative expression levels were calculated using the 2^–ΔΔCt method.

Machine learning model development

Healthy donors and patients were randomly divided into a training set and a test set at a 7:3 ratio. Using the glmnet package (version 4.1–7), variables were selected from the platelet whole-transcriptome or immune gene profiles in the training set via Least Absolute Shrinkage and Selection Operator (LASSO) for the subsequent development of logistic regression models for disease diagnosis. The performance of the models was evaluated using the test set. The area under the receiver operating characteristic curve (AUROC) was calculated and ROC curves were plotted using the pROC package (version 1.18.2).

Statistics

All statistical analyses were performed using ggplot2 (version 3.3.6) in R and GraphPad Prism (version 8.0.2). Continuous clinical variables with a normal distribution were analysed using unpaired t-test (two groups) or one-way ANOVA (multiple groups), whereas non-normally distributed clinical variables were analysed using Mann–Whitney U test (two groups) or Kruskal–Wallis test (multiple groups). Categorical clinical variables were analysed using Fisher's exact test. For flow cytometry and RT-qPCR data, paired t-test was used for paired comparisons, and Mann–Whitney U test was used for unpaired comparisons. Transcriptomic data were analysed using Mann–Whitney U test. A p value < 0.05 was considered statistically significant.

Role of funders

The funders had no role in study design, data collection, data analyses, interpretation, or writing of report.

Results

Platelet transcriptomic changes reflect immune status shifts following HSCT

To test the hypothesis that platelets could serve as immune sensors, capable of real-time monitoring of the body's immune status, we employed the haematopoietic stem cell transplantation (HSCT) model, a well-defined system characterised by gradual immune reconstitution, transitioning from innate to adaptive immunity. Innate immune cells, such as neutrophils and monocytes, typically recover within two to four weeks, while adaptive immune cells, including T cells and B cells, may take six months to two years or longer to recover.⁴⁰

To investigate whether platelet transcriptomic changes mirror immune cell dynamics during immune reconstitution, we collected peripheral blood platelets from three groups: healthy donors (HD), patients with normal short-term reconstitution (NS, 20–60 days post-HSCT), and patients with normal long-term reconstitution (NL, ≥ 180 days post-HSCT). RNA sequencing of platelet transcriptomes was performed to assess their capacity to monitor immune status (Fig. 1A), with details of sample collection and quality control shown in Figure S1A–C.

Fig. 1.

Dynamic changes in the molecular characteristics of reconstituted platelets afterHSCT. (A) Schematic of the study design for analysing platelet (PLT) immune features at different stages post-HSCT. Peripheral blood platelets were collected from 14 healthy donors (HD), 18 patients with normal short-term reconstitution (NS), and 16 patients with normal long-term reconstitution (NL). RNA sequencing was performed to analyse the changes in platelet immunity-related molecular characteristics. (B) Bar plot showing the proportion of neutrophils (NEUT%), monocytes (MONO%), and lymphocytes (LYMPH%) in the patients with NS and NL at the time of sampling. Data are presented as mean ± SEM; n ≥ 5. (C) Principal component analysis (PCA) of normalised gene expression counts in HD, NS, and NL platelets. (D) Box plots showing the scoring results for the gene sets “innate immune response in mucosa” [GO:0002227] and “adaptive immune response” [GO:0002250] in HD, NS, and NL platelets. (E–H) Expression of neutrophil-, monocyte-, T cell-, and B cell-related features in HD, NS, and NL platelets. (E) Box plots showing the scoring results for neutrophil marker genes and the “neutrophil mediated cytotoxicity” [GO:0070942] gene set in HD, NS, and NL platelets. Heatmap displaying the expression of neutrophil-related genes. (F) Box plots showing the scoring results for monocyte marker genes and the “monocyte activation” [GO:0042117] gene set in HD, NS, and NL platelets. Heatmap displaying the expression of monocyte-related genes. (G) Box plots showing the scoring results for T cell marker genes and the “T cell receptor signaling pathway” [GO:0050852] gene set in HD, NS, and NL platelets. Heatmap displaying the expression of T cell-related genes. (H) Box plots showing the scoring results for B cell marker genes and the “B cell receptor signaling pathway” [GO:0050853] gene set in HD, NS, and NL platelets. Heatmap displaying the expression of B cell-related genes. (I) Bar plots depicting RT-qPCR analysis of representative neutrophil, monocyte, T cell, and B cell genes in platelets. Data are shown as mean ± SEM; n = 5–6. NEUT% was analysed using unpaired t-test. MONO%, LYMPH% and platelet gene expression were analysed using Mann–Whitney U test. ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

We first analysed immune cell recovery based on complete blood count. Consistent with previous studies,⁴⁰ neutrophils and monocytes recovered quickly, whereas lymphocyte recovery was slower (Fig. 1B). Clinical platelet-related indicators were also evaluated. In patients with NS, platelet counts were decreased, whereas mean platelet volume (MPV), platelet distribution width (PDW), and platelet-large cell ratio (P-LCR) were elevated. Notably, these parameters returned to normal levels in patients with NL (Figure S1D).

We next performed principal component analysis (PCA), revealing distinct platelet transcriptome profiles between the groups. NS platelets displayed transcriptomes significantly different from those of NL and HD, while NL platelet transcriptomes closely resembled healthy profiles (Fig. 1C). These findings indicate that platelet transcriptomes dynamically adapt and gradually return to a normal-like state over time. We further examined the expression of gene sets related to innate and adaptive immunity. A clear distinction emerged: platelets from patients with NS exhibited high expression of innate immunity-related genes, whereas those from patients with NL showed elevated expression of adaptive immunity-related genes (Fig. 1D).

Furthermore, marker genes and active gene sets associated with innate immune cells (e.g., neutrophils and monocytes) were upregulated in NS platelets, with specific genes including neutrophil-associated DEFA3, CEACAM8, and MMP8, as well as monocyte-associated CAMP, LY6E, and LYZ (Fig. 1E and F). In contrast, marker genes and active gene sets associated with adaptive immune cells (e.g., T cells and B cells) were more prominently expressed in NL and HD platelets. For example, T cell-associated genes IL7R, CD3E, ITK, and B cell-associated genes CD79A, MS4A1, and PAX5 were significantly upregulated (Fig. 1G and H). To determine whether platelet transcriptomes dynamically reflect post-transplant immune reconstitution, key differentially expressed genes (DEGs) were validated by RT-qPCR. The results clearly showed that genes representing innate immunity, such as neutrophil/monocyte-related DEFA3 and CAMP, were highly expressed in platelets during early post-transplant recovery, whereas genes associated with adaptive immunity, including IL7R and CD79A, were upregulated in long-term reconstituted platelets and in healthy controls (Fig. 1I).

In summary, platelet transcriptomes during post-HSCT reconstitution show a clear shift from innate-dominated profiles in early recovery to adaptive-associated features in long-term platelets, reflecting the innate-to-adaptive immune transition and highlighting platelets as dynamic sensors of immune status.

Neutrophil and monocyte recovery drives the innate immune features in short-term reconstituted platelets

What drives the post-HSCT shift in platelet gene expression from innate to adaptive immunity-related profiles? Given platelets’ interactions with immune cells and the parallel between their transcriptomic changes and immune reconstitution, we hypothesised that the evolving composition and activity of reconstituted immune cells underlie this shift. To test this, we examined the association between platelet neutrophil- and monocyte-related gene expression and the abundance and activity of these innate immune cells at the short-term stage (Fig. 2A). To begin, we compared total white blood cell counts between patients with NS and NL and found no significant differences (Fig. 2B). However, neutrophil and monocyte counts were significantly higher in patients with NS than in patients with NL (Fig. 2C). By overlapping genes highly expressed in NS versus NL with known neutrophil-related genes, we identified neutrophil-related genes upregulated in NS platelets (Fig. 2D; Figure S2A). Correlation analysis revealed that 37.93% of these genes were positively associated with neutrophil counts at the time of sample collection (Fig. 2E). Expression scores of these genes in NS and NL platelets correlated positively with neutrophil counts, particularly classic neutrophil-related genes such as S100A9, S100A8, S100A12 (Fig. 2F; Figure S2B).

Fig. 2.

Neutrophils and monocytes contribute to the innate immune characteristics of short-term reconstituted platelets. (A) Schematic diagram showing the sources of neutrophil- and monocyte-related innate immune features highly expressed in NS platelets. (B) Bar plot displaying the white blood cell counts at the time of sampling in the patients with NS and NL. Data are presented as mean ± SEM; n ≥ 5. (C) Bar plot showing neutrophil and monocyte counts at the time of sampling in the patients with NS and NL. Data are presented as mean ± SEM; n ≥ 5. (D) Venn diagram illustrating the selection of neutrophil-related genes that are highly expressed in NS platelets compared to NL platelets, based on intersection with known neutrophil-related genes. (E) Pearson's correlation analysis identifying the correlation between the expression of neutrophil-related genes highly expressed in NS platelets and neutrophil counts at the time of sampling (left), and the bar plot showing the specific percentage distribution of the genes (right). (F) Scatter plot showing the Pearson's correlation between neutrophil counts and the gene set scores of neutrophil-related genes highly expressed in NS platelets that are positively correlated with the neutrophil counts at the time of sampling. (G) Venn diagram illustrating the selection of monocyte-related genes that are highly expressed in NS platelets compared to NL platelets, based on intersection with known monocyte-related genes. (H) Pearson's correlation analysis identifying the correlation between the expression of monocyte-related genes highly expressed in NS platelets and monocyte counts at the time of sampling (left), and the bar plot showing the specific percentage distribution of the genes (right). (I) Scatter plot showing the Pearson's correlation between monocyte counts and the gene set scores of monocyte-related genes highly expressed in NS platelets that are positively correlated with the monocyte counts at the time of sampling. (J and K) Changes of neutrophil and monocyte features in patients with short-term and long-term reconstitution. (J) Box plots showing the scoring results for the “neutrophil activation” [GO:0042119] and “neutrophil chemotaxis” [GO:0030593] gene sets in neutrophils from patients with NS and NL. (K) Box plots showing the scoring results for the “monocyte chemotaxis” [GO:0002548] and “monocyte differentiation” [GO:0030224] gene sets in monocytes from patients with NS and NL. Transcriptome data for neutrophils and monocytes in patients with short-term and long-term reconstitution were downloaded from GSE224714. Clinical data for white blood cell, neutrophil, and monocyte counts were analysed using unpaired t-test. Neutrophil and monocyte gene expression were analysed using Mann–Whitney U test. ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.

Similarly, we intersected genes highly expressed in NS versus NL with known monocyte-related genes, identifying monocyte-related genes upregulated in NS platelets (Fig. 2G; Figure S2C). Correlation analysis showed that 44.74% of these genes were positively associated with monocyte counts (Fig. 2H). Expression scores of these genes in NS and NL platelets also correlated positively with monocyte counts, with notable genes including CLEC12A, MYL9, CAMP (Fig. 2I; Figure S2D).

We further examined differences in the molecular features of neutrophils and monocytes during different reconstitution phases. Using data from Huo et al.,⁴¹ we observed that short-term reconstituted neutrophils exhibited stronger features of “neutrophil activation” and “neutrophil chemotaxis” compared to long-term reconstituted neutrophils. Similarly, short-term reconstituted monocytes displayed enhanced features of “monocyte chemotaxis” and “monocyte differentiation” compared to long-term reconstituted monocytes (Fig. 2J and K). These findings suggest that the immune features of reconstituted platelets are closely linked to the reconstitution dynamics of neutrophils and monocytes.

In summary, our study indicates that during the early phase after HSCT, the rapid increase and activation of neutrophils and monocytes drive the high expression of neutrophil- and monocyte-related genes in platelets.

Innate immune cells impart innate immune features to platelet transcriptomes via direct RNA transfer

Our previous analyses indicated that the upregulation of innate immune-related genes in platelets from patients with short term post-HSCT closely correlates with the recovery and activity of neutrophils and monocytes. To investigate whether this phenomenon results from the direct transfer of immune-related RNA from these cells to platelets, we fluorescently labelled the RNA in neutrophils or monocytes from patients with NS and co-incubated them with platelets. Uptake of RNA by platelets was then assessed using flow cytometry and immunofluorescence (Fig. 3A). Purity of peripheral blood neutrophils and monocytes exceeded 90%, while leucocyte contamination in platelets was < 0.1% (Figure S3A and B).

Fig. 3.

Neutrophils and monocytes transfer RNA to platelets. (A) Schematic of the experimental design for assessing RNA transfer from neutrophils and monocytes to platelets. (B) Flow cytometry gating strategy for detecting RNA transfer to platelets (left) and bar plot showing the proportion of platelets receiving RNA from neutrophils (right). Data are presented as mean ± SEM; n = 3. (C) Immunofluorescence confirming RNA transfer from neutrophils to platelets. (D) Bar plot showing the proportion of platelets receiving RNA from monocytes. Data are presented as mean ± SEM; n = 3. (E) Immunofluorescence confirming RNA transfer from monocytes to platelets. (F) Schematic illustrating the experimental design for studying neutrophil-to-platelet RNA transfer. (G) PCA showing transcriptomic changes in platelets following co-incubation with neutrophils. (H) Volcano plot depicting differentially expressed genes (DEGs) in platelets after co-incubation with neutrophils. (I) Bar plot showing functional enrichment of upregulated genes in platelets after co-incubation with neutrophils. (J) Bar plots depicting RT-qPCR analysis of representative neutrophil marker genes in platelets, with or without co-incubation with neutrophils. Data are presented as mean ± SEM; n = 4. Statistical analysis was performed using paired t-test. ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.

Flow cytometry revealed a significant increase in the proportion of fluorescent RNA-positive platelets following co-incubation with neutrophils (Fig. 3B). Immunofluorescence imaging further confirmed the presence of neutrophil-derived RNA within platelets (Fig. 3C). Similarly, RNA transfer was observed in platelets after co-incubation with monocytes (Fig. 3D and E). These results indicate that during early immune reconstitution post-HSCT, neutrophils and monocytes can indeed transfer RNA to platelets.

To assess whether immune cells can transfer their characteristic immune-related RNA to platelets, we performed RNA sequencing and RT-qPCR on platelets co-incubated with neutrophils (Fig. 3F, with leucocyte contamination < 0.1%, Figure S3C). Co-incubation induced substantial transcriptomic changes: PCA showed clear separation between co-incubated and control platelets (Fig. 3G). DEGs analysis identified 2829 upregulated and 648 downregulated genes in co-incubated platelets (Fig. 3H). Functional enrichment analysis revealed that upregulated genes in co-incubated platelets were significantly associated with immune-related processes, including “innate immune response”, “activation of immune response”, “response to bacterium”, “neutrophil migration”, and “neutrophil activation” (Fig. 3I). Scoring of neutrophil-specific functional gene sets showed significantly higher scores for “neutrophil activation” and “neutrophil chemotaxis” in co-incubated platelets (Figure S3D), and signature neutrophil genes were markedly upregulated (Figure S3E). RT-qPCR validation of several neutrophil marker genes, including DEFA3, S100A8, S100A9, S100A12, and MMP9, also showed significant upregulation in platelets after co-incubation (Fig. 3J). These results indicate that neutrophils can transfer their characteristic immune-related RNA to platelets, thereby reshaping the immune molecular profile of the platelets.

To assess whether platelets can transfer RNA to immune cells, we used neutrophils as the target and designed an RNA transfer assay (Figure S3F). Peripheral blood neutrophils used in the experiment were > 90% pure, and leucocyte contamination in platelets was < 0.1% (Figure S3G). Flow cytometry and immunofluorescence analyses detected platelet-derived RNA signals within neutrophils (Figures S3H and I), indicating that RNA transfer between platelets and immune cells occurs bidirectionally. Thus, neutrophils and monocytes can transfer RNA to platelets, with neutrophils imprinting innate immune-related RNA, reshaping the platelet transcriptome and highlighting platelets’ role as dynamic immune sensors.

Lymphocyte recovery contributes to the adaptive immune features in long-term reconstituted platelets

After establishing that neutrophils and monocytes can directly reshape the platelet transcriptome towards an innate immune profile, we next investigated whether a similar mechanism contributes to the acquisition of adaptive immune characteristics in platelets during later stages. Given that lymphocyte recovery occurs more gradually than myeloid reconstitution, we hypothesised that the adaptive immune transcriptomic signatures observed in long-term reconstituted platelets reflect the abundance and activity of recovering B and T lymphocytes.

To test this, we examined the relationship between the expression of adaptive immunity-related genes in platelets and the quantity and activity of lymphocytes post-HSCT (Fig. 4A). First, we observed that the lymphocyte count was significantly higher in the patients with NL compared to the patients with NS (Fig. 4B). Correspondingly, the neutrophil-lymphocyte ratio decreased in the patients with NL (Fig. 4C). These clinical indicators align with previous studies showing that lymphocyte recovery is slower than that of neutrophil and monocyte.⁴⁰ Based on these findings, we hypothesised that the immune characteristics of long-term reconstituted platelets might be linked to the staggered recovery of lymphocytes following HSCT.

Fig. 4.

Lymphocyte recovery contributes to the adaptive immune characteristics of long-term reconstituted platelets. (A) Schematic diagram showing the sources of lymphocyte-related adaptive immune features highly expressed in NL platelets. (B) Bar plot displaying lymphocyte counts at the time of sampling in the NS and NL platelets. Data are presented as mean ± SEM; n ≥ 5. (C) Bar plot showing the neutrophil-lymphocyte ratio at the time of sampling in the NS and NL platelets. Data are presented as mean ± SEM; n ≥ 5. (D) Venn diagram illustrating the selection of B cell- and T cell-related genes that are highly expressed in NL platelets compared to NS platelets, based on intersection with known B cell- and T cell-related genes. (E) Pearson's correlation analysis identifying the correlation between the expression of B cell- and T cell-related genes highly expressed in NL platelets and lymphocyte counts at the time of sampling (left), and the bar plot showing the specific percentage distribution of the genes (right). (F) Scatter plot showing the Pearson's correlation between lymphocyte counts and the gene set scores of B cell- and T cell-related genes highly expressed in NL platelets that are positively correlated with the lymphocyte counts at the time of sampling. (G) Scatter plot showing the top 5 Pearson's correlation coefficients for the B cell- and T cell-related genes highly expressed in NL platelets that are positively correlated with lymphocyte counts at the time of sampling. (H) Changes of B cell features in patients with short-term and long-term reconstitution. Box plots showing the scoring results for the “B cell activation” [GO:0042113], “B cell mediated immunity” [GO:0019724], “antigen processing and presentation of exogenous peptide antigen via MHC class II” [GO:0019886], and “peptide antigen assembly with MHC class II protein complex” [GO:0002503] gene sets in B cells from patients with NS and NL. Transcriptome data for B cells in patients with short-term and long-term reconstitution were downloaded from GSE224714. Lymphocyte count was analysed using unpaired t-test. Neutrophil-lymphocyte ratio and B cell gene expression were analysed using Mann–Whitney U test. ns, p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

To test this, we intersected genes highly expressed in NL platelets compared to NS with known B cell- and T cell-related genes, identifying several lymphocyte-related genes that were highly expressed in NL platelets (Fig. 4D; Figure S4A). Correlating the expression of these genes in reconstituted platelets with the lymphocyte counts of the corresponding patients at the time of sampling revealed that 61.80% of the genes were positively correlated with lymphocyte counts (Fig. 4E). The expression scores of these genes in both NS and NL platelets also showed a clear positive correlation with lymphocyte counts (Fig. 4F), with specific lymphocyte-related genes, B3GAT1, RFTN1, LCK, CARD11 and TRBC2, demonstrating particularly strong correlations (Fig. 4G).

We also utilised data from Huo et al.⁴¹ to explore molecular differences in lymphocytes during different phases of reconstitution, which could contribute to the recovery of adaptive immune features in NL platelets. Our analysis revealed that long-term reconstituted B cells exhibited stronger features of “B cell activation”, “B cell mediated immunity”, “antigen processing and presentation of exogenous peptide antigen via MHC class II”, and “peptide antigen assembly with MHC class II protein complex” compared to short-term reconstituted B cells (Fig. 4H). This suggests enhanced intrinsic activity of long-term reconstituted B cells. However, we did not observe similar enhancements in the activity of long-term reconstituted T cells (Figure S4B).

In summary, our results demonstrate that, in the long term after HSCT, the recovery and activation of lymphocytes collectively drive the high expression of adaptive immunity-related genes in platelets. This further supports the role of platelets as immune sensors, capable of accurately monitoring and reflecting the adaptive immune reconstitution process following HSCT.

Distinct transcriptomic features in patients with long-term reconstitution and their potential link to HSCT complications

Our findings suggest that platelets play a pivotal role as immune sensors, actively monitoring the immune reconstitution process during HSCT. In patients with NL, the expression levels of innate and adaptive immune genes in platelets closely resemble those observed in HD. This similarity indicates substantial restoration of immunity-related molecular features in NL platelets. However, an important question remains: do the overall transcriptomic profiles of NL platelets fully align with those of HD platelets? If not, could these differences be associated with long-term complications experienced by patients with HSCT?

To address this, we first examined whether the platelet count normalisation observed in patients with NL (Figure S1D) was accompanied by a complete molecular recovery. A comparison of transcriptomes between NL and HD platelets revealed residual differences (Fig. 5A), albeit significantly smaller than those between HD or NL platelets and NS platelets (Fig. 5B). Given that the reconstitution time range for patients with NL at the time of sampling was approximately 180–1200 days, we were curious whether the DEGs between HD and NL platelets were influenced by the reconstitution time and had a linear relationship with it. Therefore, we first analysed the correlation between the highly expressed genes in HD platelets and the reconstitution time in patients with NL. Notably, 90.91% of these genes showed no correlation with reconstitution time (Fig. 5C), and their expression scores in patients with NL similarly exhibited no significant time dependence (Fig. 5D).

Fig. 5.

Long-term reconstituted platelets exhibit differences in transcriptome compared to healthy donors, independent of reconstitution time. (A) Volcano plot showing DEGs between HD and NL platelets, with genes defined as differentially expressed if p value < 0.05 and absolute log₂FoldChange value > 1. (B) Bar plot displaying the number of DEGs between HD and NS, NL and NS, and HD and NL platelets. (C) Pearson's correlation analysis identifying genes that are downregulated in NL platelets compared to HD, and their correlation with reconstitution time at the time of sampling (left), with a pie chart showing the specific percentage distribution of the genes (right). (D) Scatter plot showing the Pearson's correlation between NL platelet low-expression gene set scores (unrelated to reconstitution time at the time of sampling) and reconstitution time. (E) Pearson's correlation analysis identifying genes that are upregulated in NL platelets compared to HD, and their correlation with reconstitution time at the time of sampling (left), with a pie chart showing the specific percentage distribution of the genes (right). (F) Scatter plot showing the Pearson's correlation between NL platelet high-expression gene set scores (unrelated to reconstitution time at the time of sampling) and reconstitution time. (G) DEGs in NL platelets that are unrelated to reconstitution time post-transplant compared to HD platelets. (H) Box plots showing elevated expression of CCR5 and PARP14 in NL platelets compared to HD platelets (top), and scatter plot showing that CCR5 and PARP14 expression in NL platelets are unrelated to reconstitution time (bottom). Statistical analysis was performed using Mann–Whitney U test, with ns p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.

Next, we analysed genes upregulated in NL platelets and their correlation with reconstitution time. Remarkably, 97.67% of these genes exhibited no correlation with reconstitution time (Fig. 5E), and their expression scores in patients with NL also showed no significant correlation (Fig. 5F). These findings suggest that the majority of the differences between NL and HD platelets are unrelated to the reconstitution time and do not normalise with prolonged reconstitution time. We further highlighted the most significantly altered genes between HD and NL platelets (Fig. 5G). Interestingly, among these, genes associated with cardiovascular disease, such as CCR5 and PARP14,42, 43, 44, 45, 46 were upregulated in NL platelets compared to HD. Importantly, their expression was independent of reconstitution time (Fig. 5H). Given the increased risk of cardiovascular complications, including atherosclerotic disease and heart failure,47, 48, 49 in patients with long term post-HSCT, the abnormal expression of these genes may offer valuable insights into the underlying mechanisms of these issues.

In conclusion, our study suggests that while platelet counts normalise in patients with long-term reconstitution, their molecular characteristics remain distinct from those of HD. These findings highlight the potential of platelets as sensors, offering a promising avenue to uncover mechanisms contributing to long-term transplantation complications.

Platelets precisely track immune state changes in aGVHD—a severe HSCT complication

It is well established that abnormalities in immune reconstitution following HSCT are often linked to a range of post-transplant complications.^40,50 These arise from disruptions in the recovery of immune cells such as neutrophils, T cells, and B cells. Building on our demonstration that platelets function as immune sensors capable of precisely monitoring normal immune reconstitution, we explored whether platelets could similarly detect abnormal immune states. We focused on two common post-HSCT complications: acute graft-versus-host disease (aGVHD) and cytomegalovirus (CMV) DNAemia, both closely associated with immune dysregulation.

We first examined platelet samples from patients with aGVHD, a frequent complication after allogeneic HSCT and a leading cause of non-relapse mortality.^51,52 Our RNA sequencing analysis revealed transcriptomic differences between patients with aGVHD and NS (complication-free), despite similar platelet counts and reconstitution times (Figure S5A). Functional enrichment of DEGs showed that pathways associated with aGVHD progression, including “neutrophil extravasation” and “lymphocyte mediated immunity”, were upregulated, whereas antigen presentation pathways, such as “peptide antigen assembly with MHC class II protein complex” and “antigen processing and presentation”, were downregulated (Fig. 6A). Gene set scoring and expression analyses confirmed upregulation of neutrophil-related genes (S100A8, S100A9) and downregulation of antigen presentation genes (HLA-DRA, HLA-DPA1) in patients with aGVHD (Fig. 6B and C). To determine whether these transcriptomic changes were related to peripheral immune cells, we examined neutrophil counts and found them to be significantly higher in patients with aGVHD (Figure S5B). Correlation analysis further showed that upregulated neutrophil-related genes positively correlated with neutrophil abundance, whereas downregulated antigen-presentation genes did not (Figure S5C), suggesting that neutrophil levels are a major factor shaping platelet molecular features. Given the critical role of neutrophil in aGVHD pathogenesis,^53,54 these results support platelets as immune sensors monitoring abnormal immune states. Additionally, this study identified potential new biomarkers for aGVHD diagnosis, including genes linked to neutrophil migration and antigen presentation.

Fig. 6.

Platelets exhibit different immune feature changes in various transplant complications, consistent with immune responses in the disease. (A) Enrichment of representative Gene Ontology: Biological Process (GO: BP) terms for DEGs between aGVHD and NS platelets. (B) Box plots showing the “neutrophil extravasation” [GO:0072672] gene set score in NS and aGVHD platelets, along with the expression of neutrophil-related genes S100A8 and S100A9. (C) Box plots showing the “peptide antigen assembly with MHC class II protein complex” [GO:0002503] gene set score in NS and aGVHD platelets, along with the expression of antigen presentation-related genes HLA-DRA and HLA-DPA1. (D) The sequential samples collected from three patients with aGVHD during diagnosis and treatment, including samples from two patients at diagnosis (Dx) and complete response (CR) and one patient at Dx and unstable partial response (unstable PR). (E) Box plots showing the scoring of DEGs that were upregulated and downregulated in aGVHD compared to NS platelets across the platelet samples from the three patients with aGVHD during diagnosis and treatment. (F) Enrichment of representative GO: BP terms for differential genes between CMV DNAemia and NS platelets. (G) Box plots showing the “antiviral innate immune response” [GO:0140374] gene set score in NS and CMV DNAemia platelets, along with the expression of related genes MX1 and DHX58. (H) Box plots showing the “interferon-mediated signaling pathway” [GO:0140888] gene set score in NS and CMV DNAemia platelets, along with the expression of related genes IFI27 and IFITM1. (I) Box plots showing the “neutrophil extravasation” and “peptide antigen assembly with MHC class II protein complex” gene set scores in NS and CMV DNAemia platelets. (J) Box plots showing the “antiviral innate immune response” and “interferon-mediated signaling pathway” gene set scores in NS and aGVHD platelets. Statistical analysis was performed using Mann–Whitney U test, with ns p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.

Finally, to evaluate whether platelet DEGs could track treatment response, we analysed sequential samples from three patients with aGVHD at diagnosis (Dx) and during therapy. Two patients provided post-treatment samples in complete response (CR), while one had an unstable partial response (unstable PR). Expression scoring of platelet transcriptomes using DEGs between patients with aGVHD and NS showed that genes upregulated in aGVHD decreased in CR and increased in unstable PR, whereas downregulated genes showed the opposite trend (Fig. 6D and E). These findings indicate that platelet transcriptomes not only reflect immune status during aGVHD but may also serve as biomarkers for monitoring treatment response.

Platelets also track immune state changes in another severe HSCT complication: CMV DNAemia

We next extended our investigation to another HSCT-associated complication: CMV DNAemia, a risk factor for non-relapse mortality.⁵⁵ Effective CMV control relies on coordinated innate and adaptive immunity. Early innate responses, including type I interferon release and NK cell activation, limit viral replication, while subsequent adaptive responses, particularly CD8⁺ T cells and their IFN-γ secretion, directly lyse infected cells to control the infection.^56,57

In our study, similar to patients with aGVHD, patients with CMV DNAemia showed no significant differences in platelet counts or reconstitution time compared to patients with NS (Figure S5D). However, platelet transcriptomes in CMV DNAemia exhibited marked changes. Antiviral pathways, such as “antiviral innate immune response” and “interferon–mediated signaling pathway”, were significantly upregulated, whereas cell cycle-related pathways, including “mitotic cell cycle” and “cell division”, were downregulated compared to NS platelets (Fig. 6F). Further scoring and gene expression analyses confirmed robust upregulation of antiviral and interferon-response genes, including MX1, DHX58, IFI27, and IFITM1 (Fig. 6G and H). To investigate whether these transcriptional changes were associated with alterations in circulating immune cells, we examined lymphocyte counts and found them to be significantly higher in patients with CMV DNAemia compared to controls (Figure S5E). Correlation analysis further showed that platelet expression of interferon-related genes positively correlated with lymphocyte abundance, whereas antiviral innate immune gene expression did not (Figure S5F), suggesting that lymphocytes may contribute to shaping platelet transcriptomic features. Given the activation of the immune system to counter viral infection, these findings further support the role of platelets as immune sensors capable of monitoring abnormal immune states in patients with CMV infection.

Finally, to assess whether platelets can distinguish immune abnormalities across different post-transplant complications, we scored immunity-related features identified in one condition within the context of the other. Features highly expressed in aGVHD, such as “neutrophil extravasation” and low expression of “peptide antigen assembly with MHC class II protein complex”, were not significantly altered in NS or CMV infection (Fig. 6I). Likewise, features upregulated in CMV DNAemia, including “antiviral innate immune response” and “interferon–mediated signaling pathway”, showed no significant changes in NS or aGVHD (Fig. 6J).

In summary, platelets not only serve as immune sensors to monitor immune state changes following HSCT but also accurately detect unique immune alterations under various post-transplant complications.

Platelets can sense and monitor immune cell activity and inflammatory states in SLE and UC

Expanding on our findings from HSCT and its complications, we investigated whether platelets function as immune sensors in other diseases. First, we analysed systemic lupus erythematosus (SLE), a complex autoimmune disorder involving dysregulated innate and adaptive immunity. In SLE, neutrophil extracellular traps release nuclear antigens that form immune complexes with autoantibodies, activating dendritic cells and macrophages to produce pro-inflammatory cytokines. B cells differentiate into plasma cells producing autoantibodies, while autoreactive CD4⁺ T cells further promote inflammation and plasma cell generation, contributing to tissue and organ damage.⁵⁸

Using platelet transcriptome data from Cornwell et al.,⁵⁹ we analysed immune cell- and inflammation-related gene expression in SLE platelets. Both innate and adaptive immunity gene sets were upregulated compared to HD (Fig. 7A and B). Specifically, genes linked to neutrophil and macrophage activation (e.g., MYD88, CXCR1, S100A9, C5AR1, TYROBP, IFNGR2) and B/T cell differentiation (e.g., STAT5B, NFAM1, LGALS1, RSAD2, SEMA4A, STAT6) were significantly elevated (Fig. 7C–F). Genes responsive to type I/II interferons, IL-1, IL-6, and IL-17 were also upregulated (Figure S6A), indicating that platelets sense and monitor inflammation in SLE.

Fig. 7.

Platelets sense immune and inflammatory activities and enable accurate disease diagnosis via machine learning in SLE and UC. (A) Box plot showing the “innate immune response” [GO:0045087] gene set score in platelets from HD and patients with SLE. (B) Box plot showing the “adaptive immune response” [GO:0002250] gene set score in platelets from HD and patients with SLE. (C–F) Expression of neutrophil-, macrophage-, B cell- and T cell-related gene sets and feature genes involved in the development of SLE in HD and SLE platelets. (C) Box plot showing the “neutrophil activation” [GO:0042119] gene set score in platelets from HD and patients with SLE. Heatmap showing the expression of relevant genes. (D) Box plot showing the “macrophage activation” [GO:0042116] gene set score in platelets from HD and patients with SLE. Heatmap showing the expression of relevant genes. (E) Box plot showing the “B cell differentiation” [GO:0030183] gene set score in platelets from HD and patients with SLE. Heatmap showing the expression of relevant genes. (F) Box plot showing the “T cell differentiation” [GO:0030217] gene set score in platelets from HD and patients with SLE. Heatmap showing the expression of relevant genes. (G) Expression of T cell-related gene sets and signature genes involved in the pathogenesis of UC in platelets from HD and patients with UC. Box plots showing the gene set scores for “T cell activation” [GO:0042110], “T cell differentiation” [GO:0030217], “T cell migration” [GO:0072678], “T cell proliferation” [GO:0042098], and “T cell receptor signaling pathway” [GO:0050852] in HD and UC platelets. Heatmap showing the expression levels of related genes. (H) A schematic diagram illustrating the development of machine learning models for SLE identification using either whole-transcriptome or immune gene profiles from platelets of HD and patients with SLE. (I) ROC curves showing the performance of SLE diagnostic models: the whole-transcriptome model achieved an AUROC of 0.881, while the immune gene-based model achieved an AUROC of 0.940. (J) Box plots displaying the expression levels of genes selected for developing the immune gene-based diagnostic model for SLE. (K) A schematic diagram illustrating the development of machine learning models for UC identification using either whole-transcriptome or immune gene profiles from platelets of HD and patients with UC. (L) ROC curves showing the performance of UC diagnostic models: the whole-transcriptome model achieved an AUROC of 0.917, while the immune gene-based model achieved an AUROC of 0.944. (M) Box plots showing the expression levels of genes selected for developing the immune gene-based diagnostic model for UC. Transcriptome data for platelets in HD and patients with SLE were downloaded from GSE226147. Transcriptome data for platelets in HD and patients with UC were downloaded from PRJNA737596. Statistical analysis was performed using Mann–Whitney U test, with ns p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001.

We further investigated platelets as immune sensors in ulcerative colitis (UC), an inflammatory bowel disease driven by gene–environment interactions and abnormal immune responses to gut microbiota. Microbial dysbiosis increases bacterial exposure, activating macrophages to produce TNF, IL-6, IL-12, and IL-23, while dendritic cells present antigens to T cells in mesenteric lymph nodes, promoting their proliferation, differentiation, and migration to inflamed sites. The resulting cytokine release and chronic inflammation drive UC development.^60,61

To assess whether platelets monitor immune status in UC, we analysed platelet transcriptomes from Xu et al.,⁶² focussing on immune cell- and cytokine-related genes. Given the strong link between T cells, inflammatory cytokines, and UC progression, we first examined T cell-related features in platelets from HD and patients with UC. Genes involved in T cell activation, differentiation, migration, proliferation, and T cell receptor signaling pathway (e.g., LCP1, FYN, DOCK8, PIK3CG, BTN3A2) were significantly upregulated in UC platelets (Fig. 7G).

We then evaluated key inflammatory cytokines and their regulatory pathways, including IL-6, IL-12, IL-23, TNF, and the JAK-STAT pathway. Both cytokine-related gene sets and individual genes were significantly elevated in UC platelets (Figure S6B), indicating that platelets can sense and monitor the inflammatory state. Thus, beyond HSCT and its complications, platelets can track disease-specific immune alterations and act as immune sensors across diverse immune and inflammatory disorders.

Platelet immune profiles enable accurate disease diagnosis via machine learning in SLE and UC

Given the capacity of platelets to function as immune sensors, we next investigated whether the immune features of platelets could be leveraged for disease state monitoring and diagnosis. To this end, we developed machine learning models using both whole-transcriptome and immune gene profiles from platelets of HD and patients with SLE to distinguish between the two groups. The immune genes were defined as those associated with neutrophils, monocytes, B cells, and T cells, as described in Fig. 2D and G, and Fig. 4D, and detailed in Table S5. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC) (Fig. 7H; detailed modelling workflow in Figure S7A). Comparative analysis revealed that the immune gene-based model achieved a test set AUROC of 0.940, outperforming the whole-transcriptome model, which reached an AUROC of 0.881 (Fig. 7I). Notably, key genes contributing to the immune gene-based model included WASHC3 (WASH Complex Subunit 3, involved in actin network regulation⁶³) and RABL3 (a RAS oncogene family-like protein, essential for lymphocyte development and function⁶⁴) (Fig. 7J). Genes incorporated into the whole-transcriptome model are listed in Figure S7B.

We further evaluated whether platelet immune features could aid in the diagnosis of UC. Following a similar approach, we built machine learning models using whole-transcriptome and immune gene profiles from the platelets of HD and patients with UC, and assessed their performance using AUROC (Fig. 7K; modelling workflow in Figure S7A). The immune gene-based model achieved a test set AUROC of 0.944, surpassing the whole-transcriptome model, which yielded an AUROC of 0.917 (Fig. 7L). Key genes selected for the immune gene-based model included TOP1 (Topoisomerase I, involved in regulating T cell-mediated anti-tumour immunity⁶⁵) and DNMT1 (DNA Methyltransferase 1, level of DNMT1 in colonic mucosa increasing with UC activity,⁶⁶ and inhibition of DNMT1 promoting Treg differentiation and alleviating UC symptoms⁶⁷) (Fig. 7M). The genes used in the whole-transcriptome model are shown in Figure S7C.

In conclusion, platelets reflect immune activity and disease states in SLE and UC, and machine learning models based on platelet immune gene profiles improve diagnostic accuracy and identify previously undefined markers, highlighting platelets as effective immune sensors for disease diagnostics.

Discussion

In this study, we investigated whether platelets act as immune sensors to monitor the immune status of the body. By analysing platelet transcriptome data at different stages following haematopoietic stem cell transplantation (HSCT), we demonstrated that platelet transcriptomes accurately reflect dynamic immune reconstitution post-HSCT, particularly the transition from innate to adaptive immunity. These changes are driven by stage-specific variations in immune cell counts and activity, which directly influence the evolution of immune molecular characteristics within the platelet transcriptome. Our analyses further confirmed that immune cells can modulate platelet transcriptomes through direct RNA transfer, contributing to the immune remodelling of platelets. Additionally, platelets exhibit a remarkable ability to monitor distinct immune state changes under different HSCT complications. Notably, platelets can sensitively detect changes in immune cell activity and inflammatory states in complex immune disorders such as systemic lupus erythematosus (SLE) and ulcerative colitis (UC). Machine learning models based on platelet immune gene profiles not only enhance diagnostic accuracy but also facilitate the discovery of previously undefined diagnostic markers. These findings confirm platelets’ role as sensitive, highly accessible immune sensors, deepen our understanding of their immune functions, and open new avenues for biomarker identification, disease monitoring and treatment.

Although the immunomodulatory roles of platelets are well-documented,² their capacity to sense and monitor immune status remains poorly understood. To explore this potential, we selected HSCT as an ideal model system. Patients with HSCT undergo a gradual immune reconstitution from innate to adaptive immunity, beginning with the recovery of innate immune cells, such as neutrophils and monocytes, and followed by adaptive immune cells, such as B cells and T cells.⁴⁰ Analysis of platelet transcriptomes at different stages after HSCT revealed that their changes accurately mirror the dynamics of immune reconstitution: in the short term, platelets predominantly express genes related to innate immune cells (e.g., neutrophils and monocytes), while in the long term, they increasingly express genes associated with adaptive immune cells (e.g., B cells and T cells). These findings strongly support the notion that platelets serve as immune sensors capable of monitoring immune status changes post-HSCT. This discovery expands our understanding of platelet immune functions from regulation to monitoring, reshaping perceptions of their role in the HSCT process.

Platelet recovery after HSCT is critical for patient outcomes. Previous studies have linked delayed platelet recovery to higher non-relapse mortality, lower survival rates, and increased risks of complications such as acute graft-versus-host disease (aGVHD).36, 37, 38^,68, 69, 70 However, these studies primarily focused on platelet counts, with limited attention to other characteristics. Our study reveals that post-HSCT platelets undergo molecular reconstitution, particularly involving immune-related genes. This suggests that the functional characteristics of platelets vary across different stages and that their dynamic immune molecular reconstitution may be pivotal for normal immune recovery. Further investigation is needed to understand the functional differences in platelets across stages and their impact on HSCT success. These findings offer a new molecular and functional perspective on the role and importance of platelets in HSCT.

To dissect how platelets monitor immune homoeostasis, we explored potential mechanisms underlying their role as immune sensors. Although platelets primarily inherit their RNA from parental megakaryocytes,^20,21 increasing evidence indicates that they can acquire RNA through interactions with other cells or via microvesicle transfer.^22,23 Platelets form tight aggregates with various immune cells,² and immune cells, such as monocytes, can transfer RNA to platelets.²² These interactions and RNA transfer mechanisms likely shape the immune molecular characteristics of platelets. Our correlation analysis showed that in short-term reconstituted platelets, 37.93% of neutrophil-related genes positively correlate with neutrophil counts and 44.74% of monocyte-related genes positively correlate with monocyte counts. In long-term reconstituted platelets, this proportion increases to 61.80%, with lymphocyte-related genes positively correlating with lymphocyte counts. These findings suggest that peripheral immune cell counts are key determinants of platelet immune molecular features post-HSCT. Furthermore, immune cell activation states influence these features, as evidenced by significantly higher neutrophil and monocyte activity in the short term and relatively low B cell activity during the same period. At the mechanistic level, we demonstrate that immune cells transfer their characteristic immune-related RNA to platelets in patients with short-term reconstitution, thereby endowing platelet transcriptomes with innate immune features and revealing the molecular basis for platelet function as immune sensors. Our findings confirm and extend previous observations of transcript uptake by platelets, providing direct evidence that human neutrophils and monocytes can deliver RNA to platelets during early post-HSCT immune reconstitution. Notably, neutrophils transfer cell type-specific immune transcripts, broadly remodelling the platelet transcriptome toward an innate immune profile. These results uncover a previously unrecognised mechanism by which platelets acquire immune information and function as dynamic immune sensors, explaining why platelet transcriptomes reflect immune activation and post-transplant complications. Collectively, our work highlights the potential of platelets as readily accessible biomarkers for immune monitoring and validates platelet transcriptomic data as an effective tool for capturing immune status, integrating signals from diverse immune cells to provide a detailed view of the immune landscape. Their accessibility and non-invasive nature position platelets as practical instruments for monitoring immune dynamics and detecting immunity-related changes.

In addition to acquiring RNA from immune cells, changes in megakaryocytes post-HSCT may also contribute to alterations in platelet immune molecular features. Previous research has identified immune-skewed megakaryocytes capable of producing platelets with immunomodulatory functions.⁷¹ Future studies should investigate whether and how the immune properties of megakaryocytes change post-HSCT to provide insights into the molecular basis of platelet immune feature alterations. Additionally, since platelets can transfer RNA to other cells, further exploration is warranted to understand the functional significance of platelet immune molecular changes in the context of HSCT.

We also examined whether platelets can monitor immune state changes under disease conditions. Transcriptome analysis of platelets from patients with aGVHD post-HSCT revealed significant upregulation of neutrophil-related genes, consistent with reports of increased neutrophil activity and molecular changes in aGVHD.^53,54,72 This finding supports platelets’ role as immune sensors in monitoring immune states during diseases. Additionally, we observed downregulation of major histocompatibility complex class II (MHC-II) genes in aGVHD platelets. While previous studies suggest that platelets enhance antigen-presenting capabilities in sepsis to suppress T cell proliferation and activity,¹² the downregulation of antigen-presenting genes in aGVHD platelets may reduce T cell suppression, potentially promoting aGVHD progression. These findings underscore the value of analysing platelet immune molecular features to monitor immune state changes, uncover previously undefined disease mechanisms, and identify new biomarkers for the diagnosis, treatment and prognosis of diverse diseases.

In patients with CMV DNAemia post-HSCT, platelets displayed high expression of antiviral response-related genes, consistent with the rapid initiation of antiviral immunity. Similar findings have been observed in platelets from patients with COVID-19,³⁴ further supporting their role as immune sensors in monitoring immune states during viral infections. Notably, scoring immunity-related genes altered in aGVHD or CMV DNAemia across another disease state confirmed that platelets can precisely monitor immune status changes under different HSCT complications. Analysis of platelet transcriptomes from patients with SLE and UC revealed their ability to monitor multiple immune cell activities and inflammatory states, further validating their function as immune sensors in complex immune disorders. Moreover, machine learning models based on platelet immune genes outperform those built on whole-transcriptome data in disease diagnosis, suggesting that platelets not only sense the body's immune status but also convey immune gene alterations that can be leveraged for diagnosing immune-related disorders. From the perspective of platelets as immune sensors, this study provides a new framework for monitoring immune status and diagnosing disease using peripheral blood. However, clinical translation requires addressing the current limitation of time-consuming sequencing workflows. Future research should focus on two directions: first, identifying core marker combinations from existing omics data and converting them into rapid detection platforms such as PCR; second, deepening understanding of platelet RNA transfer mechanisms, so that platelet-carried immune cell–specific RNA can be harnessed to develop diagnostic tools that accurately reflect subtle individual immune changes. Through these approaches, platelet transcriptome analysis could advance from basic research to clinical application, ultimately supporting personalised medicine.

Importantly, through its role as an immune sensor, the platelet transcriptome offers valuable insights into the long-term implications of HSCT. While counts of major cell types in peripheral blood recover over time, patients remain at heightened risk for chronic complications, particularly cardiovascular and respiratory diseases.^48,73 Notably, cardiovascular disease mortality in long-term survivors of HSCT is twice that of the general population.⁴⁹ The cumulative incidence of cardiovascular disease at 1, 5, and 10 years after allogeneic HSCT was 7.7%, 16.4%, and 21.9%, respectively.⁴⁷ Our analysis of patients with long-term reconstitution revealed that while platelet counts normalise, their transcriptomes exhibit significant differences compared to healthy donors. Correlation analysis indicated that over 90% of differentially expressed genes are unrelated to the time since transplantation. Key genes, such as CCR5 and PARP14, were found to be upregulated, each playing critical roles in cardiovascular disease progression. For instance, CCR5 expression is increased in atherosclerotic plaques, and its deficiency reduces lesion size and mitigates T cell and monocyte infiltration.^44,45 Furthermore, treatment with CCR5 antagonists has been shown to shrink plaques and decrease monocyte/macrophage infiltration.^42,43 Similarly, PARP14 is reported to have elevated expression in atherosclerotic lesions and, together with PARP9, regulates macrophage activation, contributing to plaque formation.⁴⁶ A detailed analysis of the genes specifically altered in patients with long-term reconstitution may thus provide previously undefined clinical insights and inform new intervention strategies to prevent long-term complications following transplantation.

In conclusion, our findings position platelets as sensitive and dynamic immune sensors that integrate signals from circulating immune cells. They reflect the evolving immune landscape following HSCT, capture disease-specific immune responses, and retain transcriptomic changes associated with chronic complications. Their accessibility and responsiveness make platelet immune profiles a powerful, minimally invasive tool for real-time immune monitoring, disease diagnosis, and therapeutic evaluation. These insights expand our understanding of platelet biology and open new avenues for precision medicine in immune-related disorders.

Contributors

L.Z. designed the experiments; D.F., J.C. and S.N. collected the blood samples; L.Z., Y.W., W.L. J.C. and D.F. performed the experiments; L.Z., J.C. and D.F. collected the clinical data; L.Z. analysed the transcriptome, experimental and clinical data, and made the figures; Y.M., P.S., C.L., S.F., W.Z. and C.H. provided the experimental support; H.W. and L.Z. wrote the manuscript; J.Z. and F.W. reviewed and edited the manuscript; H.W., J.Z. and E.J. coordinated and designed the project; H.W., J.Z. and E.J. have accessed and verified the underlying data. All authors read and approved the final version of the manuscript.

Data sharing statement

The raw data of bulk RNA sequencing have been deposited in the GSA-Human with accession codes HRA009805, HRA010325, and HRA014604. Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

Declaration of interests

H.W. is a founder and shareholder of HaemoCure lnc. C.L. is a shareholder of HaemoCure lnc. F.W, is an employee of HaemoCure lnc. The remaining authors declare no competing interests.