This study recruited participants from September 2021 to September 202 from Sichuan Cancer Hospital, comprising three groups: HD, BPN,and LAC. Inclusion criteria for BPN and LAC groups were (1): Patients with pulmonary nodules initially detected by low dose computed tomography (LDCT) who underwent surgery or histological examination at Sichuan Cancer Hospital. (2) For LAC group: Pathological diagnosis confirmed by two independent pathologists through NCCN clinical practice guidelines in oncology (Non-small cell lung cancer 2021); staging according to the 8th edition of the Union for International Cancer Control (UICC) TNM classification; no prior anti-neoplastic therapy; no history of treatments affecting platelet function (e.g., blood transfusion, aspirin); absence of other malignancies, infections, or inflammatory conditions; complete clinical information available. (3) For BPN group: Postoperative pathology confirming non-malignant status; no history of treatments affecting platelet function; absence of other malignancies, infections, or inflammatory conditions; complete clinical information available. (4) The HD group consisted of individuals with no evidence of LAC on follow-up low-dose spiral CT screening, and no other critical diseases, infections, or inflammatory conditions. This study was approved by the Medical Ethics Committee of Sichuan Cancer Hospital (approval number: SCCHEC-02-2020-043).

The work-flow chart of this study was shown in Figure 1 . Platelets and plasma were isolated from whole blood collected in purple-cap BD Vacutainers containing EDTA anti-coagulant by standard centrifugation. The nucleated blood cells were removed by centrifugation at room temperature for 20 min at 120g (Co., Ltd. Shuke, Chengdu, China), resulting in platelet-rich plasma (PRP). To avoid leukocytes and erythrocytes contamination, only used the upper 4/5 of the PRP for further collecting platelets. The platelets were isolated from the PRP by centrifugation at room temperature for 20 min at 360g and the platelet pellet was collected in 50 μl RNAlater (Sigma-Aldrich, USA). Platelet purity was determined by morphometric experiments and platelet counting by the standard of less than five white blood cells per million platelets. In addition, the upper strata platelet-poor plasma (PPP) was stored directly at -80°C for further exosome isolation. To minimize platelet RNA degradation due to temporal effects, the isolation was supposed to be completed within 2h after blood collection.

The exosomes were isolated by SEC (size exclusion chromatography) methods as described previously with minor modifications (33). Plasma exosomes are mainly identified by Transmission electron microscopy (H-7650, Hitachi Ltd., Tokyo, Japan) and Nanoparticle Tracking Analysis. The identification results were consistent with the reported exosome particle size, and the morphological characteristics proved that the exosome extraction was qualified, ( Supplementary Figure 1 ). All the samples were collected between 2021 and 2022.

A total RNA rapid extraction kit for blood liquid sample (centrifugal column type) (Bioteke, China) were used to isolate platelet total RNA, following the manufacturer’s instructions. Exosome total RNA was extracted and purified from plasma exosome using miRNeasy Serum/Plasma Advanced Kit (Qiagen, cat. No. 217204) according to the kit instruction. And the assessments of the concentration and quality and purity of platelet and plasma RNA for sequencing were evaluated using the RNA Nano 6000 Assay Kit of the Agilent Bioanalyzer 2100 System (Agilent Technologies, CA, USA). Test reports of sequenced platelets and exosomes are shown in Supplementary Table 1 .

The sequencing libraries were generated by QIAseq miRNA Library Kit (Qiagen, Frederick, MD) following manufacturer’s recommendations. Library quality was assessed by Agilent Bioanalyzer 2100 and qPCR. The clustering of the index-coded samples was performed on the acBot Cluster Generation System using TruSeq PE Cluster Kitv3-cBot-HS (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions. After cluster generation, the library preparations were sequenced on an Illumina Hiseq platform and paired-end reads were generated.

The original image Data files sequenced by Illumina HiSeq2500 platform were converted into Raw Data by Base Calling, and the results were stored in FASTQ file format. It contains the information of sequencing sequence and its corresponding sequencing quality information. To ensure the accuracy of information analysis, quality control of the original data was required to obtain high-quality sequences (namely Clean Reads). The quality control criteria of the original sequences were as follows: (1) The sequence with low quality value was removed; (2) Reads with unknown base N (N is an unidentified base) content greater than or equal to 10% were removed; (3) Reads without 3 ‘junction sequence were removed; (4) Cut off the 3 ‘joint sequence; (5) Remove sequences shorter than 15 or longer than 35 nucleotides. Then Bowtie tools soft was used, and the clean reads respectively with Silva database, GtRNAdb database, Rfam database and Repbase database sequence alignment. The ribosomal RNA (rRNA), transfer RNA (tRNA), small nuclear RNA (snRNA), small nucleolar RNA (snoRNA) and other ncRNA and repeats were filtered. Finally, the remaining reads were used to detect known miRNA and new miRNA predicted by comparing with known miRNAs from miRbase and Human Genome (GRCh38), respectively. The transcripts per million (TPM) was calculated by comparing the number of read for each miRNA.

Referring to the screening conditions in the reported reference gene validation studies and combined with the actual data of this study, the final screening conditions are determined as follows (34–37): the intra-group mean value of the coefficient of variation (CV) ≤ 20%, and the Log10 (counts)>=4. Besides, platelet and exosome differential miRNA analysis was performed by using DESeq2 R package. The screening criteria were|log2Foldchange| ≥ 0.5, P value < 0.05. The platelet differential miRNAs we selected were compared with platelet differential miRNAs that have been studied and other LAC miRNA databases, including TGGA, EMTAB8026, GSE55993, GSE68951, GSE40738, GSE137140. All data analysis was performed by using the statistical programming language R (R Foundation for Statistical Computing, http://www.Rproject.org).

A miRNA rapid extraction kit for blood liquid sample (centrifugal column type) (Bioteke, China) were used to isolate platelet miRNA. The assessments of the concentration and quality of miRNA were carried out by Thermo Scientific NanoDrop 2000 Spectrophotometer (Thermo Scientific, USA). Reverse transcription was carried out using a Mir-X™ miRNA First Strand Synthesis Kit (TaKaRa Bio, Dalian, China), following the manufacturer’s instructions. The 10μl system was selected for reverse transcription and the specific system we performed in this study as shown in Supplementary Table 2 . The design of MiRNA primers used poly (A) tailing-based method. Only forward primer needs to be designed. The reverse primer of miRNA is the general reverse primer provided in the Mir-X™ miRNA First Strand Synthesis Kit.

The primers were designed and synthesized by Tsingke Biological Technology (Beijing, China). All the primers that we used were listed as Supplementary Table 3 . Quantitative real-time polymerase chain reaction (RT-qPCR) was carried out using the ABI 7500 Real-Time PCR Detection System (Thermo Scientific; USA), in which the amplification and detection steps were combined. All the assays were performed using three biological replicates. A single qPCR reaction was performed in a 20 µL volume containing 10 µL TB Green Premix Ex Taq II(2×), 0.8 µL of each primer, 2 µL of cDNA sample, 0.4 µL ROX Reference Dye II and 6.0 µL water free of RNase and DNase. The melting curve of the qRT-PCR product, the specificity of the primer is better when the melting curve shows a single peak and the miRNAs with a cycle threshold (Ct) value less than 35 in the panel were included in the data analysis.

The average expression stability of the candidate reference miRNAs was evaluated by the comparative Delta CT method (38), BestKeeper (39), and geNorm (40), and NormFinder (41). The differential expression levels of miRNAs in platelets were used t-test or nonparametric tests, and P value < 0.05 was considered be statistically significant. The diagnostic value was evaluated by ROC analysis. The comparison of diagnostic performance between platelet differential miRNA and other biomarker and clinical models used ROC and Net Reclassification Index (NRI). All data analysis was performed by using the statistical programming language R (R Foundation for Statistical Computing, http://www.Rproject.org) and the Graphpad Prism 9.3 statistical software.

Data from the TGGA database were used to analyze the relationship between differential miRNA and clinical parameters of cancer patients. The target gene prediction of differentially expressed miRNA was implemented by the multiMiR R packages (42). The cellular component (CC), molecular function (MF), and biological process (BP) of target genes by Gene Ontology database (http://geneontology.org). Co-expression of LAC analysis for differential miRNA and target gene was performed via the web of STARBASE (https://starbase.sysu.edu.cn/). The survival analysis of the target gene was analyzed via GEPIA (http://gepia.cancer-pku.cn/index.html).

Platelets and exosomes RNA-seq results from HD (n=3) and LAC (n=3) showed that miRNAs in platelets and exosomes have significant differences ( Figure 2 ). Detailed clinical features from participants are provided in Supplementary Table 4 . Firstly, the results showed that rRNA accounted for the largest proportion of small RNA species in exosomes, while miRNA accounted for the largest proportion of small RNA species in platelets ( Figure 2A ). There was no difference in the unique known-miRNAs counts in platelets and exosomes detected by RNA-seq examination from HD and LAC groups (p>0.05, Figure 2B ). Besides, the results showed that miRNAs fragment length detected in both exosomes and platelets was mainly 22nt ( Figures 2C, D ), which the length of miRNA was consistent with the current research. Finally, platelet have more he unique known miRNAs counts than exosomes and there are significant differences in the unique known miRNAs counts between platelets and exosome groups (p=0.005, Figure 2E ).

Based on platelet RNA-seq results, we identified the top 20 miRNAs with the highest expression levels in HD and LAC groups ( Figures 3A, B ). According to the screening condition, we selected four stable and highly expressed candidate reference miRNAs (hsa-miR-146a-5p, hsa-miR-16a-5p, hsa-let-7c-5p, and hsa-let-7i-5p) in platelets for further stability verification All four candidate miRNAs were among the top 20 miRNAs with the highest expression levels in both HD and LAC group ( Figure 3C ).

RT-qPCR was used to confirm the expression levels of these four genes in LAC patients (n=16) and HDs (n=17). The baseline characteristics of all participants are listed in Table 1 . The mean of Ct values of the four reference miRNAs in the 33 subjects are shown in Supplementary Table 5 . The comparative delta CT algorithm calculated δCt by comparing the relative expression of “miRNA pairs” in each sample, used as a criterion for screening reference genes. As determined by Delta CT method analyses, the-let-7i-5p was the most stable candidate reference miRNA (SD=1.571) ( Figure 3D ).

The geNorm is a popular algorithm to determine the most stable reference genes from a set of tested candidate reference genes in each sample panel. The analysis of the GeNorm algorithm found that hsa-let-7c-5p was highly expressed and stable along with hsa-let-7i-5p ( Figure 3E ). In addition, the results of the NormFinder algorithm and the BestKeeper algorithm both showed that the most stable expression of four candidate reference miRNAs was hsa-let-7i-5p ( Figures 3F, G ). Finally, the candidate reference miRNAs were ranked in increasing order of their stability values, and platelet has-let-7i-5p was the best reference miRNA in platelets for LAC ( Table 2 ).

According to the screening criteria, we finally obtained 55 up-regulated miRNAs and 34 down-regulated miRNAs in LAC platelets. In our study, the 10 highest expression up-regulated miRNAs were hsa-miR-223-3p, hsa-miR-142-5p, hsa-miR-1-3p, hsa-miR-101-3p, hsa-miR-223-5p, hsa-miR-335-5p, hsa-miR-19a-3p, hsa-miR-19b-3p, hsa-miR-301a-3p, hsa-miR-4433b-3p, and the 10 highest expression down-regulated hsa-miR-199b-3p, hsa-miR-148b-3p, hsa-miR-148a-3p, hsa-miR-125a-5p, hsa-miR-143-3p, hsa-miR-99b-5p, hsa-miR-379-5p, hsa-miR-134-5p, hsa-miR-127-3p, hsa-miR-941. ( Figures 4A, B ). The heatmap showed that the 20 platelet differential miRNAs could clearly distinguish LAC patients from healthy individuals ( Figure 4A ).

According to the platelet differential miRNAs that have been published, hsa-miR-223-3p and hsa-miR-223-5p were replicated in our results, and this result further demonstrated the reliability of our sequencing data ( Figure 4C ).

In addition, we compared the 20 platelet differential miRNAs we selected with other LAC miRNA datasets. The results showed that most of the miRNAs differentially expressed in platelets were stably expressed in plasma, exosomes, and blood ( Figures 4D, E ). However, of the 20 miRNAs differentially expressed in platelets, 13 were also differentially expressed in tissues ( Figure 4E ). Most of the differentially expressed miRNAs in platelets were also differentially expressed in tissue further demonstrated the potential of platelet miRNAs as a biomarker for liquid biopsy.

Since low abundance in total RNA samples make miRNA analysis a challenging task (43, 44). In our study, we chose the differential miRNAs with the highest abundance rather than the Fold Change differences for preliminary verification. The candidate differential miRNAs we selected were hsa-miR-199b-3p and hsa-mir-142-5p. According to the exclusion and inclusion criteria, a total of 30 subjects were included in the first validation, including 15 LAC and 15 HD. The baseline characteristics of all participants are listed in Table 3 . Preliminary RT-qPCR results showed that the relative expression of hsa-miR-199b-3p in HD and LAC platelets was statistically different (p=0.0034), while the relative expression of hsa-mir-142-5p in HD and LAC platelets was not statistically different (p=0.23) ( Figure 5A ). Therefore, we finally selected hsa-miR-199b-3p for further large-scale verification.

To further verification, 70 LAC patients and 31 healthy subjects and 32 BPN patients were included in this study to verify whether hsa-miR-199b-3p could be a biomarker for LAC diagnosis ( Table 3 ). We used RT-qPCR methods to examine hsa-miR-199b-3p expression levels in platelets in HD group, LAC group and BPN group. The results showed that the expression levels of platelet hsa-miR-199b-3p in HD group were significantly higher than in LAC group (p=0.0004, Figure 5B ). In addition, we also compared the expression levels of hsa-miR-199b-3p in platelets between BPN group and LAC group by using RT-qPCR assay. Surprisingly, the results showed that the expression of hsa-miR-199b-3p was significantly higher in BPN group than in LAC group (p=0.0016, Figure 5C ).

In current clinics, it is very difficult to distinguish LAC stage I with only a single lesion less than 3cm from benign pulmonary nodules. We further evaluated the expression levels of platelet hsa-miR-199b-3p between LAC stage I patients with only a single lesion less than 3cm and BPN patients. To our surprise, the results showed that the expression levels of platelet hsa-miR-199b-3p in LAC stage I group and BPN group was statistically different (p=0.0098, Figure 5D ). The mean of Ct values of platelet has-miR-199b-3p in the 133 subjects were shown in Supplementary Table 6 . According to the mean value of independent samples of the two groups was compared to ensure that the test performance reached 0.8, and the number of experimental groups and control groups were qualified.

We used ROC curves to analyze the ability of platelet hsa-miR-199b-3p to distinguish benign from malignant pulmonary nodules. The analysis results showed that platelet hsa-miR-199b-3p has good diagnostic accuracy in distinguishing LAC group from BPN group and the AUC for the ROC was 0.73 (95% CI, 0.5574 to 0.7850) ( Figure 5E ), and the AUC for the ROC was 0.72 (95% CI, 0.6285 to 0.8255) in the diagnosis of BPN group and LAC stage I group ( Figure 5F ).

According to the participants with complete clinical information and laboratory information. The performance of hsa-miR-199b-3p to diagnose benign and malignant pulmonary nodules was compared with other clinical models and CEA ( Figure 5G ). The results showed that the AUC of platelet hsa-miR-199b-3p in the LAC was 0.67 (95%CI, 0.4889 to 0.8576), while the AUC of CEA in LAC was 0.55(95%CI, 0.3722 to 0.7352). The diagnostic performance to distinguish benign from malignant pulmonary nodules of other clinical models including VA model (AUC=0.58, 95%CI, 0.4034 to 0.7523), BU model (AUC=0.55, 95%CI, 0.3735 to 0.7361), and MC model (AUC=0.53,95%CI, 0.3499 to 0.7028) also presented limitation. Therefore, platelet hsa-miR-199b-3p have shown significant clinical values by showing a high sensitivity, specificity and accuracy for LAC diagnosis and benign and malignant pulmonary nodules diagnosis. Platelet hsa-miR-199b-3p had a better reclassification ability than the best clinical model (VA model) and CEA with significant NRI (respectively, 0.33, 95%CI, 0.0028-0.65; 0.34, 95%CI, 0.0017, 0.0064) ( Figure 5H ).

Finally, we established a logistic regression model with three important indicators (nodule size, CEA and TEP Has-miR-199b-3p). Then we compared the diagnostic efficiency of the model after removing the three indicators respectively. The diagnostic efficiency of diagnostic model diagnostic efficiency (AUC=0.64), however, when we are removing our indicators (TEP Has-miR-199b-3p), the diagnostic performance of the diagnostic model decreased the most (AUC=0.54) ( Supplementary Figure 2 ). Therefore, the TEP Has-miR-199b-3p is an important indicator for the diagnosis model of benign and malignant pulmonary nodules.

However, the method of distinguishing benign and malignant lung nodules according to whether lung nodules are ground glass nodules (GGNs) is better (AUC =0.71) ( Figure 5G ). Then, patients with lung nodules were divided into GGN group and non-GGN group. We found that the expression of platelet hsa-miR-199b-3p in the non-GGN group was statistically different between patients with benign and malignant pulmonary nodules (p=0.037) ( Figure 5I ). It is proved that platelet hsa-miR-199b-3p also can further play an essential role in the diagnosis of benign and malignant in non-GNN group. Finally, we also analyzed the relationship between multiple factors. We can find platelet hsa-miR-199b-3p also seem to correlate the gender ( Figure 5J ).

Correlations between the expression levels of hsa-miR-199b-3p and clinical parameters of LAC cancer subjects shown in Figures 6A–C . The expression of hsa-miR-199b-3p at different ages, genders, and stages of cancer will help further study the potential significance of hsa-miR-199b-3p as a cancer marker ( Figure 6 ). We can find that the expression levels of hsa-miR-199b-3p in solid tissue normal and primary tumor was statistically different (p <0.05) ( Figure 6B ). Further evidence of the potential significance of hsa-miR-199b-3p as a biomarker for the diagnosis of LAC. (All data comes from TCGA database). Besides, the high expression of hsa-miR-199b-3p was associated with low survival probability (p=0.0063) ( Figure 6C ).

The target gene prediction of hsa-miR-199b-3p was implemented by the multiMiR R packages. According to five different target gene prediction methods (Diana Micro, ELMMO, Miranda, Pita, Targetscan), 2593 possible target genes of hsa-miR-199b-3p were obtained. Then we integrated the information of the two databases (Mirtarbase, Tarbase), and found that 402 of the 2593 predicted target genes have been validated. We selected the top 10% predicted genes of each method for gene enrichment analysis, and the results showed that the main functions of target genes of hsa-miR-199b-3p were participated in mitotic spindle and cell adhesion molecule binding ( Figures 6D, E ). In addition, based on the five prediction methods, we found that the KTN1 gene appeared in all three prediction methods ( Figure 6F ).

By retrieving UALAN database (https://ualcan.path.uab.edu/index.html), we found that KTN1 expression was higher in cancer tissues than in para-cancer tissues ( Supplementary Figure 3 ). The complementary sequence of hsa-miR-199b-3p and the KTN1 gene shown in Figure 6G . Besides, in lung adenocarcinoma, there was a significant correlation of co-expression between the KTN1gene and hsa-miR-199b-3p (r=0.195, p=8.56e-06) ( Figure 6G ). Finally, Higher the KTN1 gene expression was associated with lower survival in LAC cancer, consistent with hsa-miR-199b-3p results ( Figure 6H ). Therefore, we thought that the KTN1gene was the most likely target gene for RNA hsa-miR-199b-3p. Then we conducted some preliminary verification on the gene expression of KTN1 in platelets of LAC and healthy donor ( Supplementary Figure 4 ). It can be preliminarily found that there are statistical differences in the expression of TEPs KTN1(P<0.0001). We found an interesting point that the expression of KTN1 in TEPs is opposite to the expression in tumor cells and tissues. Our team suspected that it may be consistent with the current mechanism hypothesis of TEPs. Mechanism hypothesis of TEPs formation is that platelets educate tumor cells by highly efficient transference of lipids, proteins and RNA through different mechanisms (28).