Diagnosis of head and neck cancer by AI-based tumor-educated platelet RNA profiling of liquid biopsies

Niles E. Wondergem; Jos B. Poell; Sjors G.J.G. In ‘t Veld; Edward Post; Steven W. Mes; Myron G. Best; Wessel N. van Wieringen; Thomas Klausch; Robert J. Baatenburg de Jong; Chris H.J. Terhaard; Robert P. Takes; Johannes A. Langendijk; Irma M. Verdonck-de Leeuw; Femke Lamers; C. René Leemans; Elisabeth Bloemena; Thomas Würdinger; Ruud H. Brakenhoff

doi:10.1172/jci.insight.186680

Diagnosis of head and neck cancer by AI-based tumor-educated platelet RNA profiling of liquid biopsies

Niles E. Wondergem,^1,2 Jos B. Poell,^1,2 Sjors G.J.G. In ‘t Veld,^3,4,5 Edward Post,^4,6,7 Steven W. Mes,^1,2 Myron G. Best,^4,6,7 Wessel N. van Wieringen,^8,9 Thomas Klausch,⁸ Robert J. Baatenburg de Jong,¹⁰ Chris H.J. Terhaard,¹¹ Robert P. Takes,¹² Johannes A. Langendijk,¹³ Irma M. Verdonck-de Leeuw,^1,14 Femke Lamers,¹⁵ C. René Leemans,^1,2 Elisabeth Bloemena,^16,15,17 Thomas Würdinger,^4,6,7 and Ruud H. Brakenhoff^1,2

Authorship note: TW and RHB contributed equally to this work.

Published November 27, 2025 - More info

View PDF

Abstract

Over 95% of head and neck cancers are squamous cell carcinoma (HNSCC). HNSCC is mostly diagnosed late, causing a poor prognosis despite the application of invasive treatment protocols. Tumor-educated platelets (TEPs) have been shown to hold promise as a molecular tool for early cancer diagnosis. We sequenced platelet mRNA isolated from blood of 101 patients with HNSCC and 101 propensity-score matched noncancer controls. Two independent machine learning classification strategies were employed using a training and validation approach to identify a cancer predictor: a particle swarm optimized support vector machine (PSO-SVM) and a least absolute shrinkage and selection operator (LASSO) logistic regression model. The best performing PSO-SVM predictor consisted of 245 platelet transcripts and reached a maximum area under the curve (AUC) of 0.87. For the LASSO-based prediction model, 1,198 mRNAs were selected, resulting in a median AUC of 0.84, independent of HPV status. Our data show that TEP RNA classification by different AI tools is promising in the diagnosis of HNSCC.

Introduction

Head and neck squamous cell carcinoma (HNSCC) arises in the mucosal linings of the upper respiratory and digestive tract, comprising the oral cavity, oropharynx, hypopharynx, and larynx. It ranks among the most prevalent cancers, affecting more than 890,000 patients annually and resulting in over 450,000 deaths in 2022 (1). Risk factors for developing HNSCC are tobacco and alcohol consumption and, particularly in the oropharynx, persistent infection with high-risk human papillomavirus (HPV) (2).

Patients with early-stage disease undergo local single modality treatment and have a 90% 5-year overall survival (OS) with good quality of life. However, most patients present with locally advanced disease and are treated with definitive chemoradiotherapy or extensive surgery with reconstruction combined with postoperative (chemo)radiotherapy, which affect quality of life while the 5-year OS remains 40%–50% (3). Despite a variety of advances in treatment modalities, the survival of HNSCC remains disappointing. Hence, early diagnosis is key to improve both prognosis of HNSCC and the quality of life of treated patients, as less invasive treatment protocols can be applied (4). Late diagnosis is often due to the mild initial symptoms and delay in recognition by primary care physicians, as early-stage HNSCC can be challenging to identify (5).

In this setting, liquid biopsies have gained increasing interest as a novel, minimally invasive means of early cancer diagnostics through the detection of biomarkers in body fluids like blood, urine, and saliva. Samples from these compartments are easy to obtain and could eliminate the need for a tumor biopsy and its inherent risk of complications. Biomarkers under investigation include circulating tumor cells (CTCs), extracellular vesicles (EVs), circulating cell-free DNA (cfDNA), circulating cell-free RNA (cfRNA) and tumor-educated platelets (TEPs) (6). Their relative abundance and straightforward isolation make TEPs a highly interesting biomarker for minimally invasive cancer diagnosis (7, 8).

Platelets are traditionally known for their function in hemostasis and wound healing, but they have more recently been recognized as protumorigenic entities in the tumor microenvironment (TME) (7). Platelets have been shown to induce prosurvival and proangiogenic signaling and participate in the epithelial-mesenchymal transition (EMT) of tumor cells, facilitating invasiveness and metastasis (9). Moreover, platelet aggregation around CTCs provides shielding from immune surveillance and reduces mechanical stress endured in the bloodstream (10). Tumor cells can induce differential splicing of platelet pre–messenger RNA (pre-mRNA) in their interaction with platelets. Additionally, platelets are able to ingest tumor-derived biomolecules such as mRNA (11, 12). These processes are proposed to constitute the tumor education of platelets, now referred to as TEPs, and provide them with a changed mRNA repertoire which can be exploited for cancer diagnostics, as shown by Best et al. (13). By performing RNA-Seq on platelets and employing a self-learning support vector machine (SVM) algorithm, they were able to discriminate patients with cancer from healthy controls with 96% accuracy (13). Results were validated in multiple follow-up studies and by other groups in various cancer types (14–19).

Crucial in the evaluation of biomarkers in cancer diagnostics, including TEPs, is the use of a well-matched control group. Particularly for patients with HNSCC, increased age, smoking habits, and alcohol consumption might affect TEP expression patterns, highlighting the importance of well-matched control groups (20, 21). Previously, a comprehensive protocol for platelet RNA-based diagnostic algorithm development was published, providing a detailed step-by-step wet and dry lab manual to be exploited by scientists in their field of interest (22). In this study, we applied this protocol and evaluated an alternative feature selection and classification strategy as compared with the original protocol, and investigated the applicability of TEPs as diagnostics in a cohort of patients with HNSCC and propensity-score matched noncancer controls.

Results

RNA of all 214 samples was sequenced with a mean read count of ~10 million reads per sample. After data quality control steps filtering out low-abundance reads (22), 5,546 of 57,736 annotated RNA transcripts remained, which were used for subsequent classification model development. In total, 6 samples had to be excluded because of poor data quality, indicating that 208 of 214 (97%) were diagnostic. To maintain a propensity-score matched 1:1 dataset, the corresponding matched sample for each sample excluded during quality control had to be excluded as well, resulting in a cohort of 202 samples containing 101 patients with HNSCC and 101 matched cancer-free controls (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.186680DS1). The clinical characteristics are depicted in Table 1. Variables — age, smoking status, packyears, and alcohol consumption — were evenly distributed among groups as a result of matching. Cases included 70% males compared with 35% in the control group (P < 0.001). This was a consequence of the design of the NET-QUBIC study where patients’ spouses were included as cancer-free controls. Since most patients with HNSCC are males, most controls were female. As we could not correct this by matching, it was corrected for by the Remove Unwanted Variation (RUV) function as described in the Methods section.

Table 1

Patient and control characteristics

PSO-SVM classification. First, we identified the most predictive TEP RNA transcripts, using particle swarm optimization (PSO) to identify the RNAs resulting in the best unsupervised hierarchical clustering. Out of 5,546 differentially expressed transcripts, 941 transcripts identified by a PSO threshold of FDR < 0.0018 were found to be particularly useful for the discrimination between cases and controls (Fisher exact test P < 0.0001; Supplemental Figure 2). These 941 transcripts were forwarded to the SVM algorithm. For training of the SVM, paired samples were randomly divided into a training set (40%, n = 41 HNSCC cases and n = 41 matched controls), evaluation set (30%, n = 30 HNSCC cases and n = 30 matched controls), and validation set (30%, n = 30 HNSCC cases and n = 30 matched controls). A general caveat in machine learning algorithms is that results are highly dependent on the composition of the training, evaluation, and validation sets. To evaluate the effect of sample distribution on model performance and selected biomarker panels, 3 SVM classification models were developed with 3 different, random distributions. Subsets were checked for significant differences in age, smoking, smoking pack years, and alcohol use to ensure that no bias was introduced by chance (Supplemental Figure 3). The first analysis (SVM-1) had 245 transcripts in the predictor and resulted in an area under the curve (AUC) of 0.87 (95% CI, 0.81–0.97) and 87% accuracy. The second analysis (SVM-2) resulted in 200 transcripts with an AUC of 0.84 (95% CI, 0.74–0.94) and 78% accuracy. The third SVM analysis (SVM-3) resulted in 200 transcripts with an AUC of 0.85 (95% CI, 0.75–0.95) and 82% accuracy (Figure 1, A–C). A Venn diagram of the selected transcripts for SVM-1, SVM-2, and SVM-3 showed an overlap of 48 RNA transcripts between the 3 analyses (Figure 1D and Supplemental Table 1).

Figure 1

Performance of learned SVMs. (A–C) ROC curves and summary of performance of training, evaluation, and validation series for learned SVMs 1–3 on the dataset of HNSCC (n = 101) and matched controls (n = 101). (D) Venn diagram showing overlap between selected biomarker panels of learned SVMs 1–3. Training set is indicated by a dashed light grey line, evaluation set by a solid grey line and validation set by a solid blue line. ROC, receiver operator characteristic; SVM, support vector machine; AUC, area under the curve; CI, confidence interval.

LASSO classification. The 3 SVM models indicated that TEP profiling was highly accurate to distinguish patients with HNSCC from matched controls. Still, the considerable variation in selected transcripts affects extrapolation to the clinical setting and prompted an alternative analysis method using a least absolute shrinkage and selection operator (LASSO). For LASSO classification, samples were randomly divided into a set for training and internal validation (66%, n = 67 HNSCC cases and n = 67 matched controls) and a set for external validation (34%, n = 34 HNSCC cases and n = 34 matched controls). Sample selection for the training and validation sets and subsequent classifier development was iterated 1,000 times, resulting in 1,000 classification models. These models reached a median AUC of 0.84 (95% CI, 0.69–0.94) and a median accuracy of 76% (Figure 2). Accuracy was dependent of disease stage: samples of early stage (TNM disease stage I/II) and advanced stage (TNM disease stage III/IV) tumors showed an detection accuracy of 68% and 86%, respectively (Table 2).

Figure 2

Performance of LASSO models. (A) ROC curve and summary of performance of validation series for 1,000 LASSO models on the dataset of HNSCC (n = 101) and matched controls (n = 101). In total, 1,198 genes were selected and used for the prediction models. (B) Box plot of the mean prediction scores of matched controls and HNSCC samples in 1,000 LASSO models. ROC, receiver operator characteristic; LASSO, least absolute shrinkage and selection operator; HNSCC, head and neck squamous cell carcinoma; AUC, area under the curve; CI, confidence interval; Acc., accuracy.

Table 2

LASSO classification accuracy

A total of 1,198 transcripts had a nonzero coefficient in at least 1 LASSO regression model. To investigate which of the 1,198 RNA transcripts were most contributing to HNSCC classification, we compared the observed frequency of transcripts with a nonzero coefficient in all selected 1,000 transcript panels to the frequency expected by random selection. In total, 316 transcripts were significantly overrepresented and are determining the prediction (FDR < 0.001, corresponding to observed frequency of ≥ 16; Supplemental Figure 4). The most predictive transcript was AKT1, which was selected in 950 of 1,000 iterations. Overexpression of AKT1 in HNSCC has been reported, and mutational activation of the PI3K/AKT/mTOR pathway is known to increase tumor cell survival (23–25). STRING analysis of all 316 transcripts showed enrichment for 41 biological processes (FDR < 0.05; Supplemental Table 2), which mainly comprised metabolic processes.

Evaluation of HPV infection on TEP RNA signatures. Oropharyngeal SCC (OPSCC) is either caused by exogenous carcinogens or HPV infection. It has been shown that platelets can take up genetic material from other cells, including mutation-harboring fragments from tumor cells (26). We hypothesized that transcripts originating from HPV could contribute to the discriminative power of TEP RNA between cases and controls. Therefore, the sequenced reads of all patients with OPSCC were mapped against the HPV genome, which yielded no hits in both the HPV⁺ and HPV^– groups, indicating that HPV-derived RNA transcripts could not be detected in TEP samples of OPSCC patients, at least not with this sequencing coverage. Of note, like human RNA, HPV RNA has a poly-A tail and is expected to be amplified during sample work-up, if present. To evaluate the potential effect of HPV infection on TEP RNA signatures in OPSCC, we compared the LASSO classification accuracy and found that HPV⁺ and HPV^– patients with OPSCC were classified with fairly comparable accuracies of 80% and 86%, respectively (Table 2), suggesting that HPV status is not of key relevance in the current study design. A LASSO based model trained to predict HPV-status resulted in a median AUC of 0.62 (95% CI, 0.35–0.89) and median accuracy of 58% after 1,000 iterations (Supplemental Figure 5), confirming that TEP profiles are not HPV specific. To assess the possibility that this poor prediction was due to the limited sample size, we developed a LASSO model on the primary cohort, randomly downsized to a sample size comparable with the OPSCC subcohort (n = 60; 30 HNSCC and 30 matched controls). We were still able to reach a median AUC of 0.78 (95% CI, 0.53–0.89) with a median accuracy of 72%. These results indicate that an HPV-specific effect on TEP signatures, if any, is limited when compared with the presence or absence of HNSCC and undetectable with our current classification methods.

Discussion

This study aimed to investigate TEP RNA profiles as a biomarker for the diagnosis of HNSCC. We analyzed a well-annotated cohort of 101 HNSCC cases containing all HNSCC subsites and all stages, propensity score matched to an equal number of noncancer controls. Matched variables included age and the classical risk factors for developing HNSCC: smoking status, smoking pack years, and alcohol use. Notably, age and smoking have previously been described to be of significant statistical contribution to TEP RNA classifiers in nonsmall cell lung cancer, underlining the importance of matching to account for the possible confounding effects of these clinical variables (14).

From a technical perspective, the use of SVM algorithms for classification tasks is well established in various fields, including cancer genomics. In this study, we identified a significant limitation: the selected gene panels may heavily depend on the distribution of samples between the training, evaluation, and validation sets. This is exemplified by the 3 SVM algorithms, which yielded different gene panels with small overlap. This could be further exaggerated by the funneling effect of PSO, which selects features based on previous results, potentially pushing, especially smaller training sets, into a specific gene panel selection. Overfitting of the model remains a critical pitfall, especially with smaller datasets. Despite this high variance in feature panel, the performance of all models was excellent, with AUCs ranging from 0.84 to 0.89, indicating the power of the TEP profiling approach. We sought to overcome the effect of sample distribution on gene panel selection by subjecting our data to a LASSO classification algorithm, enabling us to merge the training and evaluation sets into a larger training set. LASSO requires less computational power, allowing the algorithm to be iterated 1,000 times in a much shorter time with fewer cores compared with PSO-SVM. Each iteration used a different random sample distribution, which resulted in the development of 1,000 LASSO classification models reaching a median AUC of 0.84. In all LASSO iterations, 1,198 transcripts were selected, of which 316 had the most discriminating potential in the overall cohort.

From a biological perspective, it is remarkable that there was no difference between HPV⁺ and HPV^– tumors, and TEP profiles were able to predict the presence of HNSCC equally well for HPV⁺ and HPV^– cases, a perception that certainly supports clinical translation. The classification performance was affected by disease stage, which is a common finding within liquid biopsy prediction models (27). Tumor volumes have an effect on the changes in the blood, both for circulating tumor DNA and for TEPs. The precise mechanisms by which tumors affect platelets, leading to their activation, is unknown. This could relate to aberrant metabolism, hypoxia, changes in exosomes, or direct tumor-platelet interaction (28). The observation that the LASSO-selected and best predicting transcripts relate to metabolic processes might be indicative of hypoxia or metabolic influences. Still, future in vitro experiments would be required to study this in more detail. Irrespective of the precise molecular background, the HNSCC diagnostic performance of TEP profiling is encouraging (19). In suspected cases, blood samples can be drawn in primary care, where the diagnosis can be made, and patients can be referred. Not only might screening approaches in high-risk populations be of value, but disease monitoring is also an option. Ideally, liquid biopsy analyses would facilitate detection of relapses at the earliest possible stage, which would improve the possibilities for curative salvage treatment and patient survival. In patients with glioblastoma, this potential has already been demonstrated: TEP tumor scores reflected glioblastoma burden and could be used to differentiate between true and false positive progression (16).

A strength of the current study is that controls were selected from a pool of spouses, meaning that blood samples were drawn and processed at the exact moment in the same laboratory, reducing the potential confounding effect of laboratories on the resulting platelet RNA profiles. A limitation, as a consequence of the study design, is that we could not match the data for sex and, therefore, computationally had to correct the data for sex by employing RUV. Ideally, the control group would be sex matched, eliminating the need for bioinformatic corrections. Secondly, as a consequence of the inclusion protocol, 35% of cases were early stage HNSCC. While most clinical advantage would be gained in early detection of HNSCC, requiring early-stage samples for optimal classification algorithm development, our cohort reflected the distribution of disease stage at primary clinical presentation. Follow-up studies focusing on early-stage HNSCC solely should further explore the use of TEPs for early-stage detection. Finally, when studying the role of HPV, our subcohort of OPSCC was relatively small, which could have affected model performance and underestimation of the model.

In conclusion, we successfully analyzed platelet RNA from a matched cohort. We showed that 2 independent classification strategies were able to differentiate patients with HNSCC from matched cancer-free controls with promising accuracy. For future clinical implementation including monitoring for recurrent disease, early-stage classification development and large-scale longitudinal series are required.

Methods

Sex as a biological variable. Our study cohort included male and female participants. Data were corrected for sex effect.

Study participants. The study cohort included platelet samples from 107 patients with HNSCC and 107 cancer-free controls propensity score matched for age, smoking status, smoking pack years, and alcohol use, variables known to influence platelet profiles (14, 20, 21). All patients had a histologically confirmed diagnosis of primary HNSCC and were treatment-naive at the time of sample collection. Patients were staged according to the seventh edition of the American Joint Committee on Cancer staging manual, as patients were enrolled while the seventh edition was in use. The control pool consisted of patients’ spouses to minimize confounding variables that could affect classification results, such as the time of blood collection, on-bench storage time before processing, reagent, and sequencing batches. For HPV analyses, we compiled a cohort of patients with OPSCC from the matched cohort plus a surplus of 15 patients with OPSCC that had been TEP profiled but could not be included in the matched case-control cohort due to lack of a matched control. This resulted in a total of 55 patients with OPSCC, of which n = 25 were HPV^– and n = 30 were HPV⁺. Samples were obtained from the Head and Neck Cancer Collection (HNcol) study and the Netherlands Quality of life and Biomedical Cohort (NET-QUBIC) (29).

Matching. The MatchIt (v4.5.0) package was used for propensity score matching (30). Covariates age, smoking status, smoking packyears, and alcohol use were included in a logistic regression model, and controls were matched to cases in a 1:1 ratio using the nearest-neighbor method with a caliper of 0.2. There remained a sex imbalance that was corrected for in the analyses.

Sample work-up and data processing. Blood sampling, platelet RNA isolation, library preparation, sequencing, and data processing were performed as published by Best et al. (22). In brief, raw sequencing data were prepared for mapping using Trimmomatic (v0.22) and mapped using STAR (v2.3.0) (31, 32). Picard tools was used for adding read groups and SAMtools (v1.15.1) for .sam-to-.bam file conversion and data sorting, after which intron-spanning reads were selected to exclude reads from contaminating DNA molecules. Reads were summarized using HTSeq (v0.6.1) (33). To reduce noise, only those transcripts with sufficient read coverage (>30 reads in >90% of the samples) were selected for algorithm development. Samples with < 750 transcripts were excluded, as these would not have reads for all transcripts selected for the algorithm (n = 3). Data were subsequently subjected to trimmed mean of M values (TMM) normalization, and RUV correction to correct for library size and sex (34, 35). Samples with suboptimal coverage (<1 mean reads per gene) were also excluded (n = 3).

Classification model development. Classification models were developed using either a SVM or a LASSO approach. For the learned SVM, differential expression analysis was performed to compare RNA expression levels between HNSCC samples and matched cancer-free controls using 2-way ANOVA, and the total number of features was reduced by selecting the most contributive transcripts on basis of their FDR. Subsequently, the reduced biomarker panel was used for the SVM classification algorithm. The SVM was trained on a training series (40% of samples), after which performance was evaluated in an independent evaluation series (30% of samples). This process was repeated, allowing PSO of SVM parameters and selection of the most contributive transcripts, optimizing the AUC of the evaluation series after each iteration. The settings yielding the highest AUC in the evaluation set were subsequently locked and used to validate the classifier in an independent validation series (remaining 30% of samples).

For LASSO, the R-package penalized (v0.9-52) was used (36). A training/internal validation series (66% of samples) was selected to perform logistic regression to estimate regression coefficients for all transcripts that remained after preprocessing, but without FDR threshold preselection. LASSO penalizes the regression coefficients toward zero, effectively removing transcripts least associated with the outcome of interest and thereby reducing the biomarker panel. The penalty is controlled by tuning parameter λ, which was optimized for each iteration toward a maximum AUC using a 10-fold cross validation. All transcripts with a nonzero coefficient were subsequently used for the prediction of an external validation series (remaining 34% of samples).

STRING analysis. STRING v.12 (www.string-db.org) analysis was performed to identify enrichment for biological processes among the genes most contributive to TEP based HNSCC classification models (37).

Statistics. All statistical analyses were performed using R (version 4.4.3; http://www.Rproject.org). For comparison of group characteristics, 2-tailed parametric t tests were used, and the Wilcoxon Mann-Whitney test when variables were not normally distributed. Two-way ANOVA was used for differential expression analysis. P < 0.05 was considered statistically significant. SVM and LASSO model performance were summarized in a receiver operating characteristic (ROC) curve.

Study approval. The NET-QUBIC and HNcol studies were approved by the IRB of Amsterdam UMC, location Vrije Universiteit Medical Center: NET-QUBIC 2013.301(A2018.307)-NL45051.029.13; HNcol 2008/71- NL22230.029.08. All patients provided written informed consent.

Data availability. Clinical information on patients and controls and the raw sequencing data are available through the NET-QUBIC data repository upon request through https://data.onderzoek.io/kubus/ A step-by-step wet and dry lab protocol was previously published by Best et al. (22). Additional R code to perform LASSO classification is available through https://github.com/nielsevert/TEP; commit ID 8c59a51. Supporting data values for all figures are provided in the Supporting Data Values file.

Author contributions

Conceptualization was contributed by EB, TW, and RHB. Methodology was contributed by TW, SGJGIV, MGB, EP, WNV, and TK. Formal analysis was contributed by NEW, JBP, SWM, WNV, and TK. Resources were contributed by RJB, CHJT, RPT, JAL, IMVL, FL, and CRL. Writing of the original draft was contributed by NEW and RHB. Review and editing were contributed by NEW, JBP, SGJGIV, EP, SWM, MGB, WNV, TK, RJB, CHJT, RPT, JAL, IMVL, FL, CRL, EB, TW, and RBR. Visualization was contributed by NEW. Supervision was contributed by RHB. Project administration was contributed by NEW and RBR. Funding acquisition was contributed by EB, TW, and RHB.

Footnotes

Conflict of interest: TW is the inventor of relevant patents. TW is shareholder of GRAIL Inc. CRL is on the global advisory board of Merck & Co. Inc. and Rakuten Medical advisory board. RHB is on the advisory board of Nanobiotix, received grants from GenMab BV, and has collaborations with Orfenix BV and Qualix-DoT.

Reference information: JCI Insight. 2026;11(2):e186680.https://doi.org/10.1172/jci.insight.186680.

References

Bray F, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–263.
View this article via: PubMed CrossRef Google Scholar

Leemans CR, et al. The molecular landscape of head and neck cancer. Nat Rev Cancer. 2018;18(5):269–282.
View this article via: CrossRef PubMed Google Scholar

Marur S, Forastiere AA. Head and neck squamous cell carcinoma: update on epidemiology, diagnosis, and treatment. Mayo Clin Proc. 2016;91(3):386–396.
View this article via: CrossRef PubMed Google Scholar

Leung AS, et al. Optimal timing of first posttreatment FDG PET/CHJT in head and neck squamous cell carcinoma. Head Neck. 2016;38 Suppl 1:E853–E858.
View this article via: PubMed CrossRef Google Scholar

Alho OP, et al. Head and neck cancer in primary care: presenting symptoms and the effect of delayed diagnosis of cancer cases. CMAJ. 2006;174(6):779–784.
View this article via: CrossRef PubMed Google Scholar

Heitzer E, et al. Current and future perspectives of liquid biopsies in genomics-driven oncology. Nat Rev Genet. 2019;20(2):71–88.
View this article via: CrossRef PubMed Google Scholar

Best MG, et al. Tumor-educated platelets as a noninvasive biomarker source for cancer detection and progression monitoring. Cancer Res. 2018;78(13):3407–3412.
View this article via: CrossRef PubMed Google Scholar

Best MG, et al. Platelet RNA as a circulating biomarker trove for cancer diagnostics. J Thromb Haemost. 2017;15(7):1295–1306.
View this article via: CrossRef PubMed Google Scholar

Labelle M, et al. Direct signaling between platelets and cancer cells induces an epithelial-mesenchymal-like transition and promotes metastasis. Cancer Cell. 2011;20(5):576–590.
View this article via: CrossRef PubMed Google Scholar

Buergy D, et al. Tumor-platelet interaction in solid tumors. Int J Cancer. 2012;130(12):2747–2760.
View this article via: CrossRef PubMed Google Scholar

In ‘t Veld S, Wurdinger T. Tumor-educated platelets. Blood. 2019;133(22):2359–2364.
View this article via: CrossRef PubMed Google Scholar

Sol N, Wurdinger T. Platelet RNA signatures for the detection of cancer. Cancer Metastasis Rev. 2017;36(2):263–272.
View this article via: CrossRef PubMed Google Scholar

Best MG, et al. RNA-Seq of tumor-educated platelets enables blood-based pan-cancer, multiclass, and molecular pathway cancer diagnostics. Cancer Cell. 2015;28(5):666–676.
View this article via: CrossRef PubMed Google Scholar

Best MG, et al. Swarm intelligence-enhanced detection of non-small-cell lung cancer using tumor-educated platelets. Cancer Cell. 2017;32(2):238–252.
View this article via: CrossRef PubMed Google Scholar

Sheng M, et al. Identification of tumor-educated platelet biomarkers of non-small-cell lung cancer. Onco Targets Ther. 2018;11:8143–8151.
View this article via: CrossRef PubMed Google Scholar

Sol N, et al. Tumor-educated platelet RNA for the detection and (pseudo)progression monitoring of glioblastoma. Cell Rep Med. 2020;1(7):100101.
View this article via: CrossRef PubMed Google Scholar

Heinhuis KM, et al. RNA-Sequencing of tumor-educated platelets, a novel biomarker for blood-based sarcoma diagnostics. Cancers (Basel). 2020;12(6):1372.
View this article via: CrossRef PubMed Google Scholar

Lukasiewicz M, et al. Diagnostic accuracy of liquid biopsy in endometrial cancer. Cancers (Basel). 2021;13(22):5731.
View this article via: CrossRef PubMed Google Scholar

In ’t Veld SGJG, et al. Detection and localization of early- and late-stage cancers using platelet RNA. Cancer Cell. 2022;40(9):999–1009.
View this article via: CrossRef PubMed Google Scholar

Simon LM, et al. Human platelet microRNA-mRNA networks associated with age and gender revealed by integrated plateletomics. Blood. 2014;123(16):e37–e45.
View this article via: CrossRef PubMed Google Scholar

Corbin LJ, et al. Epigenetic regulation of F2RL3 associates with myocardial infarction and platelet function. Circ Res. 2022;130(3):384–400.
View this article via: CrossRef PubMed Google Scholar

Best MG, et al. RNA sequencing and swarm intelligence-enhanced classification algorithm development for blood-based disease diagnostics using spliced blood platelet RNA. Nat Protoc. 2019;14(4):1206–1234.
View this article via: CrossRef PubMed Google Scholar

Amornphimoltham P, et al. Persistent activation of the Akt pathway in head and neck squamous cell carcinoma: a potential target for UCN-01. Clin Cancer Res. 2004;10(12 pt 1):4029–4037.
View this article via: CrossRef PubMed Google Scholar

Garcia-Carracedo D, et al. Impact of PI3K/AKT/mTOR pathway activation on the prognosis of patients with head and neck squamous cell carcinomas. Oncotarget. 2016;7(20):29780–29793.
View this article via: CrossRef PubMed Google Scholar

Bussink J, et al. Activation of the PI3-K/AKT pathway and implications for radioresistance mechanisms in head and neck cancer. Lancet Oncol. 2008;9(3):288–296.
View this article via: CrossRef PubMed Google Scholar

Nilsson RJ, et al. Rearranged EML4-ALK fusion transcripts sequester in circulating blood platelets and enable blood-based crizotinib response monitoring in non-small-cell lung cancer. Oncotarget. 2016;7(1):1066–1075.
View this article via: CrossRef PubMed Google Scholar

Bettegowda C, et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci Transl Med. 2014;6(224):224ra24.
View this article via: CrossRef PubMed Google Scholar

D’Ambrosi S, et al. Platelets and tumor-associated RNA transfer. Blood. 2021;137(23):3181–3191.
View this article via: CrossRef PubMed Google Scholar

Verdonck-de Leeuw IM, et al. Advancing interdisciplinary research in head and neck cancer through a multicenter longitudinal prospective cohort study: the NETherlands QUality of life and BIomedical Cohort (NET-QUBIC) data warehouse and biobank. BMC Cancer. 2019;19(1):765.
View this article via: CrossRef PubMed Google Scholar

Ho D, et al. MatchIt: Nonparametric preprocessing for parametric causal inference. J Stat Softw. 2011;42(8):1–28.
View this article via: CrossRef Google Scholar

Bolger AM, et al. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30(15):2114–2120.
View this article via: CrossRef PubMed Google Scholar

Dobin A, et al. STAR: ultrafast universal RNA-Seq aligner. Bioinformatics. 2013;29(1):15–21.
View this article via: CrossRef PubMed Google Scholar

Anders S, et al. HTSeq--a Python framework to work with high-throughput sequencing data. Bioinformatics. 2015;31(2):166–169.
View this article via: CrossRef PubMed Google Scholar

Robinson MD, Oshlack A. A scaling normalization method for differential expression analysis of RNA-Seq data. Genome Biol. 2010;11(3):R25.
View this article via: CrossRef PubMed Google Scholar

Risso D, et al. Normalization of RNA-Seq data using factor analysis of control genes or samples. Nat Biotechnol. 2014;32(9):896–902.
View this article via: CrossRef PubMed Google Scholar

Goeman JJ. L1 penalized estimation in the Cox proportional hazards model. Biom J. 2010;52(1):70–84.
View this article via: CrossRef PubMed Google Scholar

Szklarczyk D, et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 2019;47(d1):D607–D613.
View this article via: CrossRef PubMed Google Scholar

[1] Bray F, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–263.
View this article via: PubMed CrossRef Google Scholar

[2] Leemans CR, et al. The molecular landscape of head and neck cancer. Nat Rev Cancer. 2018;18(5):269–282.
View this article via: CrossRef PubMed Google Scholar

[3] Marur S, Forastiere AA. Head and neck squamous cell carcinoma: update on epidemiology, diagnosis, and treatment. Mayo Clin Proc. 2016;91(3):386–396.
View this article via: CrossRef PubMed Google Scholar

[4] Leung AS, et al. Optimal timing of first posttreatment FDG PET/CHJT in head and neck squamous cell carcinoma. Head Neck. 2016;38 Suppl 1:E853–E858.
View this article via: PubMed CrossRef Google Scholar

[5] Alho OP, et al. Head and neck cancer in primary care: presenting symptoms and the effect of delayed diagnosis of cancer cases. CMAJ. 2006;174(6):779–784.
View this article via: CrossRef PubMed Google Scholar

[6] Heitzer E, et al. Current and future perspectives of liquid biopsies in genomics-driven oncology. Nat Rev Genet. 2019;20(2):71–88.
View this article via: CrossRef PubMed Google Scholar

[7] Best MG, et al. Tumor-educated platelets as a noninvasive biomarker source for cancer detection and progression monitoring. Cancer Res. 2018;78(13):3407–3412.
View this article via: CrossRef PubMed Google Scholar

[8] Best MG, et al. Platelet RNA as a circulating biomarker trove for cancer diagnostics. J Thromb Haemost. 2017;15(7):1295–1306.
View this article via: CrossRef PubMed Google Scholar

[9] Labelle M, et al. Direct signaling between platelets and cancer cells induces an epithelial-mesenchymal-like transition and promotes metastasis. Cancer Cell. 2011;20(5):576–590.
View this article via: CrossRef PubMed Google Scholar

[10] Buergy D, et al. Tumor-platelet interaction in solid tumors. Int J Cancer. 2012;130(12):2747–2760.
View this article via: CrossRef PubMed Google Scholar

[11] In ‘t Veld S, Wurdinger T. Tumor-educated platelets. Blood. 2019;133(22):2359–2364.
View this article via: CrossRef PubMed Google Scholar

[12] Sol N, Wurdinger T. Platelet RNA signatures for the detection of cancer. Cancer Metastasis Rev. 2017;36(2):263–272.
View this article via: CrossRef PubMed Google Scholar

[13] Best MG, et al. RNA-Seq of tumor-educated platelets enables blood-based pan-cancer, multiclass, and molecular pathway cancer diagnostics. Cancer Cell. 2015;28(5):666–676.
View this article via: CrossRef PubMed Google Scholar

[14] Best MG, et al. Swarm intelligence-enhanced detection of non-small-cell lung cancer using tumor-educated platelets. Cancer Cell. 2017;32(2):238–252.
View this article via: CrossRef PubMed Google Scholar

[15] Sheng M, et al. Identification of tumor-educated platelet biomarkers of non-small-cell lung cancer. Onco Targets Ther. 2018;11:8143–8151.
View this article via: CrossRef PubMed Google Scholar

[16] Sol N, et al. Tumor-educated platelet RNA for the detection and (pseudo)progression monitoring of glioblastoma. Cell Rep Med. 2020;1(7):100101.
View this article via: CrossRef PubMed Google Scholar

[17] Heinhuis KM, et al. RNA-Sequencing of tumor-educated platelets, a novel biomarker for blood-based sarcoma diagnostics. Cancers (Basel). 2020;12(6):1372.
View this article via: CrossRef PubMed Google Scholar

[18] Lukasiewicz M, et al. Diagnostic accuracy of liquid biopsy in endometrial cancer. Cancers (Basel). 2021;13(22):5731.
View this article via: CrossRef PubMed Google Scholar

[19] In ’t Veld SGJG, et al. Detection and localization of early- and late-stage cancers using platelet RNA. Cancer Cell. 2022;40(9):999–1009.
View this article via: CrossRef PubMed Google Scholar

[20] Simon LM, et al. Human platelet microRNA-mRNA networks associated with age and gender revealed by integrated plateletomics. Blood. 2014;123(16):e37–e45.
View this article via: CrossRef PubMed Google Scholar

[21] Corbin LJ, et al. Epigenetic regulation of F2RL3 associates with myocardial infarction and platelet function. Circ Res. 2022;130(3):384–400.
View this article via: CrossRef PubMed Google Scholar

[22] Best MG, et al. RNA sequencing and swarm intelligence-enhanced classification algorithm development for blood-based disease diagnostics using spliced blood platelet RNA. Nat Protoc. 2019;14(4):1206–1234.
View this article via: CrossRef PubMed Google Scholar

[23] Amornphimoltham P, et al. Persistent activation of the Akt pathway in head and neck squamous cell carcinoma: a potential target for UCN-01. Clin Cancer Res. 2004;10(12 pt 1):4029–4037.
View this article via: CrossRef PubMed Google Scholar

[24] Garcia-Carracedo D, et al. Impact of PI3K/AKT/mTOR pathway activation on the prognosis of patients with head and neck squamous cell carcinomas. Oncotarget. 2016;7(20):29780–29793.
View this article via: CrossRef PubMed Google Scholar

[25] Bussink J, et al. Activation of the PI3-K/AKT pathway and implications for radioresistance mechanisms in head and neck cancer. Lancet Oncol. 2008;9(3):288–296.
View this article via: CrossRef PubMed Google Scholar

[26] Nilsson RJ, et al. Rearranged EML4-ALK fusion transcripts sequester in circulating blood platelets and enable blood-based crizotinib response monitoring in non-small-cell lung cancer. Oncotarget. 2016;7(1):1066–1075.
View this article via: CrossRef PubMed Google Scholar

[27] Bettegowda C, et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci Transl Med. 2014;6(224):224ra24.
View this article via: CrossRef PubMed Google Scholar

[28] D’Ambrosi S, et al. Platelets and tumor-associated RNA transfer. Blood. 2021;137(23):3181–3191.
View this article via: CrossRef PubMed Google Scholar

[29] Verdonck-de Leeuw IM, et al. Advancing interdisciplinary research in head and neck cancer through a multicenter longitudinal prospective cohort study: the NETherlands QUality of life and BIomedical Cohort (NET-QUBIC) data warehouse and biobank. BMC Cancer. 2019;19(1):765.
View this article via: CrossRef PubMed Google Scholar

[30] Ho D, et al. MatchIt: Nonparametric preprocessing for parametric causal inference. J Stat Softw. 2011;42(8):1–28.
View this article via: CrossRef Google Scholar

[31] Bolger AM, et al. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30(15):2114–2120.
View this article via: CrossRef PubMed Google Scholar

[32] Dobin A, et al. STAR: ultrafast universal RNA-Seq aligner. Bioinformatics. 2013;29(1):15–21.
View this article via: CrossRef PubMed Google Scholar

[33] Anders S, et al. HTSeq--a Python framework to work with high-throughput sequencing data. Bioinformatics. 2015;31(2):166–169.
View this article via: CrossRef PubMed Google Scholar

[34] Robinson MD, Oshlack A. A scaling normalization method for differential expression analysis of RNA-Seq data. Genome Biol. 2010;11(3):R25.
View this article via: CrossRef PubMed Google Scholar

[35] Risso D, et al. Normalization of RNA-Seq data using factor analysis of control genes or samples. Nat Biotechnol. 2014;32(9):896–902.
View this article via: CrossRef PubMed Google Scholar

[36] Goeman JJ. L1 penalized estimation in the Cox proportional hazards model. Biom J. 2010;52(1):70–84.
View this article via: CrossRef PubMed Google Scholar

[37] Szklarczyk D, et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 2019;47(d1):D607–D613.
View this article via: CrossRef PubMed Google Scholar

Diagnosis of head and neck cancer by AI-based tumor-educated platelet RNA profiling of liquid biopsies

Article tools

Metrics

Go to

Diagnosis of head and neck cancer by AI-based tumor-educated platelet RNA profiling of liquid biopsies

Article tools

Metrics

Go to

Sign up for email alerts