RNA-Seq of Tumor-Educated Platelets EnablesBlood-Based Pan-Cancer, Multiclass,and Molecular Pathway

Article RNA-Seq of Tumor-Educated Platelets Enables Blood-Based Pan-Cancer,Multiclass,and Molecular Pathway Cancer Diagnostics

Graphical Abstract

Highlights

d Tumors‘‘educate’’platelets(TEPs)by altering th

e platelet

RNA pro?le

d TEPs provid

e a RNA biosource for pan-cancer,multiclass,

and companion diagnostics

d TEP-based liquid biopsies may guid

e clinical diagnostics and

therapy selection

d A total of100–500pg of total platelet RNA is suf?cient for

TEP-based diagnostics Authors

Myron G.Best,Nik Sol,Irsan Kooi,..., Bakhos A.Tannous,Pieter Wesseling, Thomas Wurdinger

Correspondence

t.wurdinger@vumc.nl

In Brief

Best et al.show that mRNA sequencing of tumor-educated blood platelets distinguishes cancer patients from healthy individuals with96%accuracy, differentiates between six primary tumor types of patients with71%accuracy,and identi?es several genetic alterations found in tumors.

Accession Numbers

GSE68086 Best et al.,2015,Cancer Cell28,666–676

November9,2015a2015The Authors

https://www.wendangku.net/doc/7911719864.html,/10.1016/https://www.wendangku.net/doc/7911719864.html,ell.2015.09.018

Cancer Cell

Article

RNA-Seq of Tumor-Educated Platelets Enables Blood-Based Pan-Cancer,Multiclass,

and Molecular Pathway Cancer Diagnostics

Myron G.Best,1,2Nik Sol,3Irsan Kooi,4Jihane Tannous,5Bart A.Westerman,2Franc?ois Rustenburg,1,2Pepijn Schellen,2,6 Heleen Verschueren,2,6Edward Post,2,6Jan Koster,7Bauke Ylstra,1Najim Ameziane,4Josephine Dorsman,4

Egbert F.Smit,8Henk M.Verheul,9David P.Noske,2Jaap C.Reijneveld,3R.Jonas A.Nilsson,2,6,10Bakhos A.Tannous,5,12 Pieter Wesseling,1,11,12and Thomas Wurdinger2,5,6,12,*

1Department of Pathology,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam,

the Netherlands

2Department of Neurosurgery,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam,

the Netherlands

3Department of Neurology,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam,

the Netherlands

4Department of Clinical Genetics,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam, the Netherlands

5Department of Neurology,Massachusetts General Hospital and Neuroscience Program,Harvard Medical School,14913th Street, Charlestown,MA02129,USA

6thromboDx B.V.,1098EA Amsterdam,the Netherlands

7Department of Oncogenomics,Academic Medical Center,Meibergdreef9,1105AZ Amsterdam,the Netherlands

8Department of Pulmonary Diseases,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam, the Netherlands

9Department of Medical Oncology,VU University Medical Center,Cancer Center Amsterdam,De Boelelaan1117,1081HV Amsterdam, the Netherlands

10Department of Radiation Sciences,Oncology,Umea?University,90185Umea?,Sweden

11Department of Pathology,Radboud University Medical Center,6500HB Nijmegen,the Netherlands

12Co-senior author

*Correspondence:t.wurdinger@vumc.nl

https://www.wendangku.net/doc/7911719864.html,/10.1016/https://www.wendangku.net/doc/7911719864.html,ell.2015.09.018

This is an open access article under the CC BY license(https://www.wendangku.net/doc/7911719864.html,/licenses/by/4.0/).

SUMMARY

Tumor-educated blood platelets(TEPs)are implicated as central players in the systemic and local responses to tumor growth,thereby altering their RNA pro?le.We determined the diagnostic potential of TEPs by mRNA sequencing of283platelet samples.We distinguished228patients with localized and metastasized tumors from55healthy individuals with96%accuracy.Across six different tumor types,the location of the primary tumor was correctly identi?ed with71%accuracy.Also,MET or HER2-positive,and mutant KRAS,EGFR,or PIK3CA tumors were accurately distinguished using surrogate TEP mRNA pro?les.Our results indicate that blood platelets provide a valuable platform for pan-cancer,multiclass cancer,and companion diagnostics, possibly enabling clinical advances in blood-based‘‘liquid biopsies’’.

INTRODUCTION

Cancer is primarily diagnosed by clinical presentation,radiology, biochemical tests,and pathological analysis of tumor tissue,increasingly supported by molecular diagnostic tests.Molecular pro?ling of tumor tissue samples has emerged as a potential cancer classifying method(Akbani et al.,2014;Golub et al., 1999;Han et al.,2014;Hoadley et al.,2014;Kandoth et

al.,

666Cancer Cell28,666–676,November9,2015a2015The Authors

2013;Ramaswamy et al.,2001;Su et al.,2001).In order to over-come limitations of tissue acquisition,the use of blood-based liquid biopsies has been suggested(Alix-Panabie`res et al., 2012;Crowley et al.,2013;Haber and Velculescu,2014).Several blood-based biosources are currently being evaluated as liquid biopsies,including plasma DNA(Bettegowda et al.,2014; Chan et al.,2013;Diehl et al.,2008;Murtaza et al.,2013; Newman et al.,2014;Thierry et al.,2014)and circulating tumor cells(Bidard et al.,2014;Dawson et al.,2013;Maheswaran et al.,2008;Rack et al.,2014).So far,implementation of liquid biopsies for early detection of cancer has been hampered by non-speci?city of these biosources to pinpoint the nature of the primary tumor(Alix-Panabie`res and Pantel,2014;Bette-gowda et al.,2014).

It has been reported that tumor-educated platelets(TEPs)may enable blood-based cancer diagnostics(Calverley et al.,2010; McAllister and Weinberg,2014;Nilsson et al.,2011).Blood platelets—the second most-abundant cell type in peripheral blood—are circulating anucleated cell fragments that originate from megakaryocytes in bone marrow and are traditionally known for their role in hemostasis and initiation of wound healing (George,2000;Leslie,2010).More recently,platelets have emerged as central players in the systemic and local responses to tumor growth.Confrontation of platelets with tumor cells via transfer of tumor-associated biomolecules(‘‘education’’)is an emerging concept and results in the sequestration of such biomolecules(Klement et al.,2009;Kuznetsov et al.,2012; McAllister and Weinberg,2014;Nilsson et al.,2011;Quail and Joyce,2013).Moreover,external stimuli,such as activation of platelet surface receptors and lipopolysaccharide-mediated platelet activation(Denis et al.,2005;Rondina et al.,2011),induce speci?c splicing of pre-mRNAs in circulating platelets(Power et al.,2009;Rowley et al.,2011;Schubert et al.,2014).Platelets may also undergo queue-speci?c splice events in response to signals released by cancer cells and the tumor microenviron-ment—such as stromal and immune cells.The combination of speci?c splice events in response to external signals and the capacity of platelets to directly ingest(spliced)circulating mRNA can provide TEPs with a highly dynamic mRNA repertoire, with potential applicability to cancer diagnostics(Calverley et al., 2010;Nilsson et al.,2011)(Figure1A).In this study,we charac-terize the platelet mRNA pro?les of various cancer patients and healthy donors and investigate their potential for TEP-based pan-cancer,multiclass cancer,and companion diagnostics.

RESULTS

mRNA Pro?les of Tumor-Educated Platelets Are Distinct from Platelets of Healthy Individuals

We prospectively collected and isolated blood platelets from healthy donors(n=55)and both treated and untreated patients with early,localized(n=39)or advanced,metastatic cancer (n=189)diagnosed by clinical presentation and pathological analysis of tumor tissue supported by molecular diagnostics tests.The patient cohort included six tumor types,i.e.,non-small cell lung carcinoma(NSCLC,n=60),colorectal cancer(CRC, n=41),glioblastoma(GBM,n=39),pancreatic cancer(PAAD, n=35),hepatobiliary cancer(HBC,n=14),and breast cancer (BrCa,n=39)(Figure1B;Table1;Table S1).The cohort of

healthy donors covered a wide range of ages(21–64years old, Table1).

Platelet purity was con?rmed by morphological analysis of randomly selected and freshly isolated platelet samples (contamination is1to5nucleated cells per10million platelets, see Supplemental Experimental Procedures),and platelet RNA was isolated and evaluated for quality and quantity(Figure S1A).

A total of100–500pg of platelet total RNA(the equivalent of

puri?ed platelets in less than one drop of blood)was used for SMARTer mRNA ampli?cation and sequencing(Ramsko¨ld et al.,2012)(Figures1C and S1A).Platelet RNA sequencing yielded a mean read count of$22million reads per sample.

After selection of intron-spanning(spliced)RNA reads and exclusion of genes with low coverage(see Supplemental Exper-imental Procedures),we detected in platelets of healthy donors (n=55)and localized and metastasized cancer patients (n=228)5,003different protein coding and non-coding RNAs that were used for subsequent analyses.The obtained platelet RNA pro?les correlated with previously reported mRNA pro?les of platelets(Bray et al.,2013;Kissopoulou et al.,2013;Rowley et al.,2011;Simon et al.,2014)and megakaryocytes(Chen et al.,2014)and not with various non-related blood cell mRNA pro?les(Hrdlickova et al.,2014)(Figure S1B).Furthermore, DAVID Gene Ontology(GO)analysis revealed that the detected RNAs are strongly enriched for transcripts associated with blood platelets(false discovery rate[FDR]<10à126).

Among the5,003RNAs,we identi?ed known platelet markers, such as B2M,PPBP,TMSB4X,PF4,and several long non-cod-ing RNAs(e.g.,MALAT1).A total of1,453out of5,003mRNAs were increased and793out of5,003mRNAs were decreased in TEPs as compared to platelet samples of healthy donors (FDR<0.001),while presenting a strong correlation between these platelet mRNA pro?les(r=0.90,Pearson correlation) (Figure1D).Unsupervised hierarchical clustering based on the differentially detected platelet mRNAs distinguished two sample groups with minor overlap(Figure1E;Table S2).DAVID GO anal-ysis revealed that the increased TEP mRNAs were enriched for biological processes such as vesicle-mediated transport and the cytoskeletal protein binding while decreased mRNAs were strongly involved in RNA processing and splicing(Table S3).

A correlative analysis of gene set enrichment(CAGE)GO meth-

odology,in which3,875curated gene sets of the GSEA database were correlated to TEP pro?les(see Experimental Procedures), demonstrated signi?cant correlation of TEP mRNA pro?les with cancer tissue signatures,histone deacetylases regulation,and platelets(Table2).The levels of20non-protein coding RNAs were altered in TEPs as compared to platelets from healthy individuals and these show a tumor type-associated RNA pro?le (Figure S1C).

Next,we determined the diagnostic accuracy of TEP-based pan-cancer classi?cation in the training cohort(n=175),employ-ing a leave-one-out cross-validation support vector machine algorithm(SVM/LOOCV,see Experimental Procedures;Figures S1D and S1E),previously used to classify primary and metastatic tumor tissues(Ramaswamy et al.,2001;Su et al.,2001;Vapnik, 1998;Yeang et al.,2001).Brie?y,the SVM algorithm(blindly)clas-si?es each individual sample as cancer or healthy by comparison to all other samples(175à1)and was performed175times to classify and cross validate all individuals samples.The algorithms Cancer Cell28,666–676,November9,2015a2015The Authors667

we developed use a limited number of different spliced RNAs for sample classi?cation.To determine the speci?c input gene lists for the classifying algorithms we performed ANOVA testing for differences (as implemented in the R-package edgeR),yielding classi?er-speci?c gene lists (Table S4).For the speci?c

algorithm

A B

C

D E

F G

H I

Figure 1.Tumor-Educated Platelet mRNA Pro?ling for Pan-Cancer Diagnostics

(A)Schematic overview of tumor-educated plate-lets (TEPs)as biosource for liquid biopsies.

(B)Number of platelet samples of healthy donors and patients with different types of cancer.

(C)TEP mRNA sequencing (mRNA-seq)work?ow,as starting from 6ml EDTA-coated tubes,to platelet isolation,mRNA ampli?cation,and sequencing.(D)Correlation plot of mRNAs detected in healthy donor (HD)platelets and cancer patients’TEPs,including highlighted increased (red)and decreased (blue)TEP mRNAs.

(E)Heatmap of unsupervised clustering of platelet mRNA pro?les of healthy donors (red)and patients with cancer (gray).

(F)Cross-table of pan-cancer SVM/LOOCV diagnostics of healthy donor subjects and patients with cancer in training cohort (n =175).Indicated are sample numbers and detection rates in percent-ages.

(G)Performance of pan-cancer SVM algorithm in validation cohort (n =108).Indicated are sample numbers and detection rates in percentages.(H)ROC-curve of SVM diagnostics of training (red),validation (blue)cohort,and random classi?ers,indicating the classi?cation accuracies obtained by chance of the training and validation cohort (gray).(I)Total accuracy ratios of SVM classi?cation in ?ve subgroups,including corresponding predictive strengths.Genes,number of mRNAs included in training of the SVM algorithm.

See also Figure S1and Tables S1,S2,S3,and S4.

of the pan-cancer TEP-based classi?er test we selected 1,072RNAs (Table S4)for the n =175training cohort,yielding a sensitivity of 96%,a speci?city of 92%,and an accuracy of 95%(Figure 1F).Sub-sequent validation using a separate vali-dation cohort (n =108),not involved in input gene list selection and training of the algorithm,yielded a sensitivity of 97%,a speci?city of 94%,and an accu-racy of 96%(Figure 1G),with an area un-der the curve (AUC)of 0.986to detect cancer (Figure 1H)and high predictive strength (Figure 1I).In contrast,random classi?ers,as determined by multiple rounds of randomly shuf?ing class labels (permutation)during the SVM training pro-cess (see Experimental Procedures ),had no predictive power (mean overall accu-racy:78%,SD ±0.3%,p <0.01),thereby showing,albeit an unbalanced represen-tation of both groups in the study cohort,

speci?city of our procedure.A total of 100times random class-proportional subsampling of the entire dataset in a training and validation set (ratio 60:40)yielded similar accuracy rates (mean overall accuracy:96%,SD:±2%),con?rming reproducible clas-si?cation accuracy in this dataset.Of note,all 39patients with

668Cancer Cell 28,666–676,November 9,2015a2015The Authors

localized tumors and 33of the 39patients with primary tumors in the CNS were correctly classi?ed as cancer patients (Figure 1I).Visualization of 22genes previously identi?ed at differential RNA levels in platelets of patients with various non-cancerous diseases (Gnatenko et al.,2010;Healy et al.,2006;Lood et al.,2010;Raghavachari et al.,2007),revealed mixed levels in our TEP dataset (Figure S1F),suggesting that the platelet RNA reper-toire in patients with non-cancerous disease is distinct from patients with cancer.

Tumor-Speci?c Educational Program of Blood Platelets Allows for Multiclass Cancer Diagnostics

In addition to the pan-cancer diagnosis,the TEP mRNA pro?les also distinguished healthy donors and patients with speci?c types of cancer,as demonstrated by the unsupervised hierar-chical clustering of differential platelet mRNA levels of healthy donors and all six individual tumor types,i.e.,NSCLC,CRC,GBM,PAAD,BrCa,and HBC (Figures 2A,all p <0.0001,Fisher’s exact test,and S2A;Table S5),and this resulted in tumor-speci?c gene lists that were used as input for training and validation of the tumor-speci?c algorithms (Table S4).For the unsupervised clustering of the all-female group of BrCa patients,male healthy donors were excluded to avoid sample bias due to gender-speci?c platelet mRNA pro?les (Figure S2B).SVM-based classi?cation of all individual tumor classes with healthy donors resulted in clear distinction of both groups in both the training and validation cohort,with high sensitivity and speci?city,and 38/39(97%)cancer patients with localized disease were classi?ed correctly (Figures 2B and S2C).CAGE GO analysis showed that biological processes differed between TEPs of individual tumor types,suggestive of tumor-speci?c ‘‘educational’’programs (Table S6).We did not detect suf?cient differences in mRNA levels to discriminate patients with non-metastasized from patients with metastasized tumors,suggest-ing that the altered platelet pro?le is predominantly in?uenced by the molecular tumor type and,to a lesser extent,by tumor progression and metastases.

We next determined whether we could discriminate three different types of adenocarcinomas in the gastro-intestinal tract by analysis of the TEP-pro?les,i.e.,CRC,PAAD,and HBC.We developed a CRC/PAAD/HBC algorithm that correctly classi?ed the mixed TEP samples (n =90)with an overall accuracy of 76%(mean overall accuracy random classi?ers:42%,SD:±5%,p <0.01,Figure 2C).In order to determine whether the TEP mRNA pro?les allowed for multiclass cancer diagnosis across all tumor types and healthy donors,we extended the SVM/LOOCV classi?cation test using a combination of algorithms that classi?ed each individual sample of the training cohort (n =175)as healthy donor or one of six tumor types (Figures S2D and S2E).The results of the multiclass cancer diagnostics test resulted in an average accuracy of 71%(mean overall accu-racy random classi?ers:19%,SD:±2%,p <0.01,Figure 2D),demonstrating signi?cant multiclass cancer discriminative power in the platelet mRNA pro?les.The classi?cation capacity of the multiclass SVM-based classi?er was con?rmed in the vali-dation cohort of 108samples,with an overall accuracy of 71%(Figure 2E).An overall accuracy of 71%might not be suf?cient for introduction into cancer diagnostics.However,of the initially misclassi?ed samples according to the SVM algorithms choice with strongest classi?cation strength the second ranked classi?-cation was correct in 60%of the cases.This yields an overall accuracy using the combined ?rst and second ranked classi?ca-tions of 89%.The low validation score of HBC samples can be attributed to the relative low number of samples and possibly to the heterogenic nature of this group of cancers (hepatocellular cancers and cholangiocarcinomas).

Companion Diagnostics Tumor Tissue Biomarkers Are Re?ected by Surrogate TEP mRNA Onco-signatures

Blood provides a promising biosource for the detection of com-panion diagnostics biomarkers for therapy selection (Bette-gowda et al.,2014;Crowley et al.,2013;Papadopoulos et al.,2006).We selected platelet samples of patients with distinct therapy-guiding markers con?rmed in matching tumor

tissue.

Table 1.Summary of Patient Characteristics Patient Total (n)Gender M (%)a Age (SD)b Metastasis (%)Presence (%)HBC,hepatobiliary cancer.See also Table S1.a

Indicated are number of male individuals.b

Indicated is mean age in years.

Cancer Cell 28,666–676,November 9,2015a2015The Authors 669

Although the platelet mRNA pro?les contained undetectable or low levels of these mutant biomarkers,the TEP mRNA pro?les did allow to distinguish patients with KRAS mutant tumors from KRAS wild-type tumors in PAAD,CRC,NSCLC,and HBC patients,and EGFR mutant tumors in NSCLC patients,using algorithms speci?cally trained on biomarker-speci?c input gene lists (all p <0.01versus random classi?ers,Figures 3A–3E;Table S4).Even though the number of samples analyzed is relatively low and the risk of algorithm over?tting needs to be taken into account,the TEP pro?les distinguished patients with HER2-ampli?ed,PIK3CA mutant or triple-negative BrCa,and NSCLC patients with MET overexpression (all p <0.01versus random classi?ers,Figures 3F–3I).

We subsequently compared the diagnostic accuracy of the TEP mRNA classi?cation method with a targeted KRAS (exon 12and 13)and EGFR (exon 20and 21)amplicon deep sequencing strategy ($5,0003coverage)on the Illumina Miseq platform using prospectively collected blood samples of patients with localized or metastasized cancer.This method did allow for the detection of individual mutant KRAS and EGFR sequences in both plasma DNA and platelet RNA (Table S7),indicating sequestration and potential education capacity of mutant,tumor-derived RNA biomarkers in TEPs.Mutant KRAS was de-tected in 62%and 39%,respectively,of plasma DNA (n =103,kappa statistics =0.370,p <0.05)and platelet RNA (n =144,kappa statistics =0.213,p <0.05)of patients with a KRAS mutation in primary tumor tissue.The sensitivity of the plasma DNA tests was relatively poor as reported by others (Bettegowda et al.,2014;Thierry et al.,2014),which may partly be attributed to the loss of plasma DNA quality due to relatively long blood sample storage (EDTA blood samples were stored up to 48hr at room temperature before plasma isolation).To discriminate KRAS mutant from wild-type tumors in blood,the TEP mRNA pro?les provided superior concordance with tissue molecular status (kappa statistics =0.795–0.895,p <0.05)compared to KRAS amplicon sequencing analysis of both plasma DNA and platelet RNA (Table S7).Thus,TEP mRNA pro?les can harness potential blood-based surrogate onco-signatures for tumor tissue biomarkers that enable cancer patient strati?cation and therapy selection.

TEP-Pro?les Provide an All-in-One Biosource for Blood-Based Liquid Biopsies in Patients with Cancer

Unequivocal discrimination of primary versus metastatic nature of a tumor may be dif?cult and hamper adequate therapy selection.Since the TEP pro?les closely resemble the different tumor types as determined by their organ of origin—regardless of systemic dissemination—this potentially allows for organ-speci?c cancer diagnostics.Hence we selected all healthy donors and all patients with primary or metastatic tumor burden in the lung (n =154),brain (n =114),or liver (n =127).We per-formed ‘‘organ exams’’and instructed the SVM/LOOCV algo-rithm to determine for lung,brain,and liver the presence or absence of cancer (96%,91%,and 96%accuracy,respec-tively),with cancer subclassi?ed as primary or metastatic tumor (84%,93%,and 90%accuracy,respectively)and in case of metastases to identify the potential organ of origin (64%,70%,and 64%accuracy,respectively).The platelet mRNA pro-?les enabled assignment of the cancer to the different organs with high accuracy (Figure 4).In addition,using the same TEP mRNA pro?les we were able to again indicate the biomarker status of the tumor tissues (90%,82%,and 93%accuracy,respectively)(Figure 4).DISCUSSION

The use of blood-based liquid biopsies to detect,diagnose,and monitor cancer may enable earlier diagnosis of cancer,lower costs by tailoring molecular targeted treatments,improve convenience for cancer patients,and ultimately supplements clinical oncological decision-making.Current blood-based biosources under evaluation demonstrate suboptimal sensi-tivity for cancer diagnostics,in particular in patients with localized disease.So far,none of the current blood-based bio-sources,including plasma DNA,exosomes,and CTCs,have been employed for multiclass cancer diagnostics (Alix-Pana-bie

`res and Pantel,2014;Bettegowda et al.,2014;Skog et al.,2008),hampering its implementation for early cancer detection.Here,we report that molecular interrogation of blood platelet mRNA can offer valuable diagnostics information for all cancer patients analyzed—spanning six different tumor types.Our results suggest that platelets may be employable as

an

Table 2.Pan-Cancer CAGE Gene Ontology

Top 25GO Correlations #

Lowest a

Highest a

Institute curated gene sets.CAGE,Correlative Analysis of Gene Set Enrichment;GO,gene ontology;#,number of hits per annotation;IL,interleukin;HDAC,histone deacetylase.a

Indicated are lowest and highest correlations per annotation.

670Cancer Cell 28,666–676,November 9,2015a2015The Authors

all-in-one biosource to broadly scan for molecular traces of cancer in general and provide a strong indication on tumor type and molecular subclass.This includes patients with local-ized disease possibly allowing for targeted diagnostic con?r-mation using routine clinical diagnostics for each particular tumor type.

Since the discovery of circulating tumor material in blood of patients with cancer(Leon et al.,1977)and the recognition of the clinical utility of blood-based liquid biopsies,a wealth of studies has assessed the use of blood for cancer diagnostics, prognostication and treatment monitoring(Alix-Panabie`res et al.,2012;Bidard et al.,2014;Crowley et al.,2013;Haber

A B

C

D E

Figure2.Tumor-Educated Platelet mRNA Pro?les for Multiclass Cancer Diagnostics

(A)Heatmaps of unsupervised clustering of platelet mRNA pro?les of healthy donors(HD;n=55)(red)and patients with non-small cell lung cancer(NSCLC; n=60),colorectal cancer(CRC;n=41),glioblastoma(GBM;n=39),pancreatic cancer(PAAD,n=35),breast cancer(BrCa;n=39;female HD;n=29),and hepatobiliary cancer(HBC;n=14).

(B)ROC-curve of SVM diagnostics of healthy donors and individual tumor classes in both training(left)and validation(right)cohort.Random classi?ers,indicating the classi?cation accuracies obtained by chance,are shown in gray.

(C)Confusion matrix of multiclass SVM/LOOCV diagnostics of patients with CRC,PAAD,and HBC.Indicated are detection rates as compared to the actual classes in percentages.

(D)Confusion matrix of multiclass SVM/LOOCV diagnostics of the training cohort consisting of healthy donors(healthy)and patients with GBM,NSCLC,PAAD, CRC,BrCa,and HBC.Indicated are detection rates as compared to the actual classes in percentages.

(E)Confusion matrix of multiclass SVM algorithm in a validation cohort(n=108).Indicated are sample numbers and detection rates in percentages.Genes, number of mRNAs included in training of the SVM algorithm.

See also Figure S2and Tables S4,S5,and S6.

Cancer Cell28,666–676,November9,2015a2015The Authors671

and Velculescu,2014).By development of highly sensitive targeted detection methods,such as targeted deep sequencing (Newman et al.,2014),droplet digital PCR (Bettegowda et al.,2014),and allele-speci?c PCR (Maheswaran et al.,2008;Thierry et al.,2014),the utility and applicability of liquid biopsies for clin-ical implementation has accelerated.These advances previously allowed for a pan-cancer comparison of various biosources and revealed that in >75%of cancers,including advanced stage pancreas,colorectal,breast,and ovarian cancer,cell-free DNA is detectable although detection rates are dependent on the grade of the tumor and depth of analysis (Bettegowda et al.,2014).Here,we show that the platelet RNA pro?les are affected in nearly all cancer patients,regardless of the type of tumor,although the abundance of tumor-associated RNAs seems variable among cancer patients.In addition,surrogate RNA onco-signatures of tissue biomarkers,also in 88%of localized KRAS mutant cancer patients as measured by the tumor-spe-ci?c and pan-cancer SVM/LOOCV procedures,are readily available from a minute amount (100–500pg)of platelet RNA.As whole blood can be stored up to 48hr on room temperature prior to isolation of the platelet pellet,while maintaining high-quality RNA and the dominant cancer RNA signatures,TEPs can be more readily implemented in daily clinical laboratory practice and could potentially be shipped prior to further blood sample processing.

Blood platelets are widely involved in tumor growth and can-cer progression (Gay and Felding-Habermann,2011).Platelets sequester solubilized tumor-associated proteins (Klement et al.,2009)and spliced and unspliced mRNAs (Calverley et al.,2010;Nilsson et al.,2011),whereas platelets do also directly interact with tumor cells (Labelle et al.,2011),neutrophils (Sreeramkumar et al.,2014),circulating NK-cells (Palumbo et al.,2005;Placke et al.,2012),and circulating tumor cells (Ting et al.,2014;Yu et al.,2013).Interestingly,in vivo experiments have revealed breast cancer-mediated systemic instigation by sup-plying circulating platelets with pro-in?ammatory and pro-angio-genic proteins,supporting outgrowth of dormant metastatic foci (Kuznetsov et al.,2012).Using a gene ontology methodology,CAGE,we correlated TEP-cancer signatures with publicly avail-able curated datasets.Indeed,we identi?ed widespread correla-tions with cancer tissues,hypoxia,platelet-signatures,and cytoskeleton,possibly re?ecting the ‘‘alert’’and pro-tumorigenic state of TEPs.We observed strong negative correlations with RNAs implicated in RNA translation,T cell immunity,and inter-leukin-signaling,implying diminished needs of TEPs for RNAs involved in these biological processes or orchestrated transla-tion of these RNAs to proteins (Denis et al.,2005).We observed that the tumor-speci?c educational programs in TEPs are pre-dominantly in?uenced by tumor type and,to a lesser extent,by tumor progression and metastases.Although we were not able

A

Actual Class

P r e d i c t e d C l a s s

Accuracy: 95%, p < 0.01

B

Actual Class

P r e d i c t e d C l a s s

Accuracy: 94%, p < 0.01

Actual Class

P r e d i c t e d C l a s s

Accuracy: 85%, p < 0.01 F

Actual Class

P r e d i c t e d C l a s s

Accuracy: 92%, p < 0.01

E

Actual Class

P r e d i c t e d C l a s s

Accuracy: 87%, p < 0.01

G

Actual Class

P r e d i c t e d C l a s s

Accuracy: 91%, p < 0.01

H

Actual Class

P r e d i c t e d C l a s s

Accuracy: 92%, p < 0.01

I

Actual Class

P r e d i c t e d C l a s s

Accuracy: 90%, p < 0.01

Actual Class

P r e d i c t e d C l a s s

Accuracy: 90%, p < 0.01

C

Figure 3.Tumor-Educated Platelet mRNA Pro?les for Molecular Pathway Diagnostics

Cross tables of SVM/LOOCV diagnostics with the molecular markers KRAS in (A)CRC,(B)PAAD,and (C)NSCLC patients,(D)KRAS in the combined cohort of patients with either CRC,PAAD,NSCLC,or HBC,(E)EGFR and (F)MET in NSCLC patients,(G)PIK3CA mutations,(H)HER2-ampli?cation,and (I)triple negative status in BrCa patients.Genes,number of mRNAs included in training of the SVM algorithm.See also Tables S4and S7.

672Cancer Cell 28,666–676,November 9,2015a2015The Authors

to measure signi?cant differences between non-metastasized and metastasized tumors,we do not exclude that the use of larger sample sets could allow for the generation of SVM algo-rithms that do have the power to discriminate between certain stages of cancer,including those with in situ carcinomas and even pre-malignant lesions.In addition,different molecular tumor subtypes (e.g.,HER2-ampli?ed versus wild-type BrCa)result in different effects on the platelet pro?les,possibly caused by different ‘‘educational’’stimuli generated by the different molecular tumor subtypes (Koboldt et al.,2012).Altogether,the RNA content of platelets in patients with cancer is dependent on the transcriptional state of the bone-marrow megakaryocyte (Calverley et al.,2010;McAllister and Weinberg,2014),comple-mented by sequestration of spliced RNA (Nilsson et al.,2011),release of RNA (Clancy and Freedman,2014;Kirschbaum et al.,2015;Rak and Guha,2012;Risitano et al.,2012),and possibly queue-speci?c pre-mRNA splicing during platelet circulation.Partial or complete normalization of the platelet pro-?les following successful treatment of the tumor would enable TEP-based disease recurrence monitoring,requiring the anal-ysis of follow-up platelet samples.Future studies will be required to address the tumor-speci?c ‘‘educated’’pro?les on both an (small non-coding)RNA (Laffont et al.,2013;Landry et al.,2009;Leidinger et al.,2014;Lu et al.,2005)and protein (Burkhart et al.,2014;Geiger et al.,2013;Klement et al.,2009)level and determine the ability of gene ontology,blood-based cancer classi?cation.

In conclusion,we provide robust evidence for the clinical relevance of blood platelets for liquid biopsy-based molecular diagnostics in patients with several types of cancer.Further validation is warranted to determine the potential of surrogate TEP pro?les for blood-based companion diagnostics,therapy selection,longitudinal monitoring,and disease recurrence moni-toring.In addition,we expect the self-learning algorithms to further improve by including signi?cantly more samples.For this approach,isolation of the platelet fraction from whole blood should be performed within 48hr after blood withdrawal,the platelet fraction can subsequently be frozen for cancer diag-nosis.Also,future studies should address causes and antici-

pated risks of outlier samples identi?ed in this study,such as healthy donors classi?ed as cancer patients.Systemic factors such as chronic or transient in?ammatory diseases,or cardio-vascular events and other non-cancerous diseases may also in?uence the platelet mRNA pro?le and require evaluation in follow-up studies,possibly also including individuals predis-posed for cancer.

EXPERIMENTAL PROCEDURES

Sample Collection and Study Oversight

Blood was drawn from all patients and healthy donors at the VU University Medical Center,Amsterdam,the Netherlands,or the Massachusetts General Hospital (MGH),Boston,in 6ml purple-cap BD Vacutainers containing the anti-coagulant EDTA.To minimize effects of long-term storage of platelets at room temperature and loss of platelet RNA quality and quantity,samples were processed within 48hr after blood collection.Blood samples of patients were collected pre-operatively (GBM)or during follow-up in the outpatient clinic (CRC,NSCLC,PAAD,BrCa,HBC).Nine cancer patient samples included were follow-up samples of the same patient collected within months of the ?rst blood collection (?ve samples in NSCLC,two samples in PAAD,and one sample in BrCa and HBC).Localized disease cancer patients were de?ned as cancer patients without known metastasis from the primary tumor to distant organ(s),as noticed by the physician or additional imaging and/or pathological tests.Patients with glioblastoma,a tumor that metastasizes rarely,were regarded as late-stage (high-grade)cancers.Samples for both training and validation cohort were collected and processed similarly and simultaneously.Tumor tissues of patients were analyzed for the presence of genetic alterations by tissue DNA sequencing,including next-generation sequencing SNaPShot,assessing 39genes over 152exons with an average sequencing coverage of >500,including KRAS ,EGFR ,and PIK3CA (Dias-Santagata et al.,2010).Assessment of MET overexpression in non-small cell lung cancer FFPE slides was performed by immunohistochemistry (anti-Total cMET SP44Rabit mono-clonal antibody (mAb),Ventana,or the A2H2-3anti-human MET mAb (Gruver et al.,2014)).The estrogen and progesterone status of BrCa tumor tissues and the HER2ampli?cation of BrCa tumor tissue were determined using immuno-histochemistry and ?uorescent in situ hybridization,respectively,and scored according to the routine clinical diagnostics protocol at the MGH,Boston.Healthy donors were at the moment of blood collection,or previously,not diagnosed with cancer.This study was conducted in accordance with the prin-ciples of the Declaration of Helsinki.Approval was obtained from the institu-tional review board and the ethics committee at each hospital,and informed consent was obtained from all subjects.Clinical follow-up of healthy donors

Cancer (yes/no) Primary tumor (yes/no) Mutational subtypes Cancer (yes/no) Primary tumor (yes/no) Mutational subtypes Figure https://www.wendangku.net/doc/7911719864.html,an-Focused TEP-Based Can-cer Diagnostics

SVM/LOOCV diagnostics of healthy donors (n =55)and patients with primary or metastatic tumor burden in the lung (n =99;totaling 154tests),brain (n =62;totaling 114tests),or liver (n =72;totaling 127tests),to determine the presence or absence of cancer,with cancer subclassi?ed as primary or metastatic tumor,in case of metastases the iden-ti?ed organ of origin,and the correctly identi?ed molecular markers.Of note,at the exam level of mutational subtypes some samples were included in multiple classi?ers (i.e.,KRAS ,EGFR ,PIK3CA ,HER2-ampli?cation,MET-overexpression,or triple negative status),explaining the higher number in mutational tests than the total number of included samples.TP,true positive;FP,false positive;FN,false negative;TN,true negative.Indicated are sample numbers and detection rates in percent-ages.

Cancer Cell 28,666–676,November 9,2015a2015The Authors 673

is not available due to anonymization of these samples according to the ethical rules of the hospitals.

Support Vector Machine Classi?er

For binary(pan-cancer)and multiclass sample classi?cation,a support vector machine(SVM)algorithm was used implemented by the e1071R-package.In principal,the SVM algorithm determines the location of all samples in a high-dimensional space,of which each axis represents a transcript included and the sample expression level of a particular transcript determines the location on the axis.During the training process,the SVM algorithm draws a hyperplane best separating two classes,based on the distance of the closest sample of each class to the hyperplane.The different sample classes have to be posi-tioned at each side of the hyperplane.Following,a test sample with masked class identity is positioned in the high-dimensional space and its class is‘‘pre-dicted’’by the distance of the particular sample to the constructed hyper-planes.For the multiclass SVM classi?cation algorithm,a One-Versus-One (OVO)approach was used.Here,each class is compared to all other individual classes and thus the SVM algorithm de?nes multiple hyperplanes.To cross validate the algorithm for all samples in the training cohort,the SVM algorithm was trained by all samples in the training cohort minus one,while the remaining sample was used for(blind)classi?cation.This process was repeated for all samples until each sample was predicted once(leave-one-out cross-valida-tion[LOOCV]procedure).The percentage of correct predictions was reported as the classi?er’s accuracy.To assess the predictive value of the SVM algo-rithm on an independent dataset,which is not previously involved in the SVM training process and thus entirely new for the algorithm,the algorithm was trained on the training dataset,all SVM parameters were?xed,and the samples belonging to the validation cohort were predicted.In addition,an iterative(1003)process was performed in which samples of the dataset were randomly subsampled in a training and validation set(ratio training: validation=60:40(all cancer classes)or70:30(healthy individuals),per sample class samples were subsampled in this ratio according the total size of the individual classes(class-proportional,strati?ed subsampling))and mean accuracy of all individual classi?cations was reported.Internal performance of the SVM algorithm could be improved by enabling the SVM tuning function, which implies optimal determination of parameters of the SVM algorithm (gamma,cost)by randomly subsampling the dataset used for training(‘‘inter-nal cross-validation’’)of the algorithm.Prior to construction of the SVM algo-rithm,transcripts with low expression(<5reads in all samples)were excluded and read counts were normalized as described in the Supplemental Experi-mental Procedures(differential expression of transcripts).For each individual prediction,feature selection(identi?cation of transcripts with notable in?uence on the predictive performance)was performed by ANOVA testing for differ-ences,yielding classi?er-speci?c input gene lists(Table S4).mRNAs with a LogCPM>3and a p value corrected for multiple hypothesis testing(FDR)of <0.95(pan-cancer KRAS),<0.90(CRC,PAAD,and NSCLC KRAS and HER2-ampli?ed BrCa),<0.80(PIK3CA BrCa),<0.70(NSCLC EGFR),<0.50 (triple negative-status BrCa),<0.30(MET-overexpression NSCLC),<0.10 (CRC/PAAD/HBC),<0.0001(multiclass tumor type and individual tumor class-healthy),and<0.00005(pan-cancer/healthy-cancer)were included. Internal SVM tuning was enabled to improve predictive performance.All individual tumor class versus healthy donors and molecular pathway SVMs algorithms were tuned by a(default)10-fold internal cross-validation.The pan-cancer/healthy-cancer,multiclass tumor type,and the gastro-intestinal CRC/PAAD/HBC SVM algorithms were tuned by a2-fold internal cross-valida-tion.The training cohort of the pan-cancer and multiclass tumor type,the indi-vidual tumor classes versus healthy donor tests,the gastro-intestinal CRC/ PAAD/HBC test,and all molecular pathway tests were analyzed using a LOOCV approach.To increase classi?cation speci?city in the multiclass tumor type test,additional binary and multiclass classi?ers algorithms were devel-oped,namely the pan-cancer test(Figures1F and1G),HBC-CRC,HBC-PAAD,BrCa-CRC,BrCa-CRC-NSCLC,and BrCa-HD-GBM-NSCLC tests, evaluated in both the training and validation cohort separately,which were applied sequentially to the multiclass tumor type test.Samples predicted as either condition of the supplemental classi?er were all re-evaluated using the ?lter.The latter tumor class classi?cation was regarded as the follow-up clas-si?cation.In addition,samples predicted as the all-female breast cancer class, but of male origin as determined by the gender-speci?c RNAs(Figure S2B),and samples predicted as healthy,while in the pan-cancer test predicted as cancer,were automatically assigned to the class with second predictive strength,as supplemented by the SVM output.To determine the accuracy rates of the classi?ers that can be obtained by chance,class labels of the sam-ples used by the SVM algorithm for training were randomly permutated (‘‘random classi?ers’’).This process was performed for100LOOCV classi?ca-tion procedures.P values were determined by counting the overall random classi?er LOOCV-classi?cation accuracies that yielded similar or higher total accuracy rates compared to the observed total accuracy rate.The predictive strength was also used as input to generate a receiver operating curve(ROC) as implemented in the R-package pROC(version1.7.3).Organ exams were calculated based on the compiled results of the SVM/LOOCV of the training cohort and subsequent prediction of the validation cohort,spanning in total 283samples.The pan-cancer binary SVM,the multiclass SVM,and all molec-ular pathway SVM algorithms were processed individually.Samples included for each organ exam(all healthy donors,all samples with primary tumor in a particular organ,and all samples with known metastases to the particular or-gan)were selected.Only samples with correct predictions at a particular level of the organ exam were passed to the next level for evaluation.Counts of cor-rect and false predictions in the‘‘mutational subtypes’’-stage were determined from all individual molecular pathway SVM algorithms in which the selected samples were included.

Correlative Analysis of Gene Set Enrichment Analysis

Correlative Analyses of Gene Set Enrichment(CAGE)analysis was performed in the online platform R2(R2.amc.nl).To enable analyses of RNA-sequencing read counts in a micro-array-based statistical platform,counts per million normalized read counts were voom-transformed,using sequencing batch and sample group as variables,and uploaded in the R2-environment.Highly correlating mRNAs(FDR<0.01)of a tumor type or all tumor classes combined (pan-cancer)compared to all other classes was used to generate a class-speci?c gene signature.These individual signatures were subsequently corre-lated with3,875curated gene sets as provided by the Broad Institute(http:// https://www.wendangku.net/doc/7911719864.html,/gsea).Top25ranking correlations were manually annotated by two independent researchers(M.G.B.and B.A.W.)and shared annotated terms were after agreement of both researchers reported.

ACCESSION NUMBERS

The accession number for the raw sequencing data reported in this paper is GEO:GSE68086.

SUPPLEMENTAL INFORMATION

Supplemental Information includes Supplemental Experimental Procedures, two?gures,and seven tables and can be found with this article online at https://www.wendangku.net/doc/7911719864.html,/10.1016/https://www.wendangku.net/doc/7911719864.html,ell.2015.09.018.

AUTHOR CONTRIBUTIONS

M.G.B.,B.A.T.,P.W.,and T.W.designed the study and wrote the manuscript.

E.F.S.,D.P.N.,H.M.V.,J.C.R.,and B.A.T.provided clinical samples.M.G.B., N.S.,J.T.,F.R.,P.S.,J.D.,B.Y.,H.V.,and E.P.performed sample processing for mRNA-seq.R.J.A.N.,P.S.,H.V.,E.P.,and T.W.designed and performed amplicon sequencing assays.M.G.B.,N.S.,I.K.,J.D.,B.A.W.,J.K.,N.A., E.P.,and T.W.performed data analyses and interpretation.All authors pro-vided critical comments on the manuscript.

CONFLICTS OF INTEREST

P.S,H.V.,E.P.,R.J.A.N.,and T.W.are employees of thromboDx BV.R.J.A.N. and T.W.are shareholders and founders of thromboDx BV.

ACKNOWLEDGMENTS

Financial support was provided by European Research Council E8626 (R.J.A.N.,E.F.S.,and T.W.)and336540(T.W.),the Dutch Organisation of

674Cancer Cell28,666–676,November9,2015a2015The Authors

Scienti?c Research93612003and91711366(T.W.),the Dutch Cancer Society (J.C.R.,H.M.V.,and T.W.),Stichting STOPhersentumoren.nl(M.G.B.and P.W.),the NIH/NCI CA176359and CA069246(B.A.T.),CFF Norrland (R.J.A.N.),and Swedish Research Council(R.J.A.N.).We are thankful to Esther Drees,Magda Grabowska,Danijela Koppers-Lalic,Michiel Pegtel,Wessel van Wieringen,Phillip de Witt Hamer,and W.Peter Vandertop.

Received:March23,2015

Revised:July2,2015

Accepted:September25,2015

Published:October29,2015

REFERENCES

Akbani,R.,Ng,P.K.S.,Werner,H.M.J.,Shahmoradgoli,M.,Zhang,F.,Ju,Z., Liu,W.,Yang,J.-Y.,Yoshihara,K.,Li,J.,et al.(2014).A pan-cancer proteomic perspective on The Cancer Genome https://www.wendangku.net/doc/7911719864.html,mun.5,3887.

Alix-Panabie`res,C.,and Pantel,K.(2014).Challenges in circulating tumour cell research.Nat.Rev.Cancer14,623–631.

Alix-Panabie`res,C.,Schwarzenbach,H.,and Pantel,K.(2012).Circulating tumor cells and circulating tumor DNA.Annu.Rev.Med.63,199–215.

Bettegowda,C.,Sausen,M.,Leary,R.J.,Kinde,I.,Wang,Y.,Agrawal,N., Bartlett,B.R.,Wang,H.,Luber,B.,Alani,R.M.,et al.(2014).Detection of circu-lating tumor DNA in early-and late-stage human malignancies.Sci.Transl. Med.6,224ra24.

Bidard,F.-C.,Peeters,D.J.,Fehm,T.,Nole′,F.,Gisbert-Criado,R.,Mavroudis, D.,Grisanti,S.,Generali,D.,Garcia-Saenz,J.A.,Stebbing,J.,et al.(2014). Clinical validity of circulating tumour cells in patients with metastatic breast cancer:a pooled analysis of individual patient https://www.wendangku.net/doc/7911719864.html,ncet Oncol.15, 406–414.

Bray,P.F.,McKenzie,S.E.,Edelstein,L.C.,Nagalla,S.,Delgrosso,K.,Ertel,A., Kupper,J.,Jing,Y.,Londin,E.,Loher,P.,et al.(2013).The complex transcrip-tional landscape of the anucleate human platelet.BMC Genomics14,1.

Burkhart,J.M.,Gambaryan,S.,Watson,S.P.,Jurk,K.,Walter,U.,Sickmann, A.,Heemskerk,J.W.M.,and Zahedi,R.P.(2014).What can proteomics tell us about platelets?Circ.Res.114,1204–1219.

Calverley,D.C.,Phang,T.L.,Choudhury,Q.G.,Gao,B.,Oton,A.B.,Weyant, M.J.,and Geraci,M.W.(2010).Signi?cant downregulation of platelet gene expression in metastatic lung cancer.Clin.Transl.Sci.3,227–232.

Chan,K.C.A.,Jiang,P.,Chan,C.W.M.,Sun,K.,Wong,J.,Hui,E.P.,Chan,S.L., Chan,W.C.,Hui,D.S.C.,Ng,S.S.M.,et al.(2013).Noninvasive detection of cancer-associated genome-wide hypomethylation and copy number aberra-tions by plasma DNA bisul?te https://www.wendangku.net/doc/7911719864.html,A110, 18761–18768.

Chen,L.,Kostadima,M.,Martens,J.H.A.,Canu,G.,Garcia,S.P.,Turro,E., Downes,K.,Macaulay,I.C.,Bielczyk-Maczynska,E.,Coe,S.,et al.(2014). Transcriptional diversity during lineage commitment of human blood progeni-tors.Science345,1251033.

Clancy,L.,and Freedman,J.E.(2014).New paradigms in thrombosis:novel mediators and biomarkers platelet RNA transfer.J.Thromb.Thrombolysis 37,12–16.

Crowley,E.,Di Nicolantonio,F.,Loupakis,F.,and Bardelli,A.(2013).Liquid biopsy:monitoring cancer-genetics in the blood.Nat.Rev.Clin.Oncol.10, 472–484.

Dawson,S.-J.,Tsui,D.W.Y.,Murtaza,M.,Biggs,H.,Rueda,O.M.,Chin,S.-F., Dunning,M.J.,Gale,D.,Forshew,T.,Mahler-Araujo,B.,et al.(2013).Analysis of circulating tumor DNA to monitor metastatic breast cancer.N.Engl.J.Med. 368,1199–1209.

Denis,M.M.,Tolley,N.D.,Bunting,M.,Schwertz,H.,Jiang,H.,Lindemann,S., Yost,C.C.,Rubner,F.J.,Albertine,K.H.,Swoboda,K.J.,et al.(2005).Escaping the nuclear con?nes:signal-dependent pre-mRNA splicing in anucleate plate-lets.Cell122,379–391.

Dias-Santagata,D.,Akhavanfard,S.,David,S.S.,Vernovsky,K.,Kuhlmann, G.,Boisvert,S.L.,Stubbs,H.,McDermott,U.,Settleman,J.,Kwak,E.L.,

et al.(2010).Rapid targeted mutational analysis of human tumours:a clinical platform to guide personalized cancer medicine.EMBO Mol.Med.2,146–158.

Diehl,F.,Schmidt,K.,Choti,M.A.,Romans,K.,Goodman,S.,Li,M.,Thornton, K.,Agrawal,N.,Sokoll,L.,Szabo,S.A.,et al.(2008).Circulating mutant DNA to assess tumor dynamics.Nat.Med.14,985–990.

Gay,L.J.,and Felding-Habermann,B.(2011).Contribution of platelets to tumour metastasis.Nat.Rev.Cancer11,123–134.

Geiger,J.,Burkhart,J.M.,Gambaryan,S.,Walter,U.,Sickmann,A.,and Zahedi,R.P.(2013).Response:platelet transcriptome and proteome–relation rather than correlation.Blood121,5257–5258.

George,J.N.(2000)https://www.wendangku.net/doc/7911719864.html,ncet355,1531–1539.

Gnatenko,D.V.,Zhu,W.,Xu,X.,Samuel,E.T.,Monaghan,M.,Zarrabi,M.H., Kim,C.,Dhundale,A.,and Bahou,W.F.(2010).Class prediction models of thrombocytosis using genetic biomarkers.Blood115,7–14.

Golub,T.R.,Slonim,D.K.,Tamayo,P.,Huard,C.,Gaasenbeek,M.,Mesirov, J.P.,Coller,H.,Loh,M.L.,Downing,J.R.,Caligiuri,M.A.,et al.(1999).

Molecular classi?cation of cancer:class discovery and class prediction by gene expression monitoring.Science286,531–537.

Gruver,A.M.,Liu,L.,Vaillancourt,P.,Yan,S.-C.B.,Cook,J.D.,Roseberry Baker,J.A.,Felke, E.M.,Lacy,M.E.,Marchal, C.C.,Szpurka,H.,et al.

(2014).Immunohistochemical application of a highly sensitive and speci?c murine monoclonal antibody recognising the extracellular domain of the human hepatocyte growth factor receptor(MET).Histopathology65,879–896.

Haber,D.A.,and Velculescu,V.E.(2014).Blood-based analyses of cancer: circulating tumor cells and circulating tumor DNA.Cancer Discov.4,650–661.

Han,L.,Yuan,Y.,Zheng,S.,Yang,Y.,Li,J.,Edgerton,M.E.,Diao,L.,Xu,Y., Verhaak,R.G.W.,and Liang,H.(2014).The Pan-Cancer analysis of pseudo-gene expression reveals biologically and clinically relevant tumour subtypes.

https://www.wendangku.net/doc/7911719864.html,mun.5,3963.

Healy,A.M.,Pickard,M.D.,Pradhan,A.D.,Wang,Y.,Chen,Z.,Croce,K., Sakuma,M.,Shi,C.,Zago,A.C.,Garasic,J.,et al.(2006).Platelet expression pro?ling and clinical validation of myeloid-related protein-14as a novel deter-minant of cardiovascular events.Circulation113,2278–2284.

Hoadley,K.A.,Yau,C.,Wolf,D.M.,Cherniack,A.D.,Tamborero,D.,Ng,S., Leiserson,M.D.M.,Niu,B.,McLellan,M.D.,Uzunangelov,V.,et al.;Cancer Genome Atlas Research Network(2014).Multiplatform analysis of12cancer types reveals molecular classi?cation within and across tissues of origin.

Cell158,929–944.

Hrdlickova, B.,Kumar,V.,Kanduri,K.,Zhernakova, D.V.,Tripathi,S., Karjalainen,J.,Lund,R.J.,Li,Y.,Ullah,U.,Modderman,R.,et al.(2014).

Expression pro?les of long non-coding RNAs located in autoimmune disease-associated regions reveal immune cell-type speci?city.Genome Med.6,88.

Kandoth,C.,McLellan,M.D.,Vandin,F.,Ye,K.,Niu,B.,Lu,C.,Xie,M.,Zhang, Q.,McMichael,J.F.,Wyczalkowski,M.A.,et al.(2013).Mutational landscape and signi?cance across12major cancer types.Nature502,333–339.

Kirschbaum,M.,Karimian,G.,Adelmeijer,J.,Giepmans,B.N.G.,Porte,R.J., and Lisman,T.(2015).Horizontal RNA transfer mediates platelet-induced he-patocyte proliferation.Blood126,798–806.

Kissopoulou,A.,Jonasson,J.,Lindahl,T.L.,and Osman,A.(2013).Next gen-eration sequencing analysis of human platelet PolyA+mRNAs and rRNA-depleted total RNA.PLoS ONE8,e81809.

Klement,G.L.,Yip,T.-T.,Cassiola,F.,Kikuchi,L.,Cervi,D.,Podust,V., Italiano,J.E.,Wheatley, E.,Abou-Slaybi, A.,Bender, E.,et al.(2009).

Platelets actively sequester angiogenesis regulators.Blood113,2835–2842.

Koboldt,D.C.,Fulton,R.S.,McLellan,M.D.,Schmidt,H.,Kalicki-Veizer,J., McMichael,J.F.,Fulton,L.L.,Dooling,D.J.,Ding,L.,Mardis,E.R.,et al.;

Cancer Genome Atlas Network(2012).Comprehensive molecular portraits of human breast tumours.Nature490,61–70.

Kuznetsov,H.S.,Marsh,T.,Markens,B.A.,Castan?o,Z.,Greene-Colozzi,A., Hay,S.A.,Brown,V.E.,Richardson,A.L.,Signoretti,S.,Battinelli,E.M.,and McAllister,S.S.(2012).Identi?cation of luminal breast cancers that establish

a tumor-supportive macroenvironment de?ned by proangiogenic platelets

and bone marrow-derived cells.Cancer Discov.2,1150–1165.

Cancer Cell28,666–676,November9,2015a2015The Authors675

Labelle,M.,Begum,S.,and Hynes,R.O.(2011).Direct signaling between platelets and cancer cells induces an epithelial-mesenchymal-like transition and promotes metastasis.Cancer Cell20,576–590.

Laffont,B.,Corduan,A.,Ple′,H.,Duchez,A.-C.,Cloutier,N.,Boilard,E.,and Provost,P.(2013).Activated platelets can deliver mRNA regulatory Ago2d microRNA complexes to endothelial cells via microparticles.Blood 122,253–261.

Landry,P.,Plante,I.,Ouellet,D.L.,Perron,M.P.,Rousseau,G.,and Provost,P. (2009).Existence of a microRNA pathway in anucleate platelets.Nat.Struct. Mol.Biol.16,961–966.

Leidinger,P.,Backes,C.,Dahmke,I.N.,Galata,V.,Huwer,H.,Stehle,I.,Bals, R.,Keller,A.,and Meese,E.(2014).What makes a blood cell based miRNA expression pattern disease speci?c?–a miRNome analysis of blood cell sub-sets in lung cancer patients and healthy controls.Oncotarget5,9484–9497.

Leon,S.A.,Shapiro,B.,Sklaroff,D.M.,and Yaros,M.J.(1977).Free DNA in the serum of cancer patients and the effect of therapy.Cancer Res.37,646–650.

Leslie,M.(2010).Cell biology.Beyond clotting:the powers of platelets. Science328,562–564.

Lood,C.,Amisten,S.,Gullstrand,B.,Jo¨nsen,A.,Allhorn,M.,Truedsson,L., Sturfelt,G.,Erlinge,D.,and Bengtsson,A.A.(2010).Platelet transcriptional pro?le and protein expression in patients with systemic lupus erythematosus: up-regulation of the type I interferon system is strongly associated with vascular disease.Blood116,1951–1957.

Lu,J.,Getz,G.,Miska,E.A.,Alvarez-Saavedra,E.,Lamb,J.,Peck,D.,Sweet-Cordero,A.,Ebert,B.L.,Mak,R.H.,Ferrando,A.A.,et al.(2005).MicroRNA expression pro?les classify human cancers.Nature435,834–838.

Maheswaran,S.,Sequist,L.V.,Nagrath,S.,Ulkus,L.,Brannigan,B.,Collura, C.V.,Inserra, E.,Diederichs,S.,Iafrate, A.J.,Bell, D.W.,et al.(2008). Detection of mutations in EGFR in circulating lung-cancer cells.N.Engl.J. Med.359,366–377.

McAllister,S.S.,and Weinberg,R.A.(2014).The tumour-induced systemic environment as a critical regulator of cancer progression and metastasis. Nat.Cell Biol.16,717–727.

Murtaza,M.,Dawson,S.-J.,Tsui,D.W.Y.,Gale,D.,Forshew,T.,Piskorz,A.M., Parkinson,C.,Chin,S.-F.,Kingsbury,Z.,Wong,A.S.C.,et al.(2013).Non-inva-sive analysis of acquired resistance to cancer therapy by sequencing of plasma DNA.Nature497,108–112.

Newman,A.M.,Bratman,S.V.,To,J.,Wynne,J.F.,Eclov,N.C.W.,Modlin,L.A., Liu,C.L.,Neal,J.W.,Wakelee,H.A.,Merritt,R.E.,et al.(2014).An ultrasensitive method for quantitating circulating tumor DNA with broad patient coverage. Nat.Med.20,548–554.

Nilsson,R.J.A.,Balaj,L.,Hulleman,E.,van Rijn,S.,Pegtel,D.M.,Walraven,M., Widmark,A.,Gerritsen,W.R.,Verheul,H.M.,Vandertop,W.P.,et al.(2011). Blood platelets contain tumor-derived RNA biomarkers.Blood118,3680–3683.

Palumbo,J.S.,Talmage,K.E.,Massari,J.V.,La Jeunesse,C.M.,Flick,M.J., Kombrinck,K.W.,Jirouskova′,M.,and Degen,J.L.(2005).Platelets and ?brin(ogen)increase metastatic potential by impeding natural killer cell-medi-ated elimination of tumor cells.Blood105,178–185.

Papadopoulos,N.,Kinzler,K.W.,and Vogelstein,B.(2006).The role of com-panion diagnostics in the development and use of mutation-targeted cancer therapies.Nat.Biotechnol.24,985–995.

Placke,T.,O¨rgel,M.,Schaller,M.,Jung,G.,Rammensee,H.-G.,Kopp,H.-G., and Salih,H.R.(2012).Platelet-derived MHC class I confers a pseudonormal phenotype to cancer cells that subverts the antitumor reactivity of natural killer immune cells.Cancer Res.72,440–448.

Power,K.A.,McRedmond,J.P.,de Stefani,A.,Gallagher,W.M.,and Gaora, P.O.(2009).High-throughput proteomics detection of novel splice isoforms in human platelets.PLoS ONE4,e5001.

Quail,D.F.,and Joyce,J.A.(2013).Microenvironmental regulation of tumor progression and metastasis.Nat.Med.19,1423–1437.

Rack, B.,Schindlbeck, C.,Ju¨ckstock,J.,Andergassen,U.,Hepp,P., Zwingers,T.,Friedl,T.W.P.,Lorenz,R.,Tesch,H.,Fasching,P.A.,et al.;SUCCESS Study Group(2014).Circulating tumor cells predict survival in early average-to-high risk breast cancer patients.J.Natl.Cancer Inst.106,dju066.

Raghavachari,N.,Xu,X.,Harris,A.,Villagra,J.,Logun,C.,Barb,J.,Solomon, M.A.,Suffredini,A.F.,Danner,R.L.,Kato,G.,et al.(2007).Ampli?ed expression pro?ling of platelet transcriptome reveals changes in arginine metabolic path-ways in patients with sickle cell disease.Circulation115,1551–1562.

Rak,J.,and Guha,A.(2012).Extracellular vesicles–vehicles that spread cancer genes.BioEssays34,489–497.

Ramaswamy,S.,Tamayo,P.,Rifkin,R.,Mukherjee,S.,Yeang,C.H.,Angelo, M.,Ladd,C.,Reich,M.,Latulippe,E.,Mesirov,J.P.,et al.(2001).Multiclass cancer diagnosis using tumor gene expression signatures.Proc.Natl.Acad. https://www.wendangku.net/doc/7911719864.html,A98,15149–15154.

Ramsko¨ld,D.,Luo,S.,Wang,Y.-C.,Li,R.,Deng,Q.,Faridani,O.R.,Daniels, G.A.,Khrebtukova,I.,Loring,J.F.,Laurent,L.C.,et al.(2012).Full-length mRNA-Seq from single-cell levels of RNA and individual circulating tumor cells.Nat.Biotechnol.30,777–782.

Risitano,A.,Beaulieu,L.M.,Vitseva,O.,and Freedman,J.E.(2012).Platelets and platelet-like particles mediate intercellular RNA transfer.Blood119, 6288–6295.

Rondina,M.T.,Schwertz,H.,Harris,E.S.,Kraemer,B.F.,Campbell,R.A., Mackman,N.,Grissom,C.K.,Weyrich,A.S.,and Zimmerman,G.A.(2011). The septic milieu triggers expression of spliced tissue factor mRNA in human platelets.J.Thromb.Haemost.9,748–758.

Rowley,J.W.,Oler,A.J.,Tolley,N.D.,Hunter,B.N.,Low,E.N.,Nix,D.A.,Yost, C.C.,Zimmerman,G.A.,and Weyrich,A.S.(2011).Genome-wide RNA-seq analysis of human and mouse platelet transcriptomes.Blood118,e101–e111.

Schubert,S.,Weyrich,A.S.,and Rowley,J.W.(2014).A tour through the tran-scriptional landscape of platelets.Blood124,493–502.

Simon,L.M.,Edelstein,L.C.,Nagalla,S.,Woodley,A.B.,Chen,E.S.,Kong,X., Ma,L.,Fortina,P.,Kunapuli,S.,Holinstat,M.,et al.(2014).Human platelet microRNA-mRNA networks associated with age and gender revealed by inte-grated plateletomics.Blood123,e37–e45.

Skog,J.,Wu¨rdinger,T.,van Rijn,S.,Meijer,D.H.,Gainche,L.,Sena-Esteves, M.,Curry,W.T.,Jr.,Carter,B.S.,Krichevsky,A.M.,and Breake?eld,X.O. (2008).Glioblastoma microvesicles transport RNA and proteins that promote tumour growth and provide diagnostic biomarkers.Nat.Cell Biol.10,1470–1476.

Sreeramkumar,V.,Adrover,J.M.,Ballesteros,I.,Cuartero,M.I.,Rossaint,J., Bilbao,I.,Nacher,M.,Pitaval,C.,Radovanovic,I.,Fukui,Y.,et al.(2014). Neutrophils scan for activated platelets to initiate in?ammation.Science346, 1234–1238.

Su,A.I.,Welsh,J.B.,Sapinoso,L.M.,Kern,S.G.,Dimitrov,P.,Lapp,H., Schultz,P.G.,Powell,S.M.,Moskaluk,C.A.,Frierson,H.F.,et al.(2001). Molecular classi?cation of human carcinomas by use of gene expression sig-natures.Cancer Res.61,7388–7393.

Thierry,A.R.,Mouliere,F.,El Messaoudi,S.,Mollevi,C.,Lopez-Crapez,E., Rolet,F.,Gillet,B.,Gongora,C.,Dechelotte,P.,Robert,B.,et al.(2014). Clinical validation of the detection of KRAS and BRAF mutations from circu-lating tumor DNA.Nat.Med.20,430–435.

Ting, D.T.,Wittner, B.S.,Ligorio,M.,Vincent Jordan,N.,Shah, A.M., Miyamoto,D.T.,Aceto,N.,Bersani,F.,Brannigan,B.W.,Xega,K.,et al. (2014).Single-cell RNA sequencing identi?es extracellular matrix gene expres-sion by pancreatic circulating tumor cells.Cell Rep.8,1905–1918.

Vapnik,V.N.(1998).Statistical Learning Theory(Wiley).

Yeang, C.H.,Ramaswamy,S.,Tamayo,P.,Mukherjee,S.,Rifkin,R.M., Angelo,M.,Reich,M.,Lander, E.,Mesirov,J.,and Golub,T.(2001). Molecular classi?cation of multiple tumor types.Bioinformatics17(Suppl1), S316–S322.

Yu,M.,Bardia,A.,Wittner,B.S.,Stott,S.L.,Smas,M.E.,Ting,D.T.,Isakoff, S.J.,Ciciliano,J.C.,Wells,M.N.,Shah,A.M.,et al.(2013).Circulating breast tumor cells exhibit dynamic changes in epithelial and mesenchymal composi-tion.Science339,580–584.

676Cancer Cell28,666–676,November9,2015a2015The Authors