Electron cryotomography reveals ultrastructurealterations in platelets from patients withovarian cancerRui Wanga,b, Rebecca L. Stonec, Jason T. Kaelberb,d, Ryan H. Rochata,b, Alpa M. Nickc, K. Vinod Vijayane,Vahid Afshar-Kharghanc, Michael F. Schmida,b, Jing-Fei Dongf,g,h, Anil K. Soodc,i,j,1, and Wah Chiua,b,1

aGraduate Program in Structural and Computational Biology and Molecular Biophysics, Verna and Marrs McLean Department of Biochemistry andMolecular Biology, Baylor College of Medicine, Houston, TX 77030; bNational Center for Macromolecular Imaging, Verna and Marrs McLean Department ofBiochemistry and Molecular Biology, Baylor College of Medicine, Houston, TX 77030; cDepartment of Gynecologic Oncology, The University of Texas MDAnderson Cancer Center, Houston, TX 77030; dDepartment of Molecular Virology and Microbiology, Baylor College of Medicine, Houston, TX 77030;eDepartment of Medicine, Baylor College of Medicine, Michael E. DeBakey Veterans Affair Medical Center, Houston, TX 77030; fCenter for TranslationalResearch on Inflammation, Michael E. DeBakey Veterans Affair Medical Center, Houston, TX 77030; gPuget Sound Blood Center, Department of Medicine,School of Medicine, University of Washington, Seattle, WA 98104; hDivision of Hematology, Department of Medicine, School of Medicine, University ofWashington, Seattle, WA 98104; iDepartment of Cancer Biology, The University of Texas MD Anderson Cancer Center, Houston, TX 77030; and jCenter forRNA Interference and Non-Coding RNA, The University of Texas MD Anderson Cancer Center, Houston, TX 77030

Contributed by Wah Chiu, September 30, 2015 (sent for review March 1, 2015; reviewed by Peter Frederik, Jun Liu, Kunle Odunsi, and David W. Scott)

Thrombocytosis and platelet hyperreactivity are known to be asso-ciated with malignancy; however, there have been no ultrastructurestudies of platelets from patients with ovarian cancer. Here, we usedelectron cryotomography (cryo-ET) to examine frozen-hydratedplatelets from patients with invasive ovarian cancer (n = 12) andcontrol subjects either with benign adnexal mass (n = 5) or free fromdisease (n = 6). Qualitative inspections of the tomograms indicatesignificant morphological differences between the cancer and con-trol platelets, including disruption of the microtubule marginal band.Quantitative analysis of subcellular features in 120 platelet electrontomograms from these two groups showed statistically significantdifferences in mitochondria, as well as microtubules. These structuralvariations in the platelets from the patients with cancer may becorrelated with the altered platelet functions associated with ma-lignancy. Cryo-ET of platelets shows potential as a noninvasivebiomarker technology for ovarian cancer and other platelet-relateddiseases.

electron | cryotomography | platelet | microtubule | cancer

Platelets are small anucleate multifunctional cells derivedfrom megakaryocytes. Platelets circulate in the bloodstream

and respond to vascular lesions (1). In their resting state, theyadopt a discoidal shape and have an average lifespan of 5–7 d inhumans (2). Platelets have increasingly been recognized as playingan important role in tumor growth and metastasis, in addition totheir traditional roles in hemostasis (3–6). Thrombocytosis(platelet count >450,000/μL) is found in 31% of patients withovarian cancer and is associated with a poor clinical prognosis (7,8). In addition, platelets in patients with cancer are functionallyaltered, and often adopt a hyperreactive state (9). This platelethyperreactivity helps explain the higher thrombosis risk in thesepatients (10). Furthermore, compelling preclinical and clinicaldata demonstrate that platelets actively promote tumor growthand metastasis through multiple pathways. Platelets form a phys-ical shield to protect tumor cells from natural killer cell-mediatedlysis; they facilitate the adhesion of tumor cells to the endothe-lium, allowing the critical extravasation step in the metastaticcascade to occur; and they release a plethora of bioactive mole-cules, including growth factors and cytokines stored in their se-cretory granules, that promote angiogenesis and tumor cell growth(11). It is shown in various experimental models that disruption ofplatelet–tumor interactions abolishes these effects (5, 9), improv-ing clinical outcomes of patients.The relatively high incidence of thrombocytosis in these pa-

tient populations (12), which is related to cytokine signaling fromboth tumor and nontumor tissue (8), and the implication of

platelets in cancer pathology (6, 8) suggested the possibility ofultrastructural perturbations in the platelets of patients withovarian cancer. These perturbations could provide structuralclues to improve our understanding of cancer-related hyperre-activity, thrombocytosis, or as-yet-unknown platelet defects, andcould serve as potential biomarkers for screening for ovariancancer. Prior attempts at examining platelet ultrastructure indiseases have been largely limited by methods that used plasticembedding and chemical fixation (13–15). The present investi-gation overcomes these technical limitations by using electroncryotomography (cryo-ET) to visualize human platelets in theirnaturally occurring pathophysiological states without chemicalfixation or staining (16).

ResultsAfter approval by the institutional review boards for researchon human subjects, we prospectively collected peripheral bloodsamples from patients with newly diagnosed invasive ovariancancer or benign adnexal masses (Table S1) before any chemo-therapy or surgical treatment. Age-matched healthy females were

Significance

Platelets are known to be both numerically and functionallyaltered in some patients with cancer. However, structural dif-ferences in the platelets from these patients have not beenstudied. Here we use electron cryotomography to reveal that,compared with control donors, the microtubule system and themitochondria of platelets from patients diagnosed with ovar-ian cancer are significantly different. This finding suggests thepotential of electron cryotomography as a technology to de-tect structural biomarkers of diseases affecting platelets.

Author contributions: M.F.S., J.-F.D., A.K.S., and W.C. designed research; R.W., R.L.S., V.A.-K.,and M.F.S. performed research; R.W., R.L.S., A.M.N., K.V.V., and M.F.S. contributed newreagents/analytic tools; R.W., J.T.K., R.H.R., M.F.S., J.-F.D., A.K.S., and W.C. analyzed data;and R.W., J.T.K., V.A.-K., M.F.S., J.-F.D., A.K.S., and W.C. wrote the paper.

Reviewers: P.F., Maastricht University; J.L., University of Texas Medical School at Houston;K.O., Roswell Park Cancer Institute; and D.W.S., Rice University.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

Data deposition: Representative tomograms of platelets in three different pathophysio-logical states reported in this paper have been deposited in the EMDataBank, www.emdatabank.org/ (accession nos. EMD-6471, EMD-6472, and EMD-6473).1To whom correspondence may be addressed. Email: wah@bcm.edu or asood@mdanderson.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1518628112/-/DCSupplemental.

14266–14271 | PNAS | November 17, 2015 | vol. 112 | no. 46 www.pnas.org/cgi/doi/10.1073/pnas.1518628112

recruited as controls. None of the participants had a primaryplatelet disorder or a coexisting inflammatory condition, andnone was taking medications known to interfere with plateletfunction. Average platelet counts and mean platelet volume ofthe healthy control subjects were 279.6 × 103/μL and 7.9 fL, re-spectively. Blood samples (anticoagulant: 0.38% sodium citratefinal concentration) were collected from the patients and healthysubjects after informed consent was provided. Platelet-richplasma (PRP) was obtained by centrifugation and vitrified forcryo-ET. From these frozen-hydrated samples, a total of 338 tiltseries of tomography images of individual platelets were gener-ated. Each image series was then reconstructed into a 3D volumecalled a tomogram. One hundred twenty tomograms with ade-quate contrast were selected for filtering, feature annotation,structural measurements, and statistical analysis. For the quanti-tative analysis, we required at least five annotatable tomogramsfrom each subject. In addition, platelets from these subjects wereanalyzed for aggregation induced by ADP.

Cryo-ET of Platelets from Healthy Control Subjects. We first exam-ined platelets from the healthy controls (n = 6). Fig. 1A shows anexample slice from a 3D tomogram of a typical platelet of ahealthy subject (Movie S1). Fig. 2A shows the annotated structuralfeatures of three platelets randomly selected from this healthygroup, including the plasma membrane, circumferentially coiledmicrotubule (marginal band), α and dense granules, mitochondria,and low-contrast vacuole-like (LCV) features. A fraction of theLCV was seen as empty vacuole-like structures (Figs. 1A and 2Aand Movie S1). The rest of LCV exists as narrow and tortuousinvaginations of the surface membrane, the area of which wasimpractical to estimate. We interpret this structural feature to bethe platelet surface-connected open canalicular system. Mostplatelets from healthy subjects share a similar structural pattern.The lack of a significant number of pseudopods indicates that theobserved platelets are mostly in the resting state. These structuralfeatures are consistent with the morphology of platelets fromhealthy subjects obtained by conventional 2D thin-section, plastic-embedded electron microscopy (2, 13).

Cryo-ET of Platelets from Patients. Next, we examined plateletsobtained from women with either benign adnexal mass (n = 5) orinvasive ovarian cancer (n = 12). Qualitatively, platelets frompatients with benign masses (Figs. 1B and 2B) had a similarappearance to those from the healthy subjects, including a rel-atively intact microtubule ring and abundant secretory granules.

In contrast, platelets from patients with an invasive ovariancancer (Figs. 1C and 2C and Movie S2) had a strikingly differentappearance compared with those from either healthy subjects(Figs. 1A and 2A) or patients with benign masses (Figs. 1B and2B). In many of these platelets from patients with invasive cancer,the microtubule marginal bands seen in the benign and controlplatelets were severed and depolymerized to various degrees (Figs.1C and 2C). Also unique to patients with invasive cancer wasmarked visual and quantitative heterogeneity among individualplatelets, even those from the same patient (Fig. S1).

Quantification of Platelet Tomograms. To quantitate the structuralfeatures of each platelet, we measured nine parameters (Figs. 3and 4), including number of α granules, number of densegranules, number of mitochondria, percentage of total plateletarea occupied by each of these three structures, microtubulelength, pseudopod number, and total platelet area. Tomogramswith insufficient contrast were excluded from further analysis.Platelets from individual patients with ovarian cancer are het-erogeneous (Fig. S1), which warrants the decision to measurethese parameters from at least five platelets per subject. Thesenine parameters were chosen because they are associated withplatelet structural integrity and functions and are easily rec-ognized in tomograms. For instance, α and dense granules inour tomograms have relatively large size and characteristicscattering densities because of their different biochemicalcontents. The clinical utility of this study lies in the distinctionbetween cancer and noncancer groups. Therefore, we com-bined the data from healthy donors and patients with benignmasses to form the control group for subsequent analyses be-cause we found that many of these measurements exhibited thesame pattern for both healthy and benign samples, as expectedfrom qualitative analysis (Figs. 1 and 2). Measurements from

Fig. 1. Platelets from control subjects and patients with cancer, raw cryo-electron tomogram. (A) Slice of a tomogram of a human blood platelet froma healthy female donor. This platelet has a circumferential microtubule ringand numerous secretory granules inside. The general shape is discoidal.(B) Slice of a tomogram of a human blood platelet from a patient with abenign tumor in the ovary. This platelet shares similar cytoplasmic structureswith the platelet in A, with the exception of the presence of an extendedpseudopod. (C) Slice of a tomogram of a human blood platelet from a pa-tient with invasive ovarian cancer. This platelet has more enlarged LCVstructures and an extended pseudopod. In contrast to the platelets in A andB, this platelet has fragmented and fewer microtubules.

Fig. 2. Platelets from control subjects and patients with cancer, annotatedtomogram. (A) Three randomly selected annotated platelets from healthydonors. All of them have an intact marginal band of microtubules (blue)enclosing most of their granules (α pink, dense green) and mitochondria(red) inside. The plasma membrane is gray and the LCV is yellow. (B) Threerandomly selected annotated platelets from patients with benign masses.Their structures are similar to those shown in A. (C) Three randomly selectedannotated platelets from patients with invasive ovarian cancer. Their mor-phologies appear to be more heterogeneous compared with the six plateletsin the other two panels. They seem to have more LCV, fewer and shortermicrotubule filaments, and more mitochondria.

Wang et al. PNAS | November 17, 2015 | vol. 112 | no. 46 | 14267

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

this control group (n = 11) were compared with those frompatients with invasive ovarian cancer (n = 12). Among thesemeasurements, three of the nine parameters were found toshow significant differences between control and cancer subjects:

microtubule length, mitochondria number, and percentage ofplatelet total area occupied by mitochondria (Fig. 3). Again, thesethree parameters did not differ significantly between the benignand healthy donors.

Fig. 3. Statistical analysis shows microtubule and mitochondria are discriminating parameters. (A) Microtubule strands are clearly seen in the subtomogramof the platelet. Microtubule lengths are measured and compared between the control group (healthy donors plus patients with benign mass) and the cancergroup (patients with ovarian cancer). The cancer group has significantly fewer and shorter microtubules in their platelets than the control group. (B) Mi-tochondria appear as enclosed double-membrane organelles; the inner membrane forms the characteristic cristae structure inside. Mitochondria numbers arecounted, and the percentage of the whole platelet area covered by mitochondria is measured. The cancer group has significantly more mitochondria in theirplatelets compared with the control group. They also have a higher percentage of area covered by mitochondria.

Fig. 4. The six other measured platelet parameters are found not to be significantly different between the control and cancer groups. These are (A) platelet area,(B) number of pseudopods, (C) number of α granules and percentage of granule area, and (D) number of dense granules and percentage of granule area.

14268 | www.pnas.org/cgi/doi/10.1073/pnas.1518628112 Wang et al.

Our quantification shows that the mean microtubule length wassignificantly shorter in platelets from patients with invasive ovariancancer compared with those from the control group (36.08 ±9.70 μm vs. 66.17 ± 14.42 μm; P = 0.0001; Fig. 3A). Qualitatively,microtubules of the patients were fragmented, instead of formingintact marginal band rings as in the control group (Fig. 2). Becauseall the quantifications were performed by the same investigator, itis likely that the analysis was self-consistent, but not necessarilyunbiased. However, subjectivity is reduced because we chose to-mograms having unequivocally good contrast for analysis.We also found differences in the number of mitochondria and

the percentage of the whole platelet area covered by the mito-chondria. Platelets from patients with cancer had more mito-chondria than platelets from the control group (Fig. 3B) (7.26 ±2.44 vs. 4.85 ± 1.46 mitochondria per platelet; P = 0.01). Theabsolute number of mitochondria per platelet in the control groupis consistent with previous reports (17). The mitochondria frompatients with cancer occupied a larger fraction of the area of theplatelet (3.91 ± 1.21 vs. 2.47 ± 1.18% of platelet area; P = 0.009).Most of the differences in total mitochondria area among plateletscan be explained by differences in the number of mitochondria percell (R2 = 0.80; n = 120).The other six quantitated parameters did not show statistically

significant differences (Fig. 4).

Predictive Modeling. To assess whether cryo-ET quantificationprovides enough information to predict whether a subject has amalignancy, and to assess whether the sample size in this studywas sufficient to provide predictive power, we performed dis-criminant analysis with leave-one-out cross validation (18). Pre-diction of malignancy was made using only the number ofmitochondria and the microtubule length. In 20 of 23 cases, themalignancy status was correctly predicted, yielding an accuracyof 87% in this study.

Platelet Aggregation Assay. Platelet–platelet and platelet–leuko-cyte aggregates are often detected in peripheral blood of patientswith inflammatory conditions. These aggregates are formed amongprimed and activated platelets, and their presence is widely rec-ognized as a risk for thrombosis. In addition to the detectablestructural differences between control and cancer platelets, wealso investigated their functional consequences by measuringplatelet aggregation induced by ADP at 2.5, 5, and 10 μM. Usinglower concentrations of ADP allows us to detect platelet hyper-reactivity, which was defined as ≥30% aggregation induced by2.5 μM ADP. In a previous study of 359 healthy subjects, ADP at1.5–2.5 μM was used to differentiate hypo- from hyperreactiveplatelets (19). This concentration of ADP does not induce sig-nificant platelet aggregation in most healthy subjects. Our datashow that mean levels of platelet aggregation induced by 2.5 and5 μM were significantly higher in the patients with ovarian cancercompared with controls (Fig. 5A). We also found that 75% ofpatients with ovarian cancer and 27% of our controls have hy-perreactive platelets, as defined by the above criterion (Fig. 5B),indicating an enhanced response of cancer platelets to the lowestdose of ADP. At the highest concentration of ADP (10 μM), thetwo groups become indistinguishable in their aggregation, sug-gesting the platelets from patients with ovarian cancer are notactivated before the addition of the agonist.

DiscussionThin-section electron microscopy has revealed the morphologicfeatures of platelets associated with a variety of (patho)physio-logical conditions (15), but this imaging method lacks 3D in-formation of the subcellular contents in a platelet close to itsnative state (16). Cryo-ET has emerged as a viable technology tostudy well-preserved ultrastructure of mammalian and bacterialcells in 3D (20–22). Therefore, we used cryo-ET to examine the

structures of individual platelets from patients with ovarian tu-mors. The acquired tomograms of platelets are rich in 3Dstructural information, allowing the subcellular arrangement ofvarious components in different physiological states to be seenfor the first time to our knowledge.The platelets from the healthy subjects studied here are mostly

in the resting state, as evidenced by the paucity (0.4 ± 0.7 perplatelet) of extended pseudopods called filopodia, the presence ofwhich is a structural landmark of an activated platelet. Becausethe platelet is crowded with numerous subcellular components, weonly annotated the most recognizable features in this study.Platelets from patients with cancer share many visually identifiablesubcellular features with platelets from control subjects (Fig. 4).More than 70% of α granules in all platelets were nearly spherical;the rest were irregularly shaped and morphologically heteroge-neous (23). The mean diameter of the near-spherical α granuleswas 221.34 ± 56.03 nm, in agreement with previously reportedsizes of 200–500 nm by other imaging technologies (24). Thenumbers of α and dense granules per tomogram for the controlplatelets were 43.0 ± 22.9 and 5.5 ± 1.9, respectively, which isconsistent with reported counts of 50–80 and 3–9 per platelet (25).Platelets from patients with cancer appeared qualitatively to havea higher percentage area covered by enlarged LCV structures,which are likely to be either enlarged open canalicular system orempty granules, compared with those from healthy subjects andpatients with benign masses (Figs. 1C and 2C).To quantify the visual differences of the platelets, we mea-

sured the length, area, and/or number of the observed sub-cellular features. Each tomogram has a missing wedge resultingfrom limited specimen tilt angles, resulting in inaccuracy inthe measurements along the direction of the electron beam.Therefore, our size quantifications of the identified featuresare made from their central areas instead of the volumes. Ouranalysis shows the mean platelet area did not differ significantlybetween control subjects and patients with cancer. However,three other quantitative measures showed significant differencebetween control and cancer platelets including microtubulelength, the number of mitochondria, and the percentage of thetotal platelet area occupied by mitochondria.Platelets contain fully functional mitochondria that are active in

ATP production, regulation of redox signals, and platelet apo-ptosis (26). With cryo-ET, mitochondria are easily distinguishablein the crowded cytosol from other membranous organelles be-cause of their double membrane and internal cristae. A significant

Fig. 5. Platelet aggregation assay shows a correlation between hyperac-tivity and malignancy. (A) The mean level of platelet aggregation is signif-icantly higher in patients with invasive ovarian cancer compared withcontrol subjects (Student’s t test, n = 23). In comparison, there is no signif-icant difference in maximal aggregation between the two groups. (B) Thereare significantly more patients with invasive ovarian cancer with hyperre-active platelets (black bar vs. normal platelets indicated by white bar), asdefined by aggregation at 2.5 μM ADP (χ2 test, P = 0.018), but not at 5 and10 μM of ADP compared with control subjects. CTR, control group; MT,metastatic tumor.

Wang et al. PNAS | November 17, 2015 | vol. 112 | no. 46 | 14269

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

increase in the number of, and percentage of platelet area coveredby, mitochondria (Fig. 3B) suggest they may serve as one of thestructural signatures to differentiate platelets of patients with in-vasive cancer from those of the control subjects. The number ofmitochondria per platelet is elevated by about 50% among pa-tients with malignancies, with only a small increase (<15%) in themean size of a mitochondrion.The biological cause for the increase in mitochondria remains

further to be investigated. Activated platelets can release mito-chondria in the form of microparticles (17), and mitochondria mayreplicate in circulating platelets (27). We occasionally observedsmall “platelets,” which may represent platelet microparticles,whose size can approach that of a true platelet (Fig. S1). Never-theless, it is likely that platelet mitochondrial number is largelydetermined in the megakaryocyte during thrombopoiesis. Pro-thrombopoietic factors affecting the megakaryocyte are emitted bytumor and host tissue during ovarian malignancy. This dysregulationcan cause thrombocytosis. However, thrombocytosis is observed inonly a third of patients at initial diagnosis, which limits its diagnosticpotential (8). Platelet mitochondrial number warrants further in-vestigation as a potentially more sensitive marker of megakaryocyticdysregulation, and because platelet activity affects disease progres-sion in an animal model (8).In addition, we also detect a significant reduction in the length

of microtubules. Microtubules are critical for maintaining theplatelet’s discoidal shape and participate in platelet shape changesand granule movements upon platelet activation (2).Platelet hyperreactivity, which has been reported in cancer and

other diseases of inflammation but has not been morphologicallyexplained, could be caused by the disintegration of microtubules(Fig. 3A), and in particular the disruption of the marginal band(Fig. 2 and Fig. S1). Platelet hyperreactivity means platelets areprimed to activation: they form aggregates when exposed to lowerconcentrations of agonists than normal platelets, which requirehigher concentrations of agonists to activate. Hyperreactive plate-lets are thus not activated platelets, but they become activatedmore easily upon stimulation. This hyperreactivity of platelets frompatients with cancer is supported by the aggregation assay (Fig.5A). Nine of our patients with cancer (75.0%) were identified ashaving hyperreactive platelets (Fig. 5B) compared with three of ourcontrol subjects (27.3%; P = 0.039). There may be an underlyingmechanism of cytoskeleton dysregulation in ovarian cancer causingboth hyperreactivity and reduction in microtubule length that re-quires further study.Abnormal platelet number (thrombocytosis) and hyperreactivity

of platelets have been reported in cancer (8, 28), but have notbeen useful as screening tools because of their low positive pre-dictive values. Our cryo-ET analysis demonstrates that the cancergroup has more fragmented microtubules and more mitochondriain their platelets compared with the control group. In this study,the presence or absence of a malignancy was predicted with 87%accuracy solely from information contained within the cryotomo-grams of subjects’ platelets. These findings suggest a structuralbasis for the altered platelet function associated with malignancy,and indicate that platelet ultrastructure could be useful as a bio-marker for ovarian cancer detection.To further substantiate our proof-of-principle observations

before becoming a clinical tool, larger patient numbers and ahigher-throughput protocol will be required. Recent advances incryo-ET suggest it can be used as a high-throughput method in aresearch laboratory (29–31), raising the possibility of its use inclinical laboratory. The increase in cell tomogram contrast usingthe Zernike phase plate technology in cryo-ET will likely easethe recognition and quantification of the observed subcellularfeatures (32). Software algorithms are being actively developedfor automatic feature extraction of filamentous features (33).Therefore, the ability of cryo-ET to measure fine ultrastructure

aberrations opens the possibility for this technology to be used forlarge-scale biomarker development in this and other disease states.

MethodsCryo-ET of Human Platelets. PRPwas separated from drawn blood samples andthen vitrified for cryo-ET. Women with benign adnexal masses had a finalpathological diagnosis of ovarian cystadenoma/fibroma (n = 4) or corpusluteum cyst (n = 1). Protocol was approved by Baylor College of MedicineInstitutional Review Board, protocol no. H-12278. Patients with ovariancancer had stage III–IV high-grade serous ovarian cancer or other ovarianmalignancies (n = 12). All patients provided written informed consent forparticipation in this study. All samples were obtained preoperatively, beforeany chemotherapy treatment. Blood (2.5 mL) was drawn using a 21-gaugeneedle Vacutainer brand blood collection set (Becton Dickinson) into 3.8%(wt/vol) sodium citrate polyethylene tubes. The blood was immediatelycentrifuged at 150 × g for 20 min, and the PRP was collected. Blood and PRPwere maintained at room temperature during manipulation. Quantifoilholey carbon supported transmission electron microscope grids with 3.5-μmcircular holes (Quantifoil Micro Tools GmbH) pretreated with colloidal gold(fiducial tracer, 15 nm) were glow-discharged (15 s) before use. PRP (3 μL)was applied to the grid and then blotted with calcium-free blotting paper,using a Vitrobot (Mark IV, FEI Corp), and immediately plunged into liquidethane at liquid nitrogen temperature to vitrify the platelets (1 blot, 3-swait, or 2 blot, 2-s wait).

All platelet samples were vitrified for cryo-ET study within 1 h of blood drawto prevent any potential activation thatmay disturb the platelet ultrastructure.

The frozen grids with platelets were then transferred at liquid nitrogentemperature into JEOL electron cryomicroscopes (JEM 2100, JEM 2200FS, orJEM 3200FSC; JEOL Ltd.) for imaging. Low-dose conditions were used topreserve the structural integrity. For JEM2200FS and JEM3200FSC micro-scopes, an in-column energy filter (slit = 15 eV) was applied to enhance theimage contrast by zero loss imaging. Using the SerialEM package (34), tiltseries were recorded at 200 keV/300 keV on a Gatan 4,096 × 4,096 pixel CCDcamera (Gatan Inc.). Images were collected at a defocus range of ∼8–15 μmand microscope magnification range of 8–12,000×. Total electron dose pertomogram was ∼75 electrons/Å2, as typically used for cryo-ET (35). Each tiltseries has 62–66 CCD frames with a fixed 2-degree increment.

A total of 338 tilt series of individual platelets were recorded in this study.For each individual under study, 10–20 platelets at suitable places for im-aging (e.g., not close to the grid bar) were selected without a particular bias,imaged, and processed. Platelets were selected for imaging on the basis oftheir positions on the microscopy grid, and not on internal features that arenot visible before 3D reconstruction. Of these 10–20 platelet tomograms, atleast five tomograms had adequate contrast to clearly visualize microtubulesand to distinguish double-membrane mitochondria from single-membraneα granule. In general, quantifiable tomograms were randomly selected forquantification. Some subjects yielded fewer than five visually good tomo-grams, and were not included for further tomographic reconstruction andstatistical analysis. The final data set of 120 tomograms was selected fromhealthy female controls (n = 6), women with benign adnexal masses (n = 5),and patients with newly diagnosed invasive ovarian cancer (n = 12).

Platelet Tomogram Processing and Quantification. Image stacks were alignedwith gold tracers, using the IMOD (version 4.0.27) package (36), yielding 3Dreconstructed tomograms (voxel 4k × 4k × 1k, ∼11–13 Å per pixel sampling).To enhance the contrast, tomograms were averaged by two and filtered bydenoising tools such as low-pass and nonlinear anisotropic diffusion (37) inthe EMAN2 (38) and IMOD packages. Structural features such as α granules,dense granules, mitochondria, microtubules, and membrane systems wereidentified and manually annotated using Amira (version 5.2; Visage ImagingInc.). Microtubules were clearly seen as a coiled circumferential marginalband in the control group and as partially fragmented filaments in someplatelets in the cancer group. Granules with very dark densities were an-notated as dense granules. Single membrane-bound granules with graydensities were annotated as α granules. Organelles with double membraneswere annotated as mitochondria. The inner membrane of mitochondria wasseen to form cristae-like structures.

Structural features (microtubule length, α granule number, α granulearea, dense granule number, dense granule area, mitochondria number,mitochondria area, pseudopod number) were manually quantified for eachindividual platelet tomogram, using Amira. The number of pseudopods,α and dense granules, and mitochondria was counted with IMOD. The 3Dtomogram of the whole platelet was projected in Z direction, and the areaof the whole platelet enclosed by the plasma membrane in the projectionimage was measured as platelet area. A small fraction of platelets in the

14270 | www.pnas.org/cgi/doi/10.1073/pnas.1518628112 Wang et al.

cancer group had part of their platelet area beyond the CCD frame. Forthese platelets, the area of the polygon was measured as an approximationfor the whole platelet area. For mitochondria, the area of their central slicewas measured individually in each tomogram and summed up as the areaof mitochondria. The mitochondria area was then divided by the wholeplatelet area to estimate the percentage of mitochondria area. α and densegranule areas were measured in the same way. A small fraction of α anddense granules adopt an irregular shape compared with the sphericallyshaped majority. Irregularly shaped granules were projected in the Z di-rection first, and their areas in the projection image were measured. Formicrotubules, a mask was drawn in their central slice and then was correctedby an algorithm to remove redundant annotation (i.e., measurement of thesame microtubule on consecutive sections of the tomogram; Movie S3), andthe total area of the mask was measured in Amira. The algorithm to removeredundant microtubule mask is a Python script using two functions that wasapplied offline to the manually selected 3D mask that was produced byAMIRA. This area was then divided by 250 Å (the microtubule diameter) toestimate the microtubule length (Fig. 3). No change in significance wasobserved for any comparison when the redundancy-removing algorithm wasomitted. All the quantification was performed by the same researcher tomaintain the consistency for the entire study.

Statistical Analysis of Quantified Platelet Tomograms. Unpaired, two-tailedStudent t test was performed to test for statistically significant differences inthe measured features in the platelet tomograms between the two groups(control group that comprises healthy donors and patients with benign mass;cancer group that comprises patients with ovarian cancer), with a P < 0.05considered to be significant.

Predictive Modeling.Weevaluated thepotential of cryo-ET todistinguish betweensubjects with or without malignancy. Each subject’s disease state was coded asmalignant (n = 12) or nonmalignant (n = 11). Number of mitochondria and totalmicrotubule length were selected as predictors for discriminant function analysisbased on the qualitative observations of the individual who performed

the segmentation (R.W.). Although each predictor was measured aboutfive times per subject, we collapsed these subsamples into one mean persubject per predictor, which is a conservative approach. Because thepatient sample size is small, we used leave-one-out cross validation (18).A discriminant function is built from all but one of the subjects and usedto predict the malignancy status of the omitted subject. This procedure isrepeated 23 times, leaving out one subject each time. Calculations wereperformed in JMP version 11.1 (SAS Institute).

Platelet Aggregation Assay. Blood samples were collected, using 0.38% so-dium citrate as the anticoagulant (final concentration) from control subjectsand patients with invasive ovarian cancer under an approved institutionalreview board protocol. They were centrifuged at 150 × g for 15 min at 25 °Cto collect PRP. An aliquot of PRP was loaded into a microcuvette and in-cubated for 10 min at 37 °C. Platelet aggregation was initiated by addingADP (2.5, 5, and 10 μM) and monitored for 10 min in an eight-channeloptical aggregometer (PAP8; Bio/Data Corp.). ADP-induced platelet ag-gregation was tested at multiple doses to detect platelet hyperreactivity,which was defined as ≥30% of aggregation by subthreshold concen-trations of the agonist (2.5 and 5 μM), compared with the standard doseof 10 μM that induces maximal aggregation. Platelet aggregation assayswere completed within 2 h after blood collection to prevent spontaneousplatelet aggregation.

For the aggregation assay, data were analyzed by pair comparisons tocompare levels of platelet aggregation between control subjects and pa-tients, and by χ2 test to compare the numbers of subjects who were con-sidered to have hyperactive platelets between patients and controls.

ACKNOWLEDGMENTS. This research has been supported by NIH GrantsP41GM103832, HL071895, HL085769, HL081613, and CA177909; Departmentof Defense Grants OC120547 and OC093416; an Ovarian Cancer ResearchFund Program Project Development Grant; the Bettyann Asche MurrayDistinguished Professorship; and a Baylor College of Medicine and MDAnderson Cancer Center Collaborative Award.

1. Leslie M (2010) Cell biology. Beyond clotting: The powers of platelets. Science328(5978):562–564.

2. Patel-Hett S, et al. (2008) Visualization of microtubule growth in living platelets reveals adynamic marginal band with multiple microtubules. Blood 111(9):4605–4616.

3. Gay LJ, Felding-Habermann B (2011) Contribution of platelets to tumour metastasis.Nat Rev Cancer 11(2):123–134.

4. Bambace NM, Holmes CE (2011) The platelet contribution to cancer progression.J Thromb Haemost 9(2):237–249.

5. Borsig L (2008) The role of platelet activation in tumor metastasis. Expert RevAnticancer Ther 8(8):1247–1255.

6. Holmes CE, Levis JE, Ornstein DL (2009) Activated platelets enhance ovarian cancercell invasion in a cellular model of metastasis. Clin Exp Metastasis 26(7):653–661.

7. Levin J, Conley CL (1964) Thrombocytosis Associated with Malignant Disease. ArchIntern Med 114:497–500.

8. Stone RL, et al. (2012) Paraneoplastic thrombocytosis in ovarian cancer. N Engl J Med366(7):610–618.

9. Labelle M, Begum S, Hynes RO (2011) Direct signaling between platelets and cancercells induces an epithelial-mesenchymal-like transition and promotes metastasis.Cancer Cell 20(5):576–590.

10. Jurasz P, Alonso-Escolano D, Radomski MW (2004) Platelet–cancer interactions:Mechanisms and pharmacology of tumour cell-induced platelet aggregation. Br JPharmacol 143(7):819–826.

11. Goubran HA, Stakiw J, Radosevic M, Burnouf T (2014) Platelet-cancer interactions.Semin Thromb Hemost 40(3):296–305.

12. Gungor T, Kanat-Pektas M, Sucak A, Mollamahmutoglu L (2009) The role of throm-bocytosis in prognostic evaluation of epithelial ovarian tumors. Arch Gynecol Obstet279(1):53–56.

13. White JG, Krumwiede M (2007) Some contributions of electron microscopy toknowledge of human platelets. Thromb Haemost 98(1):69–72.

14. White JG (2004) Electron microscopy methods for studying platelet structure andfunction. Methods Mol Biol 272:47–63.

15. White JG (1998) Use of the electron microscope for diagnosis of platelet disorders.Semin Thromb Hemost 24(2):163–168.

16. Luci�c V, Leis A, Baumeister W (2008) Cryo-electron tomography of cells: Connectingstructure and function. Histochem Cell Biol 130(2):185–196.

17. Boudreau LH, et al. (2014) Platelets release mitochondria serving as substrate forbactericidal group IIA-secreted phospholipase A2 to promote inflammation. Blood124(14):2173–2183.

18. Shao J (1993) Linear Model Selection by Cross-Validation. J Am Stat Assoc 88(422):486–494.

19. Yee DL, Sun CW, Bergeron AL, Dong JF, Bray PF (2005) Aggregometry detects platelethyperreactivity in healthy individuals. Blood 106(8):2723–2729.

20. Tocheva EI, Li Z, Jensen GJ (2010) Electron cryotomography. Cold Spring Harb Perspect

Biol 2(6):a003442.21. Jacobs MR (2002) In vivo veritas: In vitro macrolide resistance in systemic Streptococcus

pneumoniae infections does result in clinical failure. Clin Infect Dis 35(5):565–569.22. Jensen GJ, Briegel A (2007) How electron cryotomography is opening a new window

onto prokaryotic ultrastructure. Curr Opin Struct Biol 17(2):260–267.23. van Nispen tot Pannerden H, et al. (2010) The platelet interior revisited: Electron

tomography reveals tubular alpha-granule subtypes. Blood 116(7):1147–1156.24. Blair P, Flaumenhaft R (2009) Platelet alpha-granules: Basic biology and clinical cor-

relates. Blood Rev 23(4):177–189.25. Charafeddine AH, et al. (2012) Platelet-derived CD154: Ultrastructural localization

and clinical correlation in organ transplantation. Am J Transplant 12(11):3143–3151.26. Zharikov S, Shiva S (2013) Platelet mitochondrial function: From regulation of

thrombosis to biomarker of disease. Biochem Soc Trans 41(1):118–123.27. Schwertz H, et al. (2010) Anucleate platelets generate progeny. Blood 115(18):

3801–3809.28. Holmes CE, Ramos-Nino ME, Littenberg B (2010) An association between anti-platelet

drug use and reduced cancer prevalence in diabetic patients: Results from the Vermont

Diabetes Information System Study. BMC Cancer 10:289.29. Liu J, Wright ER, Winkler H (2010) 3D visualization of HIV virions by cryoelectron

tomography. Methods Enzymol 483:267–290.30. Suloway C, et al. (2009) Fully automated, sequential tilt-series acquisition with Leginon.

J Struct Biol 167(1):11–18.31. Nickell S, et al. (2005) TOM software toolbox: Acquisition and analysis for electron

tomography. J Struct Biol 149(3):227–234.32. Dai W, et al. (2013) Visualizing virus assembly intermediates inside marine cyano-

bacteria. Nature 502(7473):707–710.33. Rigort A, et al. (2012) Automated segmentation of electron tomograms for a quan-

titative description of actin filament networks. J Struct Biol 177(1):135–144.34. Mastronarde DN (2005) Automated electron microscope tomography using robust

prediction of specimen movements. J Struct Biol 152(1):36–51.35. Iancu CV, Wright ER, Heymann JB, Jensen GJ (2006) A comparison of liquid nitrogen

and liquid helium as cryogens for electron cryotomography. J Struct Biol 153(3):

231–240.36. Kremer JR, Mastronarde DN, McIntosh JR (1996) Computer visualization of three-

dimensional image data using IMOD. J Struct Biol 116(1):71–76.37. Frangakis AS, Hegerl R (2001) Noise reduction in electron tomographic reconstructions

using nonlinear anisotropic diffusion. J Struct Biol 135(3):239–250.38. Tang G, et al. (2007) EMAN2: An extensible image processing suite for electron

microscopy. J Struct Biol 157(1):38–46.

Wang et al. PNAS | November 17, 2015 | vol. 112 | no. 46 | 14271

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY