nondestructive tissue analysis for ex vivo and in vivo ...conventional methods for histopathologic...

SC I ENCE TRANS LAT IONAL MED I C I N E | R E S EARCH ART I C L E

CANCER D IAGNOST I CS

1Department of Chemistry, University of Texas at Austin, Austin, TX 78712, USA. 2De-partment of Biomedical Engineering, University of Texas at Austin, Austin, TX 78712,USA. 3Department of Pathology and Immunology, Baylor College ofMedicine, Houston,TX 77030, USA. 4Department of Surgery, Baylor College ofMedicine, Houston, TX 77030,USA. 5Department of Pathology, University of Texas MD Anderson Cancer Center,Houston, TX 77030, USA. 6Department of Internal Medicine, Dell Medical School, Uni-versity of Texas at Austin, Austin, TX 78712, USA.*Corresponding author. Email: [email protected]

Zhang et al., Sci. Transl. Med. 9, eaan3968 (2017) 6 September 2017

Copyright © 2017

The Authors, some

rights reserved;

exclusive licensee

American Association

for the Advancement

of Science. No claim

to original U.S.

Government Works.

http://stm.sc

Dow

nloaded from

Nondestructive tissue analysis for ex vivoand in vivo cancer diagnosis using a handheldmass spectrometry systemJialing Zhang,1 John Rector,1,2 John Q. Lin,1 Jonathan H. Young,1 Marta Sans,1 Nitesh Katta,2

Noah Giese,1 Wendong Yu,3 Chandandeep Nagi,3 James Suliburk,4 Jinsong Liu,5

Alena Bensussan,1 Rachel J. DeHoog,1 Kyana Y. Garza,1 Benjamin Ludolph,1 Anna G. Sorace,6

Anum Syed,2 Aydin Zahedivash,2 Thomas E. Milner,2 Livia S. Eberlin1*

Conventional methods for histopathologic tissue diagnosis are labor- and time-intensive and can delay decision-making during diagnostic and therapeutic procedures. We report the development of an automated and bio-compatible handheld mass spectrometry device for rapid and nondestructive diagnosis of human cancer tissues.The device, named MasSpec Pen, enables controlled and automated delivery of a discrete water droplet to a tissuesurface for efficient extraction of biomolecules. We used the MasSpec Pen for ex vivo molecular analysis of 20 humancancer thin tissue sections and 253 human patient tissue samples including normal and cancerous tissues frombreast,lung, thyroid, and ovary. The mass spectra obtained presented rich molecular profiles characterized by a variety ofpotential cancer biomarkers identified asmetabolites, lipids, and proteins. Statistical classifiers built from the histolog-ically validated molecular database allowed cancer prediction with high sensitivity (96.4%), specificity (96.2%), andoverall accuracy (96.3%), as well as prediction of benign and malignant thyroid tumors and different histologic sub-types of lung cancer. Notably, our classifier allowed accurate diagnosis of cancer inmarginal tumor regions presentingmixed histologic composition. Last, we demonstrate that theMasSpec Pen is suited for in vivo cancer diagnosis duringsurgery performed in tumor-bearing mouse models, without causing any observable tissue harm or stress to theanimal. Our results provide evidence that the MasSpec Pen could potentially be used as a clinical and intraoperativetechnology for ex vivo and in vivo cancer diagnosis.

ienc
by guest on A
pril 1, 2020em

ag.org/

INTRODUCTIONTissue assessment and diagnosis are critical in the clinical managementof cancer patients. Tissue diagnosis is particularly important duringsurgical excision of solid cancers for surgical margin evaluation. Manywomen diagnosed with breast cancer, for example, undergo breast-conserving surgery, which involves removing the lesion of interest witha rim of normal tissue and preserving the rest of the breast. One of thegreatest challenges a breast cancer surgeon faces is determining the del-icate boundary between cancerous and normal tissues to achieve nega-tivemargins for invasive and carcinoma in situwhile optimizing aestheticoutcomes (1). Similarly, optimal surgical treatment of lung carcinomasincludes complete local resection of the primary tumor (2) because ad-verse patient outcome is strongly associated with residual tumor at thebronchial resection margin (3). For high-grade serous ovarian cancer(HGSC) patients, postoperative residual disease after surgical debulkingis also negatively associated with progression-free survival and responseto adjuvant chemotherapy (4). Thus, accurate negativemargin assessmentand complete tumor excision are highly desirable across cancer surgeriesbecause they offer the greatest potential for prolonged disease-free andoverall survival (1, 5–7).

Intraoperative assessment of the extent of tumor involvement can bechallenging through conventional histopathologic analysis of frozensections. Frozen section preparation is time- and labor-intensive and

requires skilled technicians and pathologists to produce and interpretthe results. Moreover, freezing artifacts can negatively interfere withtissue structure and cellmorphology, thus complicating pathological in-terpretation. Logistically, intraoperative frozen section analysis prolongsoperative time, subjecting the patient to increased risks related to ex-tended anesthesia. Therefore, margin specimens are frequently processedpostoperatively as permanent specimens. However, when positivemargins are found during the final pathologic evaluation, the patientis subjected to additional surgical procedures for re-excision of the in-volved margin, which increases health care costs and places the patientat risk for additional surgical complications, discomfort, and anxiety (1, 8).

Molecular analysis of cancer tissues offers the exciting opportunityto incorporate cancer-specific biomarkers into clinical decision-makingfor improved cancer detection anddiagnosis. Severalmolecular imagingtechnologies have been developed and advanced to preclinical and clin-ical phases for ex vivo and in vivo tissue diagnosis. Immunohisto-chemistry protocols targeting protein biomarkers are routinely used indiagnostic pathology for postoperative evaluation of ex vivo tissue sec-tions and typing of neoplasms (9). Gene sequencing technologies arepowerful for postoperative identification of specific genetic mutationsand chromosomal translocations in ex vivo tissue samples (10). Intra-operative real-time techniques including fluorescent probes that targettumor cells are currently being implemented for in vivo tumor and sur-gical margin visualization (11). Emerging optical technologies includingRaman spectroscopy and stimulated Raman scattering microscopyhave been recently applied for intraoperative diagnosis of brain cancers(12, 13). Mass spectrometry (MS) imaging approaches have been suc-cessfully applied for molecular imaging of cancer tissues (14–17).

Within the last decade, several ambient ionization MS techniqueshave been developed for rapid molecular diagnosis of cancer tissues

1 of 11

http://stm.sciencemag.org/


http://stm.s

Dow

nloaded from

and have shown exceptional potential for clinical use (18). Desorptionelectrospray ionizationMS imaging (DESI-MSI), for example, has beenused for ex vivo cancer diagnosis and surgicalmargin evaluation of tissuesections and tissue smears (19, 20). Yet, technical incompatibilities in-cluding the use of a spray of organic solvents, high-pressure nebulizinggas, and high voltage have prevented the use ofDESI-MSI for fresh tissueand in vivo analyses. A few approaches for direct MS analysis of cancertissue specimens have been developed (18). Rapid evaporative ionizationMS, or the iKnife, combines an electrocauterization device with MS fordirect tissue analysis and classification and has been successfully usedintraoperatively for in vivo cancer diagnosis (21, 22). Ultraviolet and in-frared lasers have also been coupled with MS for characterization ofcancer tissues (23, 24). Although these approaches offer the advantageof incorporating common surgicalmethods into anMS-based diagnosticworkflow, these technologies rely on tissue damage to producemolecularions or are operationally constrained to a specific surgical modality.

Here, we describe the development and application of an automated,biocompatible, disposable handheld device, theMasSpec Pen, for direct,real-time nondestructive sampling and molecular diagnosis of tissues.We tested the MasSpec Pen for ex vivo molecular evaluation of humannormal and cancerous tissue samples from 253 patients. The massspectra obtained presented rich molecular information including diag-nosticmetabolites, lipids, and proteins. Statistical analysis using the leastabsolute shrinkage and selector operator (Lasso) technique allowed pre-diction of cancer with high sensitivity and specificity (25). In a tumor-bearingmousemodel, we demonstrate that this technology is suited forin vivo use and diagnosis of breast cancer during surgery.

by guest on April 1, 2020

ciencemag.org/

RESULTSOptimization of the MasSpec Pen design and operationWe designed the MasSpec Pen (Fig. 1A) as an automated and bio-compatible handheld sampling probe that allows gentle and time- andvolume-controlled extraction of molecules from a tissue sample usinga discrete water droplet (movie S1). Several prototypes of the systemwere engineeredwith the goal ofminimizing tissue damage, maximizingtissue-analyte extraction, and maximizing solvent transfer to the massspectrometer. The optimized systemcontains three primary components:(i) a syringe pump that is programmed to deliver a defined water vol-ume (4 to 10 ml) to the sampling probe; (ii) small diameter [inner diam-eter (ID), 800 mm] polytetrafluoroethylene (PTFE) tubing conduits,which are integrated to a fast (8ms) two-way pinch valves for controlledsolvent transport from the pump to the tissue and from the tissue to themass spectrometer; and (iii) a handheld pen-sized probe for directsampling of biological tissues.

The main component of the handheld pen-sized probe is a three-dimensional (3D) printed polydimethylsiloxane (PDMS) tip (Fig. 1B).The probe tip is designedwith three ports: an incoming port that deliversa single water droplet to the probe tip (conduit 1), a central port for gas(N2, CO2, or air) delivery (conduit 2), and an outgoing port to transportmolecular constituents in the water droplet from the tissue to the massspectrometer (conduit 3). At the probe tip, all ports combine into a smallreservoir where a single water droplet is retained and exposed to thetissue sample for a controlled amount of time (3 s), allowing efficientanalyte extraction. After the 3-s extraction period, the MasSpec Pen isremoved from the tissue. Concomitantly, conduit 3 is opened, allowingvacuum extraction of the droplet to the mass spectrometer, whereaspositive pressure from a low-pressure gas delivery (<10 psi) is providedthrough conduit 2 (Fig. 1C). The gas provided by the second tube does


not participate in the extraction process but is used instead to preventthe collapse of the system due to the vacuum used and to assist solventtransport from the tissue to the mass spectrometer. A subsequent flushstep cleans the system; this is not used for extraction of biomoleculesfrom tissues because there is no contact with the tissue during this pe-riod. Conduit 3 is directly connected to the transfer tube of a high–massresolution Orbitrap mass spectrometer so that the negative pressureof themass spectrometer vacuum system drives themovement of thedroplet from the reservoir to the mass spectrometer for ionization andmass analysis. This setup simplifies the operational steps and precludesthe use of ionization sources, although various connection and ioniza-tion methods could be coupled to our system.

The diameter of the reservoir at the probe tip determines the volumeof solvent exposed to the tissue and the spatial resolution of the device.Using current tooling, we have designed MasSpec Pen tips with sam-pling sizes ranging from 1.5 to 5.0 mm, which is determined by thereservoir diameter. At a 2.77-mm reservoir diameter, a solvent vol-ume of 10 ml is retained in the reservoir and contacts the tissue samplefor a defined time period, whereas 4.4 ml is retained in a reservoir with a1.5-mmdiameter. Contact times of 1, 3, and 5 s between the droplet andthe tissue sample were evaluated (fig. S1). The 3-s contact time wasselected for all the experiments because it allowed ease of operationby the user and yielded mass spectra with sufficient total ion intensity.A tube length of 1.5 m was used for all the conduits to allow freehandheld use of the device by an operatorwithout geometrical or spatialconstraints.

The tip design using three conduit tubes and high-speed actuatedpinch valves allowed precise control of droplet motion and showed ex-cellent performance and robustness. The entire process from samplingtomass spectral acquisition is completed in 10 s or less and is automatedusing an Arduino microcontroller so that each acquisition and analysisis individually triggered through a one-step click using a foot pedal. Sys-tem automation ensures that each solvent droplet is delivered separatelyto the inlet, yielding several mass spectra that are averaged for a finalmolecular profile of the sample. Controlled droplet delivery allowedthemass spectrometer to operatewithout any evident performance deg-radation. After each use, the MasSpec Pen was cleaned through a rapidand automated cleaning flush or by replacing the disposable tip. Con-tamination of the mass spectrometer was evaluated by installing com-mercially heated ESI source and acquiring mass spectra after theMasSpec Pen analysis. No lipid contamination was observed other thanbackground peaks commonly observed in the mass spectra.

Molecular analysis of thin tissue sections using theMasSpec PenWe tested the effectiveness of theMasSpec Pen by analyzing thinmouseand human tissue sections and pieces of tissue samples. First, 16-mm-thick tissue sections were analyzed on standard histologic glass slidesfollowing the operational steps described above for the MasSpec Pen,using pure water as the solvent. Several probe tips with different reser-voir diameters of the MasSpec Pen were tested, yielding mass spectrapresenting lipid species characteristic ofmouse brain tissue graymatter,white matter, or mixed composition for larger sampling sizes (fig. S2)(26). Figure S3 shows a representative mass spectrum obtained in thenegative ion mode using a 2.7-mm pen tip from the gray matter regionof a mouse brain tissue section and a representative background massspectrum obtained from a region of glass slide (no sample).

The negative ion mode mass spectrum obtained from the glass slideregion presented background ions from a mass/charge ratio (m/z) of

2 of 11




http://stm.sciencem

ag.org/D

ownloaded from

120 to 400, corresponding to common solvent clusters and chemicalcontaminants present on the glass slide or within themass spectrom-eter. The negative ion mode mass spectrum obtained from the graymatter region of the mouse brain tissue section presented rich mo-lecular information, including a variety of ions corresponding to de-protonated or chloride adducts of lipid species commonly detectedfrom biological tissues using solvent-based ambient ionization MStechniques (16). Peaks at high relative abundances were identifiedas fatty acids (FA) fromm/z 120 to 350, sphingolipids such as sulfa-tides from m/z 700 to 1100 and chloride adducts of ceramides (Cer)fromm/z 500 to 700, and glycerophospholipids (GL) such as glycer-ophosphoinositols (PI), glycerophosphoethanolamines (PE), glycer-ophosphoserines (PS), and doubly charged cardiolipins (CL) fromm/z 700 to 1100. In the higher mass range from m/z 1100 to 1800,GL dimers and singly charged CL were observed. A variety of peakstentatively identified as small metabolites including glutamine atm/z 145.061, glutamate at m/z 146.045, N-acetylaspartic acid atm/z 174.041, and chloride adduct of hexose at m/z 215.033 weredetected in the lower mass range from m/z 120 to 250, based on high–mass accuracy measurements and tandem MS data (table S1). Thenegative ion mode mass spectra obtained from the gray matter fromdifferent tissue sections of the same mouse brain were reproducible[relative standard deviation (RSD) = 9.3%; n = 9], comparable to whathas been reported using the samemethod for DESI-MSI (RSD = 8.0%;n = 5) (26).

In the positive ion mode, the mass spectrum obtained presentedhigh relative abundances of commonly observed molecular speciesidentified as diacylglycerols, PE, and glycerophosphocholines (fig. S4).Tentative assignments were performed using high–mass accuracymea-surements, as well as tandemMS analysis when adequate intensity offragment ions was achieved for structural interpretation (see Identi-


fication of molecular ions in the SupplementaryMaterials). Mass errorsand the m/z of fragment ions obtained by tandem MS experimentsare described for all the species identified throughout the study (tablesS1 to S5). Because isomerism of the double bonds in the FA chains ofcomplex lipids complicates precise structural assignment, FA chainswere tentatively assigned for lipid species.

To evaluate the extraction efficiency and molecular profiles ob-tained, we compared the molecular species detected in the negativeion mode using the MasSpec Pen with those observed in DESI-MSIacquired from a serial tissue section of the same mouse brain usingwater as the solvent and under analogous experimental conditions. Themass spectra obtained using the MasSpec Pen and DESI-MSI weresimilar with a calculated cosine similarity of 0.9 (see Data analysis inthe Supplementary Materials), sharing a large number of molecularspecies at comparable relative abundances and signal-to-noise ratios(fig. S5). Other solvent systems including mixtures of water with eth-anol at various ratios were also explored as solvent systems for theMasSpec Pen. Themass spectra obtained presented similar lipid speciesto those observed in those obtained using pure water, with variations intheir relative abundances (fig. S6). To ensure full biocompatibility of ourdevice, we selected water as the solvent for all subsequent MasSpec Penexperiments performed.

We then tested the capability of the MasSpec Pen to analyze 20 thintissue sections of human breast tissues (n = 5 normal breast and 5 breastductal carcinoma tissues) and thyroid tissues [n = 5 normal thyroid, 4papillary thyroid carcinoma (PTC), and 1 follicular thyroid adenoma(FTA) tissues]. The mass spectra obtained in the negative ion modefor each tissue type presented a rich variety ofmolecular ions commonlyobserved from human tissues by DESI-MSI (18), with high relativeabundances of metabolites, FA, and complex lipids. For example, themass spectra obtained for PTC tissue sections presented lipid species

Fig. 1. Schematic representation of the MasSpec Pen system and operational steps. (A) The pen-sized handheld device is directly integrated into a laboratory-built MSinterface through PTFE tubing. The integratedMS interface houses the pinch valves,microcontroller, and tubing to connect the system to themass spectrometer inlet. The systemis triggered by the user through a foot pedal. (B) TheMasSpec Pen (handheld device) is designedwith a PDMS tip and three PTFE conduits, which provide incomingwater (1) andgas (2) to the tip and anoutgoing conduit for thewater droplet (3). (C) The tip contacts the tissue for analysis. Inset shows the three conduits (1 to 3) and solvent reservoir (4)withinthe tip. When the system is triggered (t = 0 s) by using the foot pedal, the syringe pump delivers a controlled volume of water to the reservoir. The discrete water droplet interactswith the tissue to extract the molecules (t = 2 s). After 3 s of extraction, the vacuum and the gas conduits are concomitantly opened (arrows) to transport the droplet from theMasSpec Pen to the mass spectrometer through the tubing system for molecular analysis.

3 of 11



http://stmD

ownloaded from

previously identified as diagnostic markers by DESI-MSI (27), includ-ing a variety of doubly charged CL and other GL such as PI(38:4)(m/z 885.550), PI(36:4) (m/z 857.518), PE(38:4) (m/z 766.539), and PE(36:2) (m/z 742.539) (table S2). A distinct mass spectral profile was ob-tained for the normal thyroid tissue sections, presenting high relativeabundances of iodine (m/z 126.904), glutamine (m/z 145.050), ascorbicacid (m/z 175.024), C36H78O9N3I (tentatively assigned; m/z 822.472),and PI(38:4) (m/z 885.551) (Fig. 2). A series of multiply chargedmolec-ular ions at different charge states (z) includingm/z 991.091 (z = −5),m/z 1239.113 (z = −4), andm/z 1652.484 (z = −3) were detected in themass spectra obtained from all tissue sections analyzed. These ions weretentatively identified as different charge states of the protein thymosinb-4 based on high–mass accuracy measurements (fig. S7 and table S1).Principal components analysis (PCA) performed on the data obtainedfrom the human tissue sections showed separation between tumor andnormal tissues (fig. S8).

Nondestructive molecular analysis of tissue samplesThe MasSpec Pen was designed to operate directly on tissue specimensindependently of tissue stiffness and morphology. We tested theperformance of the MasSpec Pen to analyze soft tissue samples (0.1to 5 g) from different organs including mouse brain and human breast,thyroid, lung, and ovary tissues. Tissue analyses were performed underambient conditions through a simple one-step experiment, followingthe same operational steps described previously. The MasSpec Pen tipwas gently brought into contact with the surface of the tissue sample for


a period of 3 s during which extraction took place. The mass spectraobtained for a region of graymatter probed from a piece of freshmousebrain tissue were reproducible (RSD = 4.6%; n = 10) and highly similarto themouse brain tissue sectionmass spectra (cosine similarity of 0.93),indicating that the extraction process at the tissue surface occurs inde-pendently of the tissue shape and rigidity (fig. S9). MasSpec Pen analysesof human tissue samples provided similar rich molecular information,especially of tissues composed of epithelial and cancerous cells.Noncan-cerous tissue specimens that were mostly composed of soft connectivetissue such as stroma provided less abundant mass spectra. Many of thenormal breast tissue samples analyzed presented abundant fat content,which is immiscible with water and thus yielded lower total ion countsin the mass spectra when compared to breast cancer tissues or normalbreast cancer glands.

Visual and microscopic inspection of the tissue samples after theMasSpec Pen analysis revealed no detectable damage to the tissue mor-phology in the region probed. Figure 3A shows optical images obtainedfrom a lung tissue sample before, during, and after the MasSpec Penanalysis. No observable damage to the tissue was seen at the region ana-lyzed, and richmass spectra were obtained (Fig. 3B). The automated andtime-controlled operational steps of the MasSpec Pen prevent tissuedamage because the tissue is only exposed to the small water dropletand not to the vacuum used to transport the droplet from the reser-voir to the mass spectrometer. These results provide evidence thatthe MasSpec Pen can obtain rich molecular information from tissuesamples nondestructively.


.sciencemag.org/

Fig. 2. MasSpec Pen analysis of PTC and normal thyroid tissue sections. (A) A representative negative ionmodeMasSpec Penmass spectra obtained from a normal humanthyroid tissue section (average of n = 5mass spectra) and (B) a PTC tissue section (average of n = 4mass spectra) are shown. Identification of themost abundantmolecular ions isprovided. Insets shows an optical image of the H&E-stained tissue sections evaluated by histopathology.

4 of 11




http://stm.sciencem

ag.org/D

ownloaded from

Molecular diagnosis and statistical prediction of cancer inhuman tissuesWe next evaluated whether the molecular information obtained fromhuman tissue samples using the MasSpec Pen was diagnostic and pre-dictive of disease state. We analyzed a total of 253 human tissue speci-mens using the MasSpec Pen, including 95 lung samples (47 normaland 48 cancer samples including 17 adenocarcinoma, 17 squamous cellcarcinoma, and 14 cancer samples of other histologic subtypes), 57ovary samples (29 normal and 28 HGSC samples), 56 thyroid samples(27 normal, 11 FTA, and 18 PTC samples), and 45 breast samples (29normal and 16 ductal carcinoma samples). Patient demographic in-formation is provided in table S6. After the MasSpec Pen analysis, theregion analyzed was demarcated and registered through a series ofoptical images. Parallel pieces of the samples were frozen, sectioned atthe demarcated region, hematoxylin and eosin (H&E)–stained, andevaluated by histopathology to derive a diagnosis. Only samples witha predominant cell composition and clear diagnosis were used to buildmolecular databases. The histologically validated mass spectra obtainedfor the cancerous samples presented molecular species identified asseveral lipids and metabolites previously described as potential disease


markers using ambient ionization MS techniques. For the lung cancertissue, characteristicmolecularmarkers such asPI(36:1) (asm/z 863.565),PG(36:2) (m/z 773.542), PG(34:1) (m/z 747.514), and FA(18:1)(m/z 281.249) were observed (Fig. 3B and table S4). For the normallung, m/z 885.550, identified as PI(38:4), and m/z 744.552, identifiedas PE(36:1), were observed. Themass spectra obtained for breast cancertissue presented diagnostic lipidmarkers previously described byDESI-MSI (28, 29), includingm/z 885.550, identified as PI(38:4),m/z 863.565,identified as PI(36:1), m/z 773.542, identified as PG(36:2), and severalFA such asm/z 303.233, identified as FA(20:4), andm/z 281.249, iden-tified as FA(18:1) (table S5). PCA performed on the data obtained fromthe 253 human tissue samples analyzed showed separation betweencancer and normal tissues for each organ (Fig. 4).

To evaluate whether theMasSpec Pen molecular signatures are pre-dictive of cancer and normal tissues, we applied the Lasso method tobuild classification models using the histologically validated mass spec-tral database. The performance of the model was evaluated through aleave-one-patient-out cross-validation approach and was measured bysensitivity and specificity for cancer, as well as accuracy and the areaunder the curve (AUC) (Table 1). For breast cancer (n = 45), 87.5%sensitivity, 100% specificity (AUC = 1.0), and an overall accuracy of95.6% were achieved, which is comparable to the results reported usingDESI-MSI (98.2% accuracy; n = 126) (29), the iKnife (95.5% accuracy;n = 10) (22), and matrix-assisted laser desorption/ionization imagingof lipids and proteins (94.1% accuracy; n= 68) (30). ForHGSC (n= 57),100% sensitivity, 89.7% specificity, and 94.7% accuracy were achieved(AUC = 0.98), which is also similar to classification results obtained byDESI-MSI (97.1% accuracy; n = 31) (31). For lung cancer (n = 96),97.9% sensitivity, 95.7% specificity, and 96.8% accuracy were achieved(AUC = 0.97). When predicting on the basis of lung cancer histologicsubtypes, 93.8 and 92.2% accuracy was achieved for squamous cell car-cinoma and adenocarcinoma, respectively. Thyroid tumor samples in-vestigated included benign FTA andmalignant PTC samples. A classifierfor each was built, yielding 94.7% accuracy for FTA and 97.8% accuracyfor PTC. Overall, 96.4% sensitivity, 96.2% specificity, and 96.3% accuracywere achieved for all four types of cancer investigated. These resultsdemonstrate that the molecular information obtained from humantissue samples by the MasSpec Pen can be used to identify cancer andindicate that the statistical classifiers built are robust and may beused in an automated approach for rapid clinical diagnosis of tissuesamples.

Intrasample analysis of histologically distinct and cancermargin tissue regionsWe evaluated the ability of the MasSpec Pen to identify histologicallydistinct regions in a single human tissue sample that contained regionsof HGSC adjacent to normal ovarian stroma tissue. Five regions of thetissue sample were analyzed consecutively using a MasSpec Pen with a1.5-mm probe tip diameter, as demarcated in the optical image shownin Fig. 5A. A tissue section of the sample including the regions analyzedby the MasSpec Pen was subjected to H&E staining and evaluated byhistopathology. Regions 1 and 2 were diagnosed by histopathology asnormal stroma, whereas regions 4 and 5 were diagnosed as HGSC.Region 3 was in the margin between the cancer and normal stromatissue regions, presenting ~50% tumor tissue and ~50% normal stromatissue (Fig. 5A, inset). Figure 5B shows the mass spectra obtained forregions 1, 3, and 5. The spectra obtained for region 5, HGSC, presentedcharacteristic lipid markers detected in the HGSC tissues analyzedex vivo to build our statistical classifier (table S3). The mass spectra

Fig. 3. Nondestructive molecular analysis of human tissue samples using theMasSpec Pen. (A) Optical images show a lung adenocarcinoma tissue sample before,during, and after the MasSpec Pen analysis. A magnification of the tissue specimen(inset) shows no macroscopic damage to the tissue region analyzed by the MasSpecPen. (B) Negative ion mode mass spectrum obtained for the tissue region analyzedincluding the identification of the most abundant molecular ions.

5 of 11




http://stm.sciencem

ag.org/D

ownloaded from

obtained for region 1, diagnosed as normal ovarian stroma tissue, pres-ented less abundantmolecular ions as also observed for the other stromatissues analyzed ex vivo. Region 3 presented molecular profilescharacteristic of HGSCwith lower total abundance due to the contribu-tion of normal stroma tissue present within the region analyzed. Themass spectra obtained for the five regions were then evaluated by ourovarian cancer molecular classifier as an independent validation set.Our classifier correctly identified regions 1 and 2 as normal and regions3 to 5 as cancer (Fig. 5C). Similar results were obtained for a differenttissue sample with histologically distinct regions (fig. S10). These resultsshow that the molecular information obtained by the MasSpec Pen canbe used to detect cancer inmarginal regions withmixed composition ofnormal and cancer cells.

In vivo analysis of a murine model of human breast cancerduring surgeryTheMasSpec Pen was designed with biocompatible materials to ensurefull compatibility as an in vivo molecular diagnostic tool. We tested theMasSpec Pen for in vivo tissue analysis using amurinemodel of humanbreast cancer. BT474 human epidermal growth factor receptor 2–positive (HER2+) breast cancer cells were implanted subcutaneously innude athymic mice (n = 3). The tumors were grown to an average of250mm3 over a period of 4 weeks. Under anesthesia, the skin overlyingthe tumors was dissected away, and several tissue regions were analyzedfollowing the same automated experimental steps described previously,including multiple positions of the top of the tumor, the core of the tu-mor after partial tumor resection, and adjacent normal soft connectivetissue. Figure 6A shows an optical image of the animal under anesthesiabefore initiation of surgery, before analysis (after surgical removal of theskin), during theMasSpec Pen analysis, and after the analysis. Themassspectra obtained for the tumor regions presentedmanymolecular speciesobserved inhumanbreast tissue,with adistinctive profile fromwhat was


obtained for adjacent nor-mal soft connective tissueregions (Fig. 6B). Usingoptical microscopy, no ob-servable macroscopic ormicroscopic damage to thetissue regions analyzedweredetected due to MasSpecPen analyses, as evidentfrom theoptical images ob-tained of the H&E-stainedtissue sections (fig. S11).Further, no apparent ef-fects to the health of theanimals were observed dueto the MasSpec Pen analy-sis during surgery. Afterin vivo analysis, freshly ex-cised tumor specimenswere also analyzed ex vivo,yielding mass spectra withcommon lipid species tothose observed during invivo analysis despite varia-tions in relative abundances,which are likely due to thereanalysis process of the

same tissue region (fig. S12). These results suggest that the MasSpecPen is suitable for in vivo molecular evaluation and cancer diagnosis.

DISCUSSIONWe developed the MasSpec Pen as an automated and biocompatiblehandheld sampling probe that allows gentle and time- and volume-controlled extraction of molecules from a tissue sample using a discretewater droplet. Our results provide evidence that the MasSpec Pen issuited for rapid ex vivo and in vivo cancer diagnosis of tissue samples.The mass spectra obtained from the analysis of 20 thin tissue sectionsand 253 human tissue samples presented rich molecular informationthat is diagnostic of disease state. The chemical extraction process usedis gentle so that the tissue is undamaged after molecular analysis. Statis-tical classifiers built from the mass spectra obtained provided high sen-sitivity and specificity for cancer detection (>96%), including predictionof histologic subtypes of lung cancer and benign andmalignant thyroidtumors. Experiments performed in animalmodels demonstrate that thistechnology is suitable for in vivo molecular evaluation of cancer andnormal tissues without causing observable tissue harm or evident stressto the animal.

The MasSpec Pen provides rich mass spectra from biological sam-ples characterized by a variety of singly and doubly charged ions oflipids and metabolites and multiply charged protein ions, similar tothose obtained by DESI-MSI and ESI methods. A liquid-solid chemicalextraction process is used to sample molecules from the tissue using awater droplet, without assistance from a gas or a strong electric field.Our results show that by depositing a discrete water droplet onto atissue sample for a determined amount of time, efficient extractionof biomolecules is achieved while tissue integrity is preserved (Figs. 3and 6). The extraction process is similar to that reported for liquid ex-traction surface analysis (32), liquid microjunction surface sampling

Fig. 4. PCA of the data obtained from human tissue samples using the MasSpec Pen. A total of 253 patient tissue samples wereanalyzed including breast (n = 45), thyroid (n = 56), ovary (n = 57), and lung (n = 96) cancer and normal tissue samples. 3D PCA (PC1,PC2, and PC3) score plots are shown for each tissue type. The first three PCs explain the 77, 69, 51, and 87% of the total variance of breast,thyroid, lung, and ovarian data sets, respectively.

6 of 11




http://stm.sciencem

ag.org/D

ownloaded from

probe (33), and nanoDESI (34), although the latter two techniques use acontinuous flow of solvent onto a sample surface, whereas theMasSpecPen uses a single water droplet for extraction. In our current system,the water droplet containing analytes is transported as a liquid sampleto the mass spectrometer and directly introduced to an extended trans-fer tube, which is connected to the heated inlet. Vaporization andionization occur in the inlet region of the mass spectrometer, similarto what has been observed in solvent-assisted inlet ionization (35).Additional means of ionization such as electric field or laser assist-ance were not applied, although various connection methods andionization sources (ESI, atmospheric pressure chemical ionization,atmospheric pressure photoionization, and others) could be adaptedfor our system.

Biocompatibility has been recognized as a key functional require-ment of next-generation medical devices (36). The MasSpec Pen wasdesigned as a simple, disposable device made with biocompatiblematerials and chemicals for contact with living tissues. The probe tipmaterial, the conduit tubes, and the chemical used were PDMS, PTFE,and water, respectively. PDMS and PTFE are widely recognized as bio-compatible materials and have a long history of utilization in medicaldevices including catheters and long-term implants (37). Our experi-ments demonstrate that the low volume (10 ml or less) of high-puritywater used caused no effect to the tissues analyzed in vivo and ex vivo.TheMasSpec Pen has shown good compatibility for contact with livingtissues, which should enable in vivo use.

We designed the MasSpec Pen as an automatic and user-friendlydevice that could be used during routine medical diagnosis (movie S1).A drawback of many ambient ionization MSmethods is the need forgeometrical optimization and alignment between the source, the sam-ple, and the mass spectrometer inlet to achieve good ion transmission.On the other hand, the MasSpec Pen operates in a geometry-free man-ner independently of sample stiffness and morphology and does not


require optimization because all of the components required for solventdelivery, tissue sampling, and solvent retrieval are incorporated withinthe tip. The 1.5-m tube transport systems enable free motion and easeof use to the operator. The high-speed electronically and time-controlledpinch valves allowed precise control of droplet motion and showedexcellent performance, robustness, and reproducibility in the resultsobtained.

Our results suggest that the MasSpec Pen may provide the per-formance required for near real-time nondestructive molecular di-agnosis of tissues in vivo and ex vivo. Intraoperative diagnosis andsurgical margin evaluation continue to be research focuses of clinicalambient ionization MS, with efforts used to develop devices for in vivodiagnosis (18). For example, a sampling probe using a DESI-MSIsource to produce a continuous high-pressure spray of microdropletshas been proposed for tissue analysis (38), but it has not been dem-onstrated for cancer diagnosis or in vivo use. Ultraviolet and infraredlaser approaches have been reported for tissue analysis through ab-lation processes (23, 24). A resonant infrared laser ablation systemwas used to analyze a cancer tissue sample ex vivo and human fingerskin in vivo (24). To date, the iKnife has been the most successful MS-based technique demonstrated for in vivo, intraoperative diagnosis(21, 22). Yet, the electrocauterization process used by the iKnife is nec-essary for ion generation for MS analysis, resulting in thermal andmechanical damage of the analyzed tissue. Further, surgicalmodalitiesother than electrocauterization are used in oncologic surgeries, suchas ultrasonic surgical aspiration, cold knife and mechanical stapledexcisions, and laser surgery. Different from laser and electrocauteriza-tion approaches, an advantage of the proposed water-based MasSpecPen technology is its nondestructive nature, which allows diagnosis ofcancerous tissues without damage to normal tissues. Because theMasSpec Pen performs molecular diagnosis independently of any dis-section tools, it has the potential to be used inmany surgical modalities.

Table 1. Human tissue sample details and results obtained using the MasSpec Pen. Pathological diagnosis, number of patient samples, and the Lassoprediction sensitivity, specificity, accuracy, and AUC obtained using a leave-one-out cross-validation approach are shown.

Organ
Pathologic evaluation Number of patients Lasso prediction
Diagnosis
Histologic type Sensitivity Specificity Accuracy AUC
Breast
Normal 29 87.5% 100.0% 95.6% 1.00
Cancer
Ductal carcinoma 16
Lung*
Normal 47 97.9% 95.7% 96.8% 0.97
Cancer
Adenocarcinoma 17 88.2% 93.6% 92.2% 0.98
Squamous cell
17 88.2% 95.7% 93.8% 0.93
Others
14 — — — —
Ovary
Normal 29 100.0% 89.7% 94.7% 0.98
Cancer
High-grade serous 28
Thyroid†
Normal 27 — — — —
Tumor
Papillary carcinoma 18 94.4% 100.0% 97.8% 0.99
Follicular adenoma
11 90.9% 96.3% 94.7% 0.93
*Lasso prediction results for lung are shown for normal versus all cancer tissues (first row), followed by normal versus lung adenocarcinoma (middle row) andnormal versus squamous cell carcinoma (last row). †Lasso prediction results for thyroid are shown for normal versus malignant papillary carcinoma andnormal versus benign follicular adenoma.

7 of 11




http://stm.sciencem

ag.org/D

ownloaded from

Intraoperatively, we envision the MasSpec Pen to be used in conjunc-tion with a surgical resection tool, depending on the oncologic surgeryand the preference of the medical professional. Because surgeonsroutinely exchange handheld surgical instruments during surgery,with assistance from surgical technologists, we expect the MasSpecPen to be adapted into routine surgical workflows. Clinically, theMasSpec Pen could be suitable for pre- and postsurgical procedures thatrequire diagnosis of ex vivo samples (fresh tissues, tissue sections, orbiopsies) commonly examined by pathologists.

Our results show that the molecular information obtained using theMasSpec Pen is diagnostic of disease and can be used for rapid tissueclassification, cancer diagnosis, and subtyping. We tested the MasSpecPenusing 253patient tissue samples including the normal and tumorousbreast, ovary, lung, and thyroid tissues. The mass spectra presented richmolecular profiles characterized by a variety of potential cancer bio-markers. The statistical classifiers built using machine learning algo-rithms and the histologically validated molecular information providedan overall sensitivity of 96.4%, specificity of 96.2%, and accuracy of96.3% for cancer. Our classifiers allowed the identification of differ-ent histologic types of lung cancers and benign and malignant thy-roid tumors when compared to normal tissue samples and correctlypredicted ovarian cancer diagnosis in an independent test sample with


mixed histologic features. Yet, as withany technology that relies on machinelearning and statisticalmodeling of largedata sets to provide predictive diagnosis,larger sample sets are needed to increasethe training set size, improve the predic-tive accuracy of our classifiers includinginformation on tumor cell concentra-tion, and expand the histologic andmo-lecular subtypes of neoplasms andnormal tissues. Rigorous validationstudies of the statistical results usinglarge independent held-out test sets arenecessary to determine possible overfit-ting of our statistical models and tomore conclusively determine the overallperformance of the method.

Here, we demonstrated that theMasSpec Pen is effective for in vivo diag-nosis of cancer regions during murineoncological surgery. The entire procedurefrom triggering the system to data anal-ysis is performed under 10 s, and furtherimprovements are envisioned. Comparedwith the time necessary for intraoperativepathologic frozen section analysis of ex-cised species (~30min) or postoperativefinal pathologic evaluation (several days),the time required for the MasSpec Penanalysis could expedite surgical proce-dures, diagnosis, and treatment. Ourresults demonstrate that the molecularinformation obtained using a 1.5-mmsampling size allows accurate identifi-cation of cancer in marginal tumor re-gions of mixed histologic composition(Fig. 5), although further validation of

these results with larger independent sample sets is needed. We arecurrently exploring other machiningmethods to increase sampling res-olution; however, the 1.5-mm sampling size relates well with the degreeof precision achieved during surgical resection.

We envision that the MasSpec Pen will become a valuable clinicaltechnology for near real-time in vivo and ex vivo cancer diagnosis.Emerging technologies including fluorescence-guided surgery (11),Raman spectroscopy (12), optical coherence tomography (39), reflec-tance spectroscopy (40), and stimulated Raman scattering microscopy(13) have been proposed for cancer diagnosis and surgical guidance.Although powerful, many of these methods rely on injection of ex-ogenous labels that target specific cell types for tumor visualization,require tissue excision and processing for analysis, suffer frommoderatetissue specificity, or provide limited biochemical information. The ap-proach we propose here leverages on the unparalleled sensitivity andspecificity provided by MS for untargeted molecular evaluation andon its biocompatibility for disease diagnosis and clinical use. Yet, manychallenges exist in translating and integrating new technologies intothe workflow of a complex surgical and clinical environment (41). In-tegrationof theMasSpecPen into laparoscopic and robotic surgical systemsfor minimally invasive procedures and improved communicationand visualization tools will facilitate its inclusion in clinical workflows.

Fig. 5. MasSpec Pen analysis of an HGSC tissue sample with mixed histologic composition. (A) Optical image showsthe tissue sample that was analyzed at the demarcated regions (1 to 5) using a 1.5-mm-diameter MasSpec Pen. After theMasSpec Pen analysis, the tissue sample was frozen, sectioned, and H&E-stained. An optical image of the H&E-stained tissuesection obtained at region 3 is shown (inset), presenting a mixed histologic composition including cancer and adjacentnormal stroma tissue. (B) TheMasSpec Pen negative ionmodemass spectra are shown for regions 1 (normal stroma; averageof n=3mass spectra), 3 (mixture of normal stroma and cancer; average of n=3mass spectra), and 5 (cancer; average of n= 3mass spectra). (C) Table listing the pathologic diagnosis of the five regions analyzed and the Lasso prediction results.

8 of 11




http://stm.sciencem

ag.org/D

ownloaded from

Similarly, size- and cost-effective mass spectrometers with sufficient an-alytical performance for molecular evaluation are needed for dedicatedsurgical use. Advances in building and validating databases for tissueidentification, as well as automated computational methods for real-time output of predictive results, are also necessary for broad use. Fur-ther work will require careful evaluation of the long-term benefits topatients to determine the value of new technologies in clinical practice.We expect that the excellent performance and simple design andoperation of theMasSpec Pen, combinedwith its clinically desirable fea-tures, may enable its translation to the clinic for routine medical use,improving patient care and treatment.

MATERIALS AND METHODSStudy designThe objective of this study was to evaluate the potential of an MS-based probe to nondestructively analyze and diagnose cancer in hu-


man tissue samples. Here, we investi-gated the molecular profiles of humantissue samples obtained from 281 pa-tients including normal and cancerbreast, lung, thyroid, and ovary tissues.All patient samples were obtained fromthe CHTN (Cooperative Human TissueNetwork), Asterand Bioscience, and theBaylor College of Medicine Tissue Bankunder approved Institutional ReviewBoard protocol. The mass spectra ob-tained using the MasSpec Pen in tissuesamples were normalized, background-subtracted, and analyzed using a statis-tical technique to build classificationmodels. Expert, board-certified pathol-ogists (J.L., W.Y., and C.N.) evaluatedtheH&E-stained tissue sections obtainedfrom the tissue samples analyzed. Thepathologists were blind to any informa-tion about the acquisition from MSanalysis. Samples were excluded fromstatistical analysis if they were deter-mined by the pathologist to have sub-stantial heterogeneity in cell composition,which included 28 samples, resultingin a final sample set of 253 samples.The in vivo animal model experimentswere conducted under approved Insti-tutionalAnimal Care andUse Commit-tee protocol.

Design and engineering of theMasSpec PenA 3D printer (Model uPrint SE Plus)was used to print the key component,a PDMS (Dow Corning) probe tip. Thepen tips were fabricated by casting anelastomer from a negativemold designedusing SolidWorks computer-aided de-sign software and then dissolving themold away. PTFE tubing (ID, 1/32 inch;

outer diameter, 1/16 inch; Cole-Parmer) was directly inserted into theprobe tip for experiments. For more information, see the Supple-mentary Materials.

MS data acquisitionAll experiments were performed on a Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific). Full scan wascarried out at the range of m/z 120 to 1800, using a resolving powerof 140,000, a capillary temperature of 350°C, and an S-lens radiofrequency level of 100. Wild-type mouse brains were purchased fromBioreclamationIVT. Human tissue samples were obtained frozenand stored in a −80°C freezer until analysis, when they were thawedat room temperature. The tissues were placed on a glass slide andanalyzed by the MasSpec Pen using the experimental steps described.After experiments, the tissue regions analyzed were annotated andfrozen, and 16-mm tissue sections were prepared using a CryoStarNX50 cryostat. Additional tissue sections at different regions of the

Fig. 6. Intraoperative analysis of tumor and normal tissues in a murine model. (A) Experiments were performed invivo inmice under anesthesia. Optical images show the animal under anesthesia and before, during, and after theMasSpecPen analysis. (B) Representative negative ionmodemass spectra showdistinctmolecular profiles fromnormal (average ofn = 3 mass spectra) and tumor (average of n = 3 mass spectra) tissues.

9 of 11




http://stm.sciencem

ag.org/D

ownloaded from

tissue piece were obtained for the MasSpec Pen analysis. Tissuesections were H&E-stained and evaluated by histopathology afteranalysis.

In vivo mouse experimentsBT474 HER2+ cells were grown in improved minimum essential medi-um (IMEM) (Invitrogen) supplemented with 10% fetal bovine serum,1% L-glutamine, and 1% insulin to a confluency of 80 to 90% in 5% O2

and 37°C. Cells were counted via a hemocytometer and trypan blue dyeexclusion. Nude athymic female mice (n = 3) were subcutaneously im-planted with a 17b-estradiol pellet (0.72 mg, 60-day release; InnovativeResearch of America) in the nape of the neck. About 24 hours later,BT474 breast cancer cells (107) in serum-free IMEM with 20% growthfactor–reduced Matrigel were injected subcutaneously into the rightflank of the mouse (total injection of 100 ml). Tumors were monitoredweekly for growth until they reached 0.7 to 1.0 cm in diameter (averageof 250 mm3). In vivo experiments were performed during surgical re-section of tumors using murine animal models while the mice wereunder anesthesia (2% isoflurane and 98%O2). A surgical bladewas usedto open a flap of skin, leaving an estimated space of 1 to 2 cm around thetumors, and then, the skin flap was dissected from the surface of thetumor. The skin was flapped to expose the tumor and adjacent normaltissues, which were analyzed in several regions using the MasSpec Pen.Pieces of the tumor were then resected using a scalpel and analyzedex vivo. Tumor tissue pieces analyzed by the MasSpec Pen were an-notated, flash-frozen, sectioned, and subjected to H&E staining fordiagnosis.

Statistical analysisAverages of three mass spectra obtained during each 10-sMasSpec Penanalysis were used to build molecular databases. The Xcalibur raw datawere converted to Microsoft Excel spreadsheet format. The full massrange of the spectra was partitioned into bins by rounding m/z valuesto the nearest hundredth. All mass spectra were first normalizedaccording to total ion count or to the absolute intensity of m/z 885.55to account for slight fluctuations in signal intensities that may occurbetween experiments. Then, background peaks and peaks not ap-pearing in at least 10% of the samples analyzed were excluded to reducerandom noise. For each tissue section (breast or thyroid), four repre-sentative mass spectra for each tissue section analyzed were importedto MetaboAnalyst (www.metaboanalyst.ca/) for PCA using the web-site built-in function. Score and loading plots were generated throughthe website for each tissue type. For each soft tissue sample type (breast,thyroid, lung, and ovary), the data were imported to R programminglanguage. PCA was performed by centering the preprocessed datato mean zero and computing principal components using the prcompfunction in R. The first three principal components were visualizedwith the rgl and pca3d packages for R. For tissue classification, theLasso method was applied using the glmnet package in the CRAN Rlanguage library. Models generated using the Lasso are simpler tointerpret than other regularization methods because it yields “sparse”models (models that involve only a subset of the features). A math-ematical weight for each statistically informative feature is calculatedby the Lasso depending on the importance that themass spectral featurehas in characterizing a certain class (cancer versus normal or a cancersubtype versus normal). Classification was performed using a leave-one-out cross-validation approach to assess the predictive accuracywithin the training set. Performance of trained classifiers was measuredby sensitivity, specificity, accuracy, and AUC.


SUPPLEMENTARY MATERIALSwww.sciencetranslationalmedicine.org/cgi/content/full/9/406/eaan3968/DC1Materials and MethodsFig. S1. Effect of MasSpec Pen contact time on the mass spectra obtained.Fig. S2. Effect of MasSpec Pen tip diameter on the mass spectra obtained.Fig. S3. MasSpec Pen analysis of a mouse brain tissue section.Fig. S4. Positive ion mode analysis using the MasSpec Pen.Fig. S5. Comparison between MasSpec Pen and DESI-MSI mass spectra.Fig. S6. Effect of MasSpec Pen solvent systems on the mass spectra obtained.Fig. S7. MasSpec Pen detection of protein ions.Fig. S8. PCA of the data obtained for the human tissue sections including normal and tumorthyroid and breast tissue sections.Fig. S9. Comparison between the MasSpec Pen mass spectra obtained from the tissue sectionand fresh tissue piece.Fig. S10. MasSpec Pen analysis of an HGSC tissue sample with mixed histologic composition.Fig. S11. Intraoperative analysis of tumor and normal tissues in a murine model.Fig. S12. MasSpec Pen analysis of the same tissue sample in vivo and ex vivo.Table S1. Data obtained for the identification of selected negative ion mode molecular ionsfrom mouse brain tissue.Table S2. Data obtained for the identification of selected negative ion mode molecular ionsfrom human thyroid tissue.Table S3. Data obtained for the identification of selected negative ion mode molecular ionsfrom human ovarian tissue.Table S4. Data obtained for the identification of selected negative ion mode molecular ionsfrom human lung tissue.Table S5. Data obtained for the identification of selected negative ion mode molecular ionsfrom human breast tissue.Table S6. Patient demographics of the 253 human tissue samples used in this study.Movie S1. Simulation demonstrating the use of the MasSpec Pen for routine intraoperativediagnosis.

REFERENCES AND NOTES1. T. A. Buchholz, M. R. Somerfield, J. J. Griggs, S. El-Eid, M. E. H. Hammond, G. H. Lyman,

G. Mason, L. A. Newman, Margins for breast-conserving surgery with whole-breastirradiation in stage I and II invasive breast cancer: American society of clinicaloncology endorsement of the society of surgical oncology/american society for radiationoncology consensus guideline. J. Clin. Oncol. 32, 1502−1506 (2014).

2. S. J. Maygarden, F. C. Detterbeck, W. K. Funkhouser, Bronchial margins in lung cancerresection specimens: Utility of frozen section and gross evaluation. Mod. Pathol. 17,1080–1086 (2004).

3. G. Massard, C. Doddoli, B. Gasser, X. Ducrocq, R. Kessler, C. Schumacher, G. M. Jung,J. M. Wihlm, Prognostic implications of a positive bronchial resection margin.Eur. J. Cardiothorac. Surg. 17, 557–565 (2000).

4. A. M. Nick, R. L. Coleman, P. T. Ramirez, A. K. Sood, A framework for a personalizedsurgical approach to ovarian cancer. Nat. Rev. Clin. Oncol. 12, 239–245 (2015).

5. S.-S. Han, J.-Y. Jang, S.-W. Kim, W.-H. Kim, K. U. Lee, Y.-H. Park, Analysis of long-termsurvivors after surgical resection for pancreatic cancer. Pancreas 32, 271–275 (2006).

6. M. Zhang, Z. G. Li, Y. Ma, G. Y. Zhu, H. F. Zhang, Y. W. Xue, Prognostic predictors ofpatients with carcinoma of the gastric cardia. Hepatogastroenterology 59, 930–933 (2012).

7. N. Sanai, M. S. Berger, Glioma extent of resection and its impact on patient outcome.Neurosurgery 62, 753–764 (2008).

8. A. Macario, What does one minute of operating room time cost? J. Clin. Anesth. 22,233–236 (2010).

9. M. Mino-Kenudson, L. R. Chirieac, K. Law, J. L. Hornick, N. Lindeman, E. J. Mark,D. W. Cohen, B. E. Johnson, P. A. Janne, A. J. Iafrate, S. J. Rodig, A novel highly sensitiveantibody allows for the routine detection of ALK-rearranged lung adenocarcinomas bystandard immunohistochemistry. Clin. Cancer Res. 16, 1561–1571 (2010).

10. S. Jones, V. Anagnostou, K. Lytle, S. Parpart-Li, M. Nesselbush, D. R. Riley, M. Shukla,B. Chesnick, M. Kadan, E. Papp, K. G. Galens, D. Murphy, T. Zhang, L. Kann, M. Sausen,S. V. Angiuoli, L. A. Diaz Jr., V. E. Velculescu, Personalized genomic analyses for cancermutation discovery and interpretation. Sci. Transl. Med. 7, 283ra53 (2015).

11. W. Stummer, U. Pichlmeier, T. Meinel, O. D. Wiestler, F. Zanella, H.-J. Reulen; ALA-GliomaStudy Group, Fluorescence-guided surgery with 5-aminolevulinic acid for resection ofmalignant glioma: A randomised controlled multicentre phase III trial. Lancet Oncol. 7,392–401 (2006).

12. M. Jermyn, K. Mok, J. Mercier, J. Desroches, J. Pichette, K. Saint-Arnaud, L. Bernstein,M.-C. Guiot, K. Petrecca, F. Leblond, Intraoperative brain cancer detection with Ramanspectroscopy in humans. Sci. Transl. Med. 7, 274ra19 (2015).

13. D. A. Orringer, B. Pandian, Y. S. Niknafs, T. C. Hollon, J. Boyle, S. Lewis, M. Garrard,S. L. Hervey-Jumper, H. J. L. Garton, C. O. Maher, J. A. Heth, O. Sagher, D. A. Wilkinson,

10 of 11

http://www.metaboanalyst.ca/

http://www.sciencetranslationalmedicine.org/cgi/content/full/9/406/eaan3968/DC1




http://stm.sciencem

ag.org/D

ownloaded from

M. Snuderl, S. Venneti, S. H. Ramkissoon, K. A. McFadden, A. Fisher-Hubbard,A. P. Lieberman, T. D. Johnson, X. S. Xie, J. K. Trautman, C. W. Freudiger, S. Camelo-Piragua,Rapid intraoperative histology of unprocessed surgical specimens via fibre-laser-basedstimulated Raman scattering microscopy. Nat. Biomed. Eng. 1, 0027 (2017).

14. K. Chughtai, R. M. A. Heeren, Mass spectrometric imaging for biomedical tissue analysis.Chem. Rev. 110, 3237–3277 (2010).

15. C. C. Hsu, P. C. Dorrestein, Visualizing life with ambient mass spectrometry. Curr. Opin.Biotechnol. 31, 24–34 (2015).

16. C. P. Wu, A. L. Dill, L. S. Eberlin, R. G. Cooks, D. R. Ifa, Mass spectrometry imaging underambient conditions. Mass Spectrom. Rev. 32, 218–243 (2013).

17. J. L. Norris, R. M. Caprioli, Analysis of tissue specimens by matrix-assisted laserdesorption/ionization imaging mass spectrometry in biological and clinical research.Chem. Rev. 113, 2309–2342 (2013).

18. D. R. Ifa, L. S. Eberlin, Ambient ionization mass spectrometry for cancer diagnosis andsurgical margin evaluation. Clin. Chem. 62, 111–123 (2016).

19. L. S. Eberlin, I. Norton, D. Orringer, I. F. Dunn, X. H. Liu, J. L. Ide, A. K. Jarmusch, K. L. Ligon,F. A. Jolesz, A. J. Golby, S. Santagata, N. Y. R. Agar, R. G. Cooks, Ambient massspectrometry for the intraoperative molecular diagnosis of human brain tumors.Proc. Natl. Acad. Sci. U.S.A. 110, 1611–1616 (2013).

20. A. K. Jarmusch, V. Pirro, Z. Baird, E. M. Hattab, A. A. Cohen-Gadol, R. G. Cooks, Lipid andmetabolite profiles of human brain tumors by desorption electrospray ionization-MS.Proc. Natl. Acad. Sci. U.S.A. 113, 1486–1491 (2016).

21. K.-C. Schäfer, J. Dénes, K. Albrecht, T. Szaniszló, J. Balog, R. Skoumal, M. Katona, M. Tóth,L. Balogh, Z. Takáts, In vivo, in situ tissue analysis using rapid evaporative ionization massspectrometry. Angew. Chem. Int. Ed. 48, 8240–8242 (2009).

22. J. Balog, L. Sasi-Szabó, J. Kinross, M. R. Lewis, L. J. Muirhead, K. Veselkov, R. Mirnezami, B. Dezs,L. Damjanovich, A. Darzi, J. K. Nicholson, Z. Takáts, Intraoperative tissue identification usingrapid evaporative ionization mass spectrometry. Sci. Transl. Med. 5, 194ra93 (2013).

23. K.-C. Schäfer, T. Szaniszló, S. Günther, J. Balog, J. Dénes,M. Keserű, B. Dezső, M. Tóth, B. Spengler,Z. Takáts, In situ, real-time identification of biological tissues by ultraviolet and infraredlaser desorption ionization mass spectrometry. Anal. Chem. 83, 1632–1640 (2011).

24. B. Fatou, P. Saudemont, E. Leblanc, D. Vinatier, V. Mesdag, M. Wisztorski, C. Focsa,M. Salzet, M. Ziskind, I. Fournier, In vivo real-time mass spectrometry for guided surgeryapplication. Sci. Rep. 6, 25919 (2016).

25. R. Tibshirani, Regression shrinkage and selection via the Lasso. J. R. Stat. Soc. Series B Stat.Methodol. 58, 267–288 (1996).

26. A. L. Dill, L. S. Eberlin, A. B. Costa, D. R. Ifa, R. G. Cooks, Data quality in tissue analysis usingdesorption electrospray ionization. Anal. Bioanal. Chem. 401, 1949–1961 (2011).

27. J. Zhang, W. Yu, S. W. Ryu, J. Lin, G. Buentello, R. Tibshirani, J. W. Suliburk, L. S. Eberlin,Cardiolipins are molecular markers of mitochondria-rich thyroid oncocytic tumors.Cancer Res. 76, 6588–6597 (2016).

28. D. Calligaris, D. Caragacianu, X. Liu, I. Norton, C. J. Thompson, A. L. Richardson, M. Golshan,M. L. Easterling, S. Santagata, D. A. Dillon, F. A. Jolesz, N. Y. R. Agar, Application of desorptionelectrospray ionization mass spectrometry imaging in breast cancer margin analysis.Proc. Natl. Acad. Sci. U.S.A. 111, 15184–15189 (2014).

29. S. Guenther, L. J. Muirhead, A. V. M. Speller, O. Golf, N. Strittmatter, R. Ramakrishnan,R. D. Goldin, E. Jones, K. Veselkov, J. Nicholson, A. Darzi, Z. Takats, Spatiallyresolved metabolic phenotyping of breast cancer by desorption electrospray ionizationmass spectrometry. Cancer Res. 75, 1828–1837 (2015).

30. H. S. Kang, S. C. Lee, Y. S. Park, Y. E. Jeon, J. H. Lee, S.-Y. Jung, I. H. Park, S. H. Jang, H. M. Park,C.W. Yoo, S. H. Park, S. Y. Han, K. P. Kim, Y.H. Kim, J. Ro, H. K. Kim, Protein and lipidMALDIprofilesclassify breast cancers according to the intrinsic subtype. BMC Cancer 11, 465 (2011).

31. M. Sans, K. Gharpure, R. Tibshirani, J. Zhang, L. Liang, J. Liu, J. H. Young, R. L. Dood,A. K. Sood, L. S. Eberlin, Metabolic markers and statistical prediction of serous ovariancancer aggressiveness by ambient ionization mass spectrometry imaging. Cancer Res. 77,2903–2913 (2017).

32. V. Kertesz, G. J. Van Berkel, Fully automated liquid extraction-based surface samplingand ionization using a chip-based robotic nanoelectrospray platform. J. Mass Spectrom.45, 252–260 (2010).


33. V. Kertesz, M. J. Ford, G. J. Van Berkel, Automation of a surface sampling probe/electrospray mass spectrometry system. Anal. Chem. 77, 7183–7189 (2005).

34. J. Laskin, B. S. Heath, P. J. Roach, L. Cazares, O. J. Semmes, Tissue imaging usingnanospray desorption electrospray ionization mass spectrometry. Anal. Chem. 84,141–148 (2012).

35. V. S. Pagnotti, N. D. Chubatyi, C. N. McEwen, Solvent assisted inlet ionization: An ultrasensitivenew liquid introduction ionization method for mass spectrometry. Anal. Chem. 83,3981–3985 (2011).

36. M. N. Helmus, D. F. Gibbons, D. Cebon, Biocompatibility: Meeting a key functionalrequirement of next-generation medical devices. Toxicol. Pathol. 36, 70–80 (2008).

37. J. Mokkaphan, W. Banlunara, T. Palaga, P. Sombuntham, S. Wanichwecharungruang,Silicone surface with drug nanodepots for medical devices. ACS Appl. Mater. Interfaces 6,20188–20196 (2014).

38. C.-H. Chen, Z. Lin, S. Garimella, L. Zheng, R. Shi, R. G. Cooks, Z. Ouyang, Development of amass spectrometry sampling probe for chemical analysis in surgical and endoscopicprocedures. Anal. Chem. 85, 11843–11850 (2013).

39. C. Kut, K. L. Chaichana, J. Xi, S. M. Raza, X. Ye, E. R. McVeigh, F. J. Rodriguez,A. Quiñones-Hinojosa, X. Li, Detection of human brain cancer infiltration ex vivo and invivo using quantitative optical coherence tomography. Sci. Transl. Med. 7, 292ra100(2015).

40. N. Rajaram, T. J. Aramil, K. Lee, J. S. Reichenberg, T. H. Nguyen, J. W. Tunnell, Design andvalidation of a clinical instrument for spectral diagnosis of cutaneous malignancy. Appl. Optics49, 142–152 (2010).

41. J. Zhang, W. Yu, J. Suliburk, L. S. Eberlin, Will ambient ionization mass spectrometrybecome an integral technology in the operating room of the future? Clin. Chem. 62,1172–1174 (2016).

Acknowledgments: We thank the Eberlin and Milner laboratory members for valuablediscussions. L.S.E. thanks J. Brodbelt and R. Crooks (University of Texas at Austin) for reviewingthe manuscript and their mentoring. We are grateful to R. Tibshirani (Stanford University)for the assistance with the statistical analysis. We also thank T. Hooper for his help with thehardware design and E. Que for the use of microscope. Tissue samples were provided by theCooperative Human Tissue Network, a National Cancer Institute supported resource, and theBaylor College of Medicine Tissue Bank. Funding: This work was supported by the NationalCancer Institute of the NIH under award R00CA190783 to L.S.E., the startup funds provided toL.S.E. by the University of Texas at Austin, and the Cancer Prevention Research Institute ofTexas to A.G.S. (CPRIT RR160005) and to T.E.M. (CPRIT DP 150102). Author contributions:L.S.E., T.E.M., J.Z., J.Q.L., and J.R. designed the experiments. L.S.E, T.E.M., J.Z., J.R., J.Q.L., N.K., andA.Z. designed the MasSpec Pen. J.Z., J.R., J.Q.L., M.S., N.G., A.B., R.J.D., K.Y.G., B.L., L.S.E., A.G.S.,and A.S. acquired the data. L.S.E., J.Z., T.E.M., M.S., and J.S. performed the data analysis andinterpretation. W.Y., C.N., and J.L. performed the pathologic evaluation. J.H.Y., J.Z., J.Q.L.,and L.S.E. performed the statistical analysis and interpretation. L.S.E., J.Z., and T.E.M. wrote andrevised the manuscript. L.S.E. conceived and supervised the study. Competing interests:L.S.E., T.E.M., J.Z., J.L., J.R., N.K., and A.Z. are inventors on a provisional patent application62/462,524 owned by the Board of Regents of the University of Texas System that relates toa handheld probe and MS system, such as described in this study. All other authors declarethat they have no competing interests. Data and materials availability: The data for thisstudy have been deposited in Dataverse, which can be found at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XBVHGM.

Submitted 7 April 2017Accepted 17 August 2017Published 6 September 201710.1126/scitranslmed.aan3968

Citation: J. Zhang, J. Rector, J. Q. Lin, J. H. Young, M. Sans, N. Katta, N. Giese, W. Yu, C. Nagi,J. Suliburk, J. Liu, A. Bensussan, R. J. DeHoog, K. Y. Garza, B. Ludolph, A. G. Sorace, A. Syed,A. Zahedivash, T. E. Milner, L. S. Eberlin, Nondestructive tissue analysis for ex vivo and in vivocancer diagnosis using a handheld mass spectrometry system. Sci. Transl. Med. 9, eaan3968 (2017).

11 of 11

https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XBVHGM

https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XBVHGM


mass spectrometry systemNondestructive tissue analysis for ex vivo and in vivo cancer diagnosis using a handheld

Ludolph, Anna G. Sorace, Anum Syed, Aydin Zahedivash, Thomas E. Milner and Livia S. EberlinChandandeep Nagi, James Suliburk, Jinsong Liu, Alena Bensussan, Rachel J. DeHoog, Kyana Y. Garza, Benjamin Jialing Zhang, John Rector, John Q. Lin, Jonathan H. Young, Marta Sans, Nitesh Katta, Noah Giese, Wendong Yu,

DOI: 10.1126/scitranslmed.aan3968, eaan3968.9Sci Transl Med

destruction.healthy tissue during surgery in mice, without requiring specific labeling or imaging and without evidence of tissue characterizes diagnostic proteins, lipids, and metabolites. The pen could be used to rapidly distinguish tumor fromphysical contact with a tissue surface, the water droplet is transported to a mass spectrometer, which

gentlemolecular profile of tissues using a small volume water droplet and mass spectrometry analysis. After 3 s of developed a handheld pen-like device that rapidly identifies theet al.facilitate intraoperative diagnosis, Zhang

margins (absence of cancer cells at the outer edge of the excised tumor specimen) can be challenging. To Although a surgeon's goal is to remove cancer in its entirety during excision surgery, achieving negative

Is the pen mightier than the scalpel?

ARTICLE TOOLS http://stm.sciencemag.org/content/9/406/eaan3968

MATERIALSSUPPLEMENTARY http://stm.sciencemag.org/content/suppl/2017/09/01/9.406.eaan3968.DC1

CONTENTRELATED

http://stm.sciencemag.org/content/scitransmed/5/201/201ra119.fullhttp://stm.sciencemag.org/content/scitransmed/4/142/142ra94.fullhttp://stm.sciencemag.org/content/scitransmed/7/274/274ra19.fullhttp://stm.sciencemag.org/content/scitransmed/5/194/194ra93.full

REFERENCES

http://stm.sciencemag.org/content/9/406/eaan3968#BIBLThis article cites 41 articles, 14 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

registered trademark of AAAS. is aScience Translational MedicineScience, 1200 New York Avenue NW, Washington, DC 20005. The title

(ISSN 1946-6242) is published by the American Association for the Advancement ofScience Translational Medicine

of Science. No claim to original U.S. Government WorksCopyright © 2017 The Authors, some rights reserved; exclusive licensee American Association for the Advancement


http://stm.sciencem

ag.org/D

ownloaded from

http://stm.sciencemag.org/content/9/406/eaan3968

http://stm.sciencemag.org/content/suppl/2017/09/01/9.406.eaan3968.DC1

http://stm.sciencemag.org/content/scitransmed/5/194/194ra93.full




http://stm.sciencemag.org/content/9/406/eaan3968#BIBL

http://www.sciencemag.org/help/reprints-and-permissions

http://www.sciencemag.org/about/terms-service


nondestructive tissue analysis for ex vivo and in vivo ...conventional methods for histopathologic...

Documents