Prediction of Chemotherapeutic Efﬁcacy in Non Small Cell ... · Personalized Medicine and Imaging Prediction of Chemotherapeutic Efﬁcacy in Non– Small Cell Lung Cancer by Serum

Personalized Medicine and Imaging

Prediction of Chemotherapeutic Efficacy in Non–Small Cell Lung Cancer by Serum MetabolomicProfilingYanhua Tian1,2, Zhijie Wang1, Xiaohui Liu3, Jianchun Duan1, Guoshuang Feng4,Yuxin Yin5, Jin Gu6, Zhaoli Chen7, Shugeng Gao7, Hua Bai1, Rui Wan1, Jun Jiang8, Jia Liu5,Cong Zhang5, Di Wang1,2, Jiefei Han1, Xue Zhang1, Liangliang Cai1, Jie He7, and Jie Wang1

Abstract

Purpose: No validated biomarkers that could identify thesubset of patients with lung adenocarcinoma who might benefitfrom chemotherapy have yet been well established. This studyaimed to explore potential biomarkermodel predictive of efficacyand survival outcomes after first-line pemetrexed plus platinumdoublet based on metabolomics profiling.

Experimental Design: In total, 354 consecutive eligiblepatients were assigned to receive first-line chemotherapy of peme-trexed in combination with either cisplatin or carboplatin. Pro-spectively collected serum samples before initial treatment wereutilized to perform metabolomics profiling analyses under theapplication of LC/MS-MS. Binary logistic regression analysis wascarried out to establish discrimination models.

Results: There were 251 cases randomly sorted into discoveryset, the rest of 103 cases into validation set. Seven metabolitesincluding hypotaurine, uridine, dodecanoylcarnitine, choline,

dimethylglycine, niacinamide, and L-palmitoylcarnitine wereidentified associated with chemo response. On the basis of theseven-metabolite panel, a discriminant model according tologistic regression values g(z) was established with the receiveroperating characteristic curve (AUC) of 0.912 (Discovery set)and 0.909 (Validation set) in differentiating progressivedisease (PD) groups from disease control (DC) groups.The median progression-free survival (PFS) after chemothera-py in patients with g(z) �0.155 was significantly longerthan that in those with g(z) > 0.155 (10.3 vs.4.5 months,P < 0.001).

Conclusions: This study developed an effective and convenientdiscriminant model that can accurately predict the efficacy andsurvival outcomes of pemetrexed plus platinum doublet chemo-therapy prior to treatment delivery. Clin Cancer Res; 24(9); 2100–9.�2018 AACR.

IntroductionLung cancer is one of the most common fatal malignancies

worldwide, causing over one million deaths annually (1). Peme-

trexed plus platinum (cisplatin or carboplatin) doublet chemo-therapy has become a standard first-line treatment for nonsqua-mous non–small cell lung cancer (NSCLC) patients who are noteligible for targeted therapies [e.g., tyrosine kinase inhibitors(TKIs) or immunotherapy (2, 3)]. However, the efficacy of thispreferred chemotherapeutic regimen is very limited, withresponse rates of 30%–40%, and progression-free survival (PFS)of about 4–6months (3–5). Despite a large number of promisingstudies demonstrating the expression level of enzymes targeted bypemetrexed (6, 7) and enzymes involved in nucleotide excisionrepair (NER) ofDNA–platinumadducts (8) can be used to predictresponse to pemetrexed plus cisplatin treatment, the translationof these scientific data from bench to the bedside has not yetoccurred. The detection assays, tissue types and antibodies used inprevious studies, which lead to inconsistent results, needed to bestandardized in all current clinical trials (9). Besides, large clinicaltrials are needed to further validate the predictive abilities beforethese biomarkers being reliably used to guide treatment protocols(10). Unlike targeted therapies with specific gene aberrations oractivated kinases as predictive biomarkers such as mutations inEGFR or translocations of ALK (anaplastic lymphoma kinase;refs. 11, 12), cytotoxic drug-based chemotherapies do not havespecific molecular targets. Besides a cancer cell itself, pharmaco-kinetic and tumor microenvironmental issues are also known tocontribute to the clinical failure of chemotherapy and apparentdrug resistance (13). Thus, potential biomarkers that can addressthese issues relating to drug resistance and can predict the

1State Key Laboratory of Molecular Oncology, Department of Medical Oncology,National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences& Peking UnionMedical College, Beijing, China. 2Peking-Tsinghua Center for LifeSciences, Academy for Advanced Interdisciplinary Studies, Peking University,Beijing, China. 3School of Life Sciences, Tsinghua University, Beijing, China.4Center for Clinical Epidemiology & Evidence-based Medicine Beijing Children'sHospital, Capital Medical University, National Center for Children's Health,Beijing, China. 5Institute of Systems Biomedicine, School of Basic MedicalSciences, Peking University Health Science Center, Beijing, China. 6Departmentof GI Surgery III, Peking University Cancer Hospital, Beijing, China. 7Departmentof Thoracic Surgery, National Cancer Center/Cancer Hospital, Chinese AcademyofMedical Sciences &PekingUnionMedical College, Beijing, China. 8Departmentof Oncology, Affiliated Hospital of Qinghai University, Xining, P.R. China.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

Y. Tian and Z. Wang contributed equally to this article.

Corresponding Author: JieWang, State Key Laboratory of Molecular Oncology,Department of Medical Oncology, National Cancer Center/Cancer Hospital,Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing100021, China. Phone: 8601-0877-88524; Fax: 8610-8819-6562; E-mail:[email protected]

doi: 10.1158/1078-0432.CCR-17-2855

�2018 American Association for Cancer Research.

ClinicalCancerResearch

Clin Cancer Res; 24(9) May 1, 20182100

on November 4, 2020. © 2018 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 6, 2018; DOI: 10.1158/1078-0432.CCR-17-2855

http://crossmark.crossref.org/dialog/?doi=10.1158/1078-0432.CCR-17-2855&domain=pdf&date_stamp=2018-4-6

http://clincancerres.aacrjournals.org/

response of a patient to pemetrexed plus platinum doublettherapy are needed.

Metabolomics is an -omics technology that provides infor-mation that complements the genomic and proteomic profileof a sample (14). Metabolites are stable in serum and can bequantified, which presents an opportunity to establish a non-invasive approach for monitoring disease status and exploringbiomarkers to predict the efficacy of anticancer therapies (15).Mapstone and colleagues (2014) discovered and validated apanel of ten lipid metabolites from plasma that predictsAlzheimer's disease within a 2- to 3-year timeframe with over90% accuracy (16). Sreekumar and colleagues (2009) revealedthe potential role of sarcosine in prostate cancer progressionby profiling metabolites from a large cohort of clinical speci-mens (17).

We hypothesize that the capacity to respond to pemetrexedplus platinum doublet chemotherapy is a congenital feature thatrepresents a specific metabolic phenotype, and can therefore beextrapolated from analysis of the metabolic profiles of pre-chemotherapeutic serum. To explore this hypothesis, we usedLC/MS-MS to perform metabolomic profiling analysis (18) andidentified potential metabolic biomarkers that can, prior to treat-ment, predict the efficacy of pemetrexed-platinum doublet ther-apy in patients with nonsquamous NSCLC.

Patients and MethodsStudy design and patient population

This study was designed to explore the serum metabolic bio-markers predictive of efficacy of chemotherapy. Patients includ-ed in our study were assigned to receive first-line chemotherapyof pemetrexed (500 mg/m2) in combination with either cis-platin (75 mg/m2) or carboplatin (AUC ¼ 5 mg/ml.min)intravenously on day 1 every 21 days for 4–6 cycles or diseaseprogression (3). The serum samples were prospectively collect-ed before first-cycle treatment. Patients were randomly sortedinto discovery group (251 cases, 70.9%), and the rest of the

cases into validation group (103 cases, 29.1%). This study wasapproved by the medical ethics committee of Peking UniversityCancer Hospital. All patients signed written informed consentto participate.

From September 2014 to December 2016, 389 patients con-secutively diagnosed with inoperable stage IIIB or IV lung ade-nocarcinoma, who received pemetrexed plus platinum (cisplatinor carboplatin) doublet as the first-line regimen were recruitedinto this study. Patients who with symptoms associated withbacterial infection, such as fever, increased leucocytes and neu-trophils, inflammation indicated by lung imaging or microcul-ture, were not included in our study to avoid any influence ofbacterial infection on the serummetabolome. Finally, 354 serumsamples were successfully used for metabolite profiling analysis.Reasons for the patients omitted from analysis included pretreat-ment serumnot available (24 cases, 6.2%), or not completing twocycles of chemotherapy due to overt cytotoxicity (14 cases, 3.6%)or switched to molecular targeted therapy due to EGFR mutationor ALK rearrangement (7 cases, 1.8%). No further histologicsubtyping such as IASLC/AST/ERS classification was performeddue to their inconsistent prognostic value of chemo-response(19). The tumor responses to chemotherapy were evaluated perResponse Evaluation Criteria in Solid Tumors, version 1.1(RECIST v1.1).

Serum sample collectionFasting peripheral blood (3.5 mL) was collected with a serum

separator tube in the morning, one week before the first cycle ofchemotherapy treatment. The blood was centrifuged within 30minutes of venipuncture at 1, 200 g for 10 minutes at 4�C, andserum was collected and stored at �80�C.

Metabolite extractionSerum samples were thawed on ice and vortexed thoroughly.

To extract hydrophilic metabolites, 100 mL of homogeneousserum sample was mixed with 400 mL methanol (prechilled to�80�C). The mixture was vortexed for 30 seconds and incubatedfor 6–8 hours at�80�C. After centrifugation at 12,000� g at 4�Cfor 10minutes, the supernatant (300mL)was transferred to a freshtube and lyophilized under vacuum. The dried sample werereconstituted in 80-mL 80% methanol, vortexed for 30 seconds,and incubated at 4�C for 15 minutes. The samples were centri-fuged at 12,000 � g at 4�C for 20 minutes. Finally, 20 mL ofsupernatant was used for LC/MS-MS analysis.

For hydrophobic metabolite extraction, 400 mL of extractionbuffer (chloroform/methanol, 2:1, v/v) was added to 100 mL ofserum sample. The mixture was vortexed for 30 seconds andcentrifuged at 10,000 � g for 10 minutes at room temperature.Subsequently, the lower organic-phase (200 mL) was collectedinto a fresh tube and lyophilized under vacuum. The driedsamples were dissolved in 150 mL of dichloromethane/methanol(2:1, v/v), vortexed for 30 seconds, and then centrifuged at12,000 � g for 15 minutes at room temperature. Finally, 20 mLof supernatant was used for LC/MS-MS analysis.

Metabolite profiling analysisFor untargetedmetabolite screening, aQExactive orbitrapmass

spectrometer (Thermo) was used. In positive mode, 95% and50% acetonitrile was applied as mobile phase A and B, with10 mmol/L ammonium formate and 0.1% formic acid in bothphases. Atlantis HILIC Silica columns were used for separation

Translational Relevance

In clinical practice, there are currently no validated biomar-kers that can, prior to treatment, reliably indicate the intrin-sically sensitive or resistant features of a nonsquamousNSCLCpatient to pemetrexed plus platinum doublet chemotherapy.In this study, we investigated themetabolic characteristics of alarge cohort of prechemotherapeutic serum samples (354cases) and found tight associations between small metabolitesubsets and the responses of patients to this cytotoxic drugcombination. We developed an effective discriminant modelemploying a seven-metabolite panel (hypotaurine, uridine,dodecanoylcarnitine, choline, dimethylglycine, niacinamide,L-palmitoylcarnitine) that can predict the efficacy of this che-motherapeutic regimen, prior to treatment, with a sensitivityof 90.8% and specificity of 79.5%. We also identified threeone-carbon metabolism-involved metabolites including cho-line, betaine, and DMG that are potentially associated withdrug resistance to pemetrexed. This serum-based biomarkerstudy can be easily applied in clinical practice and personal-ized treatment decisions.

Chemotherapeutic Efficacy Prediction by Serum Metabolomics

www.aacrjournals.org Clin Cancer Res; 24(9) May 1, 2018 2101




with a column temperature of 35�C. Separation was initiated at1% mobile phase B with a flow rate of 300 mL/minute. Thegradient is shown in Supplementary Fig. S8A. In negative mode,mobile phase A and B were 95% and 50% acetonitrile with10 mmol/L ammonium formate. However, the pH of A and Bwas adjusted to 9.0 using ammonium hydroxide. BEH Amidecolumns (2.1 mm � 100 mm, Waters) were utilized for LCseparation in negative mode. The column temperature was35�C. The gradient was started at 5% mobile phase B with a flowrate of 250 mL/minute. The gradient is shown in SupplementaryFig. S8B. The detailed mass spectrometer parameters are listed inSupplementary Fig. S8C.

Lipid analysis was performed on a Q Exactive orbitrap massspectrometer (Thermo). Mobile phase A was prepared by dissol-ving 10 mmol/L ammonium acetate in 60% acetonitrile andmobile phase B was 10%/90% acetonitrile/isopropanol (v/v).XSelect CSHC18 columns (2.1mm�100mm,Waters)were usedfor lipid analysis with column temperature of 45�C. The gradientwas generated with flow rate of 250 mL/minute as shown inSupplementary Fig. S8D. The detailed mass spectrometer para-meters are as shown in Supplementary Fig. S8E.

All mobile phases were freshly prepared to avoid bacterialcontamination. For data quality assessment, pooled quality con-trol (QC) samples were generated by mixing equal aliquots ofrepresentative subsets of subjects. Five injections of QC sampleswere analyzed at the beginning of the sample queue for columnconditioning followed by one QC sample inserted for every tensamples. The sequence order of all the sampleswas randomized toavoid the interference from system instability.

Data processing and statistical analysisPeak extraction and alignment was performed using SIEVE

software (Thermo). Features that existed in at least 80% sam-ples of a group were retained (20). To acquire more reliablepeaks, features were further verified by TraceFinder (Thermo)based on the m/z value and retention time. The relative ionintensity was normalized to the sum of the peak area of asample. Statistical analysis was performed using SPSS 23.0software. Statistical significance was calculated using Studentt test as implemented. Survival analysis was assessed withKaplan–Meier curves via log-rank tests. All the statistical testswere two-sided and considered statistically significant at P <0.05, unless otherwise stated. Features from both ESIþ and ESI�

with statistical significance were imported into the SIMCA-Pprogram (version 12.0, Umetrics) for multivariate analysis.Partial least squares discriminate analysis (PLS-DA) wasapplied with unit variance (UV) scaling (21). The normalizedamount of each metabolite was graphed in a boxplot usingGraphPad Prism version 5.0 (GraphPad Software). The ROCcurve and binary logistic regression were applied to the serumdata. Logistic regression with the best subset selection methodwas used for variable selection, in which, the highest likelihoodscore indicated the best model under certain numbers of vari-ables. Five-fold cross validation was used to validate modelswith different variable numbers, and the model with theminimum MSE (mean squared error) was defined as the bestone. The performance of the discriminant model was charac-terized by estimating the area under the ROC curve (AUC;ref. 22). Youden index (J ¼ sensitivity þ specificity � 1), as acommonly used measure of overall diagnostic effectiveness(23), was used in conjunction with ROC analysis.

ResultsPatient sets and baseline characteristics

In total, 354 serum samples were collected from patients withhistologically confirmed lung adenocarcinoma in this study,including 251 cases as the discovery set for predictive modelestablishment and 103 cases as the validation set (Fig. 1). Allthe patients received pemetrexed plus platinum doublet[pemetrexed/cisplatin, (pem-cis), n ¼ 224 and pemetrexed/carboplatin, (pem-carbo), n ¼ 130] as a first-line therapy. Mostpatients achieved disease control, including those with a confirm-ed partial response (PR) or stable disease (SD; pem-cis, 183/224,82% and pem-carbo, 95/130, 73%). The disease control rates,stratified by chemotherapeutic regimen (pem-cis vs. pem-carbo)are presented by patient set in Table 1. The clinical characteristicsof age, gender, smoking status, stage of disease, and EasternCooperative Oncology Group (ECOG) performance status werecomparable between the discovery and validation set.

Metabolomic profiling of pretreatment serum specimensThe 251 pretreatment serum specimens of the discovery set

were subjected to nontargeted metabolomics analysis usingLC/MS-MS. Pooled quality control (QC) samples, which weretightly clustered based on principal component analysis (PCA)(Supplementary Fig. S1; ref. 24), verified the stability and repeat-ability of the sample analysis sequence. After peak alignment andremoval ofmissing values, 1,373 electrospray ionization positive-mode (ESIþ) features and 1,014 negative-mode (ESI�) featureswere obtained. Further evaluation of these metabolomic profilesrevealed a total of 379 out of 2,387 features (229 ESIþ and 150ESI�) that exhibited significant differences between the diseasecontrol (DC) group (including the PR and SD groups) and theprogressive disease (PD) group (Wilcoxon P < 0.05). Finally, 90features were selected because they had accumulation trends(either up or down) thatwere consistentwith the clinical outcomeof this chemotherapy from PR to SD to PD, where PR is the bestresponse to the therapy and PD is the worst outcome. Most ofthese selected features (85/90) were from the hydrophilic sampleextracts while the 5 from the hydrophobic extracts were notconsidered further in this study. The DC (PR and SD) and PDgroups exhibited different accumulation patterns of these 85hydrophilic features (62 ESIþ and 23 ESI�; Fig. 2A), which weretherefore used for the subsequent identification of potentialmetabolite biomarkers.

Identification of metabolites predictive of chemotherapeuticresponse

PLS-DA was used to determine the potential metabolitefeatures associated with chemotherapeutic response. On thebasis of the selected 85 features, the PLS-DA scores plotillustrated a clear separation between the DC and PD groupsin the discovery set (Fig. 2B). Values for the variable impor-tance in the project (VIP) were used to rank the importance ofdifferent features for specimen discrimination in the model(Fig. 2C). Thirty-two features with VIP value more than onewere selected for further consideration (25), including dis-criminant model establishment and identification of chemicalstructure. Three parameters of each feature, including accuratemass, retention time, and fragmentation mass spectrum wereanalyzed and queried in the following databases: The HumanMetabolome Database (HMDB, http://www.hmdb.ca/), Kyoto

Tian et al.

Clin Cancer Res; 24(9) May 1, 2018 Clinical Cancer Research2102



http://www.hmdb.ca/


Encyclopedia of Genes and Genomes (KEGG, http://www.genome.jp/kegg/), METLIN (http://metlin.scripps.edu/), andMass Bank (http://www.massbank.jp/). Ultimately, 11 meta-bolites were assigned structures, including eight with lowerlevels in the DC group compared with the PD group[hypotaurine, taurine, choline, betaine, dimethylglycine(DMG), uridine, dodecanoylcarnitine (C12:0-carnitine) andL-palmitoylcarnitine (C16:0-carnitine)], and three with elevat-

ed levels in the DC group (palmitic amide, imidazole-4-acet-aldehyde, and niacinamide; Supplementary Table S1). Thesemetabolites were unambiguously identified using tandemmass spectrometry (Supplementary Fig. S2A–S2H). The coeffi-cients of variation (CV) of these metabolites in the QC sampleswere all below 20% (an accepted tolerance level for LC/MS-MS; ref. 26). The relative abundances of these metabolites inthe DC group versus PD group are plotted in Fig. 3.

Table 1. Patient sets and baseline characteristics

Characteristics of the patients (N ¼ 354)a

Discovery set (n ¼ 251) Validation set (n ¼ 103)

CharacteristicsbPem-cis

(n ¼ 161, 64.1%)Pem-carbo

(n ¼ 90, 35.9%)Pem-cis

(n ¼ 63, 61.2%)Pem-carbo

(n ¼ 40, 38.8%)

RECIST ResultbPR

(n ¼ 51)SD

(n ¼ 81)PD

(n ¼ 29)PR

(n ¼ 31)SD

(n ¼ 35)PD

(n ¼ 24)PR

(n ¼ 22)SD

(n ¼ 29)PD

(n ¼ 12)PR

(n ¼ 14)SD

(n ¼ 15)PD

(n ¼ 11)

Age, y (P ¼ 0.134)Median (range) 59 (44–84) 61 (37–83)Sex, n (%) (P ¼ 0.114)Male 148 (60.0%) 70 (68.0%)Female 103 (40.0%) 33 (32.0%)

Smoking status, n (%) (P ¼ 0.164)Former smoker 114 (45.4%) 45 (43.7%)Never smoker 137 (54.6%) 58 (56.3%)

Disease stage, n (%) (P ¼ 0.300)IIIB 31 (12.4%) 17 (16.5%)IV 220 (87.6%) 86 (83.5%)

ECOG performance status, n (%)c (P ¼ 0.122)0 90 (35.9%) 46 (44.6%)1 161 (64.1%) 57 (55.4%)

aMedian of age was calculated by Wilcoxon two-sample test (P ¼ 0.134). Comparisons of proportions were performed with the x2 test. There were no significantdifferences between the discovery set and validation set in the characteristics of sex (P ¼ 0.114), smoking status (P ¼ 0.164), disease stage (P ¼ 0.300), ECOGperformance status (P ¼ 0.122).bPem-cis, pemetrexed-cisplatin; Pem-carbo, pemetrexed-carboplatin; PR, partial response; SD, stable disease; PD, progressive disease.cThe Eastern Cooperative Oncology Group (ECOG) performance-status scores range from 0 to 5, with higher scores indicating increasing disability. A score of 0indicates no symptoms and 1 mild symptoms.

Figure 1.

Study profile. Among the 389 patients who were eligible for the study, 45 were excluded, in which 24 patients were not available for pretreatment serum and 21patients did not complete the two cycles of treatment either because the overt side effect (n ¼ 14) or the detection of EGFR or ALK gene aberrant (n ¼ 7)during the chemotherapy and thus alteredwith tyrosine kinase inhibitor (TKI) treatment. Finally, 354 serum sampleswere randomly sorted into discovery group (251cases, 70.9%), and the rest of the cases into validation group (103 cases, 29.1%). Pem-Cis, pemetrexed-cisplatin; pem-Carbo, pemetrexed-carboplatin; PR, partialresponse; SD, stable disease; PD, progressive disease. ECOG denotes Eastern Cooperative Oncology Group.





http://www.genome.jp/kegg/

http://www.genome.jp/kegg/

http://metlin.scripps.edu/

http://www.massbank.jp/


Given that pemetrexed is an inhibitor that interferes with folatemetabolism, we examined whether any of the structurally iden-tified metabolites in our initial study are potentially associatedwith responses to pemetrexedmonotherapy. To explore this idea,

pretreatment serum specimens from 18 patients who receivedfirst-line pemetrexed monotherapy (including 6 PR, 6 SD, and 6PD) were collected and tested in targeted analyses of the previ-ously identified 11 metabolites. We found that the accumulation

Figure 2.

Examination of themetabolic profiles ofpretreatment serum specimens. A,Heatmap representation ofunsupervised hierarchical clustering ofthe 85 significantly changed ionfeatures (rows) grouped by sampletype (columns). PR, partial response;SD, stable disease; PD, progressivedisease. ESIþ and ESI� representfeatures detected in the positive andnegative electrospray ionizationmodes, respectively. Shades of yellowand blue represent increase anddecrease of an ion feature respectivelyrelative to the average levels in all thesamples (see color scale).B, Scores plotof PLS-DA of samples in the discoveryset based on the ion 85 features. Red,progressive disease (PD) samples;yellow, stable disease (SD) samples;green, partial response (PR) samples.C,Values for the variable importance inthe project (VIP) of 85 ion features areillustrated. Each column represents onefeature in the PLS-DA model (Fig. 2B).Error bars, SEM.

Tian et al.





of three metabolites, choline, betaine, and DMG, changed signif-icantly between the DC and PD specimens (SupplementaryFig. S3A and S3B).

Discriminant model establishment based on the discovery setThe optimal variable combination patterns under certain num-

bers of variables (from 1 to 11) were obtained according to thelikelihood score by using logistic regression model with the bestsubset selection method (Supplementary Table S2). Under thenumber of variables from 7 to 11, the likelihood scores werecomparable (99.269–101.54) and superior to other number ofvariables. Furthermore, 5-fold cross validationwas used to test themean squared error (MSE) of the above 11 identified variablecombination patterns. As a result, the seven-metabolite panel(including hypotaurine, uridine, C12:0-carnitine, choline,dimethylglycine, niacinamide, C16:0-carnitine) yielded the min-imal MSE, which was therefore defined as the best variable model(Supplementary Fig. S4). On the basis of this, the discriminantmodel was established according to the logistic regression valuesg(z) (Fig. 4B). The ROC analysis of the model using the seven-metabolite panel yielded the AUCof 0.9214 (an AUCof 0.939 forpem-cis and 0.896 for pem-carbo, respectively; Fig. 4A; Supple-mentary Fig. S5A and S5B).

Validation of the discriminant modelTo validate the accuracy of the discriminant model, untargeted

metabolomic analysis were performed in the aforementionedvalidation sample set composed of 103 serum specimens. Similaraccumulation trends for the seven metabolites were observed invalidation set as comparedwith discovery set (Supplementary Fig.S6). The discriminant model established from the discovery setdata distinguished the specimens of the PD group from those ofthe DC group in the validation set with an AUC of 0.9092 (0.936for pem-cis–treated patients and 0.881 for pem-carbo–treatedpatients; Fig. 4C; Supplementary Fig. S7A and S7B). Logisticregression values g(z) were obtained for the patients in thevalidation set (Fig. 4D).

Totally, for all the 354 patients both in the discovery andvalidation set, we set up the optimal cut-off value of g(z) as0.155, which was in corresponding with the maximum Youdenindex of our discrimination model. Patients with g(z) valuesgreater than 0.155 will be classified as PD cases, values equal toor less than 0.155 as DC cases. Under the cut-off value, thediscrimination model was able to distinguish all the PD patientsfromDConeswith a sensitivity of 90.8% and specificity of 79.5%.

We further analyzed the potential prediction ability of thisdiscrimination model on progression-free survival (PFS). Themedian PFS of patients who were classified as DC cases was10.3 months (95% CI, 9.7–10.9), significantly longer than 4.5months (95% CI, 4.0–5.1) of PD cases (log-rank, P < 0.001, Fig.4E). The overall survival (OS) of patients predicted as DC caseswas similar with the overall survival of patients of PD cases(median OS, 17.8 vs. 18.3 months, log-rank, P ¼ 0.534; Fig.4F), with a censoring rate of approximately 9% of all the patients.

Prognostic prediction workflowTo illustrate the prognostic prediction workflow of the discrim-

inant model, we selected three pretreatment serum sampleswhose clinical responses were typical of PR, SD, and PD, respec-tively. Metabolomics analysis of the serum samples and applica-tion of the discriminantmodel was conducted (Fig. 5A andB).Onthe basis of the relative abundance of the seven-metabolite panel,the logistic regression values g(z) of these three samples werecalculated as 0.045, 0.112, and 0.882, respectively (Fig. 5C).Under the cut-off value of g(z) of 0.155,wepredicted that samplesA and B were from DC cases (either PR or SD) and that sample Cwas fromaPD case. These predictions from themodelwere in linewith CT imaging-based clinical evaluations of these three patients(Fig. 5D).

DiscussionThis prediction modeling study was carefully designed by

following the CHARMS checklists (27), including the

Figure 3.

Relative abundance of the 11 identifiedmetabolites. The abundance of eachmetabolite was normalized to the meanlevel of all tested samples. The z-scoresare shown as box and whisker plots. Thebands inside the box represent themedian abundance, and the ends of thewhiskers represent the 5th and 95thpercentiles. Data not included betweenthe whiskers are plotted as outliers asdots. DC group samples are shown inblue, while PD group samples are in red.The first eight metabolites decreasedand the last three increased whencomparing the DC group with the PDgroup samples.






development of themodel by untargeted analysiswith adiscoveryset and quantification of themodel's predictive performance withan independent validation set by using the same analytic methodas that in the discovery set. By using metabolomics analysis in alarge lung cancer population, we have developed an effectivediscriminant model based on a seven-metabolite panel that canpredict the efficacy of pemetrexed plus platinum doublet chemo-therapy prior to treatment delivery. This predictive model couldbe easily applied by physicians to select patients who mightbenefit from pemetrexed and platinum doublet therapy, whichprovides a promising strategy to personalize this widely usedchemotherapy.

Metabolomics, due to its close relationship with the phenotypeand sensitivity tomany factors, has beenwidely used inbiomarker

discovery. As the "downstream" omics of genomic, transcrip-tomic, and proteomics, metabolomics is able to capture a plu-rality of subtle changes that reflect alterations of biological states,even when no measurable changes of genes and proteins aredetected, thus, at least theoretically, is more likely to providecandidates of potential biomarkers (28). Besides, endogenousmetabolites are fewer than genes, transcripts, and proteins andhave the same basic chemical structure as well as highly conservedpathways,whichmakes it easier to interpret themetabolomic data(29, 30). Moreover, metabolomic studies can be assessed non-invasively byusingbiofluids, such as bloodandurine, and are alsoless expensive than other omics approaches, thus are easier to betranslated into clinical practice (31, 32).Despite these advantages,metabolomics still needs to overcome some limitations that

Figure 4.

Efficacy prediction based on the seven-metabolite panel.A and C, Plots of ROCresults for distinguishing PD samplesfrom the DC samples for the discoveryset (A) and the validation set (C). TheROC curves were created by plottingthe sensitivity (i.e., true positive rate)against 1�specificity (i.e., false positiverate). The blue line in each plotrepresents the area under the curve(AUC). B and D, The logistic regressionvalues g(z) for each patient of thediscovery (B) and validation set (D).Logistic regression values g(z) werecalculated with the formula g(z) ¼ 1/(1þe�z), where z ¼ �10.312þ6.81E-07X1þ1.30E-07X2þ1.55E-07X3þ1.29E-09X4-3.10E-08X5-1.66E-07X6-4.04E-8X7 (X1, hypotaurine; X2, uridine; X3,C12:0-carnitine; X4, choline; X5, DMG;X6, Niacinamide; and X7, C16:0-carnitine). E and F, Kaplan–Meiersurvival curves for progression-freesurvival (PFS; E) and overall survival(OS; F) time of patients classified by thediscrimination model as DC cases andPD cases. Survival differences wereevaluated by log-rank test.

Tian et al.





Figure 5.

Analysis workflow of prognostic prediction. A, Typical base peak of the serum specimen in positive ion mode. B, Metabolite identification and quantification ofthe seven metabolites of hypotaurine, uridine, C12:0-carnitine, choline, DMG, niacinamide, C16:0-carnitine. C, Logistic regression values (g(z)) and outcomeprediction of the three samples. D, Typical evaluation of pemetrexed-platinum chemotherapy efficacy according to RECIST version 1.1 by comparing CT imaging inthree representative patients at baseline (pretreatment) and after two cycles of treatment (posttreatment). PD, progressive disease; PR, partial response;SD, stable disease. The tumor sizes provided under the images were evaluated by physicians.






impact its applicability to systems biology studies. First, thecomplete coverage ofmetabolomebydetection and identificationof all the metabolites is still a challenge, requiring the applicationof different detecting platforms as well as the development of themetabolome databases. Besides, as the metabolome is so highlydynamic and sensitive to a wide range of internal and externalfactors (33), the information obtained needs to be validated towarrant the consistency and reproducibility. The main reasonunderlying the frequent failure to identify biomarkers that predictpatient response to particular chemotherapy treatments is that thetargets of cytotoxic agents are not single gene or protein aberra-tions (34, 35). Instead, cytotoxic agents target general biologicalprocesses such as cell proliferation and apoptosis, which can beinfluenced by cancer-related metabolic alterations. These cancer-driven metabolic abnormalities may be representative of atumor's intrinsic features and may even be indicative of thecancer's pathogenesis, which affords a change of serum metabo-lites that are biologically relevant to the system's phenotype(36, 37). Lung adenocarcinoma patients with EGFR-sensitizingmutations had an excellent objective response (60%–80%) andsurvival outcomes (PFS, 9–13 months) after EGFR-TKIs treat-ment, meaning that these selected patients had about 70% pos-sibility obtaining response and 50% possibility with diseasecontrol timemore than about 9months once identified harboringEGFR mutation. In our study, below the settled cut-off value ofg(z)� 0.155, themedian PFS of chemotherapy was 10.3months,and intriguingly comparablewith that of EGFR-TKIs. These resultsindicated that our discriminant model had satisfactory predictivevalue to chemo-response for clinical practice. Importantly, ourformulated gauge including boundary values based on PD pos-sibility can provide integral clinical prognostic information forphysicians and be used conveniently in clinical practice.

In total, 11 out of 32 ion featureswith VIP> 1was assignedwithchemical structures based on information in databases. Thesemetabolites are mainly involved in the metabolism of aminoacids, fatty acids andpyrimidine,which are closely associatedwithcancer progression and drug resistance. Uridine, an importantcomponent of pyrimidine metabolism, plays a crucial role insynthesizing RNA, glycogen, and biomembranes (38). Taurine isan essential, sulfur-containing organic compound that has manydiverse biological functions as a neurotransmitter, a cell mem-brane stabilizer and a transport facilitator of ions such as sodium,potassium, calcium, and magnesium (39). Hypotaurine is aproduct of a reaction catalyzed by cysteamine dioxygenase andhas been reported to function as an antioxidant and protectiveagent under physiologic conditions. Hypotaurine can also beoxidized to taurine by hypotaurine dehydrogenase (40). Carni-tine is able to be acylated by the attachment of L-palmitoylcarni-tine and dodecanoylcarnitine and thus participates in the forma-tion of organic compounds (41). These metabolites, involved inbiosynthesis, solute transport, and physiologic protection, werefound in our study to be potentially associatedwith the efficacy ofpemetrexed plus platinum doublet therapy.

By verification of these metabolites in patients receiving peme-trexedmonotherapy, choline, betaine, andDMGwere found tobeassociatedwith the efficacy of pemetrexedmonotherapy. Choline,

betaine, and DMG can generate sarcosine and glycine through aseries of demethylation reactions, which have been reported toplay key roles in cancer progression (17, 42). Sarcosine andglycine are involved in the donation of carbon units in theone-carbonmetabolism (43). It is noteworthy that the anticancereffects of pemetrexed are known to occur through the inhibitionof key folate cycle enzymes, such as thymidylate synthase (TS),dihydrofolate reductase (DHFR), and glycinamide ribonucleotideformyltransferase (GARFT; ref. 44),which inhibits folate synthesisand thus influences one-carbon metabolism. Accordingly, wededuce that the relatively high accumulation of these threemetabolites in the PD group patients indicates an abundantsource of components involved in one-carbon metabolism thatmay compensate for the disruption of one-carbonmetabolism bypemetrexed treatment and thereby lead to poor clinical outcomes.

To the best of our knowledge, this is the first large-cohort studyusing metabolic biomarkers to predict the clinical response topemetrexed plus platinum doublet therapy in patients with lungadenocarcinoma. The discriminantmodel developed in our studyoffers a feasible and convenient strategy to personalize the widelyused pemetrexed plus platinum doublet chemotherapy. The highaccuracy of our established model in discriminating the possibil-ity of chemo-response warrants further development of prospec-tive clinical trials.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed by the authors

Authors' ContributionsConception and design: Y. Tian, Z. Wang, Y. Yin, J. He, J. WangDevelopment of methodology: Y. Tian, X. Liu, J. WangAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): Z. Wang, X. Liu, D. Wang, J. WangAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): Y. Tian, X. Liu, G. Feng, C. Zhang, D. Wang, J. WangWriting, review, and/or revision of the manuscript: Y. Tian, Z. Wang, J. WangAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): Z. Wang, J. Duan, Z. Chen, H. Bai, R. Wan,J. Jiang, J. Liu, J. Han, X. Zhang, L. Cai, J. WangStudy supervision: J. Gu, S. Gao, J. Wang

AcknowledgmentsWe thank all patients that were involved in this study. We also thank the

support from Lina Xu, Xueying Wang and Yupei Jiao from the MetabolomicsFacility at the Technology Center for Protein Sciences of Tsinghua University.

This work was supported by the National Natural Sciences FoundationKey Program (81630071) (81330062); National Key Research and Devel-opment Project Precision Medicine Special Research (2016YFC0902300);National High Technology Research and Development Program 863(SS2015AA020403); CAMS Innovation Fund for Medical Sciences (CIFMS2016-I2M-3-008); Aiyou foundation (KY201701); National NaturalSciences Foundation, China (81101778, 81472206, and 81702289); andBeijing Natural Science Foundation, China (7172045).

The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received September 28, 2017; revised December 18, 2017; accepted January29, 2018; published first February 6, 2018.

References1. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer

statistics, 2012. CA: A Cancer J Clin 2015;65:87–108.2. Scagliotti GV, Parikh P, vonPawel J, BiesmaB, Vansteenkiste J,ManegoldC,

et al. Phase III study comparing cisplatin plus gemcitabine with cisplatin


Tian et al.




plus pemetrexed in chemotherapy-naive patientswith advanced-stage non-small-cell lung cancer. J Clin Oncol 2008;26:3543–51.

3. Ettinger DS, Wood DE, Akerley W, Bazhenova LA, Borghaei H, CamidgeDR, et al. NCCN Guidelines insights: non-small cell lung cancer, version4.2016. J Natl Compr Canc Netw 2016;14:255–64.

4. Genova C, Rijavec E, Truini A, Coco S, Sini C, Barletta G, et al. Pemetrexedfor the treatment of non-small cell lung cancer. Expert Opin Pharmacother2013;14:1545–58.

5. Zinner RG, Fossella FV, Gladish GW, Glisson BS, Blumenschein GR Jr,Papadimitrakopoulou VA, et al. Phase II study of pemetrexed in combi-nation with carboplatin in the first-line treatment of advanced nonsmallcell lung cancer. Cancer 2005;104:2449–56.

6. Sun JM, Ahn JS, Jung SH, Sun J, Ha SY, Han J, et al. Pemetrexed pluscisplatin versus gemcitabine plus cisplatin according to thymidylatesynthase expression in nonsquamous non-small-cell lung cancer: abiomarker-stratified randomized phase II Trial. J Clin Oncol 2015;33:2450–6.

7. Shimizu T, Nakanishi Y, Nakagawa Y, Tsujino I, Takahashi N, Nemoto N,et al. Association between expression of thymidylate synthase, dihydro-folate reductase, and glycinamide ribonucleotide formyltransferase andefficacy of pemetrexed in advanced non-small cell lung cancer. AnticancerRes 2012;32:4589–96.

8. Olaussen KA, Dunant A, Fouret P, Brambilla E, Andre F, Haddad V, et al.DNA repair by ERCC1 in non-small-cell lung cancer and cisplatin-basedadjuvant chemotherapy. N Engl J Med 2006;355:983–91.

9. Bowden NA. Nucleotide excision repair: why is it not used to predictresponse to platinum-based chemotherapy? Cancer Lett 2014;346:163–71.

10. Friboulet L, Olaussen KA, Pignon JP, Shepherd FA, Tsao MS, Graziano S,et al. ERCC1 isoform expression and DNA repair in non-small-cell lungcancer. N Engl J Med 2013;368:1101–10.

11. Lynch TJ, Bell DW, Sordella R, Gurubhagavatula S,OkimotoRA, BranniganBW, et al. Activating mutations in the epidermal growth factor receptorunderlying responsiveness of non-small-cell lung cancer to gefitinib.N Engl J Med 2004;350:2129–39.

12. Kwak EL, Bang YJ, Camidge DR, Shaw AT, Solomon B, Maki RG, et al.Anaplastic lymphoma kinase inhibition in non-small-cell lung cancer.N Engl J Med 2010;363:1693–703.

13. Agarwal R, Kaye SB. Ovarian cancer: strategies for overcoming resistance tochemotherapy. Nat Rev Cancer 2003;3:502–16.

14. Beger RD. A review of applications of metabolomics in cancer. Metabolites2013;3:552–74.

15. Spratlin JL, Serkova NJ, Eckhardt SG. Clinical applications of metabolo-mics in oncology: a review. Clin Cancer Res 2009;15:431–40.

16. Mapstone M, Cheema AK, Fiandaca MS, Zhong X, Mhyre TR, MacArthurLH, et al. Plasma phospholipids identify antecedent memory impairmentin older adults. Nat Med 2014;20:415–8.

17. Sreekumar A, Poisson LM, Rajendiran TM, Khan AP, Cao Q, Yu J, et al.Metabolomic profiles delineate potential role for sarcosine in prostatecancer progression. Nature 2009;457:910–4.

18. Yuan M, Breitkopf SB, Yang X, Asara JM. A positive/negative ion-switch-ing, targeted mass spectrometry-based metabolomics platform forbodily fluids, cells, and fresh and fixed tissue. Nat Protoc 2012;7:872–81.

19. Travis WD, Brambilla E, Noguchi M, Nicholson AG, Geisinger KR,Yatabe Y, et al. International association for the study of lung can-cer/american thoracic society/european respiratory society internationalmultidisciplinary classification of lung adenocarcinoma. J Thorac Oncol2011;6:244–85.

20. SmildeAK, vanderWerfMJ, Bijlsma S, vanderWerff-van der Vat BJ, JellemaRH. Fusion of mass spectrometry-based metabolomics data. Anal Chem2005;77:6729–36.

21. Yin P, Zhao X, Li Q, Wang J, Li J, Xu G. Metabonomics study of intestinalfistulas based on ultraperformance liquid chromatography coupled withQ-TOF mass spectrometry (UPLC/Q-TOF MS). J Proteome Res 2006;5:2135–43.

22. KamarudinAN,Cox T, Kolamunnage-Dona R. Time-dependent ROC curveanalysis in medical research: current methods and applications. BMCMedRes Methodol 2017;17:53.

23. Schisterman EF, Perkins NJ, Liu A, Bondell H. Optimal cut-point and itscorresponding Youden index to discriminate individuals using pooledblood samples. Epidemiology 2005;16:73–81.

24. Want EJ, Wilson ID, Gika H, Theodoridis G, Plumb RS, Shockcor J, et al.Global metabolic profiling procedures for urine using UPLC-MS.Nat Protoc 2010;5:1005–18.

25. Huang Q, Tan Y, Yin P, Ye G, Gao P, Lu X, et al. Metabolic characterizationof hepatocellular carcinoma using nontargeted tissue metabolomics.Cancer Res 2013;73:4992–5002.

26. DunnWB, Broadhurst D, Begley P, Zelena E, Francis-McIntyre S, AndersonN, et al. Procedures for large-scalemetabolic profiling of serum and plasmausing gas chromatography and liquid chromatography coupled to massspectrometry. Nat Protoc 2011;6:1060–83.

27. Moons KGM, de Groot JAH, Bouwmeester W, Vergouwe Y, Mallett S,Altman DG, et al. Critical appraisal and data extraction for systematicreviews of prediction modelling studies: the CHARMS checklist.PLoS Medicine 2014;11:e1001744.

28. Monteiro MS, Carvalho M, Bastos ML, Guedes de Pinho P. Metabolomicsanalysis for biomarker discovery: advances and challenges. CurrMedChem2013;20:257–71.

29. Davis VW, Bathe OF, Schiller DE, Slupsky CM, Sawyer MB. Metabolomicsand surgical oncology: Potential role for small molecule biomarkers. J SurgOncol 2011;103:451–9.

30. Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, Kell DB. Metabo-lomics by numbers: acquiring and understanding global metabolite data.Trends Biotechnol 2004;22:245–52.

31. Kind T, Tolstikov V, Fiehn O, Weiss RH. A comprehensive urinary meta-bolomic approach for identifying kidney cancer. Anal Biochem 2007;363:185–95.

32. Fiehn O, Kristal B, van Ommen B, Sumner LW, Sansone SA, Taylor C, et al.Establishing reporting standards for metabolomic and metabonomicstudies: a call for participation. OMICS 2006;10:158–63.

33. Castle AL, Fiehn O, Kaddurah-Daouk R, Lindon JC. Metabolomics Stan-dards workshop and the development of international standards forreporting metabolomics experimental results. Brief Bioinform 2006;7:159–65.

34. Muhsin M, Gricks C, Kirkpatrick P. Pemetrexed disodium. Nat Rev DrugDiscov 2004;3:825–6.

35. Ahmad A, Gadgeel S. Lung Cancer and Personalized Medicine. SpringerInternational Publishing; 2016.

36. Griffin JL, Shockcor JP. Metabolic profiles of cancer cells. Nat Rev Cancer2004;4:551–61.

37. Loo JM, Scherl A, Nguyen A, Man FY, Weinberg E, Zeng Z, et al.Extracellular metabolic energetics can promote cancer progression.Cell2015;160:393–406.

38. Yamamoto T, Koyama H, Kurajoh M, Shoji T, Tsutsumi Z, Moriwaki Y.Biochemistry of uridine in plasma. Clin Chim Acta 2011;412:1712–24.

39. Ripps H, Shen W. Review: taurine: a "very essential" amino acid. Mol Vis2012;18:2673–86.

40. Fontana M, Pecci L, Dupre S, Cavallini D. Antioxidant properties ofsulfinates: protective effect of hypotaurine on peroxynitrite-dependentdamage. Neurochem Res 2004;29:111–6.

41. Reuter SE, Evans AM. Carnitine and acylcarnitines: pharmacokinetic,pharmacological and clinical aspects. Clin Pharmacokinet 2012;51:553–72.

42. Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL, et al.Metabolite profiling identifies a key role for glycine in rapid cancer cellproliferation. Science 2012;336:1040–4.

43. Locasale JW. Serine, glycine and one-carbon units: cancer metabolism infull circle. Nat Rev Cancer 2013;13:572–83.

44. Chattopadhyay S, Moran RG, Goldman ID. Pemetrexed: biochemical andcellular pharmacology, mechanisms, and clinical applications.Mol CancerTher 2007;6:404–17.






2018;24:2100-2109. Published OnlineFirst February 6, 2018.Clin Cancer Res Yanhua Tian, Zhijie Wang, Xiaohui Liu, et al. Cancer by Serum Metabolomic Profiling

Small Cell Lung−Prediction of Chemotherapeutic Efficacy in Non

Updated version

10.1158/1078-0432.CCR-17-2855doi:

Access the most recent version of this article at:

Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2018/02/06/1078-0432.CCR-17-2855.DC1

Access the most recent supplemental material at:

Cited articles

http://clincancerres.aacrjournals.org/content/24/9/2100.full#ref-list-1

This article cites 43 articles, 8 of which you can access for free at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://clincancerres.aacrjournals.org/content/24/9/2100To request permission to re-use all or part of this article, use this link



http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-17-2855

http://clincancerres.aacrjournals.org/content/suppl/2018/02/06/1078-0432.CCR-17-2855.DC1

http://clincancerres.aacrjournals.org/content/24/9/2100.full#ref-list-1

http://clincancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://clincancerres.aacrjournals.org/content/24/9/2100


Documents

Prediction of Chemotherapeutic Efﬁcacy in Non Small Cell ... · Personalized Medicine and Imaging Prediction of Chemotherapeutic Efﬁcacy in Non– Small Cell Lung Cancer by Serum