Integrative Systemic and Local Metabolomics with Impact on Survival in High Grade Serous Ovarian Cancer SUPPLEMENTARY INFORMATION Anna Bachmayr-Heyda1,*, Stefanie Aust1,*, Katharina Auer1, Samuel M. Meier2, Klaus G. Schmetterer3, Sabine Dekan4, Christopher Gerner2, Dietmar Pils5,# 1
Dept. of Obstetrics and Gynecology, Comprehensive Cancer Center (CCC), Medical University of Vienna, Vienna, Austria 2 Dept. of Analytical Chemistry, University of Vienna, Vienna, Austria 3 Dept. of Laboratory Medicine, Medical University of Vienna, Vienna, Austria 4 Dept. of Pathology, Medical University of Vienna, Vienna, Austria 5 Section for Clinical Biometrics, Center for Medical Statistics, Informatics and Intelligent Systems, Medical University of Vienna, Vienna, Austria #
[email protected]
*These authors contributed equally to this work.
Materials and Methods (extended) Patients, healthy controls, and validation cohort Serum of 65 therapy-naïve high grade serous ovarian cancer (HGSOC) patients and of 62 healthy controls–24 completely healthy women (without diagnosed severe disease) and 38 patients with benign or precancerous gynecologic diseases (comprising CIN II/III, n=22; benign ovarian cysts, n=8; uterus myomatosus, n=3; descensus uteri, n=2; chronic pelvic pain, n=2; early spontaneous abortion, n=1)–were collected from 2004 to 2014 at the Medical University of Vienna and stored at -85°C until use (patient cohorts and methods are outlined in Fig. S1). Ascites and fresh tumor samples were collected and processed as previously described (1). Clinicopathologic characteristics are given in Table S1. All patients signed an informed consent and approval for this study was obtained by the ethical review board (nos. 366/2003 and 793/2011). The validation cohort of 165 patients was selected from the gene expression study Gene Expression Omnibus acc. GSE49997 (2-4) meeting the following criteria: advanced FIGO stage and serous histology. Targeted metabolomics of serum, ascites, tissues, and blood fractions and vesicles 10 µl serum, 20 µl cell free ascites, or 10 µl tissue extract was used for targeted metabolomics. Tissues were extracted in three times (µl) 85:15 methanol:phosphate buffer (10 mM, pH 7.5) per mg tissue for three cycles of 30’ at 5,500 rpm on a FastPrep FP120 (Thermo Savant, Farmingdale, NY, USA) using ceramic beads (1.4 mm diameter). Targeted metabolomics was performed using AbsoluteIDQ p180 kits (Biocrates Life Sciences AG, Innsbruck, Austria). The kit allows the identification and (semi-)quantification of 188 metabolites by LC- and flow injection analysis (FIA)MRM: 90 glycerophospholipids (GPhLs): 38 O-acyl-O-acyl phosphatidylcholines (PC_aa...), 38 O-alkylO-acyl
phosphatidylcholines
(PC_ae...),
and
14
lyso-phosphatidylcholines
(LysoPC...);
15
sphingolipids; 40 acylcarnitines; 21 amino acids (AAs); 21 biogenic amines; and the sum of hexoses (including glucose). The samples were analyzed on an AB SCIEX QTrap 4000 mass spectrometer (Framingham, MA, USA) using an Agilent 1200 RR HPLC system (Agilent Technologies, Santa Clara, CA, USA), which were operated with Analyst 1.6.2 (AB SCIEX). The chromatographic column was obtained from Biocrates. The serum samples and additional blanks, calibration standards and quality controls were prepared according to the user manual. All amino acids and biogenic amines were derivatized with phenylisothiocyanate. The experiments were validated with the supplied software (MetIDQ, Version 5-4-8-DB100-Boron-2607, Biocrates).
The correlation coefficients of the calibration curves of amino acids and biogenic amines were all >0.95. The accuracies of methioninesulfoxide (Met-SO), histamine, symmetric dimethylarginine (SDMA), α-aminoadipic acid (α-AAA) and trans-4-hydroxyprpoline (t4-OH-Pro) were not within the tolerance window in the quality controls and some analytes were below corresponding limit of detections in most samples (one GPhL, PC aa C30:2, and following biogenic amines: c4-OH-Pro, carnosine, dopamine, nitro-tyr, and PEA) and were therefore not further evaluated. Quality controls were analyzed every 20th sample. Finally 179 metabolite concentrations were log2 transformed to obtain near normal distributions and associated to right censored overall survival data by the Significance Analysis of Microarrays (SAM) method (R-package samr v2.0 (5)) using a 5% false discovery rate (FDR) cut-off. A non-linear dimensionality reduction approach, comparable to a principal component analysis, called Isomap (R-package RDRToolbox v1.20.0 (6)) was used to condense the 43 significantly with OS-associated GPhLs to one GPhL score, using n-1 neighbors and the modified Isomap algorithm. Isolation of blood fractions Vesicle-free soluble fractions and pellets were isolated from mixtures of each five sera from cancer patients or healthy controls. Sera mixtures were brought to a density of 1.21 g/mL with KBr and centrifuged at 450,000 g for 27 hours at 10°C. KBr saturated ultracentrifugation derived soluble fractions and pellets should be vesicle-free (i.e. free of HDLs, LDLs, vLDLs, and exosomes), because these vesicles are less dense than a (nearly) saturated KBr solution (1.21 g/mL) and concentrate on top of such a solution upon a long and fast enough ultracentrifugation step. Three fractions were collected and used for targeted metabolomics: A soluble fraction (soluble_EOC and soluble_c) from the middle of the serum fraction (yellow, transparent fraction in the lower two-third range of the tube below a colorless fraction of a KBr solution), a viscose fraction from the bottom of the tube, visibly different from the remaining soluble fraction (oPellet_EOC and oPellet_c), and after rinsing of the emptied tube, the resuspended pellet (Pellet_EOC and Pellet_c). Human HDLs, LDLs, and vLDLs were obtained from Sigma-Aldrich (St. Louis, MO, USA), with following order numbers: L8039, L8292, and L7527, respectively. Exosomes were isolated from sera mixtures (diluted 1:1 with PBS) from each five patients with high or low GPhL scores and from ten healthy controls by ultracentrifugation at 110,000 g for 3 hours at 10°C (Exo_high, Exo_low, and Exo_c). Clinical laboratory and multiplexed cyto-/chemokine measurements Total cholesterol, triglycerides, low density lipoprotein (LDL), high density lipoprotein (HDL), albumin, and C-reactive protein (CRP) were measured on a cobas 8000 modular analyzer (HoffmannLaRoche, Basel, Switzerland) in an ISO9001:2008 certified laboratory (Department of Laboratory
Medicine, Medical University of Vienna) from serum and ascites. Luminex based analyses were performed from ascites and serum following instructions provided by the corresponding kits on a Bio-Plex 200 System (Bio-Rad Laboratories, Hercules, CA, USA): “Bio-Plex Pro Human Cancer Biomarker Assays: Panel 1” (n=16), “Bio-Plex Pro Human Chemokine Panel Assay” (n=40) (both BioRad Laboratories), and “Cytokine Human Magnetic 25-Plex Panel” (Life Technologies, Carlsbad, CA, USA). Correlations to the GPhL score were assessed by linear models fitted for each analyte, and significantly correlated analytes called using an empirical Bayes statistics (eBayes function, R-package limma v3.26.7 (7)) with Benjamini Hochberg correction for multiple testing (FDR). Correlations are presented as correlation plots (with Spearman’s correlation coefficients and p-values). PD-L1, PD-1, and CD8 immunohistochemistry staining Staining was performed on 4 µm formalin-fixed and paraffin-embedded (FFPE) tumor tissue sections from ovarian tumor masses (P, “primary” tumors) and peritoneal tumor masses (M, “metastases”) with a Leica BOND-III Fully Automated IHC & ISH system (Leica Biosystems Nussloch GmbH, Nussloch, Germany) using usual staining chemistry (BOND Polymer Refine Detection Kit, DS9800) and antigen retrieval Heat Induced Epitope Retrieval (HIER) for 20 min with BOND Epitope Retrieval Solution 1. Antibodies in following dilutions were used: 1:100 PD-L1 (E1L3N) XP Rabbit mAb (Cell Signaling Technology, Danvers, MA, USA), 1:50 PD-1 Cell Marque 315M-96 monoclonal mouse anti human antibody (Sigma-Aldrich, St. Louis, MO, USA), and 1:100 CD8 (clone C8/144B), DAKO CD8 M7103 monoclonal mouse anti human antibody (Agilent Technologies, Santa Clara, CA, USA), each incubated for 30 min. Following cell percentages were assessed by two independent persons, one pathologist and one pathological trained medical doctor supervised by the pathologist (discrepancies were discussed upon agreement): PD-L1 positive tumor cells, PD-L1 positive intraepithelial positive tumor infiltrating immune cells, CD8 positive intraepithelial immune cells, and PD-1 positive intraepithelial immune cells based on CD8 positive intraepithelial immune cells. Correlations of these four measures to the GPhL score were assessed by univariate and multiple linear regression models using ln-transformed values (0 replaced by 0.1) and the tissue information, P or M, as covariate. Plots are shown with linear values, Spearman’s correlation coefficients and univariate Spearman’s pvalues, and color-coded tissue types (orange, primary tumors and red, metastases). RNA-sequencing and biological interpretation Ribosomal RNA depleted total RNA (Ribo-Zero™ rRNA Removal Kit, Epicentre, Madison, WI, USA) from enzymatically detached and for EpCAM-positivity enriched tumor cells was used for 50 bp paired-end RNA sequencing to a median depth of 52 mio reads (for details see (1)). Reads were mapped to the HG19 genome, counted into the Gencode 18 gene model and normalized with R-
package limma v3.26.3 (7, 8) using voomWithQualityWeights-function and cyclic loess normalization. Genes correlated to the GPhL score were assessed by additionally using tissue (solid or floating ascites derived tumor cells) and patient information in the design matrix from 28,203 reliably expressed genes. An FDR cut-off of 5% was used for the significant gene list (320 genes) and of 20% (2,054 genes) for Signaling Pathway Impact Analysis (R-package SPIA v2.22.0, (9)). A 100 gene expression signature indicative for the GPhL score was defined as 50 most significantly positively correlated genes and 50 most significantly negatively correlated genes. Correlation to different molecular subclasses (TCGA (10) and C1-C6 (11)) was assessed by R-package genefu v2.2.0. Correlation to Yoshihara’s (12) subclassification was conducted using the 100 gene expression signature and the subclassification information from the validation cohort as provided in (2) and GSE49997. Correlations of histone expressions with the log2-fransformed number of expressed genes and the log2-fransformed number of unannotated sequencing read pairs, i.e. fragments (mapped but not annotated to the used gene model), were assessed by linear regression models using all significantly with the GPhL-score associated histones or by Spearman’s correlations of histone HIST1H4J alone (Fig. 5). Survival analysis and independent validation Survival analyses were performed by univariate and multiple Cox-regression analyses using clinicopathologic factors and relevant laboratory measures as covariates (see Table 1), mainly by the R based coxph-function. The GPhL score was trichotomized along the 33.33 and 66.67 percentiles. To avoid over-fitting of the multiple model, some minor important covariates were included into the model separately, always together with the relevant covariates and the trichotomized GPhL score. An Akaike information criterion (AIC) maximization step-down procedure (stepAIC-function (backward) from R-package MASS v7.3-45, (13)) was used to eliminate non-relevant covariates from the final Cox-regression models. The final model was illustrated by survival estimates averaging covariates and using the trichotomized GPhL score as discriminator (Fig. 4A). For validation of the impact of the GPhL score on overall survival, the 100 gene expression signature was used as surrogate marker for the GPhL score and validated in an independent and previously described cohort of serous ovarian cancer patients (1-4). The gene predictor was calculated as follows: median 50 top positively correlated genes minus median 50 top negatively correlated genes, and used as trichotomized gene predictor in univariate and multiple Cox-regression analyses. Covariates which were not available from this patient cohort were not used for Coxregressions (see Table 1). The final model was illustrated as above (Fig. 4B).
Figure S1. Scheme of used methods (top box) and samples (colored bars) and obtained results. Numbers in circles correlate to same numbers in Fig. 5, the summarized results scheme. Table S1. Characteristics of all high grade serous ovarian cancer patients and sub-cohorts used for various analyses and of healthy controls.
Ascites
Tissue
PD-L1 PD-1 FFPE
n=19 57, 44-70 14, 2 15, 4 6, 13 8, 11 20, 0-35 4
n=9 n.d.2 n.d. n.d. n.d. n.d. n.d. n.d.
n=20 56, 44-80 14, 3 18, 2 4, 16 10, 10 23, 0-30 4
Metabolomics Controls Age [median, range] Patients Age [median, range] ECOG status [0, 123] FIGO stage [III, IV] Grade [2, 3] Residual tumor [no, yes] Follow-up [median, range] Cases of death [n] 1
Serum n=62 40, 22-74 n=65 60, 26-82 53, 81 52, 13 14, 51 30, 35 29, 0-91 29
RNA-seq Tumor cells3
n=16 58, 49-80 13, 1 16, 0 5, 11 8, 8 26, 0-35 5
Cyto/chemokines (n=56) Serum Ascites
n=20 54, 44-70 13, 3 18, 2 3, 17 12, 8 24, 0-35 4
n=18 58, 44-70 13, 2 14, 4 6, 12 8, 10 20, 0-35 4
Cyto/ chemokines (n=25) Serum
n=54 62, 26-82 47, 5 46, 8 12, 42 25, 29 35, 0-91 27
Four missing. 2Not determined. Selected according tissue availability and value of the GPhL score. 3 Tumor cells enriched for EpCAM positivity after tissue digestion from ovarian tumor masses (P, “primary tumor”), from peritoneal tumor implants (M, “metastatic implants”), from ascites as single EpCAM positive cells (A, “ascites single tumor cells”) or tumor cell aggregates (S, “spheroids”).
Figure S2. Correlations of median serum concentrations of metabolite classes of 62 healthy controls and 65 HGSOC patients. Metabolite groups are: aa, amino acids (n=21); ba, biogenic amines (n=13); ac, acylcarnitines (n=40); GPhL, glycerophospholipids (n=89); SphL, sphingolipids (n=15). Size coded values in the top-right triangle represent correlation coefficients and significance levels are indicated by asterisks (p
