Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Integrated Molecular Analysis of Undifferentiated Uterine Sarcomas Reveals Clinically Relevant Molecular Subtypes

Amrei Binzer-Panchal*1, Elin Hardell*2, Björn Viklund1, Mehran Ghaderi2, Tjalling Bosse3, Marisa R. Nucci4, Cheng-Han Lee5, Nina Hollfelder1, Pádraic Corcoran1, Jordi Gonzalez-Molina2,6, Lidia Moyano-Galceran6, Debra A. Bell8, John K. Schoolmeester8, Anna Måsbäck9, Gunnar B. Kristensen10, Ben Davidson11, Kaisa Lehti6,7, Anders Isaksson†1, Joseph W. Carlson†2

*Contributed equally to this work †Contributed equally to this work

1 Science for Life Laboratory, Department of Medical Sciences, Uppsala University, 751 85 Uppsala, Sweden 2 Department of Oncology-Pathology, Karolinska Institutet, and Department of Pathology and Cytology, Karolinska University Hospital, SE-171 76 Stockholm, Sweden. 3 Department of Pathology, Leiden University Medical Center, 2300 RC Leiden, The Netherlands 4 Department of Pathology, Brigham and Women’s Hospital, Boston, MA, 02138 USA 5 Department of Pathology and Laboratory Medicine, BC Cancer, Vancouver, BC. Canada 6 Department of Microbiology, Tumor and Cell Biology, Biomedicum, Karolinska Institutet, SE-171 65, Stockholm, Sweden 7 Genome-Scale Biology, Research Programs Unit, University of Helsinki, Helsinki, Finland 8 Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, MN, 55905, USA 9 Department of Pathology, Skånes University Hospital, 222 41 Lund, Sweden 10 Department Gynecologic Oncology and Institute for Cancer Genetics and Informatics, Norwegian Radium Hospital, Oslo University Hospital, N-0424 Oslo, Norway 11 Department of Pathology, Norwegian Radium Hospital, Oslo University Hospital, N-0424 Oslo, Norway; Institute for Clinical Medicine, The Medical Faculty, University of Oslo, N-0316 Oslo, Norway

Address for Correspondence Joseph W. Carlson [email protected] Radiumhemmet P1:02 Karolinska University Hospital 17176 Stockholm

1

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Abstract:

Purpose: Undifferentiated uterine sarcomas (UUS) are rare, extremely deadly, sarcomas with no effective treatment. The goal of this study was to identify novel intrinsic molecular UUS subtypes using integrated clinical, histopathologic and molecular evaluation of a large, fully annotated, patient cohort. Experimental Design: Fifty cases of UUS with full clinicopathological annotation were analyzed for expression (n=50), copy number variation (CNV, n=40), cell morphometry (n=39) and expression (n=22). and network enrichment analysis were used to relate over- and underexpressed to pathways and further to clinicopathologic and phenotypic findings. Results: Gene expression identified four distinct groups of tumors, which varied in their clinicopathologic parameters. Gene ontology analysis revealed differential activation of pathways related to genital tract development, extracellular matrix (ECM), muscle function, and proliferation. A multivariable, adjusted Cox proportional hazard model demonstrated that RNA group, mitotic index and hormone receptor expression, influence patient overall survival (OS). CNV arrays revealed characteristic chromosomal changes for each group. Morphometry demonstrated that the ECM group, the most aggressive, exhibited a decreased cell density and increased nuclear area. A cell density cutoff of 4,300 tumor cells per mm2 could separate ECM tumors from the remaining cases with a sensitivity of 83% and a specificity of 94%. Immunohistochemical staining of MMP-14, Collagens 1 and 6 and Fibronectin revealed differential expression of these ECM related proteins, identifying potential new biomarkers for this aggressive sarcoma subgroup. Conclusions: Molecular evaluation of UUS provides novel insights into the biology, prognosis, phenotype and possible treatment of these tumors.

2

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Translational Relevance:

Undifferentiated uterine sarcomas are among the rarest and deadliest of the uterine sarcomas. This has hindered the molecular understanding of their biology, and thus limited the introduction of new therapies. This study uses a well-annotated, large cohort of UUS, combined with RNA expression, chromosomal copy number, computer assisted histological analyses and immunohistochemistry, to identify and describe four intrinsic subtypes of these tumors. These subtypes vary in their biology, clinicopathological parameters, and survival. The most aggressive, ECM, subtype was characterized by a tumor cell phenotype with distinct morphology and protein expression, which will provide means to identify these cases using current laboratory techniques. Unique chromosomal changes were significantly associated with each group. Finally, gene ontology and network enrichment analysis identified target candidates for therapy. These results, from our hypothesis-generating comprehensive approach, will open new avenues to study and stratify these tumors, with the long- term goal of developing clinical interventions that will help improve patient survival.

3

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Introduction

Undifferentiated uterine sarcomas are high-grade malignant mesenchymal tumors (1). These tumors are extremely rare, so knowledge of their biology, prognosis, and therapy has been limited to small case series with often patchy or limited follow up. They are diagnosed after exclusion of other, more common, mesenchymal tumors of the uterus and soft tissue, particularly leiomyosarcoma, low-grade and high-grade endometrial stromal sarcoma and carcinosarcoma (2).

Recent large scale genomic studies of sarcomas have revealed several important conclusions (3–5) First, they can be generally divided into translocation sarcomas, which show a diploid or near-diploid genome, and karyotypically complex sarcomas, which often show large and complex chromosomal gains and losses of genetic material (6). Second, within traditionally defined sarcoma subtypes are subtype- specific molecular characteristics that can govern biology, therapy and prognosis (3,4). Thus, although general conclusions regarding sarcoma biology can be made with these studies, further work is required to understand the subtype and location specific changes that might govern biology, prognosis and, ultimately, therapy.

Uterine sarcomas have a distinct biology from other soft tissue sarcomas. They demonstrate unique translocations, such as JJAZ-JAZF1 and YWHAE-FAM22, that are not seen at other tissue sites (7–9). Benign tissues of the female genital tract express hormone receptors, and this is retained in a subset of sarcomas (10,11). This expression has been demonstrated to confer a better prognosis in leiomyosarcomas (10). Indeed, smooth muscle tumors have a distinct biology depending upon whether they arise in the gynecologic tract or not (3,5). Previously, our group has demonstrated that division into mitotic index groups has prognostic significance for overall survival (12,13). Other studies have attempted to use atypia to subdivide UUS into “uniform” and “pleomorphic” types (14,15). To date, no large scale molecular characterization has been performed on undifferentiated uterine sarcomas.

Research into therapies for gynecologic sarcomas has been limited by the difficulty of assembling a sufficient number of cases for clinical trials. Despite limited evidence, primary therapy is complete surgical resection, if possible. There is no conclusive data regarding the use of adjuvant therapy in UUS, and the current suggestion is to use therapies indicated for soft tissue sarcomas at other sites (16). One study evaluated the use of pazopanib, a multi-targeted receptor tyrosine kinase inhibitor, in pretreated, metastatic uterine sarcomas (17). This study demonstrated clinically relevant efficacy and tolerability.

The goal of this study was to comprehensively examine the biology of a large, well- annotated cohort of UUS, in order to identify tumor intrinsic molecular subgroups with biological, clinical and potential therapeutic significance.

4

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Materials and Methods

Patient cohort and central review

This retrospective patient cohort was assembled via international collaboration from seven collaborating centers. Ethical approval was obtained from the relevant local authorities. The majority of cases were submitted to three levels of review. First, they were reviewed by the original diagnosing pathologist. Second, a central review was performed by the participating expert gynecologic pathologist. Based on this review a single representative hematoxylin and eosin (H&E) stained tumor slide was selected for the third, central, review. Finally, the selected slides were reviewed again at the Karolinska University Hospital. Mitotic rate and grade of nuclear atypia was determined as described previously (12,13). Immunohistochemistry for estrogen receptor and progesterone receptor was performed locally in clinical labs accredited for this analysis. Representative FFPE tumor material containing a minimum of 70% tumor cells was submitted to the Karolinska University Hospital for isolation of DNA and RNA. One representative tumor slide was digitally scanned (Hamamatsu NanoZoomer, 40X scan) for image analysis. Exceptions to the above protocol were Skånes University Hospital (n= 10 patients) , where no second, local, review was performed; all slides were submitted for the third, central review, and Vancouver General Hospital (n=4 patients), where no central review was performed; mitotic count and atypia review were assessed by the participating pathologist.

RNA expression arrays

RNA quality was evaluated using the Agilent 2100 Bioanalyzer system (Agilent Technologies Inc, Palo Alto, CA). 100 ng of total RNA from each sample were used to generate amplified and biotinylated sense-strand cDNA from the entire expressed genome according to the Sensation Plus™ FFPE Amplification and WT Labeling Kit (P/N 703089, Rev.4 ThermoFisher Scientific Inc., Life Technologies, Carlsbad, CA 92008 USA). GeneChip™ ST Arrays (GeneChip™ Human Gene 2.1 ST Array Plate) were hybridized, washed, stained and finally scanned with the GeneTitan™ Multi- Channel (MC) Instrument, according to the GeneTitan™ Instrument User Guide for Expression Array Plates (PN 702933, ThermoFisher, Scientific Inc., Life Technologies, Carlsbad, CA 92008 USA).

DNA copy number arrays

DNA quantity was measured using the Qubit Fluorometer. Samples with low concentration of DNA were concentrated with the use of MinElute Reaction Cleanup Kit (50) Cat nr. 28204 (QIAGEN). During this procedure, the DNA binds to a column, is rinsed with washing buffer and finally eluted in nuclease free water.

5

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

DNA Array experiments were performed according to standard protocols for Affymetrix OncoScan® Arrays (Affymetrix OncoScan® FFPE Assay Kit User Guide (P/N 703175 Rev. 2), Affymetrix Inc., Santa Clara, CA). 80 ng total genomic DNA was incubated overnight to anneal the MIP probe. Each sample was then divided into two different channels, one for AT nucleotides and another for GC nucleotides. The gaps formed after the annealing process were filled using dNTPs and relevant reagents. Exonuclease removed non-ligated MIP probes and a cleavage enzyme linearized the circular MIP probes. Then the DNA was amplified via two cycles of PCR and digested using the HaeIII enzyme. On the Oncoscan® Arrays, hybridized probes were captured by streptavidin-phycoerythrin conjugates using the GeneChip™ Fluidics Station 450 and arrays were scanned using GeneChip® Scanner 3000 7G.

The OncoScan fluorescence intensity (CEL) files were normalized with Affymetrix OncoScan Console v.1.3 with default settings. Segmentation was done with Nexus Copy Number v.9.0 from BioDiscovery with the TuScan algorithm. Allele specific copy number analysis were performed with the Tumor Aberration Prediction Suite (TAPS) 2.0 (18). Affymetrix expression array, Oncoscan and clinical metadata is available via the Gene Expression Omnibus (GEO) database repository accession number GSE119043.

Quantitative Cell Morphometry

QuPath v0.1.2 (https://github.com/qupath/qupath) open source software was used for cell detection (19). Whole scanned slides, available for the majority of cases (39/50; 78%) were reviewed, blinded to clinical and molecular parameters, and two 1.27 mm in diameter circles were selected as regions of interest (ROI). The automatic “Cell detection” was run on the regions of interest with the default settings except the “Maximum area” that was increased to 3400 µm2 due to large cells in particular samples. The results from the two ROIs were averaged to obtain a single value for the case. The results were exported as a text file and further analysed with R (http://www.r- project.org).

Immunohistochemistry

TMA Sections were deparaffinized and rehydrated (2x 10 min Tissue Clear, 2x 5 min absolute EtOH, 1x 5 min 96% EtOH, 1x 10 min 70% EtOH, 2x 5min MQ H2O). Antigen retrieval was performed using 10 mM sodium citrate pH 6 (15 min heat, 30 min cool down, 2x 5 min PBS). Endogenous peroxidase was quenched with 0.6 % H2O2 for 10 min, 2x 5 min PBS (for ImmPRESS kit) or with 0.03 % H2O2 for 10 min, 1 min H2O, 10 min PBS (for Tyramide Signal Amplification (TSA) kit).

For the ImmPRESS method, sections were blocked with 2.5% normal horse serum for 30 min (ImmPRESS, Vector laboratories, Cat # MP-7402). Sections were incubated with αMT1-MMP(LEM)=MMP14 antibody (Millipore, Cat # MAB3328) diluted 1:100 in

6

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

2.5% normal horse serum over night at 4°C in a humidity chamber. Secondary antibody incubation was performed with ImmPRESS reagent (anti-mouse IgG coupled to peroxidase, ImmPRESS, Vector laboratories, Cat # MP-7402) for 30 min at RT. Staining was revealed using DAB substrate (3 min incubation, 5 min in tap H2O). Sections were counterstained with aqueous hematoxylin (1 min incubation, rinsed with H2O), dehydrated (1x 2 min 96% EtOH, 1x 2 min absolute EtOH, 1x 5 min Tissue Clear) and mounted with Eukitt (Sigma, Cat # 25608-33-7).

For the TSA method, sections were blocked with TNB Blocking Buffer (0.1 M Tris-HCl, pH 7.5; 0.15 M NaCl; 0.5% (w/v) blocking reagent (PerkinElmer, Cat # FP1020)) for 30 min at RT. Primary antibodies were diluted in TNB Blocking Buffer as follows: Collagen 1 (Abcam, Cat # ab34710) 1:200, Collagen 6 (Abcam, Cat # ab6588) 1:200, Fibronectin (Sigma, Cat # F3648) 1:100. Primary antibody incubation was performed overnight at 4°C in a humidity chamber. After 10 min wash with TNT (0.1 M Tris-Cl, pH 7.5; 0.15 M NaCl; 0.1% (v/v) Tween 20), sections were incubated with biotinylated secondary antibody (diluted 1:200 in TNB) for 30 min, 1x 10 min TNT. Next, sections were incubated with SA-HRP (PerkinElmer, Cat # NEL750001EA, diluted 1:100 in TNB) for 30 min, 1x 10 min TNT, and with biotinylated tyramide (diluted 1:50 in amplification buffer) for 10 min, 1x 10 min TNT. Sections were incubated again with SA-HRP (diluted 1:100 in TNB) for 30 min, 1x 10 min TNT, and finally washed with PBS. Staining was revealed using AEC substrate (4 min incubation, 1 min in MQ H2O). Sections were counterstained with hematoxylin (1 min incubation, rinsed with H2O) and mounted with Aquatex (Millipore, Cat # 108562).

Immunohistochemistry Quantitation

Qupath v0.1.3, an updated pre-release version of Qupath (https://github.com/petebankhead/qupath) open source software was used for positive pixel counting on four TMAs with IHC stainings for Collagen 1, Collagen 6, Fibronectin and MMP14. TMA slides that contain 22 of the cases were dearrayed and the detected cores were manually adjusted so that the entire tissue sample was included. Tissue detection was performed with the following changes to the default settings: Threshold was set to 220, requested pixel size to 2 µm, minimum area to 15 000 µm2 and maximum fill area to 600 µm2. Subsequently positive pixels were counted with the following deviations from default: downsample factor 2.0, Gaussian sigma 0.5 µm, ‘negative’ Hematoxylin threshold 0.15 OD units and ‘positive’ DAB threshold 0.1 OD units. The results were exported as text files and further analysed with R.

Bioinformatics and statistical analysis

The RNA raw data was normalized in the free Affymetrix Expression Console™ Software provided by ThermoFisher (www.thermofisher.com) using the robust multi- array average (RMA) method (20,21). Subsequent analysis of the gene expression data was carried out in the freely available statistical computing language R.

7

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Unsupervised hierarchical clustering was done using the package “stats”. To test for differentially expressed genes between the identified groups, an empirical Bayes moderated t-test was applied using the ‘limma’ package available from the Bioconductor project (www.bioconductor.org) (22,23). To address the problem with multiple testing, the p-values were adjusted using the method of Benjamini and Hochberg (24). The Database for Annotation, Visualization and Integrated Discovery (DAVID ) v6.8 and the REVIGO (“REduce and VIsualize Gene Ontology”) tool was used to assess the over- and underexpressed genes by organizing them into ontologies and summarizing them by reducing redundant GO terms (25–27). Kaplan- Meier analyses used the R-package “rms”, Cox proportional-hazard tests the R- package “survival”. A t-test was performed using R to test for differences in the percentage of positive pixels, nuclear area and cells per area counts between RNA groups. The values included in the t-test were the mean of the two replicate cores present on the TMA or the mean of the two ROIs for each sample, when two replicates were available.

8

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Results

Patient cohort

Clinicopathologic characteristics of the patient cohort are presented in Table 1. A total of 50 cases of UUS were included, All cases were negative for the JAZF1-JJAZ1 and YWHAE-FAM22 translocations (RT-PCR: 46 cases, FISH: 4 cases). The majority of patients died within 5 years after diagnosis (34, 68%), while the remainder were alive beyond five years (14, 28%). Two patients had a follow up time less than 5 years (12 and 15 months, respectively). The mean follow-up time, including deceased patients, was 3.2 years (range 0.1-18.8). Excluding deceased patients the mean follow up time was 9.3 years (1-18.8). The majority of cases were included from the Karolinska University Hospital. The cases were almost evenly split into prognostic mitotic index groups. There was a predominance of uniform-type UUS. Estrogen and progesterone receptor status was available in the majority of cases.

RNA expression results revealed four distinct molecular groups.

RNA expression analysis was successful on all cases (n=50/50, 100%). Unsupervised clustering of RNA expression data revealed four clusters (Fig 1A), termed “Developmental”, “Leiomyosarcoma (LMS) -like”, “ECM” and “Low Proliferation” (Fig 1A, as indicated in blue, yellow, green and red, respectively). These groupings are seen in the principal component analysis (Fig. 1B, same color scheme).

Examination of the gene expression data focused on the 50 most over- and underexpressed genes within each subgroup, as determined by the fold change. The top 15 over- and underexpressed genes are shown visually in a heat map (Fig 1C). Kaplan-Meier curves (Fig 1D) showed a significant variation in overall survival between the RNA groups, with the ECM group showing the worst prognosis (log-rank test, p=0.0037). The remaining groups shared a comparatively better survival.

DNA copy number variation revealed a spectrum of chromosomal changes, with particular gains and losses associated with each RNA group.

DNA copy number variation analysis was successful in the majority of cases (n=40/50, 80%). This analysis revealed a spectrum of chromosomal copy number variation (CNV), from cases that were diploid or near diploid to cases with extensive chromosomal aberrations. This variation is summarized in Fig 2A, where the percent of the genome showing diploid DNA sections is presented. Cases to the right are diploid / near-diploid, while cases progressing to the left show increasing non-diploid sections. Cases were divided into “Low CNV” (n=15/40, 38%) and “High CNV” (n=25/40, 62%).

9

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Kaplan-Meier curves showed a tendency to decreased overall survival for the High CNV tumors compared to the Low CNV tumors (Fig. 2B). This difference was not significant (log-rank p=0.1743).

Gains and losses of chromosomal segments were analyzed for each RNA group, and this revealed a number of significant differences in chromosomal segment composition. For example, the ECM group showed an increased frequency of 4q and 7q relative gains, 6q relative loss, and LOH near the centromere of 9 (Fig 2C). Chromosomal segment compositions for the other RNA groups are presented in Supplemental Figure 1.

Overall survival depended on clinicopathologic and molecular tumor properties.

A Cox proportional hazard model was constructed, including both clinicopathologic and molecular characteristics (Table 2). The unadjusted (crude) model revealed that mitotic index group and hormone receptor expression showed a significant influence on overall survival. This has been seen previously for these cases (12,13). In the unadjusted (crude) model, RNA group assignment was close to significance, with a p- value of exactly 0.05. The adjusted model showed that these three variables had a significant impact on overall survival, with an explanatory power (r-square) of 0.43. In the adjusted model, the presence of positive hormone receptor expression (either estrogen or progesterone) was strongly protective (HR=0.21), while high mitotic index or ECM-related gene expression signature were indicators of a poor prognosis (HR=2.63 and 2.52, respectively).

Gene ontology and network enrichment analysis revealed biological differences between RNA groups.

The relationship between RNA subgroup and clinicopathologic characteristics is presented in Supplemental Table 1.

The first group contained 21 cases (42%). The GO terms seen in the overexpressed genes, as visualized by REVIGO (Fig 3A), show ontologies related to developmental pathways, particularly the gynecologic tract, positive regulation of gene expression and translation, and genes related to chromatin organization. This REVIGO analysis complements the NEA, which showed a number of pathways related to proliferation, such as peptide chain elongation. The low-expression GO terms included regulation of interferon gamma, leukocyte cell-cell adhesion, and other inflammatory pathways such as nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), leukocyte activation and response to type I interferon. Given the presence of gene ontologies related to mesenchyme and reproductive structure development, as well as embryonic and tube morphogenesis, this group was named the “Developmental” group. This group contained the highest frequency of high mitotic index cases. Most of the cases were uniform-type, few survived past five years, and most showed

10

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

negative hormone receptor staining. They were roughly evenly divided into high and low CNV groups. Of the overexpressed genes in this group, HMGA2 is a candidate biomarker.

The second group contained 10 cases (20%). The GO terms seen in the overexpressed genes in this group were related to muscle function, particularly actin cytoskeleton activation, cardiovascular system development, and circulatory system development (Fig 3B). The NEA similarly showed smooth muscle contraction as the most significant activated pathway. GO enrichment analysis of the 50 underexpressed genes in this group did not reveal any unifying ontologies. The majority of cases showed uniform-type nuclear atypia, but were otherwise roughly evenly divided in terms of survival > 5yr, hormone receptor positivity and CNV group. Given the overexpression of muscle and smooth muscle related gene ontologies this group was assigned as “leiomyosarcoma (LMS)-like”. Of the overexpressed genes in this group, MYH11, ACTG2 and MYLK, all recently described “secondary” smooth muscle markers, are candidate biomarkers.

The third group contained 8 cases (16%). The GO terms seen in the overexpressed genes in this group were related to extracellular matrix disassembly, aminoglycan catabolism and angiogenesis (Fig 3C). GO enrichment analysis of the 50 underexpressed genes in this group did not reveal any unifying ontologies. The NEA showed ECM proteoglycans, ECM-receptor interaction, and degradation of the extracellular matrix as the most significantly overexpressed pathways. All eight patients in this group died within 2 years. The cases were divided between uniform and pleomorphic type nuclear atypia, and close to evenly divided between high- and low-mitotic groups. None of these cases showed positivity with hormone receptors. There was a predominance of tumors with high CNV. There was a trend towards increased incidence of lymphovascular space invasion (LVSI) in this group versus the rest of the cases (6/7 cases, 86%, versus 13/20, 65%, p-value 0.301; note that only cases with all slides available were included in the statistics for LVSI). This group was assigned the name “ECM” group. Of the overexpressed genes in this group, MMP14, COL14A1, COL6A3 and FN1, are candidate biomarkers.

The fourth group contained 11 cases (22%). The GO terms seen in this group were based on reduced expression of genes related to mitotic nuclear division, transcription, and regulation of gene expression (Fig 3D). The NEA revealed only one significant reduced pathway, RNA polymerase III transcription. GO enrichment analysis of the 50 overexpressed genes in this group did not reveal any unifying ontologies. This group showed a correlation with low mitotic index group, with only 2 cases from the high mitotic index group (ANOVA p=0.0458). Most cases showed uniform type atypia. They were roughly evenly divided in survival > 5yrs, expression of hormone receptors and CNV high vs. low. This group was considered to represent a “low proliferation” expression profile. Thus, mitotic index appears to be a candidate marker for identification of this group.

11

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Image analysis revealed that the ECM group is characterized by reduced cell density and increased nuclear size.

In order to more closely examine the phenotypic characteristics of the cases in relation to RNA expression, image analysis of scanned H&E stained tissue sections was performed. This revealed distinct morphologic differences between the intrinsic groups identified by RNA expression analysis (Fig 4A-B). Two regions of interest were examined for each slide and averaged. Note that the image analysis algorithm had trouble reliably identifying the cytoplasmic boundary of each cell, so cell density (as determined from nuclei per unit area) was calculated. A total of 39 scanned slides were available, with each RNA group represented (DEV: 16/21 cases, LMS-LIKE: 8/10 cases, ECM: 6/8 cases and LOW PROLIFERATION: 9/11 cases). The ECM group showed a significantly decreased cell density as well as an increased nuclear area (Fig 4A,B). There was a negative correlation between cell density and deposition of ECM proteins, which was statistically significant for all 3 ECM proteins examined (Collage 1, Collagen 6 and Fibronectin, see Supplemental Table 2). Given this distinct phenotype the possibility of using a cutoff to identify ECM group cases purely from image analysis was investigated. Using a cell density cutoff of £ 4,300 cells per mm2 allowed distinction of ECM tumors from the remaining tumors with a sensitivity of 83% and a specificity of 94%.

Protein expression by immunohistochemistry revealed distinct protein expression profiles across RNA groups.

Immunohistochemistry was used to interrogate the protein expression of four ECM related proteins. The ECM group was particularly selected for evaluation due to it’s extremely poor prognosis. These proteins were selected based on overexpressed genes in the ECM group. The expression of Matrix metalloproteinase 14 (MMP-14), Collagen 1, Collagen 6 and Fibronectin were evaluated. Differences in expression were seen between the RNA groups, with highest expression seen for all these proteins in the ECM group (Fig. 4C-F), with a significant difference to the Developmental group in the Collagen 1, Collagen 6 and MMP-14. The LMS-like group showed the second highest expression, followed by the Developmental and Low Proliferation groups.

12

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Discussion

Undifferentiated uterine sarcomas are aggressive mesenchymal tumors with an unclear biology. These tumors have traditionally been grouped within the endometrial stromal sarcomas, given their lack of smooth muscle differentiation by light microscopy and immunohistochemistry. Due to their rarity, well annotated cohorts allowing evaluation of clinical, pathologic and molecular characteristics have been lacking. This has led to difficulties with developing treatment strategies that can be effective against these tumors. Recent comprehensive genomic studies of soft tissue sarcomas have revealed that, unlike epithelial malignancies, sarcomas are characterized by copy- number alterations, with low mutational burdens and few recurrently mutated genes (3). Transcriptomic diversity within sarcoma types appears to define molecular subtypes related to patient prognosis (3). A recent study using NGS methods identified numerous genomic alterations (including copy-number gains) that were targetable (28).

Previously we have demonstrated the importance of mitotic index and hormone receptor expression in the prognosis of these tumors (12,13). The goal of this study was to further that analysis using molecular methods appropriate to sarcomas. Previous genomic studies have indicated that sarcomas show variations in gene expression and chromosomal gains and losses, but that they typically show few characteristic gene mutations. A further limitation to the study of these cases is their rarity. This requires the use of archived, formalin fixed material with varying RNA and DNA quality.

Gene expression analysis of these tumors reveals four subtypes with clinical significance. In particular, tumors in the ECM group show overexpression of extracellular matrix genes and appear to be particularly deadly, with all eight patients in this group dying within 2 years of diagnosis. The Low Proliferation group correlates well with mitotic index group, thus indicating that proliferation in these tumors, as measured by the identification of mitotic figures in light microscope, has a clear prognostic and biological correlation. The LMS-like group cases that express muscle related genes and are thus possibly variants closely related to leiomyosarcoma. Finally, the Developmental group showed activation of developmental genes and differences in expression of genes related to immune function. These results indicate that these sarcomas may be modulating the host immune response using cytokine mechanisms such as interferons. Our results would indicate that further, more functional, studies of these processes may be fruitful in identifying new therapeutic targets.

Copy number variation arrays showed, surprisingly, tumors that were both near- diploid, with low CNV, and high CNV tumors with extensive gains and losses. Previously, studies have successfully divided sarcomas into translocation associated and karyotypically complex subtypes (6). Given the aggressive and high-grade nature

13

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

of these tumors, our initial hypothesis was that these tumors would all show a complex chromosomal heterogeneity. This was not the case, and roughly half the cases were diploid or near diploid. This is an important finding for future studies and may indicate that there are as yet undiscovered translocations within these high-grade tumors. Furthermore, the CNV division into high and low groups was not prognostic, indicating that even the near diploid tumors can behave aggressively. Several recent studies using array CGH in uterine smooth muscle tumors have indicated that genomic complexity is prognostic in that group of tumors (29,30). Our findings, in contrast, appear to indicate that genetic complexity is not prognostic in UUS, and thus chromosomal complexity as a prognostic marker is subtype-specific. The presence of distinct chromosomal changes that correlate with each RNA subtype is also interesting. Further study will be required to dissect the relationship between chromosomal gains and loss and gene expression.

The survival analysis revealed that, in addition to previously described prognostic markers mitotic index and hormone receptor expression, expression signature-based division to RNA group was also prognostic. This was true even in the adjusted model. Indeed, RNA group showed a hazard ratio close to that seen with mitotic index group. This result indicate that reliable prospective studies of these sarcomas will need to take gene transcription into account, and developing models of treatment and behavior of these tumors will probably require routine use of transcriptomic methods such as RNA sequencing.

Several identified pathways that are potentially targetable have been identified in the molecular analysis presented here. First, targeting ECM mechanisms may lead to active therapies in these tumors. The tumor microenvironment consists of a complex network of blood and lymphatic vessels, immune cells, cancer associated fibroblasts and extracellular matrix components. One of the most highly overexpressed genes in this group was fibronectin. This gene, which expresses a class of high molecular weight adhesive glycoproteins, has been investigated as a potential treatment target. It has a central role in ECM signaling and exists in multiple isoforms. It is possible these isoforms could be individually targeted using, for example, monoclonal antibodies (31–33). Other ECM related genes overexpressed in this group include matrix metalloproteinases, such as MMP2 and MMP14. While drugs have been developed against MMPs, initial trials were unsuccessful. There has, however, been renewed interest in developing new therapies targeting specific MMP activities (34). The overexpression of MMP has been associated in other tumors with aggressive behavior. In UUS, there was a trend to an increased incidence of LVSI in the ECM group.

The Developmental group showed reduced expression of several immune regulatory pathways, particularly nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB). This transcription factor forms protein complexes that bind and regulate target genes via consensus DNA promoter regions. It is typically considered pro-

14

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

oncogenic, stimulating proliferation, preventing apoptosis, regulating tumor angiogenesis and promoting metastasis (35). Reduction in the expression of TNF- alpha, leukocyte adhesion and leukocyte activation indicate a reduction in immunostimulatory molecules, and potentially reflect mechanisms by which this subgroup escapes host immune surveillance (36). Clearly, untangling the immunomodulatory effects of these pathways and how they can best be targeted will require further studies. However, these hypothesis generating results can serve as an indicator that help to assess if immune therapies will be useful in these tumors (37).

The increased expression of muscle related genes in the LMS-like group indicate that these tumors might best be considered a variant of leiomyosarcoma and grouped with them in clinical trials and treatment planning. During the inclusion and exclusion phase of this study the goal was to err on the side of excluding any case that could be conceivable called leiomyosarcoma. Thus, these cases would not be diagnosed as leiomyosarcoma using current methods. The RNA expression results, however, indicate that the tumors in this LMS-like subgroup express several muscle related genes, such as MYH11, ACTG2 and MYLK. The proteins expressed by these genes have recently been used to identify poorly differentiated leiomyosarcomas, and represent a several candidate biomarkers to identify these tumors (38). Finally, the Low Proliferation group appears to confirm a biologic equivalent to mitotic index group, with reduced expression of a number of proliferation related genes. Notably, two cases in the low-proliferation group also had a high mitotic index. This may be due to tumor heterogeneity - the area used for mitotic count was not necessarily the area used for molecular analysis.

Image analysis revealed morphologic differences in the most aggressive, ECM related, sarcoma subgroup. The reduced cell density and increased nuclear size demonstrates a phenotypic correlation to the RNA expression data. There was a negative correlation between ECM protein deposition and cell density, supporting the interpretation that the decrease in cell density is due to increased deposition of ECM proteins. A relationship between myxoid stromal amount and gene expression was recently identified in a comprehensive transcriptomic analysis of myxofibrosarcoma and undifferentiated pleomorphic sarcoma (3). That study indicated that those two sarcomas are not distinct entities, but rather fall along a spectrum, with varying amounts of overexpression of myxoid stromal related genes.

Surprisingly, nuclear size did not correlate with chromosomal copy number variation. Based on the immunohistochemistry results it appears that this decrease in cell density is related to an increase in ECM related proteins. Nuclear size has classically been associated with variations in ploidy, but the nuclear envelope is also affected by biomechanical forces, which in turn result from the tumor cell microenvironment. These forces may explain the increase in nuclear size in the ECM subgroup. It is important to note that, while in epithelial tumors, ECM can be produced by cancer associated fibroblasts, sarcomas are tumors of mesenchymal tissues. Hypothetically,

15

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

the malignant sarcoma cells are themselves producing ECM components because of this mesenchymal origin.

Several potential biomarkers that correlate with the most aggressive, ECM, subgroup have been identified in this work. First, image morphometry using H&E stained sections was able to identify the most aggressive subgroup and may provide a valuable tool to routine histologic examination. Routine H&E staining is readily available in all pathology labs, and morphometry methods are not difficult to set up. This morphometry method may thus allow the ECM related subgroup of UUS to be identified using readily available methods. Immunohistochemistry using ECM related proteins Collagen 1, Collagen 6 and MMP-14 also allowed identification of this subgroup, and further confirm the role of ECM related signaling in certain subgroups of UUS. These proteins warrant further analysis and consideration as potential diagnostic and prognostic biomarkers in UUS.

The findings in this study indicate several potential changes in the diagnosis of these tumors. First, overall survival appears to depend on mitotic index, hormone receptor expression and the ECM subgroup. Both mitotic index and ECM subgroup had an essentially equal contribution in the risk model for decreased overall survival (HR=2.63 vs. 2.52). Expression of hormone receptors was protective (HR=0.21). Thus, the diagnostic algorithm should incorporate these variables. Specifically, patients whose tumors are in the high mitotic index group or ECM group have an extremely high risk of death. Cell density, using a cutoff of 4300 cells per mm2, appears to be a good surrogate for ECM subgroup. Patients without these, and showing positive expression of hormone receptors, have a chance for long term survival. Thus, these diagnostic parameters allow identification of patients with an extremely poor prognosis (typically death within two years of diagnosis) and an improved prognosis (chance for long-term survival is possible).

In summary, this study provides the most detailed examination of an incredibly rare and deadly sarcoma type yet published. The results of this work indicate that routine pathologic parameters, such as mitotic index and hormone receptor immunohistochemistry, can be complemented with RNA expression analysis, to provide biologic and prognostic insights. Furthermore, these results identify several gene pathways that appear to be active in the deadly tumors, and which may be amenable to therapeutic intervention. Finally, image analysis confirms that the RNA groups show morphometric differences, particularly the most aggressive ECM group. Given that image analysis is more easily available than RNA expression analysis, this may provide a valuable diagnostic method for identifying these tumors.

16

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Disclosure of Potential Conflicts of Interest JWC received support from ThermoFisher Scientific for this study.

Funding support: JWC, EH: Radiumhemmets forskningsfonder, Stockholm läns landsting, Cancerfonden, Magnus Bergvalls Stiftelse, Thermo Fisher Scientific; BD: National Sarcoma Foundation at the Norwegian Radium Hospital.

17

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

References

1. IARC, Kurman RJ, Carcangiu ML, Herrington CS, Roung RH. WHO Classification of Tumours of Female Reproductive Organs [Internet]. Kurman RJ, Carcangiu ML, Herrington CS, Young RH, editors. IARC Press; 2014 [cited 2015 Apr 13]. 2. Lee C-H, Nucci MR. Endometrial stromal sarcoma - the new genetic paradigm. Histopathology [Internet]. 2014;1–19. 3. Abeshouse A, Adebamowo C, Adebamowo SN, Akbani R, Akeredolu T, Ally A, et al. Comprehensive and Integrated Genomic Characterization of Adult Soft Tissue Sarcomas. Cell. 2017;171:950–965.e28. 4. Barretina J, Taylor BS, Banerji S, Ramos AH, Lagos-Quintana M, Decarolis PL, et al. Subtype-specific genomic alterations define new targets for soft- tissue sarcoma therapy. Nat Genet [Internet]. Nature Publishing Group; 2010;42:715–21. 5. Guo X, Jo VY, Mills AM, Zhu SX, Lee CH, Espinosa I, et al. Clinically relevant molecular subtypes in leiomyosarcoma. Clin Cancer Res. 2015;21:3501–11. 6. Taylor BS, Barretina J, Maki RG, Antonescu CR, Singer S, Ladanyi M. Advances in sarcoma genomics and new therapeutic targets. Nat Rev Cancer. 2011;11:541–57. 7. Lee C-H, Ou W-B, Marino-Enriquez A, Zhu M, Mayeda M, Wang Y, et al. 14-3- 3 Fusion Oncogenes in High-Grade Endometrial Stromal Sarcoma. Proc Natl Acad Sci. 2012;109:929–34. 8. Croce S, Hostein I, Ribeiro A, Garbay D, Velasco V, Stoeckle E, et al. YWHAE rearrangement identified by FISH and RT-PCR in endometrial stromal sarcomas: genetic and pathological correlations. Mod Pathol [Internet]. 2013;26:1390–400. 9. Koontz JI, Soreng AL, Nucci M, Kuo FC, Pauwels P, Berghe H Van Den, et al. Frequent fusion of the JAZF1 and JJAZ1 genes in endometrial stromal tumors. 2001; 10. Davidson B, Kjæreng ML, Førsund M, Danielsen HE, Kristensen GB, Abeler VM. Progesterone Receptor Expression Is an Independent Prognosticator in FIGO Stage I Uterine Leiomyosarcoma. Am J Clin Pathol [Internet]. 2016;145:449–58. 11. Binder AK, Winuthayanon W, Hewitt SC, Couse JF, Farms T, Operations A, et al. Steroid Receptors in the Uterus and Ovary. Knobil Neill’s Physiol Reprod Fourth Ed. 2015. page 1099–193. 12. Gremel G, Liew M, Hamzei F, Hardell E, Selling J, Ghaderi M, et al. A prognosis based classification of undifferentiated uterine sarcomas: Identification of mitotic index, hormone receptors and YWHAE-FAM22 translocation status as predictors of survival. Int J Cancer [Internet]. 2015 [cited 2015 Feb 1];136:1608–18. 13. Hardell E, Josefson S, Ghaderi M, Skeie-Jensen T, Westbom-Fremer S, Cheek EH, et al. Validation of a Mitotic Index Cutoff as a Prognostic Marker in Undifferentiated Uterine Sarcomas. Am J Surg Pathol [Internet]. 2017;41:1. 14. Kurihara S, Oda Y, Ohishi Y, Kaneki E, Kobayashi H, Wake N, et al. Coincident expression of beta-catenin and cyclin D1 in endometrial stromal tumors and related high-grade sarcomas. Mod Pathol [Internet]. Nature Publishing Group; 2010 [cited 2015 Jan 2];23:225–34. 15. Kurihara S, Oda Y, Ohishi Y, Iwasa A, Takahira T, Kaneki E, et al. Endometrial

18

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

stromal sarcomas and related high-grade sarcomas: immunohistochemical and molecular genetic study of 31 cases. Am J Surg Pathol. 2008;32:1228–38. 16. El-khalfaoui K, Bois A, Heitz F, Kurzeder C, Sehouli J. Current and future options in the management and treatment of uterine sarcoma. 2014;21–8. 17. Kim H, Kim Y, Lee SJ, Lee J, Park SH, Kim H. Pazopanib monotherapy in the treatment of pretreated , metastatic uterine sarcoma : a single-center retrospective study. 2018;29:1–8. 18. Rasmussen M, Sundström M, Kultima H, Botling J, Micke P, Birgisson H, et al. Allele-specific copy number analysis of tumor samples with aneuploidy and tumor heterogeneity. Genome Biol [Internet]. BioMed Central Ltd; 2011;12:R108. 19. Bankhead P, Loughrey MB, Fernández JA, Dombrowski Y, McArt DG, Dunne PD, et al. QuPath: Open source software for digital pathology image analysis. Sci Rep. 2017;7:1–7. 20. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics [Internet]. 2003;4:249–64. 21. Li C, Wong WH. Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci U S A [Internet]. 2001;98:31–6. 22. Smyth GK. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol [Internet]. 2004;3:Article3. 23. Smyth GK. Limma: linear models for microarray data. In: Gentleman R, Carey V, Dudoit S, Irizarry S, Huber W, editors. Bioinforma Comput Biol Solut using R Bioconductor. New York: Springer; 2005. page 397–420. 24. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing [Internet]. J. R. Stat. Soc. B. 1995. page 289–300. 25. Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009;4:44– 57. 26. Huang DW, Sherman BT, Lempicki RA. Bioinformatics enrichment tools: Paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res. 2009;37:1–13. 27. Supek F, Bošnjak M, Škunca N, Šmuc T. Revigo summarizes and visualizes long lists of gene ontology terms. PLoS One. 2011;6. 28. Lucchesi C, Khalifa E, Laizet Y, Soubeyran I. Targetable Alterations in Adult Patients With Soft-Tissue Sarcomas Insights for Personalized Therapy. JAMA Oncol. 2018;4:1398–404. 29. Croce S, Ribeiro A, Brulard C, Noel J-C, Amant F, Stoeckle E, et al. Uterine smooth muscle tumor analysis by comparative genomic hybridization: a useful diagnostic tool in challenging lesions. Mod Pathol [Internet]. 2015;28:1001–10. 30. Croce S, Ducoulombier A, Ribeiro A, Lesluyes T, Noel J, Amant F, et al. Genome profiling is an efficient tool to avoid the STUMP classification of uterine smooth muscle lesions: a comprehensive array-genomic hybridization analysis of 77 tumors. Mod Pathol [Internet]. Nature Publishing Group; 2018;1–13. 31. Venning FA, Wullkopf L, Erler JT. Targeting ECM Disrupts Cancer Progression. Front Oncol [Internet]. 2015;5:1–15.

19

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

32. Bhome R, Al Saihati HA, Goh RW, Bullock MD, Primrose JN, Thomas GJ, et al. Translational aspects in targeting the stromal tumour microenvironment: From bench to bedside. New Horizons Transl Med [Internet]. Elsevier; 2016;3:9–21. 33. Kaspar M, Zardi L, Neri D. Fibronectin as target for tumor therapy. Int J Cancer. 2006;118:1331–9. 34. Vandenbroucke RE, Libert C. Is there new hope for therapeutic matrix metalloproteinase inhibition? Nat Rev Drug Discov [Internet]. Nature Publishing Group; 2014;13:904–27. 35. Xia Y, Shen S, Verma IM. NF-κB, an active player in human cancers. Cancer Immunol Res [Internet]. 2014;2:823–30. 36. van Horssen R, Ten Hagen TLM, Eggermont AMM. TNF-alpha in cancer treatment: Molecular insights, antitumor effects, and clinical utility. Oncologist [Internet]. 2006;11:397–408. 37. Galluzzi L, Vacchelli E, Pedro JB, Buqué A, Senovilla L, Baracco EE, et al. Classification of current anticancer immunotherapies. 5. 38. Demicco EG, Boland GM, Savannah KJB, Lusby K, Eric D, Ingram D, et al. Progressive loss of myogenic differentiation in leiomyosarcoma has prognostic value. Histopathology. 2015;627–38.

20

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Figure Legends

Figure 1. RNA expression analysis reveals distinct subgroups. Developmental (blue), LMS- like (yellow), ECM (green), and Low Proliferation (red). A, Unsupervised hierarchical clustering of RNA expression reveals 4 distinct groups. The squares adjacent to the case abbreviation indicate survival status (yellow: alive at 5 years, red: death prior to 5 years, black: follow up less than 5 years). B, Principal Component Analysis also demonstrates the presence of distinct expression based subgroups. C, Heat map showing the 15 most over- and underexpressed genes in each group. D, Kaplan-Meier curve showing overall survival for each RNA group.

Figure 2. DNA copy number variation analysis reveals a spectrum of chromosomal changes. A, Diploid length curve, showing the percentage of DNA segments that were diploid as a percent of the entire genome length, demonstrates that tumors had a distribution of copy number variation. B, Kaplan-Meier curve showing overall survival for each DNA group. C, Chromosomal segments showing gains (upper diagram), losses (middle diagram) and LOH (lower diagram) across the entire genome for the ECM group. These diagrams show the fraction of cases in the ECM group with segment change (gain, loss or LOH) as the positive y-axis in percent, and the fraction of cases with the same change that are not in the group as the negative y-axis. The difference between these two (i.e. the change is present in ECM group vs remainder) is then shown in the darker color. Significant differences at the p<0.05 level between ECM vs. remainder are shown in the lighter shaded regions that extend from - 100% to +100%.

Figure 3. RNA ontology, as visualized using REVIGO, for each RNA group. A, Overexpressed genes in the Developmental group. B, Overexpressed genes in the LMS-like group. C, Overexpressed genes in the ECM group. D, Underexpressed genes in the Low Proliferation group.

Figure 4. Image analysis results of morphometry (n=39) and immunohistochemistry (n=22) reveal distinct differences, particularly between the ECM Developmental groups. Boxplots showing differences in cell density (A) nuclear area (B), and immunohistochemistry for MMP- 14 (C), Collagen 1 (D), Collagen 6 (E) and Fibronectin (F) for each RNA group. Cell density is reported in cells per mm2 X 1000. Nuclear area is given in micrometer squared X 1000. Immunohistochemistry is % of the tissue that is positive. Significance test between each group is indicated by a red line with either * (p<0.05) or ** (p<0.01).

21

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Table 1. Clinicopathological characteristics of the included cases of undifferentiated Uterine Sarcoma.

Total number of cases 50

Time to last follow up - all patients (yrs) 3.2 ± 4.6 (0.1 - 18.8)

Time to last follow up - survivors (yrs) 9.3 ± 5.5 (1 - 18.8)

Long term survivors 14 (28%) (Alive minimum 5 yrs)

Deceased prior to 5 yr follow up 34 (68%)

Alive, follow up less than 5 yrs 2 (4%)

Cases included from:

Karolinska University Hospital 17 (34%)

Oslo University Hospital 13 (26%)

Skåne University Hospital 10 (20%)

Mayo Clinic 5 (10%)

Vancouver General Hospital 4 (8%)

Brigham and Women’s Hospital 1 (2%)

Mitotic index group (13):

High (>11.16 mitotic figures/mm2) 23 (46%)

Low (<11.16 mitotic figures/mm2) 27 (54%)

Nuclear atypia (15):

22

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Uniform 33 (66%)

Pleomorphic 13 (26%)

Unknown 4 (8%)

Estrogen and progesterone receptor status:

Positive 11 (22%)

Negative 28 (56%)

Unknown 11 (22%)

23

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Table 2. Cox proportional regression analysis for overall survival in relation to clinical and molecular features.

Crude results Adjusted results (single explanatory (multiple explanatory variable) variables) r-square=0.43

OS OS

Variable No. of HR p value HR p value patients

Mitotic Index

Low 27 Reference Reference

High 23 2.33 0.01** 2.63 0.01** (1.21-4.50) (1.27-5.45)

Hormone receptor expression

Negative 28 Reference Reference

Positive 11 0.17 <0.01** 0.21 .001** (0.06-0.50) (0.07-0.65)

N/A 11 0.70 0.37 0.7 0.4 (0.32-1.52) 0.31-1.61)

RNA group

Developmental 21 Reference Reference

LMS-like 10 0.46 0.09 0.48 0.14 (0.19-1.14) (0.18-1.27)

ECM 8 2.39 0.05 2.52 0.045* (1.00-5.72) (1.02-6.24)

Low 11 0.52 0.13 0.98 0.96 Proliferation 0.22-1.22) (0.39-2.39)

Nuclear atypia

Uniform 33 Reference

Pleomorphic 13 1.63 0.17

24

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

(0.81-3.26)

N/A 4 1.34 0.69 (0.31-5.78)

Cell density

Low 22 Reference

High 15 1.26 (0.62- 0.53 2.57)

N/A 13 0.81 (0.35- 0.63 1.89)

CNV group

High 25 Reference

Low 15 0.67 0.29 (0.32-1.41)

N/A 10 0.46 0.08 (0.19-1.09)

25

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Figure 1

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Figure 2

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Figure 3

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Figure 4

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on January 7, 2019; DOI: 10.1158/1078-0432.CCR-18-2792 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Integrated Molecular Analysis of Undifferentiated Uterine Sarcomas Reveals Clinically Relevant Molecular Subtypes

Amrei Binzer-Panchal, Elin Hardell, Björn Viklund, et al.

Clin Cancer Res Published OnlineFirst January 7, 2019.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-18-2792

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2019/01/07/1078-0432.CCR-18-2792.DC1

Author Author manuscripts have been peer reviewed and accepted for publication but have not yet Manuscript been edited.

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

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2019/01/07/1078-0432.CCR-18-2792. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.

Downloaded from clincancerres.aacrjournals.org on September 25, 2021. © 2019 American Association for Cancer Research.