18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License A DNA damage repair gene-associated signature predicts responses of patients with advanced soft-tissue sarcoma to treatment with trabectedin David S. Moura1, Maria Peña-Chilet1,2,3, Juan Antonio Cordero Varela1, Ramiro Alvarez-Alegret4, Carolina Agra-Pujol5, Francisco Izquierdo6, Rafael Ramos7, Luis Ortega-Medina8, Francisco Martin- Davila9, Carolina Castilla-Ramirez10, Carmen Nieves Hernandez-Leon11, Cleofe Romagosa12, Maria Angeles Vaz Salgado13, Javier Lavernia14, Silvia Bagué15, Empar Mayodormo-Aranda16, Luis Vicioso17, Jose Emilio Hernández Barceló18, Jordi Rubio-Casadevall19, Ana de Juan20, Maria Concepcion Fiaño-Valverde21, Nadia Hindi1,22,23,24, Maria Lopez-Alvarez1, Serena Lacerenza1, Joaquin Dopazo1,2,3,25, Antonio Gutierrez26, Rosa Alvarez27 , Claudia Valverde28, Javier Martinez- Trufero29 and Javier Martı́n-Broto1,22,23,24 1 Institute of Biomedicine of Seville (IBIS, HUVR, CSIC, Universidad de Sevilla), Spain 2 Clinical Bioinformatics Area, Fundación Progreso y Salud (FPS), CDCA, Hospital Virgen del Rocio, Seville, Spain 3 Bioinformatics in Rare Diseases (BiER), Centro de Investigación Biomédica en Red de Enfermedades Raras (CIBERER), FPS, Hospital Virgen del Rocio, Seville, Spain 4 Pathology Department, Miguel Servet University Hospital, Zaragoza, Spain 5 Pathology Department, Gregorio Marañon University Hospital, Madrid, Spain 6 Pathological Anatomy Service, Complejo Asistencial Universitario de León, Spain 7 Pathology Department, Son Espases University Hospital, Mallorca, Spain 8 Pathology Department, Hospital Clinico San Carlos, Madrid, Spain 9 Pathology Department, Ciudad Real General Hospital, Spain 10 Pathology Department, University Hospital Virgen del Rocio, Seville, Spain 11 Pathology Department, Canarias University Hospital, Santa Cruz de Tenerife, Spain 12 Pathology Department, Vall d’Hebron University Hospital, Barcelona, Spain 13 Medical Oncology Department, Ramon y Cajal University Hospital, Madrid, Spain 14 Medical Oncology Department, Instituto Valenciano de Oncologia, Valencia, Spain 15 Pathology Service, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain 16 Pathology Department, Hospital Universitari i Politècnic la Fe, Valencia, Spain 17 Pathology Department, Virgen de la Victoria University Hospital, Malaga, Spain 18 Pathology Department, Virgen de la Arrixaca University Hospital, Murcia, Spain 19 Medical Oncology Department, Hospital Josep Trueta, Catalan Institute of Oncology, Girona, Spain 20 Medical Oncology Department, Marqués de Valdecilla University Hospital, Santander, Spain 21 Department of Histopathology, Álvaro Cunqueiro Hospital, University Hospital Complex of Vigo, Spain 22 Medical Oncology Department, University Hospital Fundación Jimenez Diaz, Madrid, Spain 23 University Hospital General de Villalba, Madrid, Spain 24 Instituto de Investigacion Sanitaria Fundacion Jimenez Diaz (IIS/FJD), Madrid, Spain 25 INB-ELIXIR-es, FPS, Hospital Virgen del Rocı´o, Seville, Spain 26 Hematology Department, Son Espases University Hospital, Mallorca, Spain 27 Medical Oncology Department, Gregorio Marañon University Hospital, Madrid, Spain 28 Medical Oncology Department, Vall d’Hebron University Hospital, Barcelona, Spain 29 Medical Oncology Department, Miguel Servet University Hospital, Zaragoza, Spain Abbreviations AP, apurinic/apyrimidinic; AUC, area under the ROC curve; CR, complete response; DDR, DNA damage repair; DSB, double-strand break; FC, fold change; FDR, false discovery rate; FFPE, formalin-fixed paraffin-embedded; GMI, growth modulation index; HR, homologous recombination; NER, nucleotide excision repair; NHEJ, non-homologous end joining; OS, overall survival; PD, progression disease; PFS, progression-free survival; PR, partial response; SD, stable disease; STS, soft-tissue sarcoma; TMM, trimmed mean of M-values; TPM, transcripts per million; TTP, time-to-progression. Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3691 Federation of European Biochemical Societies. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. Keywords Predictive biomarkers of trabectedin represent an unmet need in advanced gene signature; predictive biomarkers; soft-tissue sarcomas (STS). DNA damage repair (DDR) genes, involved in trabectedin homologous recombination or nucleotide excision repair, had been previ- ously described as biomarkers of trabectedin resistance or sensitivity, Correspondence J. Martı´n-Broto, Instituto de Biomedicina de respectively. The majority of these studies only focused on specific factors Sevilla (IBiS), LAB 214, C/ Antonio Maura (ERCC1, ERCC5, and BRCA1) and did not evaluate several other DDR- Montaner s/n, Sevilla 41013, Spain related genes that could have a relevant role for trabectedin efficacy. In this Tel: +34 955923113 retrospective translational study, 118 genes involved in DDR were evalu- E-mail: jmartin@mustbesevilla.org ated to determine, by transcriptomics, a predictive gene signature of tra- bectedin efficacy. A six-gene predictive signature of trabectedin efficacy (Received 5 February 2021, revised 13 April was built in a series of 139 tumor samples from patients with advanced 2021, accepted 10 May 2021, available online 30 June 2021) STS. Patients in the high-risk gene signature group showed a significantly worse progression-free survival compared with patients in the low-risk doi:10.1002/1878-0261.12996 group (2.1 vs 6.0 months, respectively). Differential gene expression analy- sis defined new potential predictive biomarkers of trabectedin sensitivity (PARP3 and CCNH) or resistance (DNAJB11 and PARP1). Our study identified a new gene signature that significantly predicts patients with higher probability to respond to treatment with trabectedin. Targeting some genes of this signature emerges as a potential strategy to enhance tra- bectedin efficacy. oncogenic gene transcription, which has been described 1. Introduction as its main mechanism of action. Besides, the Sarcomas are a heterogeneous group of rare malignant transcription-coupled nucleotide excision repair (NER) tumors with mesenchymal origin that can affect soft, DNA repair machinery, in charge of repairing bulky bone, and visceral tissue. Sarcoma incidence is around adducts in DNA, is tricked by the special conforma- 50–60 new cases per year for every million people, tion of trabectedin adducts, ultimately inducing 20% being metastatic at diagnosis. Around 30–40% of double-strand breaks (DSBs) in the DNA. This pro- localized sarcomas will eventually develop metastasis. cess generates cell cycle perturbations, with a delayed In other words, a half of sarcoma population will suf- S phase progression and accumulation of cells in G2 fer, sooner or later, metastatic spread. For locally phase [6]. Additionally, trabectedin seems to affect also advanced or metastatic soft-tissue sarcoma (STS), the transcription of cytokines and growth factors in therapeutic options are scarce and the Response Eval- tumor-associated macrophages and circulating mono- uation Criteria in Solid Tumors (RECIST) responses cytes [7,8], and it sensitizes cancer cells to FAS recep- to chemotherapy are infrequent, while the median of tor programmed cell death [9]. It has been also survival ranges between 18 and 22 months [1]. While suggested that trabectedin could be a good radiosensi- doxorubicin is the backbone of first-line therapy for tizer, since it causes G2/M arrest, which is the most advanced disease, second-line treatments include pazo- sensitive cell cycle phase to radiation [10]. The high panib and chemotherapeutic agents such as trabecte- efficacy of this combination was observed in phase I/II din, eribulin, and gemcitabine combinations. The trials in patients with metastatic STS of diverse his- median progression-free survival (PFS) in first-line tologies and patients with localized myxoid liposar- therapy for STS ranges between 4.5 and 6.0 months, coma [11,12]. whereas the median PFS for most of the second-line Trabectedin treatment triggers DNA damage repair drugs is below 5 months. Around 50% of the patients (DDR). Therefore, DDR-related factors might be obtain a clinical benefit from these drugs [2,3]. potential predictive biomarkers for this drug. Few ret- Trabectedin is an antitumor drug approved for the rospective studies supported this observation and treatment of adult patients with advanced STS after reported that overexpression of genes involved in NER failure of anthracyclines and ifosfamide, or for whom machinery (e.g., ERCC1 and ERCC5) behaved as pre- are unsuited to receive these agents [4,5]. This com- dictive biomarkers for trabectedin efficacy, whereas the pound is a DNA-binding agent that interferes with overexpression of homologous recombination (HR)- 3692 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy associated factors (e.g., BRCA1) was associated with assay had to have at least 70% of tumor area; samples trabectedin resistance [13–15]. In line with this, the with less than 70% tumor or with more than 20% of potential predictive value of ERCC1 and ERCC5 was necrotic tissue underwent macro-dissection. All the also validated by our group, in a cohort of cases samples and controls were quantified in triplicate, and prospectively collected from the randomized phase II no template control was included in each run. trial that compared trabectedin plus doxorubicin ver- The RNA-Seq libraries were synthesized through sus doxorubicin alone as first line of advanced STS HTG EdgeSeq chemistry. In summary, this method [16]. Nevertheless, the majority of these studies only consists of a first step that lysed and permeabilized the focused on specific DDR-related proteins/genes, such sample, which allows the release of the RNA. Then, as ERCC1, ERCC5, and BRCA1, and did not analyze nuclease protection probes were added to the lysed several other DDR-related genes that could also have samples to hybridize with the exposed mRNA and predictive relevance for trabectedin efficacy. target-captured the transcripts. The S1 nuclease was Accordingly, the aim of this retrospective transla- added to the mix, allowing the production of a stoi- tional study was to evaluate the expression of 118 chiometric amount of target mRNA/nuclease protec- genes involved in DDR, as potential predictive tion probes duplexes. The reaction was blocked by biomarkers of trabectedin activity and to determine denaturing S1 enzyme by heat. In order to reduce whether any of these genes or a gene signature could potential biases in the run, the samples were random- predict trabectedin efficacy, in a more robust way than ized before the inclusion in the HTG EdgeSeq system. the already described expression of certain limited PCRs were set up using each hybridized sample as genes (i.e., ERCC1, ERCC5, and BRCA1). New template and specially designed tags that shared com- biomarkers or novel insights on the mechanisms mon sequences complementary to 50 -end and 30 - underlying drug activity may allow the development of sequences of the probes, as well as with common new therapeutic options for STS and a better selection adaptors that are required for cluster generation on an of patients for trabectedin treatment. Illumina sequencing platform. Additionally, each tag contains a unique barcode useful for sample identifica- tion and multiplexing. Agencourt AMPure XP (Beck- 2. Methods man Coulter, Beverly, MA, USA) was used to clean the PCR amplification product. Moreover, the libraries were quantified by quantita- 2.1. Patients and samples tive PCR, using the KAPA Library Quantification A series of 140 formalin-fixed paraffin-embedded (Roche, Basel, Switzerland) kit, according to the man- (FFPE) tumor samples, from STS patients treated with ufacturer’s instructions. All samples and controls were trabectedin in real-life setting at any line of advanced quantified in triplicate, and no template control was disease, was selected from a Spanish Group for included in every run. Library denaturation was per- Research on Sarcomas (GEIS) registry and was the formed by adding first 2 N NaOH to the library, fol- basis of this translational research. All the samples lowed by the addition of 2 N HCl. The PhiX was were collected with the informed consent signed from spiked in at a 5% (concentration of 12.5 pM). the patients, according to national regulations, and the One demultiplexed FASTQ file per sample was study protocol was approved by the Ethics Commit- retrieved from the sequencer for data processing. The tees and the institutional review board of each partici- HTG EdgeSeq host software performed the alignment pant center. Study procedures were performed in of the FASTQ files with the probe list, the results were accordance with the Declaration of Helsinki. parsed, and the output was obtained as a read counts matrix. The HTG EdgeSeq was run in the VERIP ser- vice laboratory of HTG in Tucson (HTG Molecular 2.2. Oncology Biomarker Panel Diagnostics, Tucson, AZ, USA). The HTG EdgeSeq Oncology Biomarker Panel (OBP) direct-transcriptomic assay quantitatively measures the 2.3. Data filtering and normalization expression of 2549 human RNA transcripts (https:// www.htgmolecular.com/assays/obp) associated with Negative control probes were used as quality control tumor biology. Among these RNA transcripts, 118 are to evaluate baseline performance, as previously associated with DDR mechanisms and only these described by Chadly et al. [17]. Briefly, the mean of genes were considered for the purpose of this study. negative probes for each sample was calculated, and For direct transcriptomics, all the samples used in the the difference between negative control average and Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3693 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. the mean of all negative control probes was obtained All analyses were performed with R/BIOCONDUCTOR (Δmean). Those samples with a Δmean outside the (3.10) running on R version 3.6.0, (https://www.r-pro bounds of 2 standard deviation were excluded from ject.org/). the study. Of 140 sample initially included in the study, seven were removed from the analysis. Data 2.5. Statistical analysis normalization was performed by applying trimmed mean of M-values method (TMM) using EDGER pack- The variables that followed binomial distributions age from R/BIOCONDUCTOR, adjusting for the total reads (e.g., patient demographics) were expressed as frequen- within a sample [18]. cies and percentages. Categorical variables were expressed as absolute and relative frequencies or as continuous variables as median, range (minimum–max- 2.4. Differential gene expression analysis imum). Gene expression levels were indicated as For differential gene expression analysis, samples were mean  standard deviation. The comparisons between grouped according to PFS, growth modulation index quantitative and qualitative variables were performed (GMI), or clinical benefit/response of trabectedin. PFS through U of Mann–Whitney nonparametric test. was measured from the date of initial treatment with FDR was used to adjust for multiple comparisons. trabectedin (PFS) to the event of progression or death, PFS and overall survival (OS) from trabectedin initia- whichever occurred first. Clinical benefit was defined tion were estimated according to the Kaplan–Meier as the percentage of patients achieving complete method. The log-rank test was used to determine the response (CR) or partial response (PR) and/or stable associations between the variables of interest (i.e., gene disease (SD). Response was evaluated according to expression and clinical outcomes). The P-values RECIST v.1.1. GMI was expressed as a ratio of intra- reported were 2-sided, and the statistical significance patient successive time-to-progression (TTP): GMI = was defined at P ≤ 0.05. Statistical analysis was per- TTP under trabectedin/TTP for the treatment prior to formed with SPSS 22.0 software (IBM, Armonk, NY, trabectedin. USA). For PFS analysis, the low-risk group (better progno- sis) was made up of the cases with a PFS value higher 2.6. Construction of a predictive signature based than the median (3.2 months) and the high-risk group on gene expression (worse prognosis) comprised of the samples with PFS lower than the median. For GMI analysis, the low-risk Raw counts were normalized to transcripts per million group comprised of samples with a GMI ≥ 1.33 and (TPM), as it is a more appropriate unit to compare the high-risk group the cases with a GMI < 1.33. For expression across samples than normalized counts RECIST response or clinical benefit analysis, samples from other methods such as DESEQ2 and EDGER. Of 140 were grouped as both: cases with response versus cases cases included, only one patient with PFS = 0 was with stable or progressive disease or cases with removed from the predictive gene signature analysis. response or stable disease versus cases with progres- Univariate Cox regression analysis using the survival sion disease (PD). package was performed to find individual genes whose The differential gene expression between groups expression was significantly associated (P < 0.05) with was evaluated by a negative binomial generalized log- PFS. Additionally, patients were divided, for each linear model, using the EDGER package and imple- gene, into two groups, using an optimal cutoff from menting the procedure proposed by Robinson and the survminer package which maximized the separa- Smyth [19]. The correction for multiple comparisons tion of Kaplan–Meier curves of every group, and only of Benjamini and Hochberg was applied, and a P- those genes with significantly different curves value threshold of 0.05 was set [20]. Fold change (P < 0.05) were further retained. (FC) values were obtained, along with P-values and The remaining genes were used as input to build a adjusted P-values (false discovery rate; FDR) for the gene expression signature using a multivariate Cox 118 genes evaluated. The batch effect was taken into regression applying a Lasso penalty to minimize the account by adding batch information to the model as risk of overfitting with the glmnet package. The pen- a covariable. For data visualization and later analy- alty parameter was calculated using a fourfold cross- ses, normalized log-TMM values were obtained and validation (at the minimum partial likelihood the variability due to batch effect was removed using deviance) of the complete cohort, randomly dividing the removeBatchEffect method implemented in LIMMA 75% and 25% of patients into training and testing R package. sets, respectively, 10 000 times. 3694 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy Risk scores were calculated by multiplying the diagnosed with other sarcoma subtypes. Patient demo- expression of every gene with its corresponding Cox graphics are depicted in Table 1. regression coefficient, for each patient: Risk score ¼ ∑Coxregression coefficient of Gene i 3.2. Trabectedin outcome and univariate analysis expressionof geneGene i: With a median follow-up of 45 months, the median All analyses were performed in R (version 4.0.3). PFS of trabectedin line was 3.5 months, and the med- ian of OS measured from the trabectedin initiation was 12 months. There were 124 out of 140 patients 3. Results evaluable for response by RECIST with the following distribution: 6 (4.8%) CR, 14 (11.3%) PR, 42 (33.9%) SD, and 62 (50.0%) PD as best response. The GMI 3.1. Patient demographics was ≥ 1.33 in 40 (29%) cases, < 1.33 in 97 (69%) A subset of 140 STS patients treated with trabectedin patients, whereas in 3 (2%) cases, this value was not at line of advance disease with FFPE tumor sample calculable. available for RNA expression analysis was included in In the univariate analysis, grade 3 tumors were asso- this translational study. The median age was 51 (17– ciated with a worse PFS [3.0 months (95% CI 2.2–3.8) 79) years with 54% (n = 75) being female. Among the vs 6.5 months (95% CI 2.0–11.0); P = 0.003] and 140 STS included, 81% (n = 114) were localized at worse OS [10.2 months (95% CI 5.7–14.7) vs diagnosis, with 66% (n = 93) and 34% (n = 47) having 17.5 months (95% CI 13.9–21.1); P = 0.041], and diag- a somatic or visceral primary tumor location, respec- nosis of L-sarcoma subtypes was correlated with a bet- tively. Of 140 cases, 59% (n = 83) were grade 3, 24% ter PFS [6.1 months (95% CI 3.6–8.5) vs 3.0 months (n = 33) were grade 2, and 13% (n = 18) were grade 1. (95% CI 2.0–4.0); P = 0.001] and better OS The most frequent subtypes were leiomyosarcoma [18.2 months (95% CI 13.2–23.1) vs 7.2 months (95% (n = 44; 31%), liposarcoma (n = 32; 23%), synovial CI 3.7–10.8); P = 0.001]. Somatic location was associ- sarcoma (n = 14; 10%), and undifferentiated pleomor- ated with a better PFS in our series [5.6 months (95% phic sarcoma (n = 14; 10%), while 26% (n = 36) were CI 3.2–7.9) vs 2.7 months (95% CI 2.1–3.2); P = 0.006]. Patients receiving trabectedin in first or second line had better PFS [6.3 months (95% CI 1.0– Table 1. Patient demographics. 11.5) vs 3.2 months (95% CI: 2.4–4.0); P = 0.001] (Table 2). Visceral location [HR = 1.9 (95% CI 1.3– N (%) 2.8), P = 0.002] and non-L-sarcoma [HR = 1.9 (95% Gender CI 1.3–2.9), P = 0.001] were independent factors asso- Male 65 (46) ciated with a worse PFS of trabectedin in the multi- Female 75 (54) variate analysis. Stage at diagnosis Localized 114 (81) Metastatic 26 (19) 3.3. Expression of DNA damage repair genes in Sarcoma subtype STS Leiomyosarcoma 44 (31) Liposarcoma 32 (23) The expression of DDR-related genes was heteroge- Synovial sarcoma 14 (10) neous, being DNAJB8 (log2 2.80  1.45) and APEX1 Undifferentiated pleomorphic sarcoma 14 (10) (log2 9.97  0.66) the most underexpressed and over- Other 36 (26) expressed genes, respectively, among the transcripts Grade analyzed (Table S1). 1 18 (13) 2 33 (24) Considering the expression of DDR-related tran- 3 83 (59) scripts in relation to the clinical variables grouped as: Not available 6 (4) L-sarcomas vs non-L-sarcomas and grade 3 vs grade Location 1–2, the outcome was as follows: of 118 genes ana- Somatic 93 (66) lyzed, only two genes were significantly overexpressed Visceral 47 (34) (i.e., BRIP1, DNAJB5) in L-sarcomas, compared with Median follow-up from diagnostic (months) 45 non-L-sarcomas, whereas 16 genes were significantly Median follow-up from trabectedin line (months) 12 Median age, years (range) 51 (17–79) underexpressed (i.e., DDB2, DNAJA1, DNAJA2, DNAJB2, DNAJC10, DNAJC14, DNAJC16, ERCC3, Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3695 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. Table 2. Univariate analysis of clinical factors. Table 3. Differential gene expression attending to median PFS (3.2 months). A negative fold change means that the gene is Factor PFS (95% CI) P OS (95% CI) P overexpressed in cases with PFS below the median. Sex 0.276 0.613 Gene logFC P-value FDR Female 3.2 (2.5–4.0) 13.1 (8.0–18.2) PARP3 1.738 < 0.001 < 0.001 Male 4.4 (2.5–6.3) 13.1 (4.8–21.5) POLL 1.767 < 0.001 0.002 Age 0.178 0.674 < 51 5.1 (2.2–8.0) 12.5 (5.8–19.2) DNAJB11 – 0.378 < 0.001 0.003 > 51 3.0 (2.1–3.9) 13.1 (6.2–20.1) PMS1 0.764 < 0.001 0.003 Subtype 0.001 0.001 RAD52 0.875 < 0.001 0.003 L-sarcoma 6.1 (3.6–8.5) 18.2 (13.2–23.1) PARP1 – 0.442 < 0.001 0.007 Non-L-sarcoma 3.0 (2.0–4.0) 7.2 (3.7–10.8) Grade 0.003 0.041 ATM 0.444 0.001 0.014 1 and 2 6.5 (2.0–11.0) 17.5 (13.9–21.1) NEIL1 0.773 0.001 0.021 3 3.0 (2.2–3.8) 10.2 (5.7–14.7) TDG – 0.312 0.003 0.030 Location 0.006 0.051 TOP3A – 0.451 0.003 0.030 Somatic 5.6 (3.2–7.9) 17.0 (10.2–23.9) Visceral 2.7 (2.1–3.2) 11.3 (7.6–15.0) TOP3B 1.374 0.003 0.030 Stage at 0.057 0.523 CCNH 0.620 0.003 0.030 diagnosis XRCC5 0.862 0.003 0.030 Localized 4.4 (2.1–6.7) 13.8 (8.6–19.0) Metastatic 3.0 (2.2–3.8) 5.4 (0.0–16.7) PRKDC – 0.392 0.004 0.030 Trabectedin line 0.001 0.218 DNAJC11 – 0.309 0.004 0.030 1/2 6.3 (1.0–11.5) 17.0 (9.5–24.6) DDB2 0.314 0.004 0.031 >2 3.2 (2.4–4.0) 12.0 (5.8–18.1) MUTYH 0.369 0.005 0.034 RAD23B – 0.311 0.006 0.035 DNAJB14 0.420 0.006 0.035 ERCC6, MMS19, MSH2, PMS2, SMUG1, TDG, and XPC 0.512 0.006 0.037 XRCC5) in L-sarcomas, compared with other STS DNAJC21 – 0.420 0.009 0.050 subtypes. Grade 3 tumors showed a significant overex- NEIL2 0.420 0.012 0.066 pression, compared with grade 1–2 cases, of APEX2, EXO1 – 0.522 0.017 0.087 BRIP1, DNAJC1, DNAJC11, EXO1, FEN1, MSH2, DNAJB8 0.700 0.022 0.100 MSH6, NEIL2, NEIL3, PRKDC, RAD21, RAD51, RAD54L, RPA3, TREX1, and XRCC3, whereas ATM PMS2 0.326 0.023 0.100 and XPC were overexpressed in this latter group, com- DNAJC16 0.541 0.024 0.100 pared with grade 3 tumors (Table S2). MSH6 – 0.230 0.024 0.100 ERCC5 0.242 0.025 0.100 ERCC3 – 0.184 0.026 0.100 3.4. Differential gene expression and univariate analysis APEX2 – 0.266 0.027 0.100 DNAJB6 – 0.177 0.027 0.100 Differential gene expression analysis was carried out in BRCA1 – 0.333 0.027 0.100 133 cases after data normalization. XRCC4 – 0.271 0.029 0.104 Twenty genes were significantly and differently expressed, by multiple comparisons (FDR < 0.05), for RAD21 – 0.295 0.030 0.104 the PFS of trabectedin (Table 3 and Fig. S1). High MSH4 0.512 0.034 0.114 expression of PARP3 (logFC = 1.738; FDR < 0.001), DNAJB5 0.454 0.035 0.114 POLL (logFC = 1.767; FDR = 0.002), PMS1 OGG1 0.230 0.036 0.116 (logFC = 0.764; FDR = 0.003), RAD52 (logFC = DNAJC10 – 0.241 0.038 0.119 0.875; FDR = 0.003), ATM (logFC = 0.444; FDR = DNAJA1 – 0.219 0.041 0.125 0.014), NEIL1 (logFC = 0.773; FDR = 0.021), MLH3 0.209 0.046 0.135 TOP3B (logFC = 1.374; FDR = 0.030), CCNH DNAJC7 0.348 0.048 0.137 (logFC = 0.620; FDR = 0.030), XRCC5 (logFC = 0.862; FDR = 0.030), DDB2 (logFC = 0.314; FDR = FC: fold change; FDR: false discovery rate. A 0.031), MUTYH (logFC = 0.369; FDR = 0.034), negative fold change means that the gene is overexpressed in cases with PFS below the median. DNAJB14 (logFC = 0.420; FDR = 0.035), and XPC 3696 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy (logFC = 0.512; FDR = 0.037) was associated with The overexpression of BRCA1 gene was significantly longer PFS, whereas the overexpression of DNAJB11 associated with worse PFS of trabectedin [2.5 months (logFC = −0.378; FDR = 0.003), PARP1 (logFC = (95% CI 1.6–3.3) vs 6.1 months (95% CI 2.5–9.7), −0.442; FDR = 0.007), TDG (logFC = −0.312; P = 0.001]; however, neither the expression of ERCC1 FDR = 0.030), TOP3A (logFC = −0.451; FDR = [5.2 months (95% CI 2.5–8.0) vs 3.0 months (95% CI 0.030), PRKDC (logFC = −0.392; FDR = 0.030), 2.2–3.9), P = 0.084] nor of ERCC5 [3.2 months (95% DNAJC11 (logFC = −0.309; FDR = 0.030), and CI 2.3–4.1) vs 4.4 months (95% CI 1.4–7.4), RAD23B (logFC = −0.311; FDR = 0.035) were associ- P = 0.182] showed a significant association with PFS ated with worse PFS. of trabectedin (Table S3). The univariate analysis revealed that PARP3 and Considering GMI values, grouped as GMI ≥ 1.33 vs CCNH overexpression was markedly related to a GMI < 1.33, only six genes were differentially longer PFS of trabectedin: 7.7 months (95% CI 4.6– expressed with statistical significance between both 10.9) vs 2.8 months (95% CI 2.1–3.6), P < 0.001 and groups (FDR < 0.05). Patients with a GMI ≥ 1.33 7.9 months (95% CI 4.8–11.0) vs 2.5 months (95% CI showed significantly higher expression of DNAJC16 1.9–3.0), P < 0.001, respectively. On the other extreme, (logFC = 1.163; FDR = 0.001), XPC (logFC = 0.799; DNAJB11 and PARP1 overexpression was notably FDR = 0.004), DNAJB14 (logFC = 0.611; FDR = related to a shorter PFS of trabectedin: 2.5 months 0.006), ATM (logFC = 0.523; FDR = 0.008), (95% CI 2.3–3.7) vs 8.2 months (95% CI 5.2–11.1), DNAJC7 (logFC = 0.684; FDR = 0.008), and NEIL2 P < 0.001 and 2.5 months (95% CI 1.8–3.1) vs (logFC = 0.564; FDR = 0.039), compared with those 6.4 months (95% CI 4.0–8.8), P < 0.001, respectively with a GMI < 1.33 (Table 4). (Fig. 1 and Table S3). The remaining genes with statis- Differential gene expression reached statistical signif- tical significance for the PFS of trabectedin, as well as icance in 20 genes when it was considered obtaining or their impact in OS from trabectedin initiation, are rep- not clinical benefit (CR+PR+SD vs PD) (Table 5). resented in Table S3. High expression of PARP3 (logFC = 1.538; FDR = Fig. 1. Univariate analysis. (A) PFS according to PARP3 gene expression; (B) PFS according to PARP1 gene expression; (C) PFS according to CCNH gene expression, and (D) PFS according to DNAJB11 gene expression. Groups were defined according to the median of gene expression; above and below median. Log-rank test statistical significance was defined at P ≤ 0.05. Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3697 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. Table 4. Differential gene expression attending to growth benefit of trabectedin (Table 5): ATM, DNAJB14, and modulation index. A negative fold change means that the gene is XPC (Fig. S2). Further, 14 significantly expressed overexpressed in cases with < 1.33. genes were common for PFS and clinical benefit end- Gene logFC P-value FDR points: ATM, DNAJB11, DNAJB14, PARP1, PARP3, DNAJC16 1.163 < 0.001 0.001 POLL, PRKDC, RAD23B, RAD52, TDG, TOP3A, XPC 0.799 < 0.001 0.004 TOP3B, XPC, and XRCC5 (Fig. S2). DNAJB14 0.611 < 0.001 0.006 Differential gene expression was also determined ATM 0.523 < 0.001 0.008 according to clinical variables with an independent DNAJC7 0.684 < 0.001 0.008 value for PFS in the multivariate analysis: Table S5 NEIL2 0.564 0.002 0.039 (L-sarcomas vs other subtypes) and Table S6 (visceral vs somatic location). PARP3 and PMS1, which were DNAJC8 0.402 0.004 0.068 genes associated with a better PFS of trabectedin, were MSH4 0.737 0.006 0.088 overexpressed in L-sarcomas and in tumors with a XRCC6BP1 0.912 0.009 0.121 somatic location, whereas PARP1, a gene that corre- TOP3A –0.446 0.011 0.130 lated with worse PFS of trabectedin, was highly PARP1 –0.358 0.013 0.136 expressed in non-L-sarcomas and tumors with visceral RAD52 0.606 0.016 0.158 location. BRCA2 0.349 0.022 0.200 PRKDC –0.343 0.028 0.232 0.505 0.029 0.232 3.5. Predictive gene signature of trabectedin DNAJB5 activity XRCC2 0.297 0.032 0.233 POLL 0.876 0.037 0.258 Univariate Cox regression model on a cohort of 139 ATR 0.215 0.039 0.258 patients identified eight genes whose expression was FC: fold change; FDR: false discovery rate. A significantly associated with PFS of trabectedin. Fur- negative fold change means that the gene is ther selection of genes whose Kaplan–Meier curves overexpressed in cases with < 1.33. displayed significant differences (patients were divided into two groups based on gene expression median value) yielded six final genes. Lasso-penalized Cox 0.007), POLL (logFC = 1.876; FDR = 0.007), RAD52 regression analysis was performed to construct a pre- (logFC = 0.849; FDR = 0.009), TOP3B (logFC = dictive gene signature for trabectedin. The six genes 1.818; FDR = 0.009), XRCC5 (logFC = 1.073; identified were as follows: exonuclease 1 (EXO1), FDR = 0.009), DNAJB14 (logFC = 0.463; FDR = cyclin H (CCNH), xeroderma pigmentosum group A- 0.032), XPC (logFC = 0.559; FDR = 0.036), ATM complementing protein (XPA), poly(ADP-ribose) poly- (logFC = 0.383; FDR = 0.036), MSH5 (logFC = merase 1 (PARP1), BRCA1, and apurinic/apyrimidinic 0.843; FDR = 0.043), and NEIL2 (logFC = 0.467; endodeoxyribonuclease 2 (APEX2). FDR = 0.043) were associated with clinical benefit of trabectedin, whereas overexpression of RAD21 Risk score ¼ 0:1866930  ExpressionEXO1 þ ð0:1966279Þ (logFC = −0.522; FDR = 0.007), MSH6 (logFC = ExpressionCCNH −0.369: FDR = 0.009), DNAJB11 (logFC = −0.336; þð0:2043687Þ  ExpressionXPA þ 0:2293783 FDR = 0.011), TDG (logFC = −0.332; FDR = 0.024), XRCC4 (logFC = −0.394; FDR = 0.024), PARP1 ExpressionPARP1 (logFC = −0.373; FDR = 0.036), ERCC3 (logFC = þ0:0129766  ExpressionBRCA1 þ 0:2278331 −0.242; FDR = 0.036), PRKDC (logFC = −0.380; ExpressionAPEX2 : FDR = 0.043), RAD23B (logFC = −0.315; FDR = 0.043), and TOP3A (logFC = −0.423; FDR = 0.043) The six-gene-based risk score was calculated for were associated with progressive disease after trabecte- each patient, and the optimal cutoff for the risk score din. By contrast, no different gene expression was sig- (2.146) was determined. An additional univariate Cox nificantly detected, by FDR, when the grouping was regression analysis was performed using risk scores, responders (CR+PR) vs non-responders patients resulting in a significant association with PFS (Table (Table S4). S7). Time-dependent ROC and Kaplan–Meier curves Regarding the different analyses performed, only were used to assess the predictive value of the gene sig- three genes were significantly associated with a longer nature. The AUCs (area under the ROC curve) for 3-, PFS (Table 3), GMI ≥ 1.33 (Table 4), and clinical 12-, and 24-month PFS were 0.67, 0.65, and 0.65, 3698 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy Table 5. Differential gene expression according to clinical benefit. A negative fold change means that the gene is overexpressed in cases progression disease. logFC P-value FDR PARP3 1.538 < 0.001 0.007 POLL 1.876 < 0.001 0.007 RAD21 –0.522 < 0.001 0.007 MSH6 –0.369 < 0.001 0.009 RAD52 0.849 < 0.001 0.009 TOP3B 1.818 < 0.001 0.009 XRCC5 1.073 0.001 0.009 DNAJB11 –0.336 0.001 0.011 TDG – 0.332 0.002 0.024 XRCC4 –0.394 0.002 0.024 DNAJB14 0.463 0.003 0.032 PARP1 – 0.373 0.004 0.036 XPC 0.559 0.004 0.036 ERCC3 –0.242 0.004 0.036 ATM 0.383 0.005 0.036 MSH5 0.843 0.006 0.043 PRKDC –0.380 0.006 0.043 RAD23B –0.315 0.007 0.043 NEIL2 0.467 0.007 0.043 TOP3A –0.423 0.007 0.043 DNAJC16 0.649 0.009 0.052 DNAJC21 –0.406 0.014 0.070 DNAJC11 –0.271 0.014 0.070 SLK –0.267 0.018 0.089 DNAJC10 –0.279 0.020 0.094 DNAJA1 –0.247 0.025 0.115 XRCC1 –0.252 0.031 0.133 DNAJB6 –0.178 0.031 0.133 DNAJC7 0.387 0.034 0.136 DNAJC1 –0.244 0.043 0.160 DDB2 0.226 0.044 0.160 BRCA1 –0.315 0.044 0.160 MRE11A 0.249 0.045 0.160 FC: fold change; FDR: false discovery rate. A negative fold change means that the gene is overexpressed in cases progression disease. respectively (Fig. S3). Patients in the high-risk group (Fig. 2)]. PFS rate at 6 months and at 12 months was (higher expression of EXO1, PARP1, BRCA1, and 16% and 5%, respectively, for the high-risk group and APEX2 and lower expression of CCNH and XPA) 48% and 25%, respectively, for the low-risk group. In showed a significantly worse PFS compared with the the multivariate analysis, the high-risk group was an patients in the low-risk group (lower expression of independent factor associated with PFS [HR 2.1 (95% EXO1, PARP1, BRCA1, and APEX2 and higher CI 1.4–3.2), P < 0.001]; visceral location and non-L- expression of CCNH and XPA), [2.1 months (95% CI sarcomas maintained the independent value (Table 1.8–2.4) vs 6.0 months (95% CI 4.5–7.5), P < 0.001 S8). Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3699 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. Fig. 2. Risk groups Kaplan–Meier curve. Patients were grouped according to the risk scores cutoff value (2.146). Number at risk was represented as n (%). Log-rank test statistical significance was defined at P ≤ 0.05. In the patients treated with either doxorubicin or unfavorable signature) for trabectedin efficacy in terms gemcitabine, the DDR-related gene signature was also of PFS. correlated with the PFS of these drugs. The six-gene- Even when BRCA1 (expression levels or haplotype based signature failed to significantly predict gemc- status) had been previously associated with the clinical itabine or doxorubicin activity, in terms of PFS, in our benefit of trabectedin [15,21], its predictive role was series (Table S9). weaker than the gene signature. Overexpression of XPA gene, another component of the signature and a NER-associated gene, was in line with previous reports 4. Discussion that showed a direct correlation between higher A six-gene predictive signature of trabectedin efficacy expression of NER machinery-related genes and better was built based on the transcriptomic analysis of 118 trabectedin efficacy [22,23]. Besides, and to the best of DDR-related genes and PFS, in a retrospective series our knowledge, this is the first time that the expression of 139 advanced STS patients. Previously described of EXO1, CCNH, and APEX2 had been associated predictive genes for trabectedin efficacy, as those with trabectedin efficacy. EXO1 is an XPG family related to NER machinery, such as ERCC1 and nuclease that participates in several DDR pathways, ERCC5, did not show any relevant prognostic role in such as DNA mismatch repair, HR, and non- our series. Rather, other DDR-related genes showed homologous end joining (NHEJ) [22]. Of note, the more consistent prediction of trabectedin activity, expression of EXO1 was associated with resistance to either in terms of PFS, GMI, and/or clinical benefit. cisplatin and doxorubicin via NHEJ in ovarian cancer, For this study, we decided to focus only in the rele- a tumor also sensitive to trabectedin. Nevertheless, the vance of DDR-related genes, since the DNA DSBs prognostic/predictive value of EXO1 in STS is still induced by trabectedin along with the involvement of unknown [24,25]. Cyclin H, coded by the CCNH gene, NER machinery and HR repair genes were previously is a protein that forms a cyclin-activating kinase- reported as a potential relevant mechanism of trabecte- subcomplex with CDK7 and MAT1. This subcomplex din activity in STS patients [13,15,21]. However, the is part of the transcription factor II human (TFIIH) majority of these studies were limited to few DDR- protein complex and it is involved in NER, by regulat- related genes/proteins as potential predictive biomark- ing cell cycling progression during DNA damage [26]. ers of trabectedin, in consequence many other poten- As expected in accordance with the role of cyclin H in tial predictive factors from the family of DNA repair NER pathway, a high expression of this gene was genes have been unexplored in this regard. Hence, we associated with trabectedin sensitivity in our signature. aimed to evaluate, in a broader way, the predictive APEX2 is a weak apurinic/apyrimidinic (AP) endonu- role of a large number of DDR-related genes using clease in the DNA base excision repair pathway and HTG EdgeSeq next-generation sequencing (NGS)- functions in the context of DNA damage by oxidative adapted molecular profiling. Subsequently, an estab- and alkylating agents. APEX2 also exhibits strong lished six-gene signature was able to markedly separate activities of 30 –50 exonuclease and 30 -phosphodiesterase two significant different populations (favorable or and that has been described to be necessary for the 3700 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy recovery of B-cell lymphocyte progenitors after cell cycle phase in which cells are more sensitive to tra- chemotherapy [27,28]. Moreover, APEX2 has been bectedin [39]. Accordingly, the effect of the overexpres- identified as a synthetic lethal gene, in BRCA1- and sion of each one of these proteins in cell cycle BRCA2-deficient cells [29]. APEX2 processes AP sites progression may justify, at least in part, the different at replication forks, avoiding the blocking of fork pro- results for both PARP1 and PARP3 in our study. gression and the induction of DSBs, which required Noteworthy, PARP3 seems to be overexpressed in BRCA1/2 for repair and DNA synthesis [29]. Of note, tumors that benefit from trabectedin treatment (i.e., L- trabectedin has been described to cause replication sarcomas and tumors with somatic location), while impairment and genome instability, via RNA–DNA PARP1 expression is higher in sarcomas with lower hybrid-dependent DNA damage, which is important activity of trabectedin (non-L-sarcomas and tumors for drug activity [30]. Accordingly, it is possible that with visceral location). Therefore, the expression of APEX2 may have a role in resolving replication these genes may be important biomarkers of trabecte- impairment and genome instability induced by tra- din efficacy or resistance, in specific clinical contexts bectedin, which may justify its association with lack of that depend on the histologic subtype or the location drug activity in our gene signature. This could also of the primary tumor. Besides, the association of open the option to target APEX2 as a potential thera- PARP1 high expression with worse trabectedin out- peutic strategy in combination with trabectedin if this come supports the combination of PARP1 inhibitors gene is overexpressed. Remarkably, our six-gene signa- with this chemotherapeutic agent. In line with these ture is in line with previous reports [13,15,16], wherein observations, the combination of olaparib with tra- the high expression of NER genes (e.g., XPA and bectedin has been tested in a phase 1b clinical trial in CCNH) and the low expression of HR genes (e.g., patients with advanced and non-resectable bone and EXO1 and BRCA1) would help to identify the STS STS (ClinicalTrials.gov, number NCT02398058). Of 50 patients that benefit from trabectedin treatment. patients recruited, 7 had a PR, suggesting that this Furthermore, several DDR-related genes were signif- combination has antitumor activity [40]; however, icantly associated with trabectedin activity, in terms of patient selection based on PARP1 and/or PARP3 PFS, GMI, and/or clinical benefit. Among these genes, expression levels should be considered for further clini- PARP3 overexpression significantly associated with a cal trials with this combination. better PFS and clinical benefit following trabectedin. In the study presented here, only BRCA1 high PARP3, initially described in 1999 [31], is a poly(ADP expression was significantly associated with shorter ribosyl)transferase closely related with PARP1 and PFS and OS, while ERCC1 and ERCC5 expression, PARP2, and it is highly expressed in skeletal muscle which were previously associated with good prognosis [32]. PARP3 is involved in transcriptional silencing to trabectedin [13,15,16], did not had a significant and in the maintenance of genomic integrity [33], being impact in survival. The different methodology used for part of the polycomb group complexes, which are rele- gene expression analysis could justify the distinct vant to maintain the silencing of specific critical genes results obtained in terms of survival observed for these related to development among others. More precisely, two genes; however, it is important to mention that PARP3 was shown to interact with core proteins of HTG quantitative nuclease protection assays for gene the polycomb repressive complexes 2 and 3, such as detections had been reported to have similar sensitivity the methyltransferases EZH2 and Suz12, which are rel- as qPCR [41,42]. Accordingly, our results may suggest evant in the maintenance of repression through post- that NER-associated genes ERCC1 and ERCC5 are translational methylation processes [33]. In turn, not robust and reliable biomarkers of trabectedin PARP1 expression, a gene also included in our poten- activity. Their predictive value should accordingly be tial predictive molecular signature, was significantly considered with caution; however, a further study with associated with shorter PFS and lack of clinical benefit a larger cohort of cases should validate our data. after trabectedin treatment. The distinct behavior or Among the study limitations, it is important to men- PARP1 and PARP3 genes in our study may indicate tion that the tumor samples used for this retrospective that these proteins may participate differently in some analysis were collected at the diagnostic time, and gene cellular mechanisms, which could affect the response expression levels could have been altered due to the to trabectedin. Both proteins are relevant for DNA several lines of previous treatment or during the natu- genomic integrity [34]; however, it seems that PARP1 ral progression of the disease. Moreover, another overexpression promotes cell cycle entry [35–37], important limitation is related with its initial hypothe- whereas PARP3 overexpression interferes with G1/S sis; for this analysis, only genes related with DDR cell cycle progression [38]. Of note, G1 seems to be the were selected and studied. This selection excluded, Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3701 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. from the molecular signature, genes involved in other outside the submitted work. Research funding for her cellular processes, which may also predict trabectedin Institution (clinical studies) from PharmaMar, Eli Lilly activity. Since we cannot exclude the relevance of and Company, AROG, Bayer, Eisai, Lixte, Karyo- other genes beyond DDR, further analysis, considering pharm, Deciphera, GSK, Novartis, Blueprint, Nektar, all the genes included in the OBP assay, is currently Forma, Amgen, Adaptimmune, and Daichii-Sankyo. ongoing. The absence of an independent validation JM-B: Honoraria from PharmaMar, Lilly, Bayer, and cohort, or a control cohort treated with any other Eisai; travel grants from PharmaMar; research grants drug approved for advanced STS, is other limitation from PharmaMar, Eisai, Immix BioPharma, and of this study. The validation of the predictive value of Novartis outside the submitted work. Research fund- the gene-based signature in a larger series of cases ing for his institution (for clinical studies) from Phar- could also help improve the performance of ROC maMar, Eli Lilly and Company, AROG, Bayer, Eisai, curves. Lixte, Karyopharm, Deciphera, GSK, Novartis, Blue- print, Nektar, Forma, Amgen, Adaptimmune, and Daichii-Sankyo. 5. Conclusions In summary, this translational study identified a new Author contributions molecular signature that could significantly predict patients with higher probability for trabectedin effi- DSM and JM-B designed the experiments and analy- cacy. This signature was based on wide transcriptomic sis. All the authors participated in data acquisition. analysis of DDR-related genes and provides new DSM, MP-C, JACV, and JM-B performed data analy- insights for potential targets that could enhance the sis and interpretation. DSM, MP-C, JACV, and JM-B trabectedin efficacy. Validation of this signature is in drafted the manuscript. All the authors reviewed the process with an independent series of advanced STS manuscript and approved its final version. treated with trabectedin. Finally, the potential value of specific biomarkers (e.g., PARP3) is currently being Data accessibility tested in the preclinical setting, in order to describe the mechanisms in which they participate under drug The data that support the findings of this study are treatment. available from the corresponding author (jmartin@- mustbesevilla.org) upon reasonable request. Acknowledgements References This study was funded by the Spanish Group for Research on Sarcoma (GEIS) and partially by Phar- 1 Martı́n-Broto J, Reichardt P, Jones RL & Stacchiotti S maMar. The authors would like to thank the GEIS (2020) Different approaches to advanced soft tissue data center for data management. The authors also sarcomas depending on treatment line, goal of therapy thank the donors and the Hospital Universitario Vir- and histological subtype. Expert Rev Anticancer Ther gen del Rocı́o—Instituto de Biomedicina de Sevilla 20, 15–28. Biobank (Andalusian Public Health System Biobank 2 Minchom A, Jones RL, Fisher C, Al-Muderis O, Ashley S, Scurr M, Karavasilis V & Judson IR (2010) and ISCIII-Red de Biobancos PT17/0015/0041) for Clinical benefit of second-line palliative chemotherapy part of the human specimens used in this study. David in advanced soft-tissue sarcoma. Sarcoma 2010, S. Moura is recipient of a Sara Borrell postdoctoral 264360. fellowship funded by the National Institute of Health 3 Sharma S, Takyar S, Manson SC, Powell S & Penel N Carlos III (ISCIII) (CD20/00155). (2013) Efficacy and safety of pharmacological interventions in second- or later-line treatment of Conflict of interest patients with advanced soft tissue sarcoma: a systematic review. BMC Cancer 13, 385. DSM reports institutional research grants from Phar- 4 Blay J-Y, von Mehren M, Samuels BL, Fanucchi MP, maMar, Eisai, Immix BioPharma, and Novartis out- Ray-Coquard I, Buckley B, Gilles L, Lebedinsky C, side the submitted work and travel support from Elsayed YA & Le Cesne A (2008) Phase I combination PharmaMar, Eisai, Celgene, Bayer, and Pfizer. NH study of trabectedin and doxorubicin in patients with declares Honoraria from PharmaMar and Lilly; Travel soft-tissue sarcoma. Clin Cancer Res 14, 6656–6662. Grants from PharmaMar; Research Grants from Phar- 5 Demetri GD, Chawla SP, von Mehren M, Ritch P, maMar, Eisai, Immix BioPharma, and Novartis Baker LH, Blay JY, Hande KR, Keohan ML, Samuels 3702 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy BL, Schuetze S et al. (2009) Efficacy and safety of patients with advanced non-small-cell lung cancer trabectedin in patients with advanced or metastatic (NSCLC) with XPG and/or ERCC1 overexpression and liposarcoma or leiomyosarcoma after failure of prior BRCA1 underexpression and pretreated with platinum. anthracyclines and ifosfamide: results of a randomized Lung Cancer 76, 354–361. phase II study of two different schedules. J Clin Oncol 15 Schöffski P, Taron M, Jimeno J, Grosso F, Sanfilipio 27, 4188–4196. R, Casali PG, Le Cesne A, Jones RL, Blay JY, Poveda 6 Minuzzo M, Marchini S, Broggini M, Faircloth G, A et al. (2011) Predictive impact of DNA repair D’Incalci M & Mantovani R (2000) Interference of functionality on clinical outcome of advanced sarcoma transcriptional activation by the antineoplastic drug patients treated with trabectedin: a retrospective ecteinascidin-743. Proc Natl Acad Sci USA 97, 6780– multicentric study. Eur J Cancer 47, 1006–1012. 6784. 16 Moura DS, Sanchez-Bustos P, Fernandez-Serra A, 7 Allavena P, Signorelli M, Chieppa M, Erba E, Bianchi Lopez-Alvarez M, Mondaza-Hernandez JL, Blanco- G, Marchesi F, Olimpio CO, Bonardi C, Garbi A, Alcaina E, Gavilan-Naranjo A, Martinez-Delgado P, Lissoni A et al. (2005) Anti-inflammatory properties of Lacerenza S, Santos-Fernandez P et al. (2020) CUL4A, the novel antitumor agent yondelis (trabectedin): ERCC5, and ERCC1 as predictive factors for inhibition of macrophage differentiation and cytokine trabectedin efficacy in advanced soft tissue sarcomas production. Can Res 65, 2964–2971. (STS): a Spanish Group for Sarcoma Research (GEIS) 8 Germano G, Frapolli R, Simone M, Tavecchio M, Erba Study. Cancers 12, 1128. E, Pesce S, Pasqualini F, Grosso F, Sanfilippo R, Casali 17 Chadly DM, Best J, Ran C, Bruska M, Woźniak W, PG et al. (2010) Antitumor and anti-inflammatory Kempisty B, Schwartz M, LaFleur B, Kerns BJ, Kessler effects of trabectedin on human myxoid liposarcoma JA et al. (2018) Developmental profiling of microRNAs cells. Can Res 70, 2235–2244. in the human embryonic inner ear. PLoS One 13, 9 Martinez-Serra J, Maffiotte E, Martin J, Bex T, e0191452. Navarro-Palou M, Ros T, Plazas JM, Vogler O, 18 Robinson MD & Oshlack A (2010) A scaling Gutierrez A, Amat JC et al. (2011) Yondelis(R) (ET- normalization method for differential expression 743, Trabectedin) sensitizes cancer cell lines to CD95- analysis of RNA-seq data. Genome Biol 11, R25. mediated cell death: new molecular insight into the 19 Robinson MD & Smyth GK (2008) Small-sample mechanism of action. Eur J Pharmacol 658, 57–64. estimation of negative binomial dispersion, with 10 Romero J, Zapata I, Cordoba S, Jimeno JM, Lopez- applications to SAGE data. Biostatistics 9, 321–332. Martin JA, Tercero JC, De La Torre A, Vargas JA, 20 Benjamini Y & Hochberg Y (1995) Controlling the false Moleron R & Sanchez-Prieto R (2008) In vitro discovery rate: a practical and powerful approach to radiosensitisation by trabectedin in human cancer cell multiple testing. J R Stat Soc Series B (Methodol) 57, lines. Eur J Cancer 44, 1726–1733. 289–300. 11 Gronchi A, Hindi N, Cruz J, Blay J-Y, Lopez-Pousa A, 21 Laroche-Clary A, Chaire V, Le Morvan V, Neuville A, Italiano A, Alvarez R, Gutierrez A, Rincón I, Sangalli Bertucci F, Salas S, Sanfilippo R, Pourquier P & C et al. (2019) Trabectedin and radiotherapy in soft Italiano A (2015) BRCA1 haplotype and clinical benefit tissue sarcoma (TRASTS): results of a phase I study in of trabectedin in soft-tissue sarcoma patients. Br J myxoid liposarcoma from Spanish (GEIS), Italian Cancer 112, 688–692. (ISG), French (FSG) Sarcoma Groups. 22 Larsen AK, Galmarini CM & D’Incalci M (2016) EClinicalMedicine 9, 35–43. Unique features of trabectedin mechanism of action. 12 Martin-Broto J, Hindi N, Lopez-Pousa A, Peinado- Cancer Chemother Pharmacol 77, 663–671. Serrano J, Alvarez R, Alvarez-Gonzalez A, Italiano A, 23 Takebayashi Y, Pourquier P, Zimonjic DB, Nakayama Sargos P, Cruz-Jurado J, Isern-Verdum J et al. (2020) K, Emmert S, Ueda T, Urasaki Y, Kanzaki A, Assessment of safety and efficacy of combined Akiyama SI, Popescu N et al. (2001) Antiproliferative trabectedin and low-dose radiotherapy for patients with activity of ecteinascidin 743 is dependent upon metastatic soft-tissue sarcomas: a nonrandomized phase transcription-coupled nucleotide-excision repair. Nat 1/2 clinical trial. JAMA Oncol 6, 535–541. Med 7, 961–966. 13 Italiano A, Laurand A, Laroche A, Casali P, Sanfilippo 24 He D, Li T, Sheng M & Yang B (2020) Exonuclease 1 R, Le Cesne A, Judson I, Blay JY, Ray-Coquard I, Bui (Exo1) participates in mammalian non-homologous end B et al. (2011) ERCC5/XPG, ERCC1, and BRCA1 joining and contributes to drug resistance in ovarian gene status and clinical benefit of trabectedin in patients cancer. Med Sci Monit 26, e918751. with soft tissue sarcoma. Cancer 117, 3445–3456. 25 Lorusso D, Scambia G, Pignata S, Sorio R, Amadio G, 14 Massuti B, Cobo M, Camps C, Dómine M, Provencio Lepori S, Mosconi A, Pisano C, Mangili G, Maltese G M, Alberola V, Viñolas N, Rosell R, Tarón M, et al. (2016) Prospective phase II trial of trabectedin in Gutiérrez-Calderón V et al. (2012) Trabectedin in BRCA-mutated and/or BRCAness phenotype recurrent Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3703 Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Predictive gene signature of trabectedin efficacy D. S. Moura et al. ovarian cancer patients: the MITO 15 trial. Ann Oncol 37 Yang L, Huang K, Li X, Du M, Kang X, Luo X, Gao 27, 487–493. L, Wang C, Zhang Y, Zhang C et al. (2013) 26 Adamczewski JP, Rossignol M, Tassan JP, Nigg EA, Identification of poly(ADP-ribose) polymerase-1 as a Moncollin V & Egly JM (1996) MAT1, cdk7 and cyclin cell cycle regulator through modulating Sp1 mediated H form a kinase complex which is UV light-sensitive transcription in human hepatoma cells. PLoS One 8, upon association with TFIIH. EMBO J 15, 1877–1884. e82872. 27 Guikema JE, Gerstein RM, Linehan EK, Cloherty EK, 38 Augustin A, Spenlehauer C, Dumond H, Ménissier-de Evan-Browning E, Tsuchimoto D, Nakabeppu Y & Murcia J, Piel M, Schmit A-C, Apiou F, Vonesch J-L, Schrader CE (2011) Apurinic/apyrimidinic endonuclease Kock M, Bornens M et al. (2003) PARP-3 localizes 2 is necessary for normal B cell development and preferentially to the daughter centriole and interferes recovery of lymphoid progenitors after with the G1/S cell cycle progression. J Cell Sci 116, chemotherapeutic challenge. J Immunol 186, 1943–1950. 1551–1562. 28 Lin Y, Bai L, Cupello S, Hossain MA, Deem B, 39 Erba E, Bergamaschi D, Bassano L, Damia G, Ronzoni McLeod M, Raj J & Yan S (2018) APE2 promotes S, Faircloth GT & D’Incalci M (2001) Ecteinascidin- DNA damage response pathway from a single-strand 743 (ET-743), a natural marine compound, with a break. Nucleic Acids Res 46, 2479–2494. unique mechanism of action. Eur J Cancer 37, 97– 29 Mengwasser KE, Adeyemi RO, Leng Y, Choi MY, 105. Clairmont C, D’Andrea AD & Elledge SJ (2019) 40 Grignani G, D’Ambrosio L, Pignochino Y, Palmerini Genetic screens reveal FEN1 and APEX2 as BRCA2 E, Zucchetti M, Boccone P, Aliberti S, Stacchiotti S, synthetic lethal targets. Mol Cell 73, 885–899.e886. Bertulli R, Piana R et al. (2018) Trabectedin and 30 Tumini E, Herrera-Moyano E, San Martı́n-Alonso M, olaparib in patients with advanced and non-resectable Barroso S, Galmarini CM & Aguilera A (2019) The bone and soft-tissue sarcomas (TOMAS): an open-label, antitumor drugs trabectedin and lurbinectedin induce phase 1b study from the Italian Sarcoma Group. transcription-dependent replication stress and genome Lancet Oncol 19, 1360–1371. instability. Mol Cancer Res 17, 773–782. 41 Qi Z, Wang L, He A, Ma-Edmonds M & Cogswell J 31 Johansson M (1999) A human poly(ADP-ribose) (2016) Evaluation and selection of a non-PCR based polymerase gene family (ADPRTL): cDNA cloning of technology for improved gene expression profiling from two novel poly(ADP-ribose) polymerase homologues. clinical formalin-fixed, paraffin-embedded samples. Genomics 57, 442–445. Bioanalysis 8, 2305–2316. 32 Urbánek P, Paces J, Králová J, Dvorák M & Paces V 42 Songia P, Chiesa M, Valerio V, Moschetta D, (2002) Cloning and expression of PARP-3 (Adprt3) and Myasoedova VA, D’Alessandra Y & Poggio P (2018) U3–55k, two genes closely linked on mouse Direct screening of plasma circulating microRNAs. chromosome 9. Folia Biol 48, 182–191. RNA Biol 15, 1268–1272. 33 Rouleau M, McDonald D, Gagné P, Ouellet ME, Droit A, Hunter JM, Dutertre S, Prigent C, Hendzel MJ & Poirier GG (2007) PARP-3 associates with polycomb Supporting information group bodies and with components of the DNA Additional supporting information may be found damage repair machinery. J Cell Biochem 100, 385–401. online in the Supporting Information section at the end 34 Boehler C, Gauthier LR, Mortusewicz O, Biard DS, of the article. Saliou JM, Bresson A, Sanglier-Cianferani S, Smith S, Fig. S1. Volcano plot of genes with prognostic rele- Schreiber V, Boussin F et al. (2011) Poly(ADP-ribose) vance in progression-free survival (PFS) of trabectedin. polymerase 3 (PARP3), a newcomer in cellular response Fig. S2. Venn diagram comparing the unique and to DNA damage and mitotic progression. Proc Natl shared genes between progression-free survival, growth Acad Sci USA 108, 2783–2788. modulation index and clinical benefit. 35 Carbone M, Rossi MN, Cavaldesi M, Notari A, Amati Fig. S3. Time-dependent ROC curve to assess the pre- P & Maione R (2008) Poly(ADP-ribosyl)ation is dictive value of the gene signature. implicated in the G0–G1 transition of resting cells. Oncogene 27, 6083–6092. Table S1. Expression values of DNA damage repair- 36 Simbulan-Rosenthal CM, Rosenthal DS, Luo R, related genes. Samara R, Espinoza LA, Hassa PO, Hottiger MO & Table S2. Gene significantly expressed between L-sar- Smulson ME (2003) PARP-1 binds E2F–1 coma and non-L-sarcomas and between grade 3 independently of its DNA binding and catalytic tumors vs grade 1 + 2 tumors. domains, and acts as a novel coactivator of E2F–1- Table S3. Univariate analysis of DNA damage repair- mediated transcription during re-entry of quiescent cells related genes with impact in progression-free survival into S phase. Oncogene 22, 8460–8471. to trabectedin line. 3704 Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. 18780261, 2021, 12, Downloaded from https://febs.onlinelibrary.wiley.com/doi/10.1002/1878-0261.12996 by Readcube (Labtiva Inc.), Wiley Online Library on [30/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License D. S. Moura et al. Predictive gene signature of trabectedin efficacy Table S4. Differential gene expression according to Table S7. Univariate Cox regression analysis for the objective response. values of the gene signature (risk scores). Table S5. Differential gene expression according to Table S8. Multivariate analysis. histologic subtype grouping. Table S9. Correlation between risk groups and pro- Table S6. Differential gene expression according to gression-free survival of doxorubicin or gemcitabine. tumor location. Molecular Oncology 15 (2021) 3691–3705 ª 2021 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 3705 Federation of European Biochemical Societies.