In silicolearning of tumor evolution through mutational time series

AbstractCancer arises through the accumulation of somatic mutations over time. Understanding the sequence of mutation occurrence during cancer progression can assist early and accurate diagnosis and improve clinical decision-making. Here we employ Long Short-Term Memory networks (LSTMs), a class of recurrent neural network, to learn the evolution of a tumor through an ordered sequence of mutations. We demonstrate the capacity of LSTMs to learn complex dynamics of the mutational time series governing tumor progression, allowing accurate prediction of the mutational burden and the occurrence of mutations in the sequence. Using the probabilities learned by the LSTM, we simulate mutational data and show that the simulation results are statistically indistinguishable from the empirical data. We identify passenger mutations that are significantly associated with established cancer drivers in the sequence and demonstrate that the genes carrying these mutations are substantially enriched in interactions with the corresponding driver genes. Breaking the network into modules consisting of driver genes and their interactors, we show that these interactions are associated with poor patient prognosis, thus likely conferring growth advantage for tumor progression. Thus, application of LSTM provides for prediction of numerous additional conditional drivers and to reveal hitherto unknown aspects of cancer evolution.SignificanceCancer is caused by the effects of somatic mutations known as drivers. Although a number of major cancer drivers have been identified, it is suspected that many more comparatively rare and conditional drivers exist, and the interactions between different cancer-associated mutations that might be relevant for tumor progression are not well understood. We applied an advanced neural network approach to learn the sequence of mutations and the mutational burden in colon and lung cancers, and to identify mutations that are associated with individual drivers. A significant ordering of driver mutations is demonstrated, and numerous, previously undetected conditional drivers are identified. These findings broaden the existing understanding of the mechanisms of tumor progression and have implications for therapeutic strategies.

Download Full-text

In silico learning of tumor evolution through mutational time series

Proceedings of the National Academy of Sciences ◽

10.1073/pnas.1901695116 ◽

2019 ◽

Vol 116 (19) ◽

pp. 9501-9510 ◽

Cited By ~ 4

Author(s):

Noam Auslander ◽

Yuri I. Wolf ◽

Eugene V. Koonin

Keyword(s):

Time Series ◽

Tumor Progression ◽

Cancer Progression ◽

Clinical Decision Making ◽

Short Term Memory ◽

Complex Dynamics ◽

Clinical Decision ◽

Tumor Evolution ◽

Cancer Evolution ◽

Driver Genes

Cancer arises through the accumulation of somatic mutations over time. Understanding the sequence of mutation occurrence during cancer progression can assist early and accurate diagnosis and improve clinical decision-making. Here we employ long short-term memory (LSTM) networks, a class of recurrent neural network, to learn the evolution of a tumor through an ordered sequence of mutations. We demonstrate the capacity of LSTMs to learn complex dynamics of the mutational time series governing tumor progression, allowing accurate prediction of the mutational burden and the occurrence of mutations in the sequence. Using the probabilities learned by the LSTM, we simulate mutational data and show that the simulation results are statistically indistinguishable from the empirical data. We identify passenger mutations that are significantly associated with established cancer drivers in the sequence and demonstrate that the genes carrying these mutations are substantially enriched in interactions with the corresponding driver genes. Breaking the network into modules consisting of driver genes and their interactors, we show that these interactions are associated with poor patient prognosis, thus likely conferring growth advantage for tumor progression. Thus, application of LSTM provides for prediction of numerous additional conditional drivers and reveals hitherto unknown aspects of cancer evolution.

Download Full-text

Clonal approaches to understanding the impact of mutations on hematologic disease development

Blood ◽

10.1182/blood-2018-11-835405 ◽

2019 ◽

Vol 133 (13) ◽

pp. 1436-1445 ◽

Cited By ~ 6

Author(s):

Jyoti Nangalia ◽

Emily Mitchell ◽

Anthony R. Green

Keyword(s):

Somatic Mutations ◽

Clonal Selection ◽

Single Cells ◽

Driver Mutations ◽

Great Promise ◽

Tumor Evolution ◽

Disease Development ◽

Hematopoietic Tissue ◽

The Impact

Abstract Interrogation of hematopoietic tissue at the clonal level has a rich history spanning over 50 years, and has provided critical insights into both normal and malignant hematopoiesis. Characterization of chromosomes identified some of the first genetic links to cancer with the discovery of chromosomal translocations in association with many hematological neoplasms. The unique accessibility of hematopoietic tissue and the ability to clonally expand hematopoietic progenitors in vitro has provided fundamental insights into the cellular hierarchy of normal hematopoiesis, as well as the functional impact of driver mutations in disease. Transplantation assays in murine models have enabled cellular assessment of the functional consequences of somatic mutations in vivo. Most recently, next-generation sequencing–based assays have shown great promise in allowing multi-“omic” characterization of single cells. Here, we review how clonal approaches have advanced our understanding of disease development, focusing on the acquisition of somatic mutations, clonal selection, driver mutation cooperation, and tumor evolution.

Download Full-text

Mutational landscape of primary, metastatic, and recurrent ovarian cancer reveals c-MYC gains as potential target for BET inhibitors

Proceedings of the National Academy of Sciences ◽

10.1073/pnas.1814027116 ◽

2018 ◽

Vol 116 (2) ◽

pp. 619-624 ◽

Cited By ~ 11

Author(s):

Charles Li ◽

Elena Bonazzoli ◽

Stefania Bellone ◽

Jungmin Choi ◽

Weilai Dong ◽

...

Keyword(s):

Ovarian Cancer ◽

Somatic Mutations ◽

Driver Mutations ◽

Cancer Evolution ◽

Metastatic Tumors ◽

Driver Genes ◽

Primary Tumors ◽

Mutational Landscape ◽

Recurrent Tumors ◽

Bet Inhibitors

Ovarian cancer remains the most lethal gynecologic malignancy. We analyzed the mutational landscape of 64 primary, 41 metastatic, and 17 recurrent fresh-frozen tumors from 77 patients along with matched normal DNA, by whole-exome sequencing (WES). We also sequenced 13 pairs of synchronous bilateral ovarian cancer (SBOC) to evaluate the evolutionary history. Lastly, to search for therapeutic targets, we evaluated the activity of the Bromodomain and Extra-Terminal motif (BET) inhibitor GS-626510 on primary tumors and xenografts harboring c-MYC amplifications. In line with previous studies, the large majority of germline and somatic mutations were found in BRCA1/2 (21%) and TP53 (86%) genes, respectively. Among mutations in known cancer driver genes, 77% were transmitted from primary tumors to metastatic tumors, and 80% from primary to recurrent tumors, indicating that driver mutations are commonly retained during ovarian cancer evolution. Importantly, the number, mutation spectra, and signatures in matched primary–metastatic tumors were extremely similar, suggesting transcoelomic metastases as an early dissemination process using preexisting metastatic ability rather than an evolution model. Similarly, comparison of SBOC showed extensive sharing of somatic mutations, unequivocally indicating a common ancestry in all cases. Among the 17 patients with matched tumors, four patients gained PIK3CA amplifications and two patients gained c-MYC amplifications in the recurrent tumors, with no loss of amplification or gain of deletions. Primary cell lines and xenografts derived from chemotherapy-resistant tumors demonstrated sensitivity to JQ1 and GS-626510 (P = 0.01), suggesting that oral BET inhibitors represent a class of personalized therapeutics in patients harboring recurrent/chemotherapy-resistant disease.

Download Full-text

Clonal haematopoiesis in chronic ischaemic heart failure: prognostic role of clone size for DNMT3A- and TET2-driver gene mutations

European Heart Journal ◽

10.1093/eurheartj/ehaa845 ◽

2020 ◽

Cited By ~ 1

Author(s):

Birgit Assmus ◽

Sebastian Cremer ◽

Klara Kirschbaum ◽

David Culmann ◽

Katharina Kiefer ◽

...

Keyword(s):

Heart Failure ◽

Somatic Mutations ◽

Prognostic Significance ◽

Gene Mutations ◽

Amplicon Sequencing ◽

Driver Mutations ◽

Driver Gene ◽

Roc Curve Analysis ◽

Driver Genes ◽

Clone Size

Abstract Aims Somatic mutations of the epigenetic regulators DNMT3A and TET2 causing clonal expansion of haematopoietic cells (clonal haematopoiesis; CH) were shown to be associated with poor prognosis in chronic ischaemic heart failure (CHF). The aim of our analysis was to define a threshold of variant allele frequency (VAF) for the prognostic significance of CH in CHF. Methods and results We analysed bone marrow and peripheral blood-derived cells from 419 patients with CHF by error-corrected amplicon sequencing. Cut-off VAFs were optimized by maximizing sensitivity plus specificity from a time-dependent receiver operating characteristic (ROC) curve analysis from censored data. 56.2% of patients were carriers of a DNMT3A- (N = 173) or a TET2- (N = 113) mutation with a VAF >0.5%, with 59 patients harbouring mutations in both genes. Survival ROC analyses revealed an optimized cut-off value of 0.73% for TET2- and 1.15% for DNMT3A-CH-driver mutations. Five-year-mortality was 18% in patients without any detected DNMT3A- or TET2 mutation (VAF < 0.5%), 29% with only one DNMT3A- or TET2-CH-driver mutations above the respective cut-off level and 42% in patients harbouring both DNMT3A- and TET2-CH-driver mutations above the respective cut-off levels. In carriers of a DNMT3A mutation with VAF ≥ 1.15%, 5-year mortality was 31%, compared with 18% mortality in those with VAF < 1.15% (P = 0.048). Likewise, in patients with TET2 mutations, 5-year mortality was 32% with VAF ≥ 0.73%, compared with 19% mortality with VAF < 0.73% (P = 0.029). Conclusion The present study defines novel threshold levels for clone size caused by acquired somatic mutations in the CH-driver genes DNMT3A and TET2 that are associated with worse outcome in patients with CHF.

Download Full-text

Profiling the Genomic Landscape at the Early Stages of CLL: Low Genomic Complexity and Paucity of Driver Mutations in MBL and Indolent CLL

Blood ◽

10.1182/blood.v128.22.3214.3214 ◽

2016 ◽

Vol 128 (22) ◽

pp. 3214-3214 ◽

Cited By ~ 1

Author(s):

Andreas Agathangelidis ◽

Viktor Ljungström ◽

Lydia Scarfò ◽

Claudia Fazi ◽

Maria Gounari ◽

...

Keyword(s):

Deep Sequencing ◽

Research Funding ◽

Somatic Mutations ◽

Driver Mutations ◽

Whole Genome ◽

Sequencing Analysis ◽

Driver Genes ◽

Hematopoietic Stem ◽

Genomic Complexity ◽

Targeted Deep Sequencing

Abstract Chronic lymphocytic leukemia (CLL) is preceded by monoclonal B cell lymphocytosis (MBL), characterized by the presence of monoclonal CLL-like B cells in the peripheral blood, yet at lower numbers than those required for the diagnosis of CLL. MBL is distinguished into low-count (LC-MBL) and high-count (HC-MBL), based on the number of circulating CLL-like cells. While the former does not virtually progress into a clinically relevant disease, the latter may evolve into CLL at a rate of 1% per year. In CLL, genomic studies have led to the discovery of recurrent gene mutations that drive disease progression. These driver mutations may be detected in HC-MBL and even in multipotent hematopoietic progenitor cells from CLL patients, suggesting that they may be essential for CLL onset. Using whole-genome sequencing (WGS) we profiled LC-MBL and HC-MBL cases but also CLL patients with stable lymphocytosis (range: 39.8-81.8*109 CLL cells/l) for >10 years (hereafter termed indolent CLL). This would refine our understanding of the type of genetic aberrations that may be involved in the initial transformation rather than linked to clinical progression as is the case for most, if not all, CLL driver mutations. To this end, we whole-genome sequenced CD19+CD5+CD20dim cells from 6 LC-MBL, 5 HC-MBL and 5 indolent CLL cases; buccal control DNA and polymorphonuclear (PMN) cells were analysed in all cases. We also performed targeted deep-sequencing on 11 known driver genes (ATM, BIRC3, MYD88, NOTCH1, SF3B1, TP53, EGR2, POT1, NFKBIE, XPO1, FBXW7) in 8 LC-MBL, 13 HC-MBL and 7 indolent CLL cases and paired PMN samples. Overall similar mutation signatures/frequencies were observed for LC/HC-MBL and CLL concerning i) the entire genome; with an average of 2040 somatic mutations observed for LC-MBL, 2558 for HC-MBL and 2400 for CLL (186 for PMN samples), as well as ii) in the exome; with an average of non-synonymous mutations of 8.9 for LC-MBL, 14.6 for HC-MBL, 11.6 for indolent CLL (0.9 for PMN samples). Regarding putative CLL driver genes, WGS analysis revealed only 2 somatic mutations within NOTCH1, and FBXW7 in one HC-MBL case each. After stringent filtering, 106 non-coding variants (NCVs) of potential relevance to CLL were identified in all MBL/CLL samples and 4 NCVs in 2/24 PMN samples. Seventy-two of 110 NCVs (65.5%) caused a potential breaking event in transcription factor binding motifs (TFBM). Of these, 29 concerned cancer-associated genes, including BTG2, BCL6 and BIRC3 (4, 2 and 2 samples, respectively), while 16 concerned genes implicated in pathways critical for CLL e.g. the NF-κB and spliceosome pathways. Shared mutations between MBL/CLL and their paired PMN samples were identified in all cases: 2 mutations were located within exons, whereas an average of 15.8 mutations/case for LC-MBL, 8.2 for HC-MBL and 9 for CLL, respectively, concerned the non-coding part. Finally, 16 sCNAs were identified in 9 MBL/CLL samples; of the Döhner model aberrations, only del(13q) was detected in 7/9 cases bearing sCNAs (2 LC-MBL, 3 HC-MBL, 2 indolent CLL). Targeted deep-sequencing analysis (coverage 3000x) confirmed the 2 variants detected by WGS, i.e. in NOTCH1 (n=1) and FBXW7 (n=1), while 4 subclonal likely damaging variants were detected with a VAF <10% in POT1 (n=2), TP53 (n=1), and SF3B1 (n=1) in 4 HC-MBL samples. In conclusion, LC-MBL and CLL with stable lymphocytosis for >10 years display similar low genomic complexity and absence of exonic driver mutations, assessed both with WGS and deep-sequencing, underscoring their common low propensity to progress. On the other hand, HC-MBL comprising cases that may ultimately evolve into clinically relevant CLL can acquire exonic driver mutations associated with more dismal prognosis, as exemplified by subclonal driver mutations detected by deep-sequenicng. The existence of NCVs in TFBMs targeting pathways critical for CLL prompts further investigation into their actual relevance to the clinical behavior. Shared mutations between CLL and PMN cells indicate that some somatic mutations may occur before CLL onset, likely at the hematopoietic stem-cell level. Their potential oncogenic role likely depends on the cellular context and/or microenvironmental stimuli to which the affected cells are exposed. Disclosures Stamatopoulos: Novartis: Honoraria, Research Funding; Janssen: Honoraria, Other: Travel expenses, Research Funding; Gilead: Consultancy, Honoraria, Research Funding; Abbvie: Honoraria, Other: Travel expenses. Ghia:Adaptive: Consultancy; Gilead: Consultancy, Honoraria, Research Funding, Speakers Bureau; Abbvie: Consultancy, Honoraria; Janssen: Consultancy, Honoraria, Speakers Bureau; Roche: Honoraria, Research Funding.

Download Full-text

Driver somatic mutations identify distinct disease entities within myeloid neoplasms with myelodysplasia

Blood ◽

10.1182/blood-2014-03-560227 ◽

2014 ◽

Vol 124 (9) ◽

pp. 1513-1521 ◽

Cited By ~ 158

Author(s):

Luca Malcovati ◽

Elli Papaemmanuil ◽

Ilaria Ambaglio ◽

Chiara Elena ◽

Anna Gallì ◽

...

Keyword(s):

Somatic Mutations ◽

Clinical Decision Making ◽

Myeloproliferative Neoplasms ◽

Clinical Decision ◽

Driver Mutations ◽

Myeloid Neoplasms ◽

Key Points ◽

Distinct Disease ◽

Disease Entities

Key Points Different driver mutations have distinct effects on phenotype of myelodysplastic syndromes (MDS) and myelodysplastic/myeloproliferative neoplasms (MDS/MPN). Accounting for driver mutations may allow a classification of these disorders that is considerably relevant for clinical decision-making.

Download Full-text

ecTMB: a robust method to estimate and classify tumor mutational burden

10.1101/602060 ◽

2019 ◽

Author(s):

Lijing Yao ◽

Yao Fu ◽

Marghoob Mohiyuddin ◽

Hugo YK Lam

Keyword(s):

Somatic Mutations ◽

Clinical Decision Making ◽

Patient Survival ◽

Clinical Decision ◽

Statistical Framework ◽

Mutational Burden ◽

Mutation Model ◽

Tumor Mutational Burden ◽

Multiple Challenges

AbstractTumor Mutational Burden (TMB) is a measure of the abundance of somatic mutations in a tumor, which has been shown to be an emerging biomarker for both anti-PD-(L)1 treatment and prognosis. Nevertheless, multiple challenges still hinder the adoption of TMB for clinical decision-making. The key challenges are the inconsistency of TMB measurement among assays and a lack of meaningful threshold for TMB classification. We describe a powerful and flexible statistical framework for estimation and classification of TMB (ecTMB). ecTMB uses an explicit background mutation model to predict TMB robustly and consistently. In addition to the two known TMB subtypes, TMB-high and TMB-low, we discovered a novel TMB subtype, named TMB-extreme, which was significantly associated with patient survival outcome. This discovery enabled ecTMB to classify samples to biologically and clinically relevant subtypes defined by TMB.

Download Full-text

Shifts in Intra-Clonal Dynamics Rather Than Novel Mutations Are the Main Engine Driving Tumor Evolution in Relapsed CLL

Blood ◽

10.1182/blood.v118.21.284.284 ◽

2011 ◽

Vol 118 (21) ◽

pp. 284-284

Author(s):

Dan-Avi Landau ◽

Petar Stojanov ◽

Michael S Lawrence ◽

Carrie Sougnez ◽

Aaron McKeena ◽

...

Keyword(s):

Copy Number ◽

Somatic Mutations ◽

Selective Pressure ◽

Allelic Frequency ◽

Tumor Evolution ◽

Novel Mutations ◽

Number Variation ◽

Cancer Drivers ◽

Relapsed Disease ◽

Allelic Fraction

Abstract Abstract 284 Tumor evolution is a complex process, and is the biologic underpinning of disease progression, resistance to therapy and relapse. Using whole-exome sequencing (WES) of sequential samples from patients with relapsed chronic lymphocytic leukemia (CLL) treated with conventional chemotherapy, we studied genetic tumor evolution of cancer relapse. We performed WES using paired-end reads on DNA from two peripheral blood-derived CLL tumor samples at least one year apart and on germline DNA for 20 patients. Here we report the analysis of tumor exomes from the first seven patients, of whom 6 had relapsed disease after chemotherapy and one untreated patient without intervening therapy between samples. All samples had a tumor purity that exceeded 90%. Sequencing coverage was >86% of target territory, with 132x depth obtained for all samples. In total, 187 coding region mutations (124 nonsynonymous, 63 synonymous) were identified (median: 21 somatic mutations/patient; range: 10–64), not including Ig gene mutations which were >80% clonal and remained clonally stable in our cohort. We measured the abundance of specific mutations in each patient tumor to assess clonality. An allelic frequency of 0.3–0.6 likely represents heterozygous mutations in most or all tumor cells (‘clonal') while a frequency of <0.3 represents mutations in a subset of tumor cells (‘subclonal'). Overall, 118 (63%) somatic mutations were clonal, and their allelic frequency remained unchanged in the relapse samples. 65 (35%) mutations were subclonal (average allelic frequency 0.13±0.075). Ten subclonal mutations, found in 3 of 7 initial samples, evolved into clonal mutations in the relapse samples, compared with only a single opposite occurrence where a clonal mutation became subclonal (p< 0.005, FDR q<0.01). The remaining 4 of 7 tumors showed only minor shifts in allelic frequencies over time, and included the individual who did not receive chemotherapy between samples. In Patient A, a subclone with three mutations appeared to expand to become the dominant clone, with a change in allelic fraction from an average of 0.17 (0.14–0.23) to an average of 0.43 (0.41–0.46) (p<0.000001). Two of three mutations were non-silent and are likely cancer drivers: NRAS (Q61R, found in 38/38 samples in COSMIC- Catalogue of Somatic Mutations in Cancer, Sanger Institute), and a cancer related gene PLK1. The third mutation is likely a passenger mutation as it was a synonymous mutation in ADAM18. In Patient B, a subclone containing a novel, recently identified driver in CLL, SF3B1, became the dominant clone that included additional mutations in cancer-related genes, CSMD1 and KIAA1199 (change in allelic fraction from an average of 0.16 (0.12–0.18) to an average of 0.37 (0.35–0.38) (p<0.001)). In another example, Patient C, a TP53 mutation increased in allelic frequency from 0.18 in the initial sample to 0.69 in the relapse sample (p<0.005). Analysis of copy number variation (CNV) by CapSeg (a novel algorithm that examines CNV from WES) revealed this change in allelic frequency to be coupled with a ploidy change in del(17p) from 0.8 to 0.5, consistent with a loss of both alleles. Only one sample demonstrated the appearance of novel mutations with relapse (Patient C), with 19 new mutations (13 non-silent, 3 Silent) of a total of 46 appearing at relapse. All however were subclonal, and thus less likely to have driven tumor relapse. A comparison of the 10 mutations that were selected by chemotherapy to all other mutations demonstrated an enrichment in mutations seen in the COSMIC database (p<0.05), which hints at a higher proportion of cancer drivers in this set. Our ongoing analyses are focused on the association of clinical features with copy number variation and changes in gene expression. In summary, our analysis of serial exomes from seven patients provided important insights into the genetic evolution of CLL under the selective pressure of chemotherapy. We demonstrate a significant change in clonal dynamics in one half of treated patients, which suggests that relapsed disease following treatment is driven by expansion of subclones under the selective pressure of chemotherapy rather than by novel mutagenesis. This observation may have clinical implications, as it suggests that pre-treatment WES may allow not only for the delineation of current genetic abnormalities, but through investigation of subclonal mutations, may also predict genetic evolution in future relapse. Disclosures: No relevant conflicts of interest to declare.

Download Full-text

CancerGeneNet: linking driver genes to cancer hallmarks

Nucleic Acids Research ◽

10.1093/nar/gkz871 ◽

2019 ◽

Vol 48 (D1) ◽

pp. D416-D421 ◽

Cited By ~ 4

Author(s):

Marta Iannuccelli ◽

Elisa Micarelli ◽

Prisca Lo Surdo ◽

Alessandro Palma ◽

Livia Perfetto ◽

...

Keyword(s):

Large Fraction ◽

Driver Mutations ◽

Gene Products ◽

Driver Genes ◽

Cancer Driver ◽

Cancer Hallmarks ◽

Disease Phenotypes ◽

Regulatory Circuits ◽

Cancer Drivers ◽

Cancer Phenotypes

Abstract CancerGeneNet (https://signor.uniroma2.it/CancerGeneNet/) is a resource that links genes that are frequently mutated in cancers to cancer phenotypes. The resource takes advantage of a curation effort aimed at embedding a large fraction of the gene products that are found altered in cancer cells into a network of causal protein relationships. Graph algorithms, in turn, allow to infer likely paths of causal interactions linking cancer associated genes to cancer phenotypes thus offering a rational framework for the design of strategies to revert disease phenotypes. CancerGeneNet bridges two interaction layers by connecting proteins whose activities are affected by cancer drivers to proteins that impact on the ‘hallmarks of cancer’. In addition, CancerGeneNet annotates curated pathways that are relevant to rationalize the pathological consequences of cancer driver mutations in selected common cancers and ‘MiniPathways’ illustrating regulatory circuits that are frequently altered in different cancers.

Download Full-text

Mutational likeliness and entropy help to identify driver mutations and their functional role in cancer

10.1101/354324 ◽

2018 ◽

Author(s):

Giorgio Mattiuz ◽

Salvatore Di Giorgio ◽

Lorenzo Tofani ◽

Antonio Frandi ◽

Francesco Donati ◽

...

Keyword(s):

Cancer Progression ◽

Somatic Mutations ◽

Driver Mutations ◽

Cancer Evolution ◽

Loss Of Function ◽

Driver Genes ◽

Cancer Driver ◽

Cancer Genomes ◽

Passenger Mutations ◽

Mutational Processes

AbstractAlterations in cancer genomes originate from mutational processes taking place throughout oncogenesis and cancer progression. We show that likeliness and entropy are two properties of somatic mutations crucial in cancer evolution, as cancer-driver mutations stand out, with respect to both of these properties, as being distinct from the bulk of passenger mutations. Our analysis can identify novel cancer driver genes and differentiate between gain and loss of function mutations.

Download Full-text