Abstract
Background/Aim: Endocrine therapy is the standard treatment for hormone receptor-positive (HR+) breast cancer (BC). Yet, it is accompanied by treatment-related toxicities, leading to poor treatment adherence, high relapse, and low rates of survival. While pharmacogenomic variants have the potential to guide personalized treatment, their predictive value is inconsistent across published studies. Materials and Methods: To systematically assess the literature’s current landscape of pharmacogenomics of endocrine therapy-related adverse drug effects, systematic searches in MEDLINE, Embase, Cochrane CENTRAL, Google Scholar and PharmGKB databases were conducted. Results: We identified 87 articles. Substantial heterogeneity and variability in pharmacogenomic effects were evident across studies, with many using data from the same cohorts and predominantly focusing on the Caucasian population and postmenopausal women. Meta-analyses revealed Factor V Leiden mutation as a predictor of thromboembolic events in tamoxifen-treated women (p<0.0001). Meta-analyses also found that rs7984870 and rs2234693 were associated with musculoskeletal toxicities in postmenopausal women receiving aromatase inhibitors (p<0.0001 and p<0.0001, respectively). Conclusion: Overall, the current body of evidence regarding the potential role of pharmacogenomics in endocrine therapy-related toxicity in BC remains largely inconclusive. Key concerns include the heterogeneity in toxicity definitions, lack of consideration for genotype-treatment interactions, and the failure to account for multiple testing. The review underscores the necessity for larger and well-designed studies, particularly with the inclusion of premenopausal women and non-Caucasian populations.
- Systematic review
- meta-analysis
- adverse drug reactions
- side effects
- drug safety
- toxicity
- breast cancer
- endocrine therapy
- tamoxifen
- aromatase inhibitors
- pharmacogenomics
- pharmacogenetics
- personalized medicine
- review
Breast cancer (BC) is the most commonly diagnosed cancer globally (1). Seventy to eighty per cent of BC cases are hormone receptor-positive (HR+) cases, for which endocrine therapy, including tamoxifen and aromatase inhibitors (AIs), plays a pivotal role in preventing recurrence and improving survival. While AIs are preferred for postmenopausal women with HR+ early BC, both treatments significantly reduce relapse rates and increase survival rates when administered for 5-10 years (2). Despite advances, BC remains the leading cause of cancer-related death in women, largely due to recurrence and metastasis (3). Adverse drug effects (ADEs) related to endocrine therapy are key predictors of poor adherence and persistence, affecting 30%-70% of patients (4).
Efforts to enhance BC survivorship including interventions to mitigate endocrine therapy-related ADEs have been made. Emerging evidence suggests that specific genomic variations may impact the clinical toxicity outcomes of patients with BC undergoing endocrine therapy. Previous reviews on the pharmacogenomics of toxicities related to endocrine therapy have almost exclusively been narrative in nature, concentrating on individual genes or specific toxicities for a given endocrine agent. Identifying variants that predispose individuals to a wider range of toxicity outcomes from various endocrine drugs would be more clinically useful as it gives patients more options for alternating between agents. Therefore, a comprehensive systematic assessment of pharmacogenomics of endocrine therapy-related toxicities is crucial to improve the existing evidence base for both researchers and clinicians. Herein, we conducted a systematic review to provide a comprehensive analysis of the current landscape and critically evaluate the methodological quality of the studies identified.
Materials and Methods
Data sources and search strategy. We systematically searched MEDLINE, Embase and Cochrane CENTRAL databases to identify pharmacogenomic studies investigating toxicities or ADEs related to endocrine therapy in BC. We also searched Google Scholar to capture reports from journals not indexed in these databases. Our database-specific search strategies are detailed in Tables S1-S3. To ensure comprehensive coverage, we examined clinical annotations and phenotypes in PharmGKB concerning toxicities and ADEs associated with endocrine treatment (5).
In addition to peer-reviewed articles, efforts were made to include unpublished data and non-peer-reviewed studies to minimize publication bias. This involved searching for dissertations, theses and subsequent or follow-up reports from conference abstracts. In cases of missing information or the need for clarification, we reached out to original investigators and sought any subsequent unpublished results. To ensure comprehensiveness, we performed retrospective reference harvesting of influential studies. To identify studies with potentially relevant data, we reviewed the study methodology in full-text articles.
Eligibility criteria. Eligibility criteria for study inclusion were defined a priori based on the PICO framework (Population, Intervention, Comparison, Outcome), to ensure transparency in the selection process. Refer to Table I for details.
The inclusion and exclusion criteria in this study.
Study selection. The search outcomes from different databases were extracted and compiled using Mendeley. Duplicates were manually eliminated. Abstract and title screening was independently performed by two reviewers (KM, LJ), and irrelevant items were excluded. Relevant records underwent a full-text examination according to our pre-determined criteria for inclusion. Consensus decisions were made and articles not deemed relevant were excluded.
Data extraction. Key study characteristics, including sample size, year of publication, population description, study authors, study design, interventions(s), gene, genomic variant(s), toxicity outcomes and menopausal status were extracted and documented. Data for meta-analysis were collected from eligible studies.
Quality assessment. The methodological rigor of the studies was assessed using a validated 15-item checklist designed specifically for genetic research, which incorporates crucial methodological aspects vital to genetic studies (6). This checklist was adapted from the STrengthening the REporting of Genetic Association Studies (STREGA), an extension of the Strengthening the Reporting of OBservational Studies in Epidemiology (STROBE) Statement (7). To assess the overall quality of the studies included in this review, summary scores were computed.
Quantitative data synthesis and statistical analysis. Quantitative data from eligible studies were combined via meta-analysis. Consideration was given to both clinical and statistical aspects to determine appropriateness. A meta-analysis was conducted by combining studies that examined the same treatments and variants with similar measures of related toxicity outcomes and excluding those with different or tangentially related outcomes. Due to their diverse definitions, we did not meta-analyze studies on CYP2D6 genotype-predicted metabolizer phenotypes or activity scores. We discussed their findings descriptively instead. In cases where multiple studies used data from the same cohort, only the study with the largest analysis and/or longest follow-up period was included to prevent overrepresenting patient data in the meta-analysis. When necessary, effect sizes were transformed into a uniform metric for meta-analysis after being calculated using standard procedures. We employed random-effects models to address the anticipated heterogeneity resulting from the broad range of study characteristics that satisfied our eligibility criteria. The Bonferroni correction was applied to the p-values to account for multiple hypothesis testing. Forest plots were generated to visualize overall patterns. Stata/SE version 16.0 (StataCorp, College Station, TX, USA) was used for the statistical analyses. Both I2 and p-values for Q-statistic are displayed in forest plots.
Results
Eighty-seven studies fulfilled our inclusion criteria. Having screened 2,050 publications from various databases, 63 studies met the inclusion criteria. An additional 32 records were identified from PharmGKB and Google Scholar searches. Having excluded abstracts with insufficient data, 87 studies (encompassing 45,630 patients) were included with sample sizes ranging from 24 to 4,580 (8-94). The PRISMA flow chart of the systematic literature search and the selection process of studies are depicted in Figure 1.
The PRISMA flow chart of systematic literature search and selection process. Preferred reporting items for systematic reviews and meta-analyses (PRISMA) flow diagram demonstrating the screening and selection stages of pharmacogenetic studies of adverse drug effects related to endocrine therapy in breast cancer.
Aromatase inhibitors and musculoskeletal adverse effects were the most examined. Third-generation AIs were the most studied treatment modality, representing 87% of total associations (Figure S1). Musculoskeletal and vasomotor ADEs (MS-ADEs and VM-ADEs) were the most examined among the 87 studies. Forty-seven studies examined MS-ADEs (12, 14, 33-37, 39-41, 47, 48, 16, 49-52, 55, 56, 58, 60, 65, 67, 18, 69-71, 73, 76-78, 80-82, 19, 83-85, 89-92, 21, 27-29, 32), while 32 investigated VM-ADEs (8, 11, 29-31, 36, 38, 44-46, 53, 55, 13, 58, 59, 62-64, 66, 68, 72, 76, 80, 15, 87, 94, 18-20, 22, 24, 25) (Figure S2). Yet, MS-ADEs accounted for 48.4% of all associations, followed by overall toxicities, which were examined by ten studies and comprised 34.4% of the total associations (Figure S3).
Heterogeneity in definitions and measures of toxicity outcomes. It should be noted that there was considerable heterogeneity in the definitions and measures of toxicity outcomes, ranging from specific toxicity endpoints to general ADEs. Toxicity outcomes measurement methods varied widely across studies including biomarker changes, time to onset of toxicity, and treatment discontinuation due to ADEs (21, 34, 35, 61, 73, 79, 82, 85, 91). Some studies used composite outcomes or analyzed declined health-related quality-of-life and/or their impact on treatment discontinuation (18, 58, 61, 74).
Premenopausal women and non-Caucasian populations were underrepresented. Early-stage BC was the most examined, though some studies did not specify the BC stage and few analyzed data from patients with advanced or metastatic BC (10, 33, 87). Premenopausal women and non-Caucasian populations were notably underrepresented in the studies. Most associations, 89%, were in postmenopausal women, with just 0.9% in premenopausal women (Figure S4). The majority of studies (86.6%) were conducted in high-income countries, mainly in the USA and Europe, and predominantly Caucasian cohorts (Figure S5).
Most research focused on candidate gene studies. The majority of studies were candidate gene studies, with only three genome-wide association studies identified (16, 40, 91) all investigating MS-ADEs associated with AIs, with two of them using data from the same MA.27 trial (16, 40). Almost all included studies were retrospective with only one randomized genotype-guided study identified (87).
ESR1 and CYP19A1 were the most investigated. The included studies explored variants from 58 genes, with ESR1 and CYP19A1 being the most studied, representing 23% and 19% of total associations, respectively (Figure S6). Most investigations focused on single nucleotide variants (SNVs) and genotype-predicted metabolizer phenotypes, with a few examining haplotypes (11, 19, 37, 45, 51, 52, 67, 81), and deletions namely the UGT2B17 deletion (61, 70, 74, 78, 80).
Across all identified studies, most studies used germline DNA, seven studies utilized DNA samples extracted from FFPE BC tissues (10, 20, 26, 41, 45, 47, 50), and three did not explicitly specify the source of their DNA samples (11, 48, 73).
The overwhelming majority of associations were not statistically significant. Among the 87 studies, there were various toxicity outcomes, genomic variants, and genetic models, resulting in a total of 4,423 associations, the vast majority of which (94.3%) were non-significant. Approximately 10% of studies solely reported positive results, while 22 studies did not report any significant results (8, 10, 24, 30, 31, 33, 38, 40, 42, 53, 59, 61, 12, 63, 64, 69, 70, 72, 76-78, 84, 87, 14, 90, 92, 16-18, 20, 22, 23) (Figure S7). Amongst the individual endocrine agents examined, more significant associations were found for tamoxifen (43%). However, aromatase inhibitors collectively showed more significant associations (55.5%) compared to selective estrogen receptor modulators (SERMs) such as tamoxifen (Figure S1).
Notably, the majority of studies in this review overlooked potential interactions. Out of 87 studies, only 13 included genotype-treatment interactions in their analyses. Additionally, while most studies adjusted their analyses for patient risk factors, inconsistencies existed regarding which covariates to adjust for (Figure S8).
Data duplication was evident across studies. It is noteworthy that most studies were small, with 37 publications (42.5%) using data from the same clinical trial(s) or cohort(s). For instance, six studies drew data from the ELPh trial (34, 49, 57, 73, 78, 91), six studies used data from the TAM trial (13, 14, 27, 36, 86, 94), four studies used data from the MAP.3 trial (61, 70, 74, 80), three studies used data from the BIG 1-98 trial (26, 47, 50), three studies used data from the MA.27 study (16, 40, 71), three studies used data from the B-ABLE cohort (35, 51, 67), two studies used data from the TEAM trial (41, 45), two studies used data from IBIS-I trial (8, 24) and two publications used data from the same longitudinal study (32, 54). Moreover, two studies were performed in the same cohort (81, 90), and one study (58) expanded on a previous pilot study (18) for statistical power.
Key characteristics of studies with significant findings are summarized in Table II.
Characteristics of studies which reported statistically significant findings.
The included studies had overall high-quality scores. The studies generally scored high in quality with an average score of 88.24% on the STREGA and STROBE checklists. It should be noted that discrepancies were observed in reporting biases, adherence to Hardy-Weinberg equilibrium and ethnic classification across studies. Only 56 studies met bias reporting criteria, 47 reported Hardy-Weinberg equilibrium, and only 21 of the 45 studies addressing mixed ethnicities met this criterion (Table S4).
Only three meta-analyses showed a significant summary effect size. We conducted 44 meta-analyses involving 30 studies. Only three meta-analyses yielded significant results, detailed in Table III, with corresponding forest plots in (Figure 2, Figure 3 and Figure 4). Meta-analysis of four studies (8, 15, 46, 53) showed that the Factor V Leiden (FVL) mutation significantly increased the risk of thromboembolic events in tamoxifen-treated women OR=3.47 (1.95, 6.17), p<0.0001. Even when one study using a broader definition for thromboembolic events was excluded (15), the association persisted between the FVL mutation and venous thromboembolism in tamoxifen-treated patients OR=2.55 (1.13, 5.75), p=0.024 (Figure 5). Furthermore, ESR1 PvuII (rs2234693) and RANKL rs7984870 were significantly associated with MS-ADEs in postmenopausal women treated with AIs, OR=1.64 (1.25, 2.14), p<0.0001 and OR=1.45 (1.18, 1.79), p<0.0001, respectively. Due to the small number of studies in each meta-analysis (<10), funnel plots were not created, and asymmetry tests were not performed. Pooled estimates for all meta-analyses performed are provided in Table S5.
Meta-analyses with a statistically significant summary effect size.
Meta-analysis of Factor V Leiden mutation and thromboembolic events in tamoxifen-treated patients. Meta-analysis of the odds ratios (OR) and 95% confidence intervals (CI) of four studies examined Factor V Leiden mutation (rs6025) and thromboembolic events in patients taking tamoxifen. Individual and pooled odds ratios from studies are reported in the Forest plot. Squares represent study-specific effect estimates and the size of the square reflects the study-specific weight (i.e., the inverse of the variance). The diamond represents the summary effect estimate with a 95% confidence interval, and the horizontal lines indicate 95% confidence intervals.
Meta-analysis of ESR1 PuvII and musculoskeletal adverse effects in postmenopausal women treated with third-generation aromatase inhibitors. Meta-analysis of the odds ratios (OR) and 95% confidence intervals (CI) of three studies examined ESR1 PuvII (rs2234693) and musculoskeletal adverse effects in postmenopausal women taking third-generation aromatase inhibitors. Individual and pooled odds ratios from studies are reported in the Forest plot. Squares represent study-specific effect estimates and the size of the square reflects the study-specific weight (i.e., the inverse of the variance). The diamond represents the summary effect estimate with a 95% confidence interval, and the horizontal lines indicate 95% confidence intervals.
Meta-analysis of RANKL rs7984870 and musculoskeletal adverse effects in postmenopausal women treated with third-generation aromatase inhibitors. Meta-analysis of the odds ratios (OR) and 95% confidence intervals (CI) of three studies examined RANKL rs7984870 and musculoskeletal adverse effects in postmenopausal women taking third-generation aromatase inhibitors. Individual and pooled odds ratios from studies are reported in the Forest plot. Squares represent study-specific effect estimates and the size of the square reflects the study-specific weight (i.e., the inverse of the variance). The diamond represents the summary effect estimate with a 95% confidence interval, and the horizontal lines indicate 95% confidence intervals.
Meta-analysis of Factor V Leiden mutation and venous thromboembolism in tamoxifen-treated patients. Meta-analysis of the odds ratios (OR) and 95% confidence intervals (CI) of three studies examined Factor V Leiden mutation (rs6025) and venous thromboembolism in patients taking tamoxifen. Individual and pooled odds ratios from studies are reported in the Forest plot. Squares represent study-specific effect estimates and the size of the square reflects the study-specific weight (i.e., the inverse of the variance). The diamond represents the summary effect estimate with a 95% confidence interval, and the horizontal lines indicate 95% confidence intervals.
The complete list of associations between genomic variants and endocrine therapy-related ADEs is provided in (Table S6). Variants associated with ADEs were annotated with font colors: black for increased risk, green for decreased risk, and red for no significant association.
Discussion
Interventions to reduce ADEs associated with endocrine therapy are an integral part of the effort to improve BC survivorship. Research indicates that certain genetic variants can modulate the toxicity outcomes associated with endocrine therapy in patients with BC. Using a comprehensive search strategy and large-scale meta-analyses, this review provides an extensive analysis of the available evidence regarding the pharmacogenomics of endocrine therapy-related toxicities, advancing our understanding of pharmacogenomics research overall.
While the majority of identified associations were statistically non-significant, three meta-analyses yielded significant results. The FVL mutation can serve as a predictive genetic marker of thromboembolic events in patients with BC on tamoxifen, who can be switched to alternative endocrine agents or potentially targeted for monitoring to improve adherence and consequently overall survival rates. While meta-analyses showed that PuvII (rs2234693) is associated with MS-ADEs in postmenopausal women who had received AI treatment, no significant associations were observed in other studies with either MS-ADEs-related discontinuation of Exemestane and/or Letrozole under any genetic model (34) or with MS-ADEs in patients taking endocrine agents (56). Similarly, while meta-analyses demonstrated that postmenopausal women carrying RANKL rs7984870 who received AIs had an increased risk of MS-ADEs, another study did not find significant alterations in the MS-ADEs symptom cluster (73). The evidence for these two variants with regard to MS-ADEs is therefore inconsistent and larger studies are needed to provide more robust findings.
Almost all included studies were retrospective with only one randomized genotype-guided study identified (87), in which the study’s authors concluded that ADEs did not vary significantly between increased dose and regular dose arms. It is noteworthy that the study focused on common ADEs related to tamoxifen and safety was not the primary endpoint. The majority of studies were candidate gene studies, with only three genome-wide association studies. Larger GWAS studies are necessary, given the small sample sizes in the three GWAS studies included in this review compared to typical GWAS on polygenic traits.
The majority of the studies were small, with approximately half of the publications utilizing data from the same cohort(s) or clinical trial(s). The substantial overlap in participants from the same trial or cohort across studies, especially the ones that yielded statistically significant results, can introduce biases and result in overestimation of pharmacogenomic effect estimates (95). The dependence occurs when the same participant samples are used in individual studies to estimate multiple effect sizes for the same or interrelated toxicity outcomes, leading to dependent sampling errors (96).
The broad range of characteristics among the eligible studies resulted in significant heterogeneity in both methodological and clinical aspects. The substantial heterogeneity in the definitions and measures of toxicity outcomes observed across studies emphasizes the need for using standardized toxicity-related symptom measurements and definitions of toxicity outcomes. This enables accurate interpretation and facilitates the pooling of effect estimates via meta-analysis.
Overall, the studies included in this review scored high in quality. Yet, it is vital to emphasize the need for greater attention to reporting biases and Hardy-Weinberg equilibrium across studies as well as addressing mixed ethnicities when applicable. Some investigators potentially introduce outcome reporting bias by grouping heterogeneous ADEs with varying degrees of seriousness and severity together. The selective reporting of findings observed, with some studies only reporting positive results, and the incomplete outcome reporting by some authors (14, 57, 82, 86) suggest outcome reporting bias and potentially exacerbate publication bias (97, 98). Additionally, while most studies adjusted their analyses for patient risk factors, inconsistencies existed regarding which covariates to adjust for, raising concerns about the robustness of the analyses. Hence, efforts should be made to improve data dissemination, transparency, and complete reporting among investigators to facilitate further statistical analyses (99).
Furthermore, the authors of most studies have not explored the potential genotype-treatment interactions in regression analyses, which is recommended in best practice guidelines (100, 101). The adverse implications of disregarding interaction effects or effect modifications in statistical analyses are well-documented in the literature (102). Thus, caution is required when interpreting the findings from associations reported across those studies that have not incorporated interaction terms in their statistical model.
Only a small fraction, less than 1%, of the associations investigated in the studies focused on premenopausal women. However, this proportion does not correspond with the incidence of BC cases in premenopausal compared to postmenopausal women (30.9% vs. 69.1%) (103). Due to the abrupt disruption in estrogen levels associated with systemic endocrine treatment, premenopausal women with ER-positive BC are more susceptible to endocrine therapy-related ADEs (104) and thereby more likely to discontinue their endocrine therapy regimen (105). Furthermore, the dense tissue of the breasts in premenopausal women makes it more challenging for clinicians to detect issues from mammography and so BC in premenopausal women is most frequently diagnosed at a later stage with more aggressive disease (106). Thus, this population faces more intensive therapy and lower survival rates compared to postmenopausal women, underscoring the necessity for further pharmacogenomics research in this population (104). It is noteworthy that premenopausal BC in less developed countries constitutes a significantly larger proportion of all incident BC cases and deaths compared with higher-income countries (103).
Most studies were performed in high-income countries or in cohorts of predominantly Caucasian or white people. Although BC incidence is notably lower in less developed countries compared to more developed ones (107), BC remains the most common cause of cancer-related death especially in middle and low-income countries. While this is largely attributed to recurrence and metastasis, early diagnosis and access to treatment continue to be significant challenges in low-income and middle-income countries (103). Furthermore, although black female patients have a lower incidence of BC compared to white or Caucasian women, their BC-related mortality rate is 40% higher. Moreover, in contrast to White and Asian patients, Black patients exhibit a higher prevalence of overweight or obesity and are more frequently receiving chemotherapy agents such as taxanes (108). Women who reported lower adherence to treatment behaviors were also more likely to be Black (as opposed to White) and had a higher medication-related ADE burden (109). In order to improve the generalizability of findings to other under-represented ethnic groups and decrease racial disparities and health inequalities, more inclusive research using larger cohorts with more diverse ethnic populations is required (110). To address the deficiency in pharmacogenetic data among African populations, initiatives such as the African Pharmacogenomics Network (APN) are essential (111). These efforts are crucial for advancing comprehensive pharmacogenetic studies to identify variants in pharmacogenes, which could be leveraged to reduce ADEs and enhance therapeutic efficacy (112).
Finally, given that multiple factors, such as ethnicity, age, and BMI can influence inter-individual variability (113), incorporating both genetic and non-genetic determinants of ADEs associated with endocrine agents has the potential to improve the precision of predicting individual responses to these agents (114).
This review has a few limitations. First, the retrospective nature of the vast majority of included studies and relying largely on published data may introduce bias despite our comprehensive searches. Although some studies were post-hoc analyses of randomized control trials, ADEs were rarely the primary endpoints and were often analyzed retrospectively. This could have resulted in false positives and an overestimation of effect estimates. Second, limiting the search to English-only publications may introduce language bias, potentially leading to an over-representation of studies from Western countries and patients of Caucasian backgrounds in the data. Third, the meta-analyses included a small number of studies with a limited number of patients and therefore heterogeneity among the studies cannot be eliminated. Finally, the inclusion of four dissertations and theses (61, 62, 70, 74) may raise concerns due to their perceived variability in their design quality compared to journal articles. Although it has been perceived that these non-traditional sources are not generally subject to the same peer-review procedures (115), this view has been challenged (116). Dissertations and theses usually undergo rigorous appraisal, and they scored high as per our quality assessment criteria. Moreover, the inclusion of non-traditional sources helps reduce publication bias and enhances the representation of relevant research particularly in rapidly evolving fields such as pharmaco-genomics.
Conclusion
The existing evidence on pharmacogenomics in breast cancer endocrine therapy-related toxicity is largely inconclusive and should move beyond small-scale studies and post-hoc analyses of clinical trials. To improve the evidence in this context, it is crucial to conduct carefully designed research and larger cohort studies, particularly involving premenopausal women and non-Caucasian populations. This study further underscores the imperative of incorporating both hypothesis-free and expanded candidate gene approaches in future research efforts.
Acknowledgements
This study received support from the University of Exeter Sanctuary Scholarship. For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) license to any Author Accepted Manuscript version arising from this submission.
Footnotes
Data Availability
All data relevant to the present study are included in the article or uploaded as supplementary information.
Supplementary Material
To access the supplementary materials, visit: https://doi.org/10.6084/m9.figshare.25872487.v3
Conflicts of Interest
The Authors declare that they have no conflicts of interest that are directly relevant to the content of this study.
Authors’ Contributions
KM, MW, VM and LJ conceptualized the study and designed the research. KM and LJ conducted the research and graphical overviews. MW performed the statistical analysis and generated the graphs.
- Received May 21, 2024.
- Revision received June 17, 2024.
- Accepted June 18, 2024.
- Copyright © 2024 The Author(s). Published by the International Institute of Anticancer Research.
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY-NC-ND) 4.0 international license (https://creativecommons.org/licenses/by-nc-nd/4.0).
