This systematic review provides an overview of hypothesis-generating pharmacogenetic studies in osteosarcoma patients of the last 10 years. In addition, replication studies of all top-hit associations of the studies were identified in a structured manner to give a more complete idea of the evidence that is present. Treatment response and doxorubicin-induced cardiotoxicity are the most extensively studied phenotypes. Chemotherapy-induced nephrotoxicity, hepatotoxicity and bone marrow toxicity were examined, but only in small patient cohorts. The influence of genetic variants on cisplatin-induced ototoxicity was not investigated in any of the hypothesis-generating studies.
The only GWAS that was performed in relation to treatment response to chemotherapeutic treatment in osteosarcoma patients was the GWAS by Koster et al. (40). In that study, two variants in in the GLDC gene were found to be associated to overall survival. As previously mentioned, Lin et al. successfully replicated the association of rs55933544 with decreased overall survival in their patient cohort (76). Noteworthy, the TT genotype of this variant was not related to GLDC expression, but it was associated to lower expression of interleukin-33 (IL33) (40, 76). On top of that, Kang and colleagues genotyped common variants in the IL33 gene and found that the A-allele of rs1048274 was associated to survival in a osteosarcoma patient cohort of Chinese ancestry (77). The patient cohort of Koster et al. consisted of European and Brazilian subjects and they did not identify statistically significant associations with common variants in IL33. Differences in linkage disequilibrium structures between populations allowed for different variants on the same locus to be associated to survival of patients with osteosarcoma, indicating that not GLDC, but IL33 is causal for decreased survival through the variant. This emphasizes that studying populations of different ethnicities helps in fine mapping the genetic background that causes a phenotype. IL33 was previously associated to prognosis in other cancers (78–80) and it is known to have pro- and anti-tumorigenic properties mediated through immune cells (81). In osteosarcoma, IL33 plays a role in osteosarcoma cell viability in in vitro experiments mediated through the PI3K/AKT pathway (82, 83). However, the exact role and the effects of genetic variants remains to be found.
A pathway approach including genes linked to osteosarcoma treatment was used by most of the included studies, so consequently members of the ABC transporter family were included in the pharmacogenetic investigations and sometimes found to be associated to the outcomes of interest. These genes code for membrane-bound proteins which participate in the movement of most drugs and their metabolites across cell surface and cellular organelle membranes. Defects in these genes can be important in terms of cancer therapy and pharmacokinetics (84). As indicated in the results of this review, ABCC2 variants were repeatedly associated to toxicities in these studies, however Figure 2 shows that these associations were scarcely replicated. Doxorubicin and methotrexate are both transport substrates for ABCC2, which caused these variants to be studied, but this does not necessarily explain the causative functional background of the association that is observed. On the other hand, in the pathway approach study by Caronia et al. it was found that per T-allele of the ABCC3 rs4148416 variant, patients have an 8-fold higher risk of death, and 6 times lower risk on event-free survival, and this association was consistently replicated in two other cohorts (27–29). ABCC3 codes for multidrug resistance protein 3 (MRP3) and is an important transporter of bile salts, but is also involved in efflux of methotrexate from liver and kidney cells (85, 86).
The necessity of routine MTX plasma concentration measurement during treatment with MTX has allowed several research groups to study genetic variation involved MTX pharmacokinetics. This was not one of the clinical outcome measures of interest of this systematic review, and therefore these studies were excluded as shown in Figure 1, however, genetic variants that are associated to high MTX plasma levels may also give increased risk for toxicity and genetic variants that are associated to low MTX plasma levels may also predispose to a suboptimal treatment response. Lui et al. found three variants localized in ABCG2 to be associated with methotrexate clearance in patients with osteosarcoma, namely rs13120400, rs13137622, rs12505410 (26). Rs13120400 was the most significant variant, and the CC genotype of this variant was previously also associated to increased response to methotrexate in psoriasis patients (87). In addition, in the study by Hegyi et al., which included 59 osteosarcoma patients, ABCG2 rs2231142 was found to be significantly associated with a longer half-life time of methotrexate (55). However, this variant is not in LD with variants identified by Lui et al.. ABCG2 codes for the breast cancer resistance protein (BCRP) and has an important role in the transport of methotrexate out of the liver and kidney and knockdown of ABCG2 increased the bioavailability of methotrexate in mice (86, 88). Both ABCC3 and ABCG2 may be of interest for further investigation with a larger patient cohort to relate it to both pharmacokinetic parameters and clinical outcomes of treatment.
Six variants were found to be associated with doxorubicin-induced cardiotoxicity in a cohort containing osteosarcoma patients and were replicated minimally once in an independent patient cohort, namely CELF4 rs1786814, HAS3 rs2232228, RARG rs2229774, SLC22A17 rs4982753, SLC22A7 rs4149178, SLC28A3 rs7853758. The associations with HAS3 rs2232228 and CELF4 rs1786814 identified by Wang et al., consisted of gene-environment interactions, which means the variant effect is larger in patients that received a higher dose of doxorubicin. This emphasizes the importance of doxorubicin dose in the development of cardiotoxicity and in the effect size of genetic variant. CUGBP Elav-like family member 4 (CELF4) is involved in splicing of TNNT2, which codes for cardiac troponin T (cTnT). cTnT plays a role in Ca2+ signaling and contraction of the heart muscle and is a biomarker for myocardial damage (89). Whereas the embryonal TNNT2 splicing variant, carrying an additional exon 5, is usually downregulated in adults, patients with CELF4 rs1786814 GG genotype express both the adult and embryonal TNNT2 splicing variant. This results in a temporally split myofilament response to calcium, decreasing the ventricular pumping efficiency, and thereby increasing the risk on dilated cardiomyopathy and cardiotoxicity (52, 90). In addition, pathogenic variants in TNNT2 are an established cause of hypertrophic and dilated cardiomyopathy (91). Adult and pediatric patients with cancer who developed chemotherapy-induced cardiomyopathy have an increased prevalence of pathogenic variants in sarcomere genes compared to controls, indicating that genetics involved in susceptibility to cardiomyopathy, such as mutations in sarcomere genes, may also be of importance in doxorubicin-induced cardiotoxicity (92). HAS3 encodes for hyaluronan which is a component of the extracellular matrix and is involved in tissue remodeling after cardiac damage. In addition, hyaluronan reduces cardiac injury caused by ROS, which is an important element of doxorubicin-induced cardiac damage. Furthermore, the RARG gene codes for retinoic acid receptor gamma and binds to the topoisomerase IIβ (Top2b) promotor to repress its expression. Top2b is a target of doxorubicin mediated DNA damage, and if Top2b expression is low in cardiac tissue due to repression by RARG, the tissue is less susceptible to damage caused by doxorubicin (54). Despite that replication studies of the association of the RARG variant with cardiotoxicity were inconsistent, a functional study in iPSC-derived cardiomyocytes showed that the variant RARG increases sensitivity to doxorubicin-induced cardiomyopathy (93). Lastly, three variants in genes of the solute carrier transporter family were associated to doxorubicin-induced cardiomyopathy. A variant downstream of SLC22A17 was associated to cardiotoxicity. SLC22A17 is ubiquitously expressed, also in the heart, and plays a role in iron transport and homeostasis. Accumulation of iron in mitochondria can cause doxorubicin-induced cardiotoxicity, however the exact role of SLC22A17 in this process is not studied (94). Secondly, SLC22A7 encodes for the organic anion transporter 2 (OAT2), which is highly expressed in liver and kidney and is known to play an important role in clearance of medicines but is not previously indicated in transport of cisplatin, doxorubicin or methotrexate. Lastly, a variant in SLC28A3 was associated with cardiotoxicity. SLC28A3 codes for the sodium-coupled nucleoside transporter 3 (CNT3). Only for CNT3, it was established to transport doxorubicin, indicating it may play a role in doxorubicin pharmacokinetics (95). Despite that this was not shown for OAT2, it does have considerable overlap in substrates with CNT3 and transports several nucleoside-based drugs, for example 5-fluorouracil and zidovudine (96).
While cisplatin-induced ototoxicity is one of the most prevalent adverse effects of cisplatin treatment, it was not investigated in any of the studies that were identified in this systematic review. The most recent GWAS on cisplatin-induced ototoxicity that included patients with osteosarcoma, was performed in 2009 and was therefore excluded from this review (97). In this study, Ross et al. identified genetic variants in the TPMT and COMT gene that are associated to cisplatin-induced hearing loss in pediatric patients with cancer. However, replication studies are very contradictive, as shown in a meta-analysis by Thiessen et al. in 2018 (17, 18). Whereas this was the only GWAS that included patients with osteosarcoma, work from other patient cohorts treated with cisplatin may be applicable to patients with osteosarcoma too. Xu et al. identified that the A-allele of ACYP2 rs1872328 gives an increased risk to cisplatin-induced ototoxicity in a pediatric brain tumor cohort (14). In 2020, Clemens et al. showed in a meta-analysis of 5 studies with a total of 1418 pediatric patients that this association was also found in cohorts containing patients with osteosarcoma (OR (95%CI) = 3.94 (1.04-14.93), p = 0.04). In addition, they showed a significant association of SLC22A2 rs316019 with cisplatin induced hearing loss in a meta-analysis of 4 studies (OR (95%CI) = 1.46 (1.07-2.00), p = 0.02). Whereas these results are significant, the heterogeneity between studies remained an obstacle (I2=66% for ACYP2 rs1872328 and 44% for SLC22A2 rs316019). In conclusion, in order to find variants associated with cisplatin-induced ototoxicity, the priority would be to perform a GWAS with large patient cohorts to find reliable results. In the meantime, ACYP2 rs1872328 and SLC22A2 rs316019 could be studied further to find out their true potential for clinical practice.
In general, the quality of reporting data in the included studies was good, as shown with the quality assessment using STREGA guidelines. However, there was minimal reporting on follow-up and missing data. All data in the studies was collected retrospectively and there was no prospective follow-up and, therefore, it is sensible that nothing was reported on patients that were lost to follow-up. It was also not reported how missing data was handled in statistical analysis or what kind of analysis was used for the main comparison regarding the missing data. However, it is assumed that these studies did a complete case analysis, because that is most conventional in retrospective genetic association studies. Therefore, this does not compromise the quality of these articles.
The aim of this review was to identify variants that were discovered to be associated to a phenotype in a hypothesis-free manner and might form the basis for future pharmacogenomic studies in osteosarcoma. The most evident method to identify articles that describe these variants would be to limit the systematic search to GWA studies, as they are the textbook example to hypothesis-free research. However, the number of GWASs is limited in pediatric oncology cohorts and even more so in osteosarcoma cohorts. In addition, when a broad range of ADME genes is studied, it does not restrict itself to only genes that are previously indicated in the pathway of a drug. In order to include all literature with a hypothesis-free component, a boundary has been set to articles that assess more than 10 variants in more than 5 genes. The smallest studies in this review, by Hattinger et al., Hegyi et al., and Windsor et al. (45, 29 and 36 variants in cohorts of 57, 59 and 58 osteosarcoma patients), showed the poorest reproducibility (Table S8, S9, S10) (31, 32, 55). Possibly, due to the small number of patients in these studies, the power was too low to perform multiple testing correction causing the authors to report false-positive findings.
Not only in the case of small studies, but also in larger studies, replication remains laborious. Our structured search shows that many replication studies have been performed but the majority do not confirm the findings from the discovery study with significant results (see figure 2). Obvious explanations for this are the general heterogeneity between cohorts, different ethnicities, phenotypes and treatment regimens. In addition, power in discovery studies is often too low to correct for multiple testing, leading to false-positive findings. Insufficient power in replication studies may also stand in the way of confirming true-positive findings. As a solution for that, meta-analysis of the discovery and replication cohorts could increase the total power, however in this review there was too much heterogeneity between studies perform reliable meta-analyses. To study treatment response different outcome parameters, such as overall survival, event-free survival, disease-free survival, progression-free survival, histological response and tumor necrosis were used, making it impossible to combine in a meta-analysis. For toxicity outcomes, there were large differences in grading systems and the exact definitions for which patients are considered cases. In pediatric patient cohorts, cardiotoxicity is defined in as fractional shortening below a limit that varies among studies. In adults, cardiotoxicity is defined as a decrease of ejection fraction below the lower limit of normal or a large absolute reduction in ejection fraction, making it impossible to combine pediatric and adult patient cohorts in meta-analysis. The introduction of more sensitive imaging tools may allow for earlier detection of cardiac dysfunction and homogenize phenotypes, for example using global longitudinal strain (GLS), 3D volumetric echocardiographic or MRI. Lastly, confidence intervals of discovery and replication studies in this review do not always overlap, and therefore, it is already to be predicted that meta-analyses will be very heterogenous. In the future, the consistency of phenotypes would be improved if research groups with similar interests would collaborate to clearly define the phenotypes together, in order to perform powerful analyses in larger patient cohorts.