Genome-wide study identifies novel genes associated with bone toxicities in children with acute lymphoblastic leukemia

Abstract

Bone toxicities are common among pediatric patients treated for acute lymphoblastic leukemia (ALL) with potentially major negative impact on patients’ quality of life. To identify the underlying genetic contributors, we conducted a genome-wide association study (GWAS) and a transcriptome-wide association study (TWAS) in 260 patients of European-descent from the DFCI 05–001 ALL trial, with validation in 101 patients of European-descent from the DFCI 11–001 ALL trial. We identified a significant association between rs844882 on chromosome 20 and bone toxicities in the DFCI 05–001 trial (P = 1.7×10⁻⁸). In DFCI 11–001 trial, we observed a consistent trend of this variant with fracture. The variant was an eQTL for two nearby genes, CD93 and THBD. In TWAS, genetically predicted ACAD9 expression was associated with increased risk of bone toxicities, which was confirmed by meta-analysis of the two cohorts (meta-P = 2.4×10⁻⁶). In addition, a polygenic risk score (PGS) of heel quantitative ultrasound speed of sound was associated with fracture risk in both cohorts (meta-P = 2.3×10⁻³). Our findings highlight the genetic influence on treatment-related bone toxicities in this patient population. The genes we identified in our study provide new biological insights into development of bone adverse events related to ALL treatment.

Graphical Abstract

Genome-wide association study identified a significant association between rs844882, located near THBD and CD93 on chromosome 20, and bone toxicities in the DFCI 05–001 trial. THBD and CD93 may contribute to the development of fracture and osteonecrosis in patients received ALL therapy through their regulation by glucocorticoid.

Introduction

Bone toxicities, including fracture and osteonecrosis, occur at a high rate among pediatric patients treated for acute lymphoblastic leukemia (ALL). In the Dana-Farber Cancer Institute (DFCI) ALL consortium 05–001 trial, 25% of patients experienced fracture and 10% experienced osteonecrosis during or after treatment(1). These adverse events can have detrimental impact on patient’s quality of life, especially when considering the active behavior typical in this age group. Therefore, identification of patients at high risk of bone toxicities prior to starting ALL treatment may inform tailored close monitoring, screening, and prophylactic measures to prevent the occurrence of these adverse events.

Because bone mineral density (BMD) and related bone pathologies, including osteoporosis and fracture, have strong genetic heritability, germline genetic variants may have predictive value for these phenotypes(2). In the last two decades, a number of genome-wide association studies (GWAS) have been conducted for BMD and risk of osteoporosis and fracture, mostly in the older non-Hispanic White populations without cancer. These studies have identified many associated risk variants(3–7). Although each individual variant has only a small effect, polygenic risk scores (PGS) built to aggregate the total effect of a large number of variants may provide clinically meaningful risk stratification. This has been demonstrated in a recent study of fracture risk, which trained and tested a PGS for heel quantitative ultrasound speed of sound (SOS) for the purpose of predicting risk of osteoporotic fracture in an adult population in the United Kingdom(8).

Our group recently reported lower risk of fracture and osteonecrosis among Black and Hispanic pediatric patients treated for ALL in comparison to non-Hispanic Whites(1). Our genetic analysis revealed that the lower risk was attributed to the admixture of African ancestry in the Black and Hispanic patient groups. The findings provide evidence that genetic variations may play a role in shaping the risk of bone toxicities and contribute to the observed racial and ethnic disparities among children treated for ALL. In the present study, we investigate these research questions by performing a GWAS and a transcriptome-wide association study (TWAS) of bone toxicities in two clinical trials for pediatric ALL. Unlike previous GWAS focusing on osteonecrosis in pediatric ALL patients(9–12), we pooled fracture and osteonecrosis together in order to identify the genetics shared by these two bone toxicity events.

Methods

Patient population

Samples and clinical data were derived from DFCI ALL Consortium Protocols 05–001 (NCT00400946) and 11–001 (NCT01574274). DFCI 05–001 enrolled newly diagnosed ALL patients aged 1 to 18 years. Patients who achieved complete remission after induction therapy were randomly assigned to intravenous pegaspargase or intramuscular native E coli L-asparaginase. DFCI 11–001 included newly diagnosed ALL and lymphoblastic lymphoma patients aged 1 to 21 years who were randomly assigned to intravenous pegaspargase or calaspargase. Dexamethasone and other chemotherapy agents were identical between the two trials. The trials have been described in detail in previous publications(13–16). Treatment-related bone toxicity (fracture and osteonecrosis) was assessed in DFCI 05–001 and 11–001 as described in a previous publication(1). This study was approved by Institutional Review Board at Roswell Park Comprehensive Cancer Center.

Genotyping, data quality control, and imputation

Genotyping of DFCI 05–001 (484 patients) and 11–001 (167 patients) was performed by the Genomic Shared Resource at Roswell Park Comprehensive Cancer Center using the Illumina OmniExpress Beadchip array as described in previous publication(1). We separated samples from each cohort into different populations (European, Asian, African, Hispanic, and Other) based on trial-reported race and ethnicity (Table 1) and performed sample-level and variant-level quality control (QC) within each population. In order to increase the sample size of European population in DFCI 11–001, we included all samples with European ancestry estimate ≥ 0.9 regardless of self-reported race and ethnicity.

Table 1.

Patient characteristics of the genotyped cohort of DFCI 05–001 and DFCI 11–001 trial.

Characteristic	DFCI 05–001 (N = 449)	DFCI 11–001 (N = 166)	P
Age, y			0.68
<10	332 (74)	120 (73)
≥10	117 (26)	46 (28)
Sex			0.27
Male	252 (56)	102 (61)
Female	197 (44)	64 (39)
Final risk group *			0.93
Standard risk	248 (55)	92 (55)
High risk/Very high risk	201 (45)	73 (44)
Missing	0 (0)	1 (1)
Population †			3.32 × 10−5
European	294 (65)	94 (57)
Asian	15 (3)	6 (4)
African	20 (4)	3 (2)
Hispanic	74 (16)	20 (12)
Other	46 (10)	43 (26)
Fracture ‡			0.70
Yes	110 (24)	43 (26)
No	339 (76)	123 (74)
Osteonecrosis ‡			0.37
Yes	45 (10)	22 (13)
No	404 (90)	144 (87)

Data are No. (%).

^*The patient with missing information was excluded from the Fisher’s exact test.

^†Population was defined based on self-reported race/ethnicity. Hispanic population corresponds to patients with self-reported ethnicity "Hispanic or Latino". The remaining patients were separated into European, Asian, and African populations if the self-reported race was “White”, “Asian”, and “Black or African American” respectively. Patients with self-reported race as “Other” or “More than one race” were in the Other population.

^‡The corresponding p-values were from competing risk model for comparing time to event between the two cohorts. Death and recurrence were considered competing risks.

Samples were removed if the missing rate was >5%, the typed and reported sex did not match, there were abnormal inbreeding coefficients, or cryptic relatedness. Duplicate samples with higher sample missing rate were filtered out. A total of 449 and 166 patients from DFCI 05–001 and 11–001 respectively passed these QC steps. Population outliers were further removed from the European population by using EIGENSTRAT(17), leaving 260 and 101 European-descent patients in DFCI 05–001 and 11–001, respectively. In variant-level QC, variants were removed if the missing rate was >2%, if there were inconsistent genotypes between duplicate samples, or if variants violated Hardy-Weinberg equilibrium (P ≤ 1×10⁻⁶). The remaining samples and variants within the European population after QC were used for imputation, which was performed using the University of Michigan Imputation Server(18) with the Haplotype Reference Consortium (HRC) reference panel(19). Eagle2 was used for pre-phasing, while minimac3 and minimac4 were used for imputation of DFCI 05–001 and 11–001 respectively. We also performed imputation for all DFCI 05–001 patients that passed QC using the 1000G Phase 3 reference panel. Eagle2 and minimac4 were used for pre-phasing and imputation, respectively. Variants with imputation quality Rsq < 0.3 were excluded from further analysis.

Polygenic risk score calculation

The PGS weight file for SOS (ID: PGS000657, 21,716 variants) was downloaded from the PGS Catalog(20). The PGS weights for 25OHD (143 variants) based on summary statistics from meta-analysis of UKB and SUNLIGHT Consortium were obtained from the supplemental data of the publication(21). As both PGS were developed using genotype data from populations of European descent, we calculated PGS only for individuals of European-descent using imputation dosage from the HRC reference panel. Variants with MAF > 1% were utilized for PGS calculation.

Measurement of plasma vitamin D concentration

25-Hydroxyvitamin D2 (25OHD2) and D3 (25OHD3) were measured in 243 DCFI 05–001 patients of European-descent with available plasma samples collected at Day 18 of induction therapy using liquid chromatography-mass spectrometry performed by Heartland Assays (Ames, IA). For patients with 25OHD2 concentration below detection limit (<1.5 ng/mL), we set the 25OHD2 concentration as 0.75 ng/mL. Total 25OHD concentration was calculated by adding 25OHD2 and 25OHD3 concentration together.

Statistics

For GWAS analysis, we used regression models for the subdistribution hazard of the cumulative incidence function to relate imputation dosage of each variant to time to any bone toxicity event (fracture or osteonecrosis), with death, recurrence, second malignancy and induction failure as competing risk factors, while adjusting for age, sex, final risk group (Standard Risk vs High Risk/Very High Risk), and the principal components deemed significant by the Tracy-Widom statistic derived from EIGENSTRAT. Final risk group was included as a covariate to control for the dose and duration of glucocorticoid exposure as patients in different risk groups received different doses of glucocorticoids. Induction events, including death and/or induction failure, were considered as the competing risk events at time 0. The “cmprsk”, “survival” and “GWASTools” packages in R were used to perform single variant association analysis within the European population. Only autosomal variants that had MAF > 5% within the European population of the study cohort and had significant allele frequency difference between European and African populations based on allele counts from the Genome Aggregation Database (gnomAD)(22) (Fisher exact test P < 1×10⁻⁴) were tested. Variants with P-value < 5×10⁻⁸ in the GWAS were considered genome-wide significant.

To investigate variant associations in racial and ethnic minority patients, we performed separate analysis of the 155 non-European patients in the DFCI 05–001 trial, and tested the association between bone toxicities and variant dosage from imputation using the 1000G Phase 3 reference panel. The competing risk regression models included age, sex, and final risk group as covariates.

As the test statistic from competing risk models can be anti-conservative when the minor allele count of a variant is low, we tested the significance of association in DFCI 11–001 using permutation. We first calculated the log pseudo-likelihood ratio test statistic ${\lambda }_{LR}=-2(l_0-l)$, where $l$ and $l_0$ are the log pseudo-likelihoods from the competing risk model described with and without the variant dosage. We then permuted the phenotypes 10,000 times while leaving the covariates intact. The covariates included age, sex, final risk group and the significant principal components. For each permutation, we calculated ${\lambda }_{LR,i}$ by running the competing risk model on the permuted phenotypes and the intact covariates (with and without the variant dosage). At the end, the permutation $P=\sum _{i=1}^{10000}{I(\lambda }_{LR,i}\ge {\lambda }_{LR})/10000$. METAL was used to perform a sample size weighted meta-analysis from p-values in DFCI 05–001 and 11–001 trials(23).

Transcriptome-wide association study

Imputation dosages of variants with MAF > 5% and Rsq ≥ 0.8 in the DFCI 05–001 European population were used as input to PrediXcan(24) with a whole-blood prediction model trained in 922 whole-blood samples from Depression Genes and Networks (DGN)(25). We used a whole-blood prediction model as the genome-wide significant SNP rs844882 was a strong eQTL only in whole blood based on GTEx data(26). Predicted gene expression was tested for association with time to any bone toxicity event using the same model in GWAS (see “Statistical analysis” section above). A total of 11,525 genes were tested, yielding a transcriptome-wide significance cutoff 4.3×10⁻⁶.

Data availability

The genetic and phenotypic data of DFCI 05–001 and 11–001 are available at dbGaP under accession number phs002317.v2.p1.

Results

GWAS of bone toxicities

To identify genetic loci associated with ALL treatment-related bone toxicities (fracture and osteonecrosis), we performed a GWAS on the 260 European-descent patients from DFCI 05–001. As African ancestry was found to be protective of bone toxicities in DFCI 05–001(1), we focused our analysis on the variants whose allele frequencies (AF) were different between African and European populations, and identified a genome-wide significant locus on chromosome 20 (sub-distribution hazard ratio (SHR) = 2.86, P = 1.7×10⁻⁸ for the C allele of an imputed SNP rs844882, C allele frequency = 0.0591, Rsq = 0.9685, Figures 1 and 2, Supplemental Figure 1, see Methods). The C allele of rs844882 was also associated with higher risk of fracture (SHR = 2.64, P = 6.7×10⁻⁵) and osteonecrosis when analyzed separately (SHR = 2.83, P = 4.7×10⁻⁴) (Figure 3A). The risk C allele is less common in African population than non-Finnish European population according to the Genome Aggregation Database (gnomAD)(22) (AF = 0.0098 and 0.0552 respectively in gnomAD v3.1.2), consistent with our prior observation that African ancestry had significant protective effect on bone toxicities(1). We further evaluated the association in 155 non-European patients in DFCI 05–001. The C allele of rs844882 was associated with a suggestively increased risk of both fracture and osteonecrosis (SHR=1.47 and 6.16 respectively, P=0.59 and 0.08 respectively, Supplemental Figure 2), although the p-values were not statistically significant, likely due to small sample size and lower allele frequency (AF=0.0360, Rsq=0.7075). rs844882 was a strong eQTL in whole blood for the two nearest genes, CD93 and THBD (P = 2.2×10⁻²⁶ and 3.1×10⁻¹² respectively) based on GTEx data(26), where the risk C allele was associated with decreased levels of gene expression (Supplemental Figure 3).

Figure 1. Manhattan plot for GWAS of bone toxicities in DFCI 05–001.

The red horizontal line corresponds to genome-wide significance cutoff (5×10⁻⁸).

Figure 2. LocusZoom plot for the top GWAS locus.

The 1Mb region centered on the lead variant rs844882 was plotted using LocusZoom.js(55). LD structure was based on 1000 genomes EUR population.

Figure 3. Estimated cumulative incidence curves of fracture and osteonecrosis by rs844882 genotypes for European-descent patients in DFCI 05–001 (A) and 11–001 (B).

Patients with imputation dosage ≥1.5 were considered having genotype TT. Patients with imputation dosage <0.5 were considered having genotype CC. Patients with imputation dosage in-between were considered having genotype CT. None of the European-descent patients in DFCI 11–001 had imputation dosage <0.5. Death and relapse were competing events for bone toxicity.

To replicate our finding, we tested the association between rs844882 and bone toxicities in the DFCI 11–001 trial with 101 patients of European-descent. We observed consistent effect direction as DFCI 05–001, but the association was not statistically significant (SHR=1.23, permutation P = 0.65). Meta-analysis across the two cohorts yielded a p-value of 5.1×10⁻⁷. In DFCI 11–001, we observed a trend consistent with DFCI 05–001 that the C allele of rs844882 increased the risk of fracture (SHR = 1.78, permutation P = 0.24, Figure 3B). However, no such effect was found for osteonecrosis in DFCI 11–001 (SHR=0.54, permutation P = 0.54).

Transcriptome-wide association study of bone toxicities

In TWAS analysis of bone toxicities in DFCI 05–001, one gene ACAD9 reached transcriptome-wide significance and its expression was significantly associated with higher risk of bone toxicities (SHR = 1.36, P = 3.1×10⁻⁶, Figure 4). Although not transcriptome-wide significant, the genetically predicted expression levels of THBD and CD93, the two genes nearest to rs844882, were associated with lower risk of bone toxicities (SHR = 0.64, P = 6.5×10⁻⁴ and 3.4×10⁻³ respectively), consistent with our GWAS finding. In replication analysis in the DFCI 11–001 trial, genetically predicted expression of ACAD9 was also associated with a higher risk of bone toxicities, although it did not reach statistical significance (SHR = 2.24, P = 0.15). Meta-analysis across the two cohorts yielded a stronger association for ACAD9 (meta-P = 2.4×10⁻⁶). The genetically predicted expression of THBD and CD93 was also associated with lower risk of fracture (SHR = 0.63 and 0.59, P = 0.10 and 0.06 respectively), but not with osteonecrosis (P = 0.62 and 0.68 respectively).

Figure 4. Manhattan plot for TWAS of bone toxicities in DFCI 05–001.

The red horizontal line corresponds to transcriptome-wide significance cutoff (4.3×10⁻⁶).

Analysis of PGS with bone toxicities in DFCI 05–001 and 11–001 trials

Two selected PGS for bone health-related phenotypes, namely heel quantitative ultrasound speed of sound (SOS)(27) and circulating concentrations of total 25-hydroxyvitamin D (25OHD)(21), were calculated and tested in association with bone toxicities. In the DFCI 05–001 trial, the PGS of SOS was significantly associated with lower risk of fracture (P = 0.047) but not with osteonecrosis (P = 0.83), while the PGS of 25OHD was not associated with either bone toxicities (Table 2). Similarly, in the DFCI 11–001 trial, the PGS of SOS was significantly associated with lower risk of fracture (P = 0.01) but not osteonecrosis (P = 0.50), and the PGS of 25OHD was not associated with either bone toxicities (Table 2). Meta-analysis across the two cohorts yielded stronger association between PGS of SOS and fracture (meta-P = 2.3×10⁻³).

Table 2.

Risk of fracture and osteonecrosis by PGS in the genotyped European-descent patients of the DFCI 05–001 and 11–001 trial.

		DFCI 05–001		DFCI 11–001
Bone toxicity	PGS	SHR	P	SHR	P
Fracture	SOS	0.58	0.047	0.39	0.01
	25OHD	2.63	0.10	0.35	0.08
Osteonecrosis	SOS	1.10	0.83	0.64	0.50
	25OHD	1.43	0.66	1.02	0.98

Death, recurrence, second malignancy and induction failure were considered competing risks.

Adjusted covariates included age, sex, and final risk group.

SHR: subdistribution hazard ratio.

Significant P-values (≤ 0.05) were in bold.

Association of plasma 25OHD concentration with bone toxicities in DFCI 05–001

As vitamin D is essential for bone health and THBD and CD93 are under transcriptional regulation by vitamin D receptor(28–31), we examined plasma total 25OHD and 25-Hydroxyvitamin D3 (25OHD3) concentration in relation to bone toxicities in DFCI 05–001. Both total 25OHD and 25OHD3 levels were significantly correlated with the PGS of 25OHD (Pearson correlation coefficient = 0.26 and 0.25 respectively; P = 3.6×10⁻⁵ and 7.2×10⁻⁵ respectively). However, neither total 25OHD concentration nor 25OHD3 concentration associated with bone toxicities (Supplemental Table 1).

Discussion

Our GWAS for ALL treatment-related bone toxicities in DFCI 05–001 identified a genome-wide significant locus near CD93 and THBD. The effect C allele of the top variant rs844882 conferred risk for both fracture and osteonecrosis. Interestingly, as the rs844882 C allele is more rare in African population than European population, the associations with this single variant were consistent with our prior finding that African ancestry was protective of both fracture and osteonecrosis in DFCI 05–001. Although not statistically significant, we observed a consistent trend of the rs844882 C allele increased risk of both fracture and osteonecrosis in DFCI 05–001 patients of non-European populations. rs844882 is a strong eQTL in whole blood for both CD93 and THBD with the risk C allele associated with reduced expression of both genes. Consistently, we found higher predicted gene expression of CD93 and THBD associated with lower risk of bone toxicities in TWAS of DFCI 05–001 even though they did not reach transcriptome-wide significance. In the small replication cohort DFCI 11–001, we observed a trend of the rs844882 C allele being a risk factor for fracture, albeit not statistically significant. In addition, predicted expression of both CD93 and THBD was associated with lower risk of fracture at borderline significance in DFCI 11–001.

We acknowledged the lack of statistical significance in our replication study, which was likely due to the small sample size of DFCI 11–001 and the heterogeneity between the two DFCI cohorts. Out of the 101 European-descent patients in DFCI 11–001, only 34 and 15 had fracture and osteonecrosis respectively. As rs844882 has relatively low frequency (MAF ~5%), it is more susceptible to unstable risk estimates due to limited sample size. There were also differences in asparaginase formulation and dosing between the two DFCI protocols. In DFCI 05–001, patients were randomized to receive either weekly intramuscular injections of the shorter acting L-asparaginase or intravenous pegaspargase every two weeks. In DFCI 11–001, patients were randomized to receive intravenous pegaspargase every two weeks or intravenous calaspargase every three weeks. Overall, patients in DFCI 11–001 likely had greater exposure to asparaginase activity, which potentially could have potentially impacted the results. Although treatment arms did not affect fracture and osteonecrosis outcomes in both DFCI 05–001 and 11–001 (13, 16), asparaginase formulation and dosing may modify the genetic effect by influencing dexamethasone pharmacokinetics. However, our study was not powered to investigate such interaction effects.

Thrombomodulin, encoded by THBD, forms a complex with thrombin to activate protein C that functions as a physiologic anticoagulant, and to convert thrombin’s function from procoagulant to anticoagulant(32). Therefore, our findings suggest that increased THBD expression, resulting in increased anticoagulant activity, may lead to improved blood flow to the bone and a subsequent reduction in the risk of fracture (and osteonecrosis). Importantly, THBD is a primary target gene of glucocorticoid receptor(33). Its expression is regulated by glucocorticoids in a dose-dependent manner: low-dose dexamethasone treatment (100 nM) was reported to induce THBD expression(33), while high-dose dexamethasone treatment (1 mM) inhibited THBD expression(34). ALL patients are treated with high-dose glucocorticoids, which is considered a predominant risk factor to fracture and osteonecrosis(35–37). Our finding supports a possible mechanism that high-dose glucocorticoid administration results in bone toxicities through THBD inhibition (Figure 5). More studies are warranted to further investigate the hypothesis. If proved true, preventive approaches such as monitoring thrombomodulin levels as well as anticoagulant prophylaxis can be tested for minimizing bone toxicities during ALL treatment.

Figure 5. Schematic illustrating potential mechanisms contributing to fracture and osteonecrosis development in patients received ALL therapy.

The other gene in our GWAS locus, CD93, is involved in intercellular adhesion and in the clearance of apoptotic cells. Recent studies revealed CD93 plays important roles in vascular angiogenesis(38–42). CD93 overexpression was found to drive immature and dysfunctional vasculature within solid tumors(38, 40), and CD93 knockdown inhibited retinal angiogenesis and reduced normal endothelial sprouting during angiogenesis(39, 40). CD93 expression also responds to glucocorticoid treatment in a dose-dependent manner: CD93 expression in a B-ALL cell line was increased 16–24 hours after low-dose dexamethasone treatment (50 nM)(43), while CD93 expression was down-regulated in the paraventricular nucleus of the hypothalamus from male rats under high-dose corticosterone exposure for 12 days(44). We speculate that high-dose glucocorticoid therapy suppresses CD93 expression and impedes normal vascular angiogenesis, resulting in impaired bone turnover and subsequent bone toxicity events in ALL patients (Figure 5). Future studies are needed to further validate THBD and CD93’s involvement in bone toxicities under high-dose glucocorticoid treatment.

The genes identified in our GWAS do not coincide with those identified in the four previous GWAS of osteonecrosis in pediatric ALL patients (NWD2(9); BMP7 and PROX1-AS1(10); GRIN3A(11); ACP1(12)). This may be due to differences in phenotype definition (our GWAS included fracture in addition to osteonecrosis) and patient ancestry (our GWAS was restricted to European ancestry, whereas the previous GWAS included patients of multiple ancestries). In addition, heterogeneity in the study population (e.g., treatment medications, cumulative dose, and administration schedule) likely contributes more to the absence of common findings, given that the genes among previous GWAS were also distinct from one another(9–12). Despite the lack of shared genes, we noticed that both our genes and those from previous GWAS are involved in angiogenesis. NWD2 is a paralog of AAMP, which promotes endothelial cell migration and angiogenesis(45). BMP7 is known to induce angiogenesis(46–49), whereas GRIN3A inhibits angiogenesis(50).

Our TWAS in DFCI 05–001 found higher expression of ACAD9 significantly associated increased risk of bone toxicities. Meta-analysis across DFCI 05–001 and DFCI 11–001 resulted in an even stronger association. ACAD9 is essential for proper assembly of mitochondrial respiratory chain complex I and also plays a role in fatty acid oxidation(51). ACAD9 deficiency can cause muscle weakness, heart problems, and intellectual disability. It remains to be investigated how ACAD9 contributes to bone toxicities from ALL treatment.

We observed higher value of SOS PGS in significant association with lower risk of fracture in both DFCI 05–001 and DFCI 11–001, which suggested that patients genetically predisposed to higher bone density were protected from ALL treatment-related fracture. This observation supports the commonly held belief that low bone mineral density increases fracture risk(35–37, 52, 53) (Figure 5). Furthermore, our observation suggests that this PGS could be utilized to identify patients who are particularly susceptible to fractures during or after treatment, thereby necessitating the implementation of a personalized preventive strategy.

Although vitamin D is known to be beneficial for bone health, we did not find significant association between either PGS of vitamin D or measured plasma vitamin D concentration and bone toxicities in DFCI 05–001. Consistent with our finding, measured vitamin D concentration was not associated with bone toxicities in DFCI 11–001, either(54).

In conclusion, our study revealed that genetics plays a role in bone toxicities caused by ALL treatment among pediatric patients. The novel genes identified in this study provide additional insight on how high-dose glucocorticoid therapy may impact bone toxicity events and provide clues for developing new preventative strategies to minimize cancer treatment-related bone toxicity risk. Additional validation studies are warranted to demonstrate the impact of our findings on other ALL treatment regimens, considering the variability in chemotherapy schedules and dosages across pediatric ALL clinical trials.