Characterization of Cell-Free DNA Size Distribution in Osteosarcoma Patients

Osteosarcoma is the most common type of bone cancer in children and adolescents (1). More than 60% of patients with osteosarcoma are children ages 0 to 16, with the majority being between the ages of 10 and 14 (2). The mortality of osteosarcoma is estimated to be 23% in patients with localized disease and as high as 74% in patients with distant metastasis (3). Even with a greater understanding of the molecular profile of osteosarcoma, patient outcomes have not changed in decades (4). Most patients with osteosarcoma who fail to respond to chemotherapy develop pulmonary metastasis, which is a significant cause of death, within 1.3 years after initial diagnosis (5).

The analysis of cell-free DNA (cfDNA), a fragment of DNA released into the bloodstream by normal and cancer cells during the natural process of cell death, has been actively investigated as a valuable tool for cancer management. A growing body of research has revealed distinct nongenetic properties of cfDNA released by cancer cells, including the size distribution of cfDNA fragments (6). The circulating tumor DNA (ctDNA), cfDNA fragment derived from tumor cells, is reported to be shorter than in nontumor cells. The length of cfDNA fragments in plasma from healthy volunteers is generally observed at approximately 167 bp (7), whereas the cfDNA obtained from patients with cancer is more fragmented compared with healthy donors (8). The regulation of chromatin remodeling and epigenetic modification has been proposed as a major mechanism contributing to the generation of various lengths of cfDNA fragments in cancer (9). Hypomethylation is a common epigenetic alteration found in cancers across the genome, resulting in cancer cells with more open chromatin structure than normal cells (10). As a result, during apoptosis, nucleosome-bond cancer DNA becomes more accessible to endonucleases, producing more alternative cleavage sites, resulting in shortened circulating tumor DNA. This phenomenon resembles the primary mechanism governing fetal cfDNA production in pregnant women (9). Recently, genome-wide analysis of cfDNA fragmentation has demonstrated a unique fragmentation profile of cfDNA among various cancer types with a high sensitivity (>90%) and specificity (98%) for the detection of breast, colorectal, gastric, lung, ovarian, and pancreatic cancers (11). Furthermore, the size of cfDNA fragment has been proposed as a predictive factor in a variety of cancers. Shortened fragments lower than 167 bp have been linked to poor clinical outcomes in patients with advanced pancreatic cancer, breast cancer, and renal cell carcinoma (12–14).

In this study, we aim to examine the size distribution of cfDNA detected in patients with osteosarcoma compared with healthy controls as well as the relationship between the length of cfDNA fragments in patients with osteosarcoma and poor prognosis. The distribution of the sizes of cfDNA fragments was evaluated by an automated high-resolution capillary electrophoresis platform. We then analyzed an association of cfDNA size profiles in osteosarcoma patients versus healthy donors to assess the sensitivity and specificity for diagnosing patients with osteosarcoma. Furthermore, we performed statistical analysis to define the best cut-off of cfDNA length for assessing the prognostic power in the osteosarcoma cohort. Finally, we validated the source of short cfDNA detected in patients with osteosarcoma through an analysis of whole genome sequences of representative cfDNA and paired tissue samples.

Biological samples from a total of 65 individuals were comprised in this study: 15 healthy donors and 50 patients with osteosarcoma. Fifteen healthy donor samples were gathered from volunteers at Chiang Mai University (CMU). The inclusion criteria for healthy donors included (i) healthy donors with very good physical and psychologic status (normal BMI, no debilitation, no under-nutrition, no jaundice, nonsmoking, no or minimal alcohol use, and no metal instability); (ii) free of medical conditions; (iii) not receiving permanent medication; (iv) no use of oral contraceptives; (v) no current pregnant; (vi) no chronic illness; and (vii) not recovered from an infectious disease in the previous 2 weeks. For patients with osteosarcoma, a retrospective cohort study was conducted from January 1, 2012, to June 1, 2021, at Maharaj Nakorn Chiang Mai Hospital under the Suandok Repository Unit (SRU) System, the official human biobank at Chiang Mai University (Ethics approval code ORT-2563–07122). The blood samples from 50 cases of patients diagnosed with osteosarcoma were collected during the diagnosis prior to the treatment. Clinicopathologic characteristics, including Enneking staging, tumor volume, percentage of tumor necrosis, and survival status, were retrieved from CMU hospital records and pathology reports (Table 1). All subjects supplied written informed consent for participation in compliance with the Declaration of Helsinki. All experimental procedures were approved by the Research Ethics Committee, Faculty of Medicine, CMU (Ethics approval code FAC-MED-2564–08180).

A 10 mL whole peripheral blood sample was collected from each patient with K2-EDTA tubes. Plasma was isolated within 2 hours after blood collection using a centrifugation protocol; centrifuged at 1,600 × g, 10 minutes to separate plasma and followed by high-speed centrifugation at 6,000 × g for 10 minutes to remove leukocytes and cell debris. The aliquots of plasma were stored at −80°C until processed for cfDNA extraction. Freeze–thaw cycles should be avoided. cfDNA was extracted from 2 mL of plasma using the QIAamp Circulating Nucleic Acid Kit (Qiagen) with optimized manufacturer's protocols and eluted in 20 μL of TBE buffer. Tumor-derived gDNA from fresh-frozen tissue and blood DNA (PBMC) were extracted using the salting-out chloroform extraction method. All purified DNA was quantified by Nanodrop2000 (Thermo Fisher Scientific).

The cfDNA fragment size was determined using the QIAxcel Advanced System with a QIAxcel DNA High Resolution Kit, according to the manufacturer's instructions. The cfDNA fragment size pattern was analyzed by QIAxcel screengel software and displayed as an electropherogram. The cfDNA fragment size and concentration were calculated based on the marker 100 bp to 2.5 kb and the alignment marker 15 bp/3 kb. The average main peak of each sample was defined by an independent three-time analysis. For survival analysis, we detected the cfDNA peaks, including main peaks and subpeaks (secondary peaks), to stratify patients with osteosarcoma. Samples with at least one cfDNA peak (main peaks and subpeaks) shorter than or equal to 150 bp are considered positive, whereas samples with all cfDNA peaks longer than 150 bp are considered negative (Supplementary Fig. S1). The cfDNA size distribution was measured in triplicate for each patient; individuals with cfDNA peaks shorter than or equal to 150 bp from two out of three measurements were assigned as positive.

The whole-genome library preparation and sequencing workflow was performed by the Macrogen Corporation. The cfDNA whole-genome sequencing libraries were prepared from 10 ng of cfDNA using the TruSeq Nano DNA (350) Kit, according to the manufacturer's instructions. Briefly, cfDNA was processed without further fragmentation or size selection. The amplification and barcoding were performed after adapter ligation by light PCR cycles (6–10 cycles). The 1 μg of gDNA extracted from matched tumor tissue and blood control samples (PBMC) were used for sequencing library preparation. cfDNA library fragments were sequenced utilizing 150-bp paired-end reads on Novaseq 6000 platforms (Illumina) at 14× coverage. Concurrently, gDNA library fragments were deep sequenced at 30× and 60× coverage for PBMCs and tissue samples, respectively.

The following sequence data were analyzed using a recommendation suggested Genome Analysis Toolkit (GATK) pipeline: the quality of FASTQ files was checked by the FASTQC (RRID:SCR_014583; ref. 15). After the removal of contaminating adapter sequences, paired-end sequence reads were aligned to the human reference genome (GRCh38) using BWA-mem (RRID:SCR_010910; ref. 16). Format conversion and de-duplication were marked using MarkDuplicates (Picard Tools), and these were excluded from downstream analysis along with reads of low mapping quality and supplementary alignments. The Mutect2 variant caller was used to identify somatic SNV and Indel mutations (17). Genomic DNA from matched whole blood was utilized to pair to discriminate between somatic and germline abnormalities. Somatic variant calls with less than 1% mutant allelic frequency in the matching blood control sample but at least 1% allelic frequency and three reads validating variant alleles in tumor samples were accepted.

The final formula has been developed, where t is the segment mean values of each segment bin, p belongs to the proportion of each segment bin, and α represents the overall CNV signal throughout the whole genome.

For statistical analysis, STATA version 16 (Serial No. 501606204774) and GraphPad Prism version 9.4.1(GPS-1145384-ELPE-5E7B2; RRID:SCR_002798) were used in this study. The amount of cfDNA concentration and average fragment size between patients with osteosarcoma and healthy controls were compared by the Wilcoxon Mann–Whitney U test. The optimal cut-off points for cfDNA level and cfDNA fragment size were defined using ROC curve analysis. To estimate the association of cfDNA level, size, and overall survival, Kaplan–Meier survival curves with log-rank tests were applied. Although the univariate and multivariable analyses were determined using Cox proportional hazards regression. For all statistical tests, P-value < 0.05 was considered to be statistically significant.

The data generated in this study are publicly available in NCBI BioProject database ID PRJNA917431.

To investigate the cfDNA level and cfDNA fragment size pattern, 50 patients with osteosarcoma and 15 healthy donors were included in this study, in which the osteosarcoma cases had either locally advanced stage IIb (n = 38), or metastatic tumors or stage III (n = 12).

We demonstrated that the plasma cfDNA concentration in patients with osteosarcoma was significantly higher than in healthy donors (Fig. 1A). The cfDNA level in stage III patients was statistically greater than in stage IIb patients. The median cfDNA level of healthy controls was 6.9 ng/mL plasma (range 2.25–17.85 ng/mL plasma; IQR, 4.4), whereas the median cfDNA concentration was 25.1 ng/mL plasma (range 7.1–665 ng/mL plasma; IQR, 20.2) in patients with osteosarcoma with stage IIb, and 53.3 ng/mL plasma (range 17.9–300.4 ng/mL plasma; IQR, 51) in stage III patients. Because of the size range limitation of the DNA ladder, capillary electrophoresis was unable to determine the total cfDNA level in 2 of 50 patients with osteosarcoma.

By using automated capillary-based electrophoresis, we found different size distributions of cfDNA fragments from patients with osteosarcoma and healthy controls (Supplementary Fig. S2). The size of cfDNA fragments was significantly shorter in patients with osteosarcoma than in healthy controls (Fig. 1B). The median cfDNA fragment size distribution was 177 bp (range 174–184 bp; IQR, 5.0) in 15 healthy controls, 165 bp (range 128–175 bp; IQR, 6.8) in stage IIB, and 151 bp (range 116–177 bp; IQR, 30.3) in stage III (Fig. 1B). Furthermore, the length of cfDNA fragments differed significantly between stage IIb and stage III patients.

Numerous studies found that increasing plasma cfDNA concentrations were related to the size of tumors (20–22). In this study, we also found that the cfDNA level in patients with a high tumor burden was significantly higher than the median cfDNA level in patients with low tumor burden (Fig. 1C). The median cfDNA level in patients with high tumor burden (>500 cm³) was 42.8 ng/mL plasma (IQR, 141.3; 18.3–665 ng/mL plasma), whereas the median cfDNA level in patients with a low tumor burden (≤500 cm³) was 28.3 ng/mL plasma (IQR, 19.6; 7.1–95 ng/mL plasma). However, no correlation between the tumor mass and the size of cfDNA fragments was observed (Fig. 1D).

These findings revealed a considerable difference in cfDNA levels and fragment size distribution between patients with osteosarcoma and healthy controls, as well as across different stages of disease.

The diagnostic performance of cfDNA tested in our cohort was examined by comparing cfDNA levels or fragment size of patients with osteosarcoma with healthy controls. We performed ROC curve analysis to identify the best cut-off of the level and the fragment size of cfDNA. The cfDNA level at a cut-off value of 13.9 ng/mL plasma showed a sensitivity of 85% and a specificity of 93% to diagnose patients with osteosarcoma (area under the ROC curve; AUC, 0.955; 95% CI, 0.908–1.000; P-value < 0.0001).

The analysis of cfDNA fragment size at a cut-off value of 173 bp demonstrated the strong predictive power of cfDNA size distribution in differentiating osteosarcoma cases from healthy donors with a high specificity of 100% and a sensitivity of 94% (AUC = 0.987; 95% CI, 0.967–1.000, P value < 0.0001). The combination of cfDNA levels and sizes in condition “or” was used to identify patients and healthy donors. All volunteers who met either the cut-off level or fragment size were classified as patients with osteosarcoma. The integration of cfDNA with level or fragment size enhanced diagnostic efficiency to sensitivity of 100%; and specificity 100% (AUC, 1.000; P value < 0.0001; Fig.1E).

Furthermore, the results are consistent when we performed subgroup analysis, with 94% sensitivity of stage IIb and 91% sensitivity of stage III, respectively (Fig.1F). Overall, these findings suggest that the analysis of cfDNA fragment size distribution is applicable for diagnosing patients with stages IIB and III disease.

All patient characteristics data are summarized in Table 1. ROC curves were used to define cut-off points for cfDNA levels and fragment size; cfDNA levels >30 ng/mL plasma; peak of short cfDNA fragments ≤150 bp (Fig. 2A). An example of the cfDNA peak identification has been provided in Supplementary Fig. S1.

The prognostic value of cfDNA analysis was performed in 50 cases of patients with osteosarcoma. Overall, patient survival ranged from 0.7 to 111.3 months (median = 27.8) after initial diagnosis. Their 1- and 5-year survival rates were 76% and 50%, respectively.

Univariate analysis of overall survival demonstrated that the patients with peak of short cfDNA fragments (≤150 bp) had a significantly shorter overall survival rate (HR, 5.092; 95% CI, 1.19–21.68; P = 0.028; Table 2; Fig. 2B), whereas the plasma cfDNA level had no significant value with overall survival (HR, 1.55; 95% CI, 0.68–3.51; P = 0.293; Table 1; Fig. 2C). The 5-year survival rates of the patients with peak of short cfDNA fragments were 36.1%, whereas the 5-year survival rates of the patients with long cfDNA fragments were 81.1%. Furthermore, we found that levels of cfDNA were not different between the group of patients with short and long cfDNA fragments (Fig. 2D).

Multivariable analysis demonstrated that short cfDNA fragments (HR, 9.03; 95% CI, 1.13–72.2; P = 0.038) and advanced stages (HR, 4.00; 95% CI, 1.26–12.67; P = 0.019) were independent indicators of poor prognosis (Table 2). These findings indicate that a short cfDNA fragment is a potential prognostic marker for osteosarcoma.

Following the finding that the size of cfDNA fragments could have an impact on predicting the prognosis of patients with osteosarcoma, we further validated whether these small cfDNA fragments are mainly derived from cancer cells. The mutation profiles of plasma cfDNA samples derived from 3 patients with osteosarcoma, including OS07, OS08, and OS10, were examined using a whole-genome sequencing platform. Following Mutect2 variant calling, the cfDNA fragments carrying somatic variants (mutant cfDNA) was categorized from wild-type cfDNA (nonmutant cfDNA) using Jvarkit (Biostar322664; RRID:SCR_002580). The length distribution of cfDNA fragments was compared between mutant and nonmutant cfDNA. The results showed that the majority of mutant and nonmutant cfDNA from OS07 and OS10 samples had a main peak at 167 bp, whereas the OS08 sample had a main peak at 148 bp (Fig. 3). Notably, short cfDNA fragments, ranging from 140 to 160 bp, contained a higher proportion of mutant cfDNA fragments than nonmutant cfDNA fragments in all samples. These findings indicate that short cfDNA fragments might be a major source of tumor-derived cfDNA.

According to several accumulative studies, the average cfDNA fragment length released from noncancerous cells is 167 bp (7, 23–25). In addition, we found that 150 bp of cfDNA was an independent prognostic marker for patients with osteosarcoma. Therefore, to enrich tumor-derived cfDNA fragments feature, we applied an in silico size selection approach to sort cfDNA length at a cut-off of 150 bp for structural variant analysis (Fig. 4A).

The CNV signals of the OS07 sample were undetectable, with an estimated TF of 0.14 (Fig. 4B). Using in silico size selection of short cfDNA (≤150 bp length), we detected an increased CNV signal up to 0.55 with a TF of 0.35, whereas the signal was decreased in cfDNA longer than 150 bp. Similarly, in OS10, the TF and CNV signals were greatly improved after short cfDNA selection. Interestingly, the OS08 sample, which contained a large proportion of 148 bp cfDNA, exhibited the strongest TF and CNV signals compared with the OS07 and OS10 samples, even in the nonfiltering experiments. Further, in silico size selection of 150 bp of cfDNA resulted in higher CNV signals of 0.94 and a TF of 1.0.

We demonstrated that in silico size selection significantly increased the amplitude of detectable CNV signal by 2.3-fold compared with without size selection (P = 0.03; Fig. 4C). Furthermore, we found that the concordance rate of CNV detected in cfDNA compared with their matched tumors was significantly improved in short cfDNA fractions. Particularly in the OS07 sample, the rate of concordance was elevated to 77% in short cfDNA with less than or equal to 150 bp length (Fig. 4D). The results also showed a high concordance percentage of 95% between the short cfDNA of OS08 and OS010 samples and their matched tumors. Additional CNV spots were detected on short cfDNA fragments in all three samples, in which eight distinct CNV spots were detected in short cfDNA of OS08 samples.

These findings indicate that short cfDNA fragments with less than or equal to 150 bp are major sources of genetic alterations in osteosarcoma blood samples. The enrichment of short cfDNA fragments could potentially improve the sensitivity of detecting osteosarcoma mutations on a genome-wide scale.

In this study, we demonstrated that the length of cfDNA fragments in patients with osteosarcoma was significantly shorter and more fragmented than in healthy individuals. Similar findings have been reported in a wide range of cancer types, but for the first time in osteosarcoma, to our knowledge (6, 11, 12). The results showed a high-performance diagnostic value of cfDNA fragment size analysis at a cut-off of 173 bp, with a sensitivity and specificity of 94% and 100%, respectively. In comparison with cfDNA levels, the cfDNA size analysis had a higher performance for differentiating osteosarcoma from healthy donors. The average cfDNA level in healthy people was 7.7 ng/mL of plasma, which is consistent with previous studies that reported the cfDNA levels in healthy individuals are less than 10 ng/mL of plasma and rarely exceed 30 ng/mL of plasma (26, 27). The average of 48 osteosarcoma cases increased to 67.8 ng/mL of plasma, which is the same range as previously reported in patients with osteosarcoma, from 7.2 to 1130.2 ng/mL (28, 29). By using the cfDNA level cut-off at 13.9 ng/mL plasma, we found seven cases of osteosarcoma with false-negative results and one false-positive healthy donor. Notably, all seven cases were from stage IIb patients, implying that the cfDNA level may be limited to advanced osteosarcoma. Remarkably, combining cfDNA level or fragment size increases diagnostic sensitivity and specificity up to 100%.

The length of cfDNA fragments is related to nucleosome occupancy, in which the core of the nucleosome is wrapped by 147 bp of DNA string with 20 bp of linker DNA, yielding a 167 bp cfDNA fragment in healthy people (6). In this study, we observed a slightly distinct median size of cfDNA fragments at 177 bp in healthy donors using an automated electrophoresis. Similar findings were reported in other studies using a microfluidics automated electrophoresis-based platform, in which the size of cfDNA was measured to be between 164 and 185 bp (14, 30, 31). We found that an automated electrophoresis-based technology can effectively differentiate healthy from patients with osteosarcoma with 100% specificity, in which the cfDNA fragments in 50 cases of patients with osteosarcoma are in a range of 116 to 177 bp compared with 174 to 184 bp in healthy donors. With three independent experiments, this approach can precisely quantify the length of cfDNA with a mean deviation of 3 bp. The consistency between the results of cfDNA measurement using automated electrophoresis and whole-genome sequencing further verified the power of automated electrophoresis in identifying the size distribution of cfDNA (Supplementary Fig. S3).

Our findings clearly demonstrated that a shortened cfDNA fragment size of 150 bp was an independent prognostic marker even when we assessed cfDNA fragment size in the presence of the other key clinical factors such as tumor volume, stage, and chemoresponse in the regression model. Patients with short cfDNA fragments (≤150 bp) had a worse overall survival rate (Fig. 2E). No significant difference was found in our analysis using continuous variables instead of the cfDNA peak cutoff of 150 bp (data not shown). We further demonstrated that short fragments lower than 150 bp were a source of cfDNA released by cancer cells and a source of genetic alterations, which may be related to the aggressiveness of osteosarcoma. Peneder and colleagues has recently shown a relationship between short cfDNA fragments and epigenetic regulation using region-set enrichment analysis (LOLA software; ref. 29). Short/long (S, 100–150 bp/L, 151–200 bp) ratio analysis of cfDNA fragments across the genome in patients with Ewing sarcoma (EwS) revealed that regions with short cfDNA fragments (higher S/L ratios) were enriched for EwS-specific open chromatin, showing peaks of enhancer-associated histone H3K27 acetylation in EwS tumors and EwS-specific DNase I hypersensitive sites.

Furthermore, we found that in silico size selection of short cfDNA fragments potentially enriched tumor-derived cfDNA with CNV mutations that resemble parental osteosarcoma tissues. The size selection for 150 bp cfDNA enhanced tumor cfDNA fraction detection and increased CNV signals up to 2.3-fold compared with nonfilter cfDNA detection. The strong CNV signals were not detected when cfDNA fragments longer than 150 bp were enriched, showing a relationship between short cfDNA fragments and genetic aberrations. Several studies have shown that under genetic-based analysis, the fraction of cfDNA fragments shorter than 150 bp increased proportionately to the detected tumor cfDNA in various types of cancer (32–37), supporting the importance of shortened cfDNA fragments as a source of genetic alterations. The approaches for in vitro and in silico cfDNA size selection have been applied to improve the detection of tumor-derived cfDNA in patients with melanoma, ovarian cancer, colorectal cancer, and cholangiocarcinoma (33). The diagnostic performance AUC increased to 0.9 after 90 to 150 bp cfDNA size selection, compared with AUC 0.69 from baseline genetic detection. Smith and colleagues also demonstrated that the strategy of in silico cfDNA size selection at 90 to 150 bp enhanced the detection of tumor-derived cfDNA detection sensitivity in patients with renal tumors from 8.3% to 85% (38). Sorting of 150 bp cfDNA fragments also increased the detection signal in the plasma of patients with ovarian cancer after chemotherapy treatment by 6.4×. Furthermore, the analysis of size-selected cfDNA allowed more accessibility to identify clinically actionable mutations (33). For instance, the enrichment of short cfDNA fragments in the range of 90 to 150 bp significantly improves the detection of tumor cfDNA, especially in the low-variant allele frequencies, including EGFR exon 20 p. T790M, BRAF exon 15 p.V600, KRAS exon 2 p.G12, and KRAS exon 2 p.G13 codons in multiple types of cancers, including melanoma, colorectal adenocarcinoma, and pancreatic ductal adenocarcinoma (32). According to our findings, a germline or tumor profile is not required for cfDNA size analysis by automated capillary electrophoresis. Furthermore, using our pipeline, TF estimation based on CNV analysis can be performed without the use of germline or tumor data. This makes the cfDNA size distribution test very attractive for a wide range of applications.

Our findings provide evidence for the notion that the identifying size distribution patterns have a high potential for being used as diagnostic and prognostic indicators for osteosarcoma. The cfDNA size analysis approach applied in this study might be an attractive tool for screening, diagnosing, as well as monitoring patients with osteosarcoma with precision, high throughput, and cost-effectiveness. The current findings are based on a sample size of a limited population. Further validation of these findings in independent cohorts is required. A multicenter validation cohort might be warranted for our follow-up studies.

Furthermore, we proposed that in silico cfDNA size selection is a promising strategy for enhancing sensitivity for detecting tumor-derived cfDNA in osteosarcoma. An in-depth investigation into the area of fragmentomics of cfDNA might bring new knowledge of osteosarcoma biology and contribute to the development of novel therapies.

No author disclosures were reported.

S. Udomruk: Conceptualization, data curation, formal analysis, investigation, methodology, project administration, validation, visualization, funding acquisition, writing–original draft, writing–review and editing. A. Phanphaisarn: Data curation, formal analysis, writing–review and editing. T. Kanthawang: Resources, investigation, writing–review and editing. A. Sangphukieo: Formal analysis, software, validation, writing–review and editing. S. Sutthitthasakul: Formal analysis, methodology, validation. S. Tongjai: Resources, validation, writing–review and editing. P. Teeyakasem: Resources, data curation. P. Thongkumkoon: Formal analysis, validation, writing–review and editing. S. Orrapin: Investigation, validation, writing–review and editing. S. Moonmuang: Investigation, validation, writing–review and editing. J. Klangjorhor: Validation, writing–review and editing. A. Pasena: Resources, data curation. P. Suksakit: Resources. S. Dissook: Formal analysis, validation. P. Puranachot: Software, writing–review and editing. J. Settakorn: Resources. T. Pusadee: Resources. D. Pruksakorn: Conceptualization, supervision, resources, validation, funding acquisition, writing–review and editing. P. Chaiyawat: Conceptualization, supervision, methodology, funding acquisition, project administration, resources, validation, visualization, writing–original draft, writing–review and editing.

This research project was supported by Fundamental Fund 2023, Chiang Mai University, the Faculty of Medicine, Chiang Mai University, and Genomics Thailand Project, Health Systems Research Institute (HSRI), Thailand. We would like to thank the patients and their parents for participating in this study. Figures 2E and 4A were created with BioRender.com.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.