Pediatric Chordoma: A Tale of Two Genomes

Abstract

Little is known regarding the genomic alterations in chordoma, with the exception of loss of SMARCB1, a core member of the SWI/SNF complex, in poorly differentiated chordomas. A TBXT duplication and rs2305089 polymorphism, located at 6q27, are known genetic susceptibility loci. A comprehensive genomic analysis of the nuclear and mitochondrial genomes in pediatric chordoma has not yet been reported. In this study, we performed whole exome and mitochondrial DNA (mtDNA) genome sequencing on 29 chordomas from 23 pediatric patients. Findings were compared with that from whole genome sequencing datasets of 80 adult skull base chordoma patients. In the pediatric chordoma cohort, 81% percent of the somatic mtDNA mutations were observed in NADH complex genes, which is significantly enriched compared to the rest of the mtDNA genes (p=0.001). In adult chordomas, mtDNA mutations were also enriched in the NADH complex genes (p<0.0001). Furthermore, a progressive increase in heteroplasmy of non-synonymous mtDNA mutations was noted in patients with multiple tumors (p=0.0007). In the nuclear genome, rare likely germline in-frame indels in ARID1B, a member of the SWI/SNF complex located at 6q25.3, were observed in five pediatric patients (22%) and four patients in the adult cohort (5%). The frequency of rare ARID1B indels in the pediatric cohort is significantly higher than that of the adult cohort (p=0.0236, Fisher’s exact test), but they were both significantly higher than that in the ethnicity-matched populations (p<5.9e-07 and p<0.0001174, respectively).

Implications:

germline ARID1B indels and mtDNA aberrations appear important for chordoma genesis, especially in pediatric chordoma.

Introduction

Chordoma is a rare type of sarcoma that arises from remnants of the embryonic notochord within the axial skeleton that may be located anywhere along the spine, from the skull base to the sacrum. The incidence has been reported to be approximately one per 1 million in the USA, yet only five percent of chordomas occur in pediatric patients.[1] Generally, three types of chordomas are recognized based on histologic features: conventional, poorly differentiated, de-differentiated. Children are more likely to have a chordoma located at the skull base, making surgical resection challenging, and complete resection frequently impossible.[1] Residual tumor is often treated with high-dose radiation therapy which can cause significant morbidity. Chordomas have been reported to have an aggressive clinical course in young patients, with a median progression-free survival of just 28 months.[1, 2] An improved understanding of the disease pathophysiology is required to develop more effective therapeutic approaches.

Poorly differentiated pediatric chordoma carries a dismal prognosis and has been associated with loss of SMARCB1, a subunit of the SWI/SNF chromatin remodeling complex. Germline or somatic driver mutations of other chordoma subtypes have yet to be clearly defined.[3, 4] Tarpey et al. performed whole-genome and whole-exome sequencing of 104 sporadic chordomas which revealed potential driver mutations in several candidate genes, including CDKN2A, LYST, PIK3CA, PTEN, SETD2, PBRM1and ARID1A.[5] SETD2 is a chromatin-remodeling gene, as are PBRM1 and ARID1A, which also encode subunits of the SWI/SNF complex. The hypothesis that these genes are potential drivers of tumorigenesis is supported by the Le et al. study that also implicated CDKN2A, PTEN and SETD2 as the genes most frequently affected by copy number alterations in chordoma.[6] Notably, Bai et al. performed whole-genome sequencing (WGS) of 80 skull base chordomas and also identified frequent alterations in PBRM1, in addition to mutations in SETD2 and CDKN2A.[7] Finally, Mattox et al. analyzed cell-free DNA (cfDNA) from 32 chordoma patients, and demonstrated frequent mutations in PBRM1 and ARID1A as well as in PTEN and TP53.[8] These studies suggest that although the landscape of driver somatic mutations seems to be diverse, deleterious mutations in chromatin-remodeling genes in general, and SWI/SNF complex genes specifically, are frequently altered in differentiated chordomas, as well as poorly differentiated pediatric chordoma.

Even though loss of SMARCB1 protein expression is a defining feature of poorly differentiated pediatric chordoma, germline mutations in SMARCB1 or any other cancer predisposition locus have yet to be found. A duplication at 6q27, which contains the TBXT gene, was first shown to segregate with affected individuals in four multiplexed families with three or more chordoma cases.[9] Subsequently, a common SNP in the TBXT gene, rs2305089, was found to be strongly associated with an increased risk for chordoma.[10] The association was replicated in a study of familial and sporadic chordomas by Kelly et al. [11], and later in a study in Korean patients by Sa et al. [12] and a study of Iranian patients by Jalessi et al. [13]. An association of the SNP, rs2305089, with the diagnosis of chordoma was also reported by Bettegowda et al. [14] However, other than the reported differential expression of the TBXT gene between chordomas and notochords [15], it is unclear how the duplication of the TBXT gene, and the very common missense variant rs2305089 (MAF = 0.489 according to gnomAD v4.0.0) could be causal for this extremely rare cancer. A recent study by Yepes et al. identified 34 potentially germline pathogenic variants in a number of candidate genes and pathways from 138 chordoma patients of European ancestry and the previously mentioned 80 Chinese skull base chordoma patients. [7] [16] These variants nonetheless seemed to lack a common biological mechanism as seven were loss-of-function variants and 27 were missense variants.

More importantly, genomic studies of chordoma have been limited to the nuclear genome. In our previous studies, we established that mitochondrial DNA (mtDNA) mutations contribute to tumorigenesis of multiple pediatric cancers.[17] Among pediatric central nervous system (CNS) tumors, high-grade gliomas had a significantly higher number of mtDNA mutations than low-grade gliomas.[18] One of two chordoma patients had three loss-of-function (LoF) mutations, two in MT-ND4 and one in MT-CO2. Both chordoma cases harbored the same stop-gain mutation in MT-ND4, m.10971G>A. Notably, Erlandson et al. and Murad et al. reported the altered ultrastructure of mitochondrial-associated endoplasmic reticulum membrane as a distinct feature of chordomas.[19, 20], implicating a role for altered mitochondrial function in this disease.

Most notably, the patients in previous chordoma genomic studies were largely adults. There have been few genomic studies of pediatric chordomas, likely due to their rare occurrence in the population. Although they account for just five percent of all chordomas, they are more likely to be associated with an underlying genetic risk factor compared to adult chordomas. In the present study we therefore performed combined exome and mtDNA genome sequencing of a multi-institutional cohort of pediatric chordomas. These data were compared with both nuclear germline DNA variants and mtDNA variants using the tumor-normal paired whole genome sequencing data reported by Bai et al.[7] as a comparator.

Materials and Methods

Pediatric chordoma cohort.

We performed whole-exome sequencing (WES) and mitochondrial DNA (mtDNA) genome sequencing of 29 distinct tumors from 23 patients. The genomic study of these de-identified tumor samples was approved by the Children's Hospital Los Angeles (CHLA) Institutional Review Board (CHLA-19-00336). Frozen tumor tissue DNA was extracted using the Puregene Core Kit A from Qiagen (Hilden, Germany). Sequencing libraries were generated using the Agilent SureSelect Human All Exon V6 kit plus mtDNA genome capture kit. Libraries were sequenced using Illumina Nextseq 500. Raw FASTQ data files are being submitted to dbGaP under the study name “Whole Exome and mtDNA Genome Sequencing of Pediatric Chordomas” (PI: Gai).

Variant analysis.

Since most of the pediatric cohort patients had only tumor samples available, we employed three independent methods that minimize the need for patient-matched normal controls in order to distinguish between somatic and germline origin of detected variants. The Dragen Somatic Small Variant Caller v3.9.3 was used to process mapped and aligned reads to the human reference genome GRCh38 that includes the rCRS mitochondrial reference genome, identifying SNVs and indels. This is achieved through local de novo assembly of haplotypes within active regions. For tumor-only variant calling, the Dragen pipeline can mark potential germline variants in the INFO field as “Germline Status” using population databases. This is facilitated by the “—vc-enabl-germline-tagging” flag.

Additionally, SGZ (Somatic-Germline-Zygosity) represents a computational approach for determining the origin (somatic vs. germline) and zygosity status (homozygous, heterozygous, or sub-clonal) of variants identified through massively parallel sequencing (MPS) in cancer specimens. [21] SGZ makes these determinations by modeling the allele frequency (AF) of each alteration, considering factors such as tumor content, tumor ploidy, and local copy number.

Lastly, the UNMASC method incorporates data from public databases that catalogue canonical somatic and germline variants. It also utilizes information on adjacent positions from normal controls. [22]

For each nuclear variant we studied, we only reported those confirmed by all three variant callers. If all three callers agreed on the “germline” or “somatic” status of a variant, we labeled it as either 'Germline' or 'Somatic.' When only two callers agreed, we labeled the variant as 'Likely Somatic' or 'Likely Germline.'

Nuclear variants were annotated using the Ensembl Variant Effect Predictor (VEP) GRCh38 v109. We selected rare nuclear DNA variants based on the following criteria: a Variant Allele Frequency (VAF) ranging from 0.05 to 0.95 in the sequenced samples, and an allele frequency less than 0.005 in the population as determined by gnomAD (version 4.0.0). Only rare coding and splice variants were considered. Upstream regulatory and downstream variants were excluded from our analysis.

Mitochondrial DNA variants were also annotated using the VEP GRCh38 v109. Germline mtDNA variants and somatic variants were discriminated analytically in pediatric cancer patients, germline variants were nearly always homoplasmic and frequently seen in the reference population. Somatic or tumor-only mtDNA mutations, however, were almost exclusively heteroplasmic and rarely reported in the reference population.[25] In this study, we report the mtDNA variants that a) were heteroplasmic and b) were observed in less than 1% of population as a heteroplasmic or homoplasmic variant according to gnomAD (v.4.0.0)

Skull base chordoma cohort.

Whole-genome sequencing (WGS) data of 93 chordomas and matched WGS data of the normal tissues of 80 skull based chordoma patients were obtained from the database of Genotypes and Phenotypes (dbGaP) (Accession #: phs002301.v1.p1). Germline nuclear DNA variants were called using the Illumina Dragen germline pipeline v3, and somatic mtDNA variants were called using the Illumina Dragen somatic pipeline v3. mtDNA variants with variant allele frequency (VAF) >95% were excluded given that they likely represent germline variants. Variants with frequency <3% were excluded to eliminate low VAF variants that were potential false-positive calls of mtDNA somatic mutations.

Statistical analysis.

A variety of techniques were used to determine statistical significance. The Kruskal-Wallis test was used to compare continuous variables in three different ways. Fisher's exact test and χ² tests were used to assess the independence of categorical variables. Tableau and the package maftools in R (version 4.1.2) were used to visualize groups of mutations (version 2021.4.11). Age and the number of variations were correlated using Pearson's correlation coefficients. Every test of the hypothesis was two-sided.

Data Availability.

Whole-exome sequencing data of the 29 chordomas from 23 pediatric patients have been submitted to the Database of Genotypes and Phenotypes (dbGaP) under the study “Whole Exome and mtDNA Genome Sequencing of Pediatric Chordomas”.

Results

Mitochondrial genome results: the NADH complex genes are enriched for mutations in chordoma.

Genes with frequent mtDNA mutations in pediatric chordoma cohort.

The pediatric chordoma cohort included 29 distinct tumors from 23 patients of three ethnic groups (Europeans, East Asians, Mixed Americans) with an average age of 10.5 years (Table 1). These included 23 primary tumors and six recurrent tumors from six academic medical centers. We sequenced the mtDNA genome of the 29 samples as part of our WES platform as previously described.[17, 18] We obtained a median mtDNA coverage of 1670X for all tumor samples in the pediatric chordoma cohort. Of 681 mitochondrial variants, 83% were homoplasmic, of which 96% were known variants. In contrast, of 115 heteroplasmic variants, only 53% were known variants and only 18% were annotated as common heteroplasmic variants based on the gnomAD database.

Table 1.

Pediatric chordoma cohort overview. * Patients' ancestry data was not directly collected. Instead, their ancestry genotype was identified using the somalier algorithm on whole exome sequence data (https://github.com/brentp/somalier). This process categorized patients into five super populations: Africans (AFR), Admixed Americans (AMR), East Asians (EAS), Europeans (EUR), and South Asians (SAS).

Age (yrs)	Average	10.48
	Standard deviation	6.06
Sex	Male	n=9
	Female	n=14
Specimen	FFPE	n=3
	Frozen	n=26
Tumor location	Back/Spine	n=6
	Brain/Skull	n=23
Recurrence	Primary Tumor	n=23
	Recurrent Tumor	n=6
	Normal	n=1
Ancestry genotype *	Admixed Americans	n=8
	East Asians	n=5
	Europeans	n=10
Total	Patients	n=23
Total	Samples	n=29

To assess the likely functional consequences of these mutations, we therefore investigated the somatic mutations grouped by the mitochondrial complex that the proteins belong to. In the pediatric chordoma cohort, 78% of all heteroplasmic mutations in the mitochondrial protein coding genes were observed in three genes MT-ND5, MT-ND1, and MT-ND4 which are all part of mitochondrial complex 1, also known as the NADH complex (Figure 1). The enrichment of non-synonymous variants in the NADH complex was significant (p = 0.001). Other notable mtDNA mutations included a nonsense mutation in MT-ND5, a frameshift mutation in MT-ND6, and a nonsense mutation in MT-CYB. In contrast, synonymous mutations were randomly distributed among the mitochondrial complexes (Supplementary Data).

Figure 1.

Oncoplot displaying non-synonymous heteroplasmic mtDNA variants in the pediatric chordoma cohort. Mitochondrial variants are color coded based on VEP variant classifications. Dot size corresponds to variant allele frequency. The right side displays the percentage of samples in which each variant is observed. The number of variants per sample is shown at the bottom.

Strikingly, we observed that non-synonymous mtDNA mutations frequently arose at the same genomic locus, analogous to hotspot mutations that accumulate in cancer driver genes (Supplementary Figure 1). The most frequent recurrent mutation was a single transitional T>C mutation at position 10983. This missense variant which was observed in 32% of chordoma patients resulted in a Leu75Pro mutation in MT-ND4. Both gnomAD v4.0.0 and MitoMap reported this variant at 0% frequency in all populations. An A>C transversion at mtDNA position 13651 was observed in 23% of patients. This mutation resulted in a Thr439Pro amino acid change in MT-ND5. The frequency of this variant is reported to be less than 0.0001 in East Asian, Latino and European populations and 0% for all other populations (gnomAD v4.0.0). The third most frequent mtDNA mutation was m.503A>C. This mutation was located within the D-Loop region, but outside of the hyper-variable clusters. The variant was observed in less than 0.000002 of the population (gnomAD v4.0.0).

Frequent mtDNA mutations in skull base chordoma cohort.

To test the generalizability of the mtDNA findings in the pediatric chordoma cohort, we mined the published dataset of paired tumor/normal WGS of 80 Chinese patients with 93 skull base chordoma tumors.[7] We analyzed mtDNA mutation signatures in this mostly adult cohort (Figure 2). As in the pediatric cohort, tumors were significantly enriched for heteroplasmic non-synonymous mutations in the NADH complex (p<0.0001), which included 5 frameshift mutations in MT-ND1, MT-ND2, MT-ND4, and MT-ND5. Again, in contrast to nonsynonymous alterations, synonymous variants were randomly distributed among the mitochondrial complexes.

Figure 2.

Oncoplot displaying heteroplasmic non-synonymous mtDNA variants in the skull base chordoma cohort from Bai et al.

While some of the heteroplasmic mtDNA mutations were only found in the tumors, and not in the matched normal samples of the skill base chordoma cohort, some heteroplasmic mtDNA mutations were found in both. Variant allele frequencies, i.e., heteroplasmy level, of all heteroplasmic mtDNA mutations in tumor and matched normal samples were hence evaluated, including in patients with multiple tumor samples. A progressive increase in variant allele frequency or heteroplasmy among missense mtDNA mutations was noted; however, the same effect was not observed for synonymous mtDNA heteroplasmic mutations (Figure 3, Supplementary Table 2).

Figure 3. Variation in Mitochondrial DNA Allele Frequencies Across Multiple Samples per Patient.

Lines connect identical variants found in different samples from the same patient, illustrating the dynamic nature of variant frequencies over time or across conditions. Dots denote unique variants present in only one sample per patient. Color indicates the mitochondrial gene affected. For selected variants, line labels provide variant IDs. Panel A: missense variants, Panel B: synonymous variants. strike signifies a statistically significant difference (p < 0.05) in VAFs. Statistical method: One-Way ANOVA, pairwise comparison; Pairwise comparison: Games-Howell post-hoc test; p.value adjustment for multiple comparisons: Holm–Bonferroni method.

Nuclear genome results: rare in-frame indels in ARID1B are potential predisposition factors for chordomas.

Rare ARID1B in-frame indels are frequently observed in pediatric chordoma.

The pediatric chordoma cohort consisted of 23 primary tumors and six recurrent tumors from six academic medical centers (Table 1). The genetic ancestries of these patients were determined to be Admixed Americans (8), East Asians (5), and Europeans (10). For one of the patients (patient 4) primary tumor tissue, first relapse tissue, second relapse tissue and matched normal samples were available. For the same patient 4, we had a matched normal skin sample that allowed us to assess whether variants were germline or somatic. Analysis was limited to the COSMIC Cancer Gene Census, which was a set of 743 well-curated driver genes across human cancers, at the time of the study, which included not only somatically mutated genes but also cancer predisposition genes (https://cancer.sanger.ac.uk/census). We also limited our analysis to rare variants with an overall and ethnicity-matched population frequency less than 0.5% based upon gnomAD v4.0.0. Overall, 285 non-synonymous nuclear genome variants were identified from the 29 chordomas (Supplementary Table 1). Genes with the highest number of nonsynonymous variants were ATM, TPR, FAT1 and ARID1B. TPR, and FAT1 genes had only missense variants. Excluding missense variants, 17 COSMIC genes had a total of 4 frameshift, 16 in-frame deletion, 6 in-frame insertion, and 7 stop-gain variants (Figure 4). These included two stop-gain mutations in the SETD2 gene. To prioritize genes of interest, the mutation rate for each gene was compared to the background (or passenger) mutation rate. This dn/ds analysis identified ARID1B as the only significant gene with statistically significant higher recurrence of non-synonymous variants versus synonymous variants, more than expected by random chance, with a false discovery rate (FDR) of q < 0.1 (q = 0.023) (Supplementary Table 3).

Figure 4.

Oncoplot of recurrently mutated nuclear genes in the pediatric chordoma cohort. Variants are color coded based on VEP variant classifications. Dot size corresponds to variant allele frequency. The right side displays orange bars indicating the percentage of samples in which each variant is observed. The number of variants per sample is shown at the bottom. Patient age and tumor source (primary or recurrent) are indicated at the bottom.

There were five ARID1B in-frame indels found in five different patients (patient 4, patient 2, patient 17, patient 3, and patient 15) that were rare in the general population but also in the ethnicity-matched sub-populations according to gnoMAD v4.0.0. All were found exclusively in exon 1 of ARID1B. These variants were rare but present in the general population, suggesting their germline origin. The allele frequencies of these ARID1B variants in most tumor samples centered around 0.5, including all three tumor samples of patient 4 and both tumor samples of patient 2 (Table 2), also consistent with germline indels. To further evaluate their likely germline or somatic status, we ran three variant callers (Dragen, SGZ, and UNMASC) that were all designed to distinguish germline vs acquired variants using different supporting data elements. Specifically, Dragen uses de novo assembly of haplotypes within active regions and references of population databases. SGZ makes the determinations by modeling the allele frequency (AF) of each alteration, considering factors such as tumor content, tumor ploidy, and local copy number. [21] The UNMSAC method incorporates data from public databases that catalogue canonical somatic and germline variants. It also utilizes information on adjacent positions from normal controls. [22] All variants were deemed to be ‘Germline’, as agreed by all three methods, or ‘Likely Germline’ as agreed by two of the three methods. To further test the origin, we obtained a skin sample from patient 4 which harbored the ARID1B indel that was deemed to be ‘Likely Germline’, and confirmed this variant was indeed germline in origin by Sanger sequencing. It is thus likely that ARID1B variants in tumors of other patients represent rare heterozygous germline variants as well. In total, 5 of the 23 patients (21.7%) were shown to have rare ARID1B in-frame indels, which was significantly higher (p < 5.9e-07) than what was reported in gnoMAD v4.0.0 when compared to the highest frequency of these 5 variants in any ethnicity-matched population (i.e. 6-156778847-GGGCGGCGGC-G in East Asians, which was 0.0064). This enrichment remained significant even after applying Bonferroni correction for all 743 COSMIC Cancer Gene Census (CGC) genes examined (p < 0.00043837).

Table 2.

ARID1B mutations identified in the pediatric chordoma cohort. Table includes patients with multiple samples. Variant caller results: G: Germline, S: Somatic, L G: Likely Germline; L S: Likely Somatic. Ancestry: EUR: European (non-Finnish), EAS: East Asian, AMR: Admixed American. Consequence: infrm: In Frame, INS: Insertion, Del: Deletion. Variant ID format: [Chr6] Start Position: Reference Allele > Altered Allele.

			Variant caller result (Germline/Somatic)						gnomAD 4.0 Allele Frequency
Patient ID	Variant id (chr6:)	VAF	Dragen	FMI SGZ	UNMASK	Consensus	Consequence	Ancestry	exome	genome	within ancestry
P_04	156778052:C>CGCGGCGGCA	0.39	G	S	G	L G	infrm_INS	EUR	0.0002	0.0003	0.0003
		0.42	G	G	G	G	infrm_INS	EUR	0.0002	0.0003	0.0003
		0.47	G	S	G	L G	infrm_INS	EUR	0.0002	0.0003	0.0003
		0.59	G	G	S	L G	infrm_INS	EUR	0.0002	0.0003	0.0003
P_02	156778847:GGGCGGCGGC>G	0.48	G	G	G	G	infrm_DEL	EAS	0.0019	0.0027	0.0064
		0.53	G	G	G	G	infrm_DEL	EAS	0.0019	0.0027	0.0064
P_17	156778198:C>CCCA	0.49	G	G	G	G	infrm_INS	EUR	0.0001	0.0001	0.0001
P_03	156778982:AGGCGGC>A	0.39	G	G	S	L G	infrm_DEL	EAS	0.0002	0.0002	0.0007
P_15	156778031:GTCCTCC>G	0.42	S	G	S	L G	infrm_DEL	AMR	0.0000	0.0000	0.0002

Rare ARID1B indels were also enriched in the skull base chordoma cohort.

In order to further examine ARID1B in the setting of chordoma, we mined the paired tumor/normal WGS data from 80 Chinese patients with 93 skull base chordomas.[7] While Bai et al. did not specifically report any ARID1B mutations, their study focused on tumor-only mutations since they had matched normal samples for all patients.[7] To distinguish tumor-only and germline variants, we re-processed the tumor and normal WGS data sets separately. A total of eight patients had three distinct and rare in-frame indels in ARID1B (rs770512547, rs1046394316, and rs773423003) in both the tumor and normal samples (Figure 5, Supplementary Table 4). The overall MAF of rs773423003 was very low (0.0014), but quite common in East Asians (0.049), which is very similar to the frequency of this variant in the Chinese chordoma patients (4/80). We therefore excluded this variant from being considered rare and for further enrichment analysis. Counting only the other two variants (rs770512547, rs1046394316), the percentage of skull base chordoma patients (4/80 or 5%) with rare ARID1B indels was still significantly higher than that of the East Asians (p<0.0001174). The frequency of rare germline ARID1B in-frame indels in our pediatric chordoma cohort, however, was still significantly higher than from that of the skull base adult cohort (p=0.0236, Fisher’s exact test).

Figure 5.

ARID1B variants in matched tumor and germline samples in the skull base chordoma cohort. Variants are color coded based on VEP variant classifications. Dot size corresponds to variant allele frequency.

As described above, neither Bai et al. or Tarpey et al. reported any ARID1B indels, likely because they focused on somatic mutations and the patients were mostly adults. [5, 7] Mattox and colleagues did find, however, that one of their 32 chordoma patients had an ARID1B non-synonymous mutation in cfDNA.[8] It is unclear it was somatic or germline.

We previously validated a comprehensive DNA and RNA sequencing panel, OncoKids, for pediatric cancers[23]. The OncoKids panel includes all exonic regions of ARID1B. We retrospectively reviewed data from 1039 OncoKids samples. A total of 15 patients were found to have a total of 16 ARID1B in-frame indels that all fell into a 604bp region (chr6:157,099,421-157,100,024) in the first exon of ARID1B: rs750911844 (1), rs587779744 (2), rs587779747 (4), rs1388229420 (1), rs1469262535 (1), and rs747790383 (7). These are all very rare variants with MAFs of 0.0000431-rs750911844, 0.0000928- rs587779744, 0.002162- rs587779747, 0.00005 - rs1388229420, 0.000008 - rs1469262535, and 0.004513 - rs747790383 according to gnomAD v4.0.0.

OncoKids samples with these rare ARID1B in-frame indels were from a mixture of multiple tumor types: rs750911844 (Diffuse Glioma), rs587779744 (Acute Myeloid Leukemia, Unknown), rs587779747 (Acute Myeloid Leukemia, B-Lymphoblastic Leukemia/Lymphoma, Osteosarcoma, Diffuse Glioma), rs1388229420 (Osteosarcoma), rs1469262535 (Osteosarcoma), rs747790383 (Encapsulated Glioma - 2, Ependymoma, Rhabdomyosarcoma, Neuroblastoma, B-Lymphoblastic Leukemia/Lymphoma, Wilms Tumor). Together, these findings indicate that ARID1B in-frame indels within the first exon may be oncogenic, and associated with multiple types of pediatric cancers, but are particularly associated with pediatric chordoma.

Discussion

Mitochondrial DNA mutations as drivers of skull base chordoma and pediatric chordoma.

Characterization of mtDNA mutations in specific tumor types as well as pan-cancer sequencing studies have established a strong contribution of mtDNA mutations to many subtypes of adult and pediatric cancers.[17, 24] The mtDNA study of the Pan-Cancer Analysis of Whole Genomes (PCAWG) consortium argued strongly for the oncogenic effect of mtDNA loss-of-function mutations in both the initiation and clonal evolution of some cancer types.[25] This is consistent with earlier findings in renal oncocytoma, [26] which were later functionally validated by direct base-editing of the mtDNA genome.[27]

In both the primarily adult skull base chordoma cohort and the pediatric chordoma cohort, we found a large number of non-synonymous heterozygous mtDNA mutations. These mutations cluster within the NADH complex (complex 1) subunit genes, which is consistent with previous pan-cancer studies, including our own landscape study of mtDNA mutations in pediatric malignancies.[17, 24, 25, 28] In addition to the enrichment of mtDNA mutations in NADH complex genes, our findings suggest that there are mtDNA mutation hotspots in pediatric chordomas, and that many of the heteroplasmic mtDNA mutations were present in the matched normal sample but with increased heteroplasmy in the tumor samples, arguing strongly for the significant contributions of mtDNA mutations to chordoma development, progression, and/or response to therapy.

Understanding the oncogenic nature of the mtDNA mutations in chordoma, however, is complicated by the likely heterogeneity of mtDNA mutation heteroplasmy at the single-cell level, as shown by Almeida et al. in colorectal cancer.[29] This is a limitation of this study and other studies which are based on bulk tumor sequencing only.

Mitochondria-associated ER membranes.

It was noted over five decades ago that the distinct ultrastructure of the mitochondria-associated ER membranes was a unique characteristic of chordoma.[19, 20] While obviously implicating altered mitochondrial function, the mitochondria-associated ER membranes are known to play significant roles in cellular functions such as autophagy [30], including that of cancer cells [31], in a specific type of breast cancer [32], and in chordoma specifically.[33] The abnormal mitochondria-associated ER membranes in chordoma may provide a mechanistic explanation for the frequent mtDNA mutations observed in tumors. Disruption of the mitochondrial cristae structure was recently discovered to cause the accumulation of deleterious mtDNA mutations, perhaps due to an inability to clear them from the cytoplasm.[34] This is consistent with an earlier report by Chapman et al. describing how disruption of the mitochondrial membrane could lead to defects in mtDNA maintenance, integrity and copy number regulation.[35] It is therefore plausible that tumor initiation leads to abnormal mitochondrial and mitochondria-associated ER membranes, which fail to maintain mtDNA integrity, which then leads to the accumulation and the increased heteroplasmy of mtDNA mutations.

Mitochondrial Unfolded Protein Response (UPR^mt) pathway.

Abnormal mitochondrial and mitochondria-associated ER membranes potentially explain the accumulation of mtDNA mutations in chordomas. It might be a Catch-22 scenario, however, given the presence of the mitochondrial unfolded protein response or UPR^mt pathway. The UPR^mt pathway is a stress response pathway that regulates many cellular functions including that of cancer cells.[36, 37] ATF5, a transcription factor that plays a vital role in many cancers, also mediates the UPR^mt pathway.[38] Lin and colleagues made a breakthrough discovery that deleterious mtDNA mutations can activate the UPR^mt pathway.[39] Instead of being a by-product of abnormal mitochondrial membranes, deleterious mtDNA mutations could instead activate the UPR^mt pathway and through this, regulate other cellular functions. It is worth noting that the functional impact of these deleterious mtDNA mutations is anticipated to go far beyond their impact on the functions of the genes that they belong to, but rather as molecular signals that regulate a wide range of cellular functions, likely through the UPR^mt pathway. This may be the primary mechanism with which the mtDNA mutations regulate tumorigenesis, instead of affecting cellular metabolism alone.

ARID1B as a potential predisposition gene for pediatric chordoma.

Our findings strongly suggest that rare germline in-frame indels contribute to the development of pediatric chordoma. ARID1B is located at 6q25.3 and is only 9Mb from the TBXT gene located at 6q27 where both the duplication that segregated with chordoma disease status in four multiplexed families and the very common rs2305089 SNP that is strongly associated with increased risk for chordoma are located.[9] Again, other than reported differential expression of the TBXT gene between chordomas and notochords,[15] it is unclear how the duplication of the TBXT gene and a very common variant could predispose to an extremely rare cancer, chordoma. In our pediatric chordoma cohort, 16 of the 23 patients carried the rs2305089 allele, which is not statistically significant compared to gnomAD v.3.1.2 (p = 0.13267) likely because of the small sample size, although six of 23 patients were homozygous for rs2305089, which is statistically significant (p = 0.025755). Given the realization that the ARID1B gene is in close disequilibrium, it is therefore possible that rs2305089 is a tagging SNP while the true oncogenic variants are the ARID1B in-frame indels.

In terms of gene function, ARID1B, which encodes a subunit of the SWI/SNF complex, is also a stronger candidate than the TBXT gene. The deficiency of the SWI/SNF chromatin remodeling complex is widely implicated in human cancers.[40] Specifically regarding chordomas, poorly differentiated chordoma is characterized by the loss of SMARCB1, and sequencing studies have repeatedly identified pathogenic mutations in other subunits of SWI/SNF complexes, namely PBRM1 and ARID1A.[3-5, 7, 8, 41] The candidate ARID1B variants are all in-frame indels, instead of loss-of-function variants. This seems to be inconsistent with the typical inactivating mutations seen in several of the SWI/SNF genes in poorly differentiated chordoma, or other childhood cancers overall.[4, 41-43] Gain-of-function variants of some SWI/SNF subunits, such as in ARID1A, however suggest a dominant mechanism in tumor development.[44, 45] While the loss of ARID1B was thought to cause 6q25 microdeletion syndrome,[46] the synthetic lethality of ARID1A knockout and ARID1B deficiency in cancer implies that ARID1B deficiency alone is unlikely to be sufficient for tumorigenesis. Aberrant localization of the ARID1B protein is associated with activated ERK signaling, clearly through a gain-of-function mechanism.[47] It is therefore conceivable that these rare ARID1B in-frame indels also act through a gain-of-function mechanism. Compared with the loss of SMARCB1 and its significant role in poorly differentiated chordoma, or the loss and gain-of-function of ARID1A, depending on the biological context, it will therefore be very interesting to explore clinical phenotypes of patients carrying ARID1B indels. Given the limitation of having access to only de-identified tumor samples in this study, we will have to rely upon future studies to answer this question.

Chromatin remodeling and mitochondrial DNA mutations.

As our findings suggest oncogenic roles for both ARID1B in-frame indels and mtDNA mutations, it is important to understand the potential relationships between chromatin remodeling and mitochondria. Chromatin remodeling requires ATP and is regulated by oxygen abundance [48-50]. Remodeling of chromatin regulates mitochondrial functions such as Ca²⁺ flux, coenzyme synthesis and mitochondrial quality control.[51, 52] The connection between chromatin remodeling, mtDNA mutations, and the UPR^mt pathway was revealed in a seminal study by Tian and colleagues, who demonstrated that the mitochondrial stress response, through the UPR^mt pathway, resulted in chromatin remodeling across the genome, parallel to the effects of altering the SWI/SNF complex.[53] Both ARID1B indels and mtDNA mutations in chordoma could both lead to altered chromatin remodeling, ultimately regulating tumor initiation and progression.

Functional studies to further explore the roles of ARID1B indels and mtDNA mutations were beyond the scope of this study, as we only had limited formalin fixed and frozen pediatric chordoma samples from the patients. Future studies based on patient-derived cell lines and mouse models will be required to further explore these hypotheses.