From Histology to Multi-Omics: Review of Chordoma Classification and Its Clinical Implications

1. Introduction

Chordoma is a rare primary malignant neoplasm of the axial skeleton arising from remnants of the embryological notochord, with a reported annual incidence of approximately 0.088 per 100,000 persons in the United States. From 2004 to 2014, a total of 3,670 chordomas were diagnosed in the US, with the most common location being cranial (38.7%), followed by sacral (34.3%) and spinal (27.0%), and incidence increases with age with a clear male predominance[1]. On average, patients are diagnosed in their late fifties, and the incidence varies between countries and presumably between races[2]. The 5-, 10-, and 15-year overall survival probabilities following chordoma diagnosis are approximately 0.74, 0.58, and 0.48, respectively, with no significant change in the hazard of death observed over the 2003–2019 period[3].

The clinical management of chordoma is challenging and centers primarily on surgical resection followed by high-dose radiotherapy. Gross-total resection has an established relationship with progression-free and overall survival; however, the tumor’s anatomical location frequently interferes with attempts at complete resection, and no consensus currently exists on the most effective adjuvant radiotherapy modality[4]. Chordoma is unresponsive to available chemotherapies and to doses of radiation that can be safely delivered with conventional radiotherapy, which substantially limits treatment options[5]. The advent of particle therapy — in particular proton and carbon ion irradiation — has meaningfully expanded the therapeutic armamentarium, as these modalities permit dose escalation to levels exceeding 70 Gy while sparing adjacent critical structures[6]. Nevertheless, even with multimodal therapy, local recurrence remains the most common failure pattern, translating to adverse overall survival[7]. For patients who fail surgery and radiotherapy, there remains an urgent unmet need for new therapeutic options.

In this study, we present a review of the evolving molecular and pathological classification of chordoma, examining the histological, radiomic, clinical, genomic, epigenomic, transcriptomic, and proteomic frameworks that have been proposed to stratify this disease into clinically and biologically meaningful subtypes. Accurate classification is prerequisite to improved prognostication, identification of actionable therapeutic targets, and rational clinical trial design — goals that are particularly pressing given the rarity of chordoma and the absence of approved systemic therapies. The contribution of molecular features to chordoma classification remains significantly less advanced than in the official classification systems for other neoplasms.

2. Materials and Methods

2.1. Search

A comprehensive systematic search of English literature was conducted on the 1st of February 2026. PubMed^® and CENTRAL were accessed and searched for the following phrases: ‘chordoma classification’, ‘chordoma DNA sequencing’, ‘chordoma RNA sequencing’, ‘chordoma methylation’, and ‘chordoma copy number’. No additional restrictions or Boolean logic was set.

2.2. Eligibility Criteria and Study Selection

Studies were screened; any doubts between the reviewers were resolved through a discussion. The selection process had two steps. First the abstracts of the identified articles were accessed. Review articles, case reports, articles pertaining to other pathologies and in languages other than English were excluded. Second, full-length articles were accessed and searched for relevance to chordoma classification. This approach is summarized in Table 1.

2.3. Data Extraction

Data were summarized in a table, with predefined columns: area of interest, year of publication, technology, number of patients, and presence of validation cohort. The data were organized to enable meaningful comparisons and synthesis of findings. It is available in the Supplementary Table S1.

2.4. Assessment of Risk of Bias of Individual Studies

Given that the studies were conducted in a wide variety of places and times, entities diagnosed as chordoma may differ between studies (see Results, Section 3.1), the risk of bias for each included study could result from factors beyond the researchers’ control. Nineteen studies included more at least 100 patients, 39 included at least50 patients, and 67 at least 25 patients. Fourteen studies applied a validation dataset. These two identifiable factors, related to risk in clustering studies are included in the Supplementary Table S1.

2.5. Results Presentation

Given that the nature and variety of included studies, the most productive approach was a qualitative review. R 4.5.2 along with ggplot2 were utilized to summarize and visualize the results. Both the search results and visualization program are available as supplementary materials. Generative AI was utilized (Sonnet 4.6) for grammar and language checks and to discuss scientific clarity, correctness, and accuracy of the claims made in this work.

3. Results

In total queries resulted in 400 unique articles, of which 296 passed the initial abstract screening and after full-text review, 106 were deemed relevant to the chordoma classification. In this process, additional 10 studies were included from the references of the selected articles. See Figure 1. for more granular information.

Overall, most of the articles (349/406 = 400 identified in search + 6 identified during review) were published after 2000 with trend towards more publications most pronounced with the chordoma RNA query (see Figure 2a,b). Most common cause for study rejection was no included classification of chordoma cases (see Figure 2c). Of the 108 included studies, 103 were published after 2000 with the visible growth mainly in articles reviewed in section 3.6, dealing with gene expression patterns in chordoma (see Figure 2d). Recurring words (>10 occurrences) in the titles of the queried articles are depicted in Figure 2e. The number of times these words appeared in titles of included articles are color-coded, showing consistency of coverage.

3.1. Key Differential Diagnoses

Before establishing classification patterns within a neoplasm entity, it is crucial to define lesions representing such neoplasm. According to the WHO Classification of Tumors of Soft Tissue and Bone, the key distinguishing features of chordoma are lobular growth pattern, cells embedded in rich basophilic mucoid or myxoid matrix, and physaliphorous cells (large, polygonal cells with abundant, pale, bubbly or vacuolated cytoplasm).[11]

First primary distinction is between chordoma and chondrosarcoma as they represent distinct pathological entities with different embryonic origins and anatomical predilections. Chordomas arise from the remnants of the primitive notochord, whereas chondrosarcomas originate from the cartilaginous matrix of the skull base synchondroses[12]. An important diagnostic differentiator is their anatomical location, as chordomas are predominantly midline structures involving the clivus, while chondrosarcomas are typically situated off-midline, most frequently at the petro-occipital fissure[13]. Furthermore, while chondrosarcomas are often classified as histological Grade I or II and exhibit a relatively indolent clinical course, chordomas are characterized by a more locally invasive and aggressive growth pattern[14]. Ultimately, these biological differences translate into divergent survival outcomes; chondrosarcomas are associated with significantly better long-term prognoses, with 5-year survival rates often reaching 80% to 90%, whereas chordomas generally demonstrate lower 5-year survival rates of approximately 65%[12].

Radiographic characteristics and clinical outcomes further distinguish these two malignancies. Chondrosarcomas frequently exhibit characteristic intratumoral calcifications on computed tomography, a feature less common in chordomas[14]. Quantitative analysis using diffusion-weighted magnetic resonance imaging indicates that chondrosarcomas typically have higher apparent diffusion coefficient values than chordomas, reflecting the former’s lower cellularity and higher water content within the cartilaginous matrix[15]. Histopathologically, the two entities diverge significantly. Chordomas lack the chondroid-like cells embedded in the lacunae typical for chondrosarcoma. They also tend to be positive for brachyury and cytokeratin (e.g., cytokeratin 19) staining.[11] Chondrosarcomas tend to be affected by IDH1/2 mutations more often than chordomas[16] although these mutations in chordomas were also reported[10].

Parachordoma is an exceedingly rare soft tissue tumor (~40 reported cases)[17] that mimics axial chordoma but arises in extra-axial locations. Historically recognized as chordoma periphericum or chordoid sarcoma, the neoplasm typically presents as a slow-growing, painless mass, often involving deep fascial tissues. Microscopic examination reveals a lobulated architecture consisting of nests and cords of epithelial-like rounded cells with abundant, often vacuolated cytoplasm, occasionally forming structures that resemble the embryonal notochord. The extracellular matrix generally appears myxoid, chondroid, or hyaline in consistency. Immunohistochemical analysis is essential for definitive diagnosis, as the neoplastic cells demonstrate a bimodal differentiation indicated by strong expression of vimentin alongside focal expression of epithelial markers such as cytokeratin (peptides 8 and 19) and EMA, as well as variable expression of S-100 protein. These epithelium-specific immunoreactivity pattern is critical for differentiating parachordoma from extraskeletal myxoid chondrosarcoma, which lacks such markers[18]. Parachordomas do not express brachyury, contrary to extra-axial chordomas.[19] Current literature suggests that parachordoma is a mesenchymal neoplasm with a generally good prognosis following surgical excision[17].

Another clinically important entity to be distinguished from chordoma is benign notochordal cell tumor (BNCT) also known as ecchordosis physaliphora. It is a congenital, histologically benign hamartomatous remnant of the embryological notochord typically located along the craniospinal axis, most frequently at the level of the clivus[20,21]. It is reported in up to 2% of autopsies.[20] Radiologically, the lesion is characterized as a well-circumscribed, T2-hyperintense, non-enhancing mass, often distinguished from other retroclival lesions by a bony stalk or pedicle projecting from the clivus[21,22]. Although historically regarded as an asymptomatic incidental finding, BNCT can precipitate life-threatening complications, including spontaneous cerebrospinal fluid rhinorrhea and recurrent bacterial meningitis associated with midline clival defects[21]. Histopathologically, BNCT comprises physaliphorous cells with vacuolated cytoplasm and a very low Ki-67 proliferation index of less than 1%, yet significant radiological and histological overlap with low-grade chordomas suggests a spectrum of disease rather than a strict benign-malignant dichotomy [22,23]. Therefore, creation of a separate entity Atypical Notochordal Cell Tumor to encompass the in-between lesions was proposed[24].

3.2. Histological Variants of Chordoma

According to the WHO Classification of Tumors of Soft Tissue and Bone, vast majorityof chordoma cases represent conventional chordoma that is characterized by the lobular growth pattern, along with cells embedded in mucoid/myxoid matrix, and physaliphorous cells. However, there exist 2 rarer forms of chordoma, dedifferentiated chordoma (DC), and poorly-differentiated chordoma (PDC)[11]. Further subtypes within the realm of classical chordoma can be described. In a recent study evaluating the ultrastructural features of clival chordomas these tumors were stratified into cell-dense (CDT) and matrix-rich (MRT) subtypes using electron microscopy. Their analysis revealed that the CDT variant is associated with a significantly higher Ki-67 proliferation index, an increased risk of early recurrence, and higher mortality rates compared to MRT chordomas[25].

There exists some controversy regarding chondroid chordoma (ChoC). It is a morphological variant of conventional chordoma characterized by a biphasic growth pattern containing both conventional chordoma elements and areas of cartilaginous differentiation[11,26]. While historically debated and occasionally misclassified as low-grade chondrosarcoma due to morphological similarities[27], its classification as a true chordoma was definitively established through immunohistochemical profiling. Unlike low-grade chondrosarcomas, which are mesenchymal tumors lacking epithelial markers, the cells within both the conventional and cartilaginous components of chondroid chordomas consistently express brachyury and epithelial markers such as cytokeratin, epithelial membrane antigen, and carcinoembryonic antigen[28]. Consequently, immunohistochemistry serves as the critical diagnostic tool to differentiate true chondroid chordomas from chondrosarcomas. Accurate clinicopathological classification is essential because, the ChoC is generally associated with a more favorable prognosis and better overall survival, compared to classic chordoma[28]. In the newest WHO Classification ChoC is described as a variant of conventional chordoma[11].

Dedifferentiated chordoma is a rare and highly aggressive malignancy, comprising less than one percent of all chordomas[29]. It is histologically defined by a biphasic appearance in which a typical conventional chordoma is directly juxtaposed with a high-grade sarcomatous component[29]. During the process of dedifferentiation, the sarcomatous regions characteristically lose the expression of key diagnostic lineage markers, notably brachyury and cytokeratin, which remain conserved in the adjacent conventional tumor cells[29]. Clinically, the onset of dedifferentiation portends an extremely poor prognosis, characterized by rapid disease progression and a median overall survival of approximately twenty months[29]. The molecular landscape of DC is distinct from its conventional counterpart. The transition is frequently associated with recurrent mutations in TP53, which is found in a significant subset of these tumors and serves as a major driver of the aggressive phenotype with a prognostic value[30]. It was shown that TP53 loss of function can be limited to the dedifferentiated parts of the tumor[29]. Moreover, it was shown that even in the cohort of conventional chordomas hTERT expression, linked to TP53 disfunction was a prognostic factor[31]. Furthermore, dedifferentiation may arise de novo or secondary to radiation therapy[32]. This behavior in skull-base tumors was characterized by unique epigenetic shift and specific somatic alterations, including PIK3CA mutations[32]. Additionally, a distinct Polycomb-type dedifferentiation pathway has been identified in a specific subset of skull base dedifferentiated chordomas[33]. This variant is often driven by homozygous deletions of EED, leading to a targeted loss of H3K27 trimethylation exclusively within the malignant sarcoma-like component, often resulting in a histology that mimics malignant peripheral nerve sheath tumors[33].

Poorly differentiated chordoma represents another rare and aggressive subtype of chordoma, frequently located at the skull base or cervical spine that occurs often in the pediatric population[34,35,36]. In the adult population, reports of extra-axial PDC were made[37,38]. Unlike conventional chordoma, poorly differentiated chordomas exhibit an atypical, high-grade morphology consisting of cellular proliferations of epithelioid to rhabdoid cells with prominent nucleoli and vesicular nuclei, often lacking the hallmark bubble-like vacuolization[39]. A defining molecular feature of this entity is the recurrent loss of SMARCB1 (INI1) expression, a core subunit of the SWI/SNF chromatin remodeling complex, which serves as a critical diagnostic marker[39,40]. Despite the loss of SMARCB1, these tumors maintain the strong nuclear expression of brachyury. This expression profile reliably distinguishes poorly differentiated chordomas from other INI1-deficient tumors, i.e., atypical teratoid/rhabdoid tumors, which are consistently brachyury-negative[34]. The clinical progression of poorly differentiated chordoma is notably more rapid than that of conventional chordoma, characterized by a high propensity for early local recurrence, metastatic spread, and increased mortality[34]. The identification of SMARCB1 loss has not only refined the diagnostic criteria for these tumors but has also provided a mechanistic rationale for targeted epigenetic therapies. Specifically, SMARCB1 deficiency leads to unopposed oncogenic activity of Enhancer of Zeste Homolog 2 (EZH2), prompting the clinical investigation of EZH2 inhibitors, such as tazemetostat, which have demonstrated sustained antitumor responses in patients with this specific molecular profile[41].

3.3. Radiological and Surgical Classifications

For diagnostic purposes, majority of patients with chordoma undergo MRI. Capitalizing on this, radiological classification of skull base chordomas was proposed. It utilizes normalized signal intensity ratios derived from T1-weighted FLAIR, T2-weighted, and contrast-enhanced MRI sequences compared against the pons[42,43]. The ratio of tumor-to-pons signal intensity on T2 sequences (RT2) and enhanced T1 sequences can serve as a primary prognostic indicator, identifying high-grade tumors, which are associated with increased tumor blood supply, greater intraoperative blood loss, and progression risk[42]. These imaging characteristics correlate with the underlying electron microscopic ultrastructure, where MRT chordomas typically exhibit higher RT2 values and more favorable long-term outcomes compared to CDT chordomas[43]. Furthermore, advanced radiomic analysis of T2-weighted imaging features has demonstrated clinical utility in constructing predictive signatures for four-year recurrence probability, with high-dimensional 3D features achieving accuracy rates of approximately 85% in distinguishing high-risk tumor profiles before surgical intervention[44].

Radiogenomics and radiomics provide emerging non-invasive methodologies for the diagnostic and prognostic evaluation of chordomas by correlating macroscopic imaging characteristics with microscopic, molecular, and epigenetic profiles[45,46,47]. In sacral lesions, 3D computed tomography radiomics, facilitates the accurate differentiation of chordomas from giant cell tumors[45]. For skull base chordomas, MRI radiomics can predict both clinical outcomes and some DNA copy-number alterations (CNA), i.e., losses of chr1p and chr9[46]. Furthermore, radiogenomic profiling has linked a specific prognostic DNA methylation profile to a 14-feature MRI signature[47].

Surgical classification schemes of skull base and clival chordomas are primarily utilized to guide operative strategy and optimize resection margins, which correlate strongly with long-term progression-free survival[48]. Recent retrospective analysis revealed that the sphenoclival region is being the most affected, and the pattern of bone invasion is predominantly endophytic[48]. More specific staging systems categorize clival chordomas into distinct types (I – dorsal clivus, II – ventral clivus, III – inferior third of the clivus, and IV - paramedian)[49]. These classification frameworks are critical for determining the suitability of minimally invasive endoscopic endonasal approaches versus traditional, open cranial base approaches[50]. By applying these classification schemes preoperatively, surgeons can better predict the feasibility of achieving marginal or gross total resection, identify highly probable sites of residual tumor tissue such as the cavernous sinus, petrous apex, or parapharyngeal space, and ultimately minimize surgical morbidity while maximizing therapeutic efficacy[48,49,50].

3.4. Genomic Landscape of Chordoma

Although majority of chordoma cases are sporadic, familial susceptibility was described[51,52,53]. Our current understanding of genetic susceptibility to chordoma implicates a complex, polygenic landscape involving both rare and common germline alterations. The foundational driver of chordoma risk in familial predisposition cases involves germline duplications and specific sequence variants, in the TBXT gene[51]. A high frequency of the common TBXT susceptibility variant rs2305089 was additionally confirmed in 97.8% of patients in a French cohort, significantly elevated compared to the general population[54]. Beyond TBXT-related mechanisms, rare germline variants in homologous recombination genes, notably BRCA2 and PALB2, have been identified in familial and sporadic cohorts, highlighting DNA repair deficiency as a pathogenic factor and potential therapeutic vulnerability[52]. Results of broader DNA sequencing indicates additional rare loss-of-function and missense germline variants distributed across diverse oncogenic and developmental signaling networks, including the SWI/SNF, PI3K/AKT/mTOR, and Sonic Hedgehog pathways[53]. Furthermore, common single nucleotide polymorphisms, e.g in the LGALS3 gene, contribute to this multifactorial susceptibility by elevating the risk of skull base chordoma and predicting reduced progression-free survival following treatment[55].

Chordoma is characterized by a low overall somatic mutation burden relative to most other cancer types[10,30,56]. Whole-genome and whole-exome sequencing studies have consistently identified a limited repertoire of recurrently mutated driver genes. Alterations in SWI/SNF chromatin remodeling complex genes are among the most frequent and best-characterized events, with PBRM1 mutations or structural variants reported across skull-base, spinal, and sacral cohorts, and co-occurring alterations in SETD2 and ARID1A further implicating defective chromatin remodeling as a central oncogenic mechanism[10,54,56,57,58]. Activating mutations in PI3K pathway genes, predominantly PIK3CA, occur in approximately 10–16% of tumors across multiple cohorts and represent potentially actionable targets[54,56,59]. LYST, encoding a lysosomal trafficking regulator, was identified as a candidate novel chordoma driver gene through recurrent truncating mutations in approximately 10% of sporadic cases; a speculative mechanistic link to the lysosomal biology of notochordal cells has been proposed, though this requires further functional validation[56]. Early targeted panel sequencing identified mutations in KDR and KIT in small cohorts, though these remain infrequent and their functional driver significance is uncertain[60].

It was observed that half of chordoma cases lack any driver mutation present[10,56]. However, CNA are frequent and chromosomal instability (CIN) was identified as prognostic marker[8,9,59]. The changes are consistent between primary and recurrent or metastatic samples, suggesting a foundational role in tumor development.[10] More specifically, homozygous deletion of the CDKN2A/B (on chr9q) locus is recurrent and was more commonly observed in spinal and recurrent tumors in two independent cohorts[10,54,61].There is some controversy surrounding the prognostic significance of these alterations: Horbinski et al. observed worse overall survival in chordomas with 9p loss of heterozygosity, while homozygous 9p21 deletion only trended toward significance without reaching it[62]. Meanwhile, Passeri et al. observed the homozygous loss of CDKN2A/B to confer worse prognosis and be exclusive with mutations affecting the SWI/SNF complex[54]. However, Bai et al. found that CDKN2A/B status was not independently associated with chordoma specific survival, but it was associated with worse progression-free survival[10]. Since CDKN2A/B genes encode p16 (INK4A) and p15 (INK4b) negative regulators of cyclin dependent kinases 4 and 6 (CDK4/6) high frequency of chromosome 9p loss sparked an interest in targeting cell cycle regulation in chordoma. A non-randomized phase II trial PMO-1601 showed a modest antitumor activity in of CDK4/6 inhibitor palbociclib in CDKN2A/B-deleted tumors[63].

Beyond that, recurrent losses of arm- or chromosome include deletions of 1p, 3, 4, 5, 9p, 9q, 10, 13q, 14q, 18, and 22q, while gains of 1q and 7 are also observed[10,59,64,65,66,67,68,69,70,71,72]. Deletion of 22q, which harbors SMARCB1, was associated with worse disease-specific and recurrence-free survival, and the combination of PBRM1 point mutations with 22q deletion further strengthened this prognostic association[10]. However, loss of INI1 staining in conventional chordoma was not associated with worse prognosis in another cohort[73]. Nonetheless, homozygous SMARCB1 is linked with the PDC[34,35,36,37,38,39,40,41] and it is not yet clear if the heterozygous loss is a transition state towards this pathology. Somatic deletions of 14q and 18p were associated with persistent or recurrent disease in a North American cohort [59]. Genomic diversity by tumor site is well-documented: sacral chordomas harbored driver gene mutations, TBXT amplifications, and deletions of 5p, 5q, and 9p more frequently than clival tumors[10,59,74]. We have recently proposed a 4-group classification scheme for CNA in chordoma[75]:

• C1 for chromosomally stable tumors.
• C9 for tumors with predominant chromosome losses e.g., chr9q; these tumors tend to also have a 22q loss.
• C7 for tumors with predominant chromosome gains especially chr7,
• C2 for tumors with both gains and losses (i.e., both chr9q loss and chr7 gain); gain of chr2 seems to be characteristic for that cluster. Notably, gain of chr2 was independently associated with higher recurrence rate in skull bas chordoma[76].

3.5. Classifiers, Based on DNA-Methylation

Accumulating evidence indicates that aberrant DNA methylation is a significant epigenetic feature of chordoma pathobiology. Rinner et al. provided early evidence of this by identifying 20 differentially methylated genes compared to peripheral blood — including hypermethylation of the tumor suppressor genes RASSF1, HIC1, and KL — with a multigene methylation-based classifier able to distinguish chordoma from healthy blood DNA[77]. Marucci et al. demonstrated that MGMT promoter methylation is present in a significant proportion of recurring clival chordomas, while remaining consistently unmethylated in non-recurring cases[78]. Since the efficiency of temozolomide treatment has been previously shown as related to MGMT promoter methylation in other caner types[79] this observation raised the possibility that temozolomide may have a role as adjuvant therapy in recurrent chordoma cases. Alholle et al. performed genome-wide DNA methylation profiling of 26 chordomas and normal nucleus pulposus samples, identifying 8,819 significantly differentially methylated loci; functional analysis indicated that genes affected by cancer-specific methylation changes were involved in networks including cancer disease, nervous system development and function, and cellular proliferation[80].

DNA methylation profiling has emerged as a reliable molecular tool for the classification and prognostic stratification of chordomas. Across independent cohorts, unsupervised clustering of genome-wide methylation data consistently identifies two distinct epigenetic subtypes[8,81,82]. One subtype, variously termed immune-infiltrated or Chordoma I, is characterized by relative global hypermethylation, higher immune cell infiltration including cytotoxic T lymphocytes, B cells, neutrophils, and macrophages, and lower tumor purity[8,81]. The other subtype — cellular or Chordoma C — displays global hypomethylation with CpG island hypermethylation, higher tumor purity, and greater chromosomal instability[8,81]. CDKN2A/B locus deletions are associated with the subtypes, though the specific cluster association differs between studies: Baluszek et al. found 9p deletions in 9 of 10 Chordoma C tumors, while Huo et al. reported RB1 and CDKN2A/B deletions predominantly in their immune-infiltrated Cluster 1[82]. Both Huo et al. and Zucatto et al. reported worse prognosis in the immune-infiltrated subtype while we have not observed such effect in our cohort[8,81,82].

Beyond subtype classification, methylation profiling has shown translational promise in two additional directions. Plasma cell-free DNA methylomes, obtained via cfMeDIP-seq, were shown to distinguish chordomas from clinical mimics such as meningiomas and spinal metastases and leave-one-out models correctly assigned all twelve paired tumors to their tissue-based methylation subtype using plasma-derived signals alone, demonstrating the feasibility of non-invasive prognostication[81]. A three-CpG-site prognostic risk score, incorporating loci mapped to BATF, ACTR3C, and FGFBP2, achieved AUC values of 0.775, 0.795, and 0.904 for 3-, 5-, and 10-year survival respectively[82]. In the broader sarcoma context, the DKFZ Sarcoma Classifier — a random forest tool trained on 1,077 reference samples across 54 bone and soft tissue subtypes — correctly classified chordomas in 85% of cases (75/88 samples) in an independent external validation cohort of 986 bone and soft tissue tumors, making chordoma one of the entities most reliably identified by this approach[83].

3.6. Gene Expression Patterns in Chordoma

Chordomas universally express simple epithelium cytokeratins 8, 18, and 19 alongside vimentin, EMA, and S100 while lacking desmin and squamous-type cytokeratins, reflecting their notochordal origin[84,85,86,87]. Pediatric cases more frequently exhibit p53 expression, INI1 loss, and elevated MIB-1 labeling indices, features associated with more aggressive clinical behavior[85]. Brachyury expression levels correlate with shorter progression-free survival and with upregulation of PI3K/Akt pathway gene expression in skull base chordoma, and somatic copy number gain of the brachyury gene locus has been identified in a subset of cases[88,89]. EGFR, c-Met, and HER2/neu are detectable by immunohistochemistry across chordomas, with most tumors showing strong expression of both c-Met and EGFR[90,91]. Overexpression of MET in chordoma is not caused by gene fusions as commonly observed in sarcomas[92]. Tamborini et al. observed that PDGFRB is highly expressed and phosphorylated in chordoma, while PDGFRA and KIT have lower expression levels but nonetheless are activated by phosphorylation. Since no gain-of-function mutations nor gene amplification were identified in that study, an autocrine or paracrine ligand-driven activation loop was proposed as the underlying cause[93]. PDGFR-β expression additionally correlates with invasive behavior via the mTOR signaling pathway and predicts shorter progression-free and overall survival in patients with clival chordoma who undergo non-total resection[94]. Finally, high PALB2 expression independently predicts shorter progression-free survival in skull base chordoma and promotes proliferation, migration, and invasion in vitro, implicating dysregulation of DNA homologous recombination repair as a contributor to chordoma pathobiology[95].

Gene expression profiling has revealed significant molecular heterogeneity in chordoma across anatomical subtypes and patient cohorts. RNA sequencing of skull base chordomas identified two immune subtypes differing in the degree of macrophage and T cell infiltration, with higher CD68 and CD163 expression correlating with shorter progression-free and overall survival[96]. Similar transcriptomic analysis of skull base chordomas distinguished two molecular subtypes: CC1 tumors were associated with somatic mutations in chromatin remodeling genes such as PBRM1 and SETD2, while CC2 tumors exhibited upregulation of Sonic Hedgehog pathway genes, with expression of markers such as PTCH1 showing prognostic significance[97]. Gene expression patterns are also linked to stromal component of chordomas; these tumors can be classified into stroma-rich and stroma-poor cases (similar to CDT and MRT types, described earlier). Stroma-rich chordomas are more often enhanced in MRI studies, have higher expression of genes associated with tumor metastasis and progression, and show tendency towards poorer prognosis[98]. Here exists some inconsistency regarding the prognostic value as Bai et al. reports MRT to have longer expected survival [43].

Recent single-cell RNA sequencing (scRNA-seq) studies have corroborated those while substantially advancing understanding of chordoma biology, heterogeneity, and therapeutic targeting. First, Duan et al. identified varied tumor microenvironment with numerous infiltrating M2-like macrophages and T/NK lymphocytes and marked TGF-β signaling, mainly between macrophages and fibroblasts[99]. These findings were confirmed by Zhang et al. who additionally identified a stem-like malignant subpopulation — marked by CTSL expression — associated with radioresistance through a telomere-end packaging mechanism. The same study described a partial epithelial-mesenchymal transition (p-EMT) program, driven via TGF-β, that localized to the invasive tumor edge, predicted worse progression-free and overall survival, and was pharmacologically targetable with the TGF-β inhibitor[100]. Extending this research, scRNA-seq analysis on 14 chordoma samples identified VEGFR and TGF-β as co-dominant therapeutic targets. Subsequently engineered dual VEGFR/TGF-β CAR-T cells demonstrated superior and sustained cytotoxicity over VEGFR-only CAR-T cells against chordoma cell lines and patient-derived organoids[101]. Complementarily, a basement membrane-related gene signature comprising five genes, involved in this mechanism — ITGB3, SMOC1, UNC5B, COL14A1, and COL13A1 — was independently associated with recurrence-free survival, immune cell infiltration patterns favoring immunosuppression and differential drug sensitivity, with ITGB3 knockdown functionally impairing chordoma cell proliferation and migration via the PI3K-Akt pathway, confirming the relevance of TGF-β signaling[102].

Bridging the stroma and immune microenvironment research, multi-omics profiling including spatial transcriptomics and multiplexed immunofluorescence across large patient cohorts demonstrated that a specific inflammatory cancer-associated fibroblast subset, while spatially distant from tumor cells, correlates with malignant phenotypes and adverse outcomes, with functional experiments confirming its role in promoting chordoma progression[103]. Furthermore, Huo et al. applied scRNA-seq alongside T- and B-cell receptor sequencing to compare primary and recurrent chordomas, finding that recurrent tumors displayed reduced antigen-presenting cell activity, fewer plasma cells, and contracted BCR clonotype diversity. Fibronectin 1 (FN1), upregulated in recurrent disease and secreted by tumor cells, macrophages, and lymphocytes, promoted chordoma invasion and proliferation in vitro and in vivo; the study further identified exhaustion of CD8⁺GZMK⁺ T cells and M2-like macrophage polarization as key immunosuppressive mechanisms associated with recurrence[104]. Additionally, chordoma progression is driven by metabolic-immune crosstalk via the BACH1/ANGPTL4/SDC4 signaling pathway, wherein cholesterol-metabolic tumor-associated macrophages induce stemness and cholesterol accumulation in tumor budding-like cell subpopulations[105]. These studies build on earlier immune research in the field which showed that PD-L1 expression is predominantly localized to tumor-infiltrating macrophages and lymphocytes rather than chordoma cells in vivo, although expression remains inducible in chordoma cell lines[106]. Furthermore, the anti-apoptotic protein survivin (BIRC5) maintains cell cycle progression in these tumors, and its pharmacological inhibition results in G2/M phase arrest, increased polyploidy, and apoptosis[107].

Several other molecular markers have been investigated individually for their prognostic and therapeutic relevance in chordoma through gene expression studies. Two independent studies implicated developmental gene expression in worse prognosis in skull-base chordomas; they both built a signature of genes, associated with fetal development, and found it to be prognostic in chordoma. However, the composition of signatures differed[108,109]. CDK12 expression, assessed by tissue microarray in 56 chordoma specimens, was significantly elevated in recurrent and metastatic disease and independently predicted shorter overall and progression-free survival; siRNA-mediated CDK12 knockdown attenuated chordoma cell growth in vitro[110]. Similarly, high CDK9 expression correlated with recurrence and poor clinical outcomes, and selective pharmacological inhibition with LDC000067 reduced cell proliferation, induced apoptosis, and suppressed colony and spheroid formation[111]. In skull-base chordomas, TGFB1 mRNA levels were significantly higher in hard-type versus soft-type tumors and in female versus male patients across a cohort of 57 cases, with elevated expression associating with tumor progression[112]. The expression of two genes encoding RNA-binding proteins that regulate post-transcriptional gene expression, SAM68 and IMP3 were examined in sacral chordoma and conferred shorter progression-free survival[113,114].

Non-coding RNAs (ncRNAs) have also emerged as significant regulators of chordoma pathobiology. Integrated miRNA-mRNA profiling comparing chordoma tissues with fetal notochord identified 33 significantly dysregulated miRNAs and 2,791 differentially expressed mRNAs, whose expression was associated with these miRNAs. Pathway analysis implicating the MAPK signaling as the most overrepresented oncogenic pathway[115]. Complementary experiment in sacral chordoma and nucleus pulposus samples identified even broader miRNA-mRNA regulatory network associated with tumorigenesis and immune modulation. Notably, it revealed a previously uncharacterized miRNA/mRNA axis involved in stemness regulation, with molecular components proposed as candidate therapeutic targets[116]. The role of individual noncoding RNAs was also described; miR-1 — previously shown to suppress chordoma cell growth by targeting MET — has been demonstrated to carry prognostic significance in chordoma patients, supporting a functional role for this miRNA in disease progression[117].

Proteomic analyses of chordoma have progressively advanced from early differential profiling to large-scale multi-omics integration, collectively revealing molecular heterogeneity that carries both prognostic and therapeutic implications. An iTRAQ-based quantitative proteomic study of clivus chordoma classified tumors by degree of bone invasion into endophytic and exophytic subtypes. The results indicated 250 differentially expressed proteins and implication of TGFβ1 downregulation and reduced PTEN expression in aggressive bone infiltration, with the PI3K/AKT/mTOR pathway proposed as a mediator[118]. Next, the results of two tandem mass tag-based proteomic experiment studiess were published. First identified ASNS as a significantly upregulated protein in rapidly recurring skull base chordoma and subsequently validated it by immunohistochemistry in a 187-patient tissue microarray cohort, where high ASNS expression independently predicted shorter recurrence-free survival[119]. Second study included 102 chordoma patient samples and described three molecular subtypes — bone microenvironment-dominant, mesenchymal-derived, and mesenchymal-to-epithelial transition-mediated — with distinct biological features encompassing osteoclastogenesis and immunogenicity, oxidative phosphorylation, and receptor tyrosine kinase activation, respectively. This study matched these subtypes to theranostic therapies with denosumab, S-Gboxin, and anlotinib and validated in patient-derived xenograft models[120]. The largest and most comprehensive multi-omics study to date integrated whole-exome sequencing, RNA sequencing, proteomics, and phosphoproteomics across 187 skull-base chordoma tumors. It pointed out CIN as a prognostic predictor, with CIN-high status correlating with elevated DNA replication stress, E2F3 transcriptional activity, and CDK4-mediated RB1 phosphorylation. Chromosome 1q gain was further associated with mitochondrial pathway upregulation and poor outcomes, while immune subtyping revealed a cold immune subtype linked to chr9p/chr10q loss. Moreover, proteomics-based classification identified invasive subtypes with potential therapeutic vulnerabilities including RPRD1B as a target in radiotherapy-resistant disease[9].

4. Discussion

This review synthesizes evidence from 108 studies encompassing 6,349 individuals to provide a comprehensive, multi-domain classification framework for chordoma across pathological, clinical, and molecular dimensions. First, we delineated the principal differential diagnostic entities and the immunohistochemical and radiological tools used to distinguish them. Second, we described the established histological variants of chordoma, each carrying distinct molecular features and prognostic implications. Third, we reviewed radiological and surgical classification frameworks, including emerging radiomic approaches that link imaging phenotypes to underlying molecular and epigenetic profiles. Fourth, we characterized the genomic landscape of chordoma, highlighting its low somatic mutation burden, the predominance of chromatin remodeling gene alterations, and recurrent CNAs. Next, we reviewed DNA methylation-based classifiers that demonstrated reproducibility across independent patient cohorts. Finally, we examined transcriptomic and proteomic datasets — including single-cell and spatial transcriptomic analyses — that have collectively uncovered tumor microenvironment heterogeneity, immunosuppressive mechanisms, and novel therapeutic vulnerabilities. Together, these six interrelated domains reflect the progressive shift in chordoma research from purely morphological toward integrated molecular classification, with direct implications for prognostication, therapeutic targeting, and clinical trial design in this rare and treatment-refractory malignancy.

There exists an urgent need to consolidate the various classification approaches into clinically consistent groups that can be acted upon. Some clear patterns emerge from the reviewed articles. First is associated with the CNA – the CIN score is clearly prognostic, the CDKN2A/B loss is associated with DNA methylation pattern, and worse prognosis in some form[8,10,54,62,82]. The importance of cell cycle in chordoma biology is underpinned by the studies showing CDK9 and CDK12 expression to be prognostic[110,111] and some of the studies on DNA copy number profiling that showed prognostic relevance of chromosome 9p loss[10,54]. Beyond CDKN2A/B loss, patient stratification, based on CNA, is linked to transcriptomic profile, and immune cells infiltration[75]. Hopefully, further research will allow to establish if there is a place for theranostic use of CDKN2A/B status[63].

Second theme is already mentioned dysfunction of the SWI/SNF chromatin remodeling complex. At the histological level, homozygous loss of SMARCB1 defines PDC and is mechanistically linked with EZH2 activity, providing the rationale for tazemetostat treatment[34,35,36,37,38,39,40,41]. While a distinct Polycomb-type dedifferentiation pathway driven by homozygous EED deletion has been identified in a subset of skull base DC[33]. Heterozygous deletion of chr22q, which harbors SMARCB1, is additionally observed in conventional chordoma and independently associates with worse disease-specific and recurrence-free survival in one cohort[10], raising the unresolved question of whether heterozygous loss represents a transition state toward poorly differentiated disease. At the genomic level, somatic mutations in PBRM1, SETD2, and ARID1A are among the most frequent driver events across skull base, spinal, and sacral cohorts [10,54,56,57,58], and CDKN2A/B deletion and SWI/SNF mutations appear mutually exclusive in at least one cohort[54], suggesting these represent parallel oncogenic routes rather than convergent ones. Another dichotomy is reflected transcriptomically - RNA sequencing identified two molecular subtypes — CC1 tumors enriched for PBRM1 and SETD2 mutations, and CC2 tumors characterized by Sonic Hedgehog pathway upregulation[97].

This leads to TGF-β signaling that also emerges as a recurrent mechanistic theme across multiple classification domains in chordoma. At the microenvironmental level, single-cell analyses have identified marked TGF-β signaling predominantly between macrophages and fibroblasts within the tumor stroma[99], and a partial epithelial-mesenchymal transition program, driven by TGF-β, was localized to the invasive tumor edge, independently predicting worse progression-free and overall survival[100]. Extending this, VEGFR and TGF-β were identified as co-dominant therapeutic targets in chordoma, with dual VEGFR/TGF-β CAR-T cells demonstrating superior cytotoxicity over VEGFR-only constructs against chordoma cell lines and patient-derived organoids[101]. Moreover, proteomic analysis of bone-invasive chordoma implicated TGF-β1 downregulation and reduced PTEN expression in this behavior[119]. Collectively, these findings suggest TGF-β signaling can be poised as a mechanism, linking stromal composition, immune suppression, and invasive phenotype.

The final overarching theme is variability in immune microenvironment of chordoma, consistently evident across transcriptomic, epigenomic, and proteomic classification frameworks, emerging as a major axis of biological and prognostic heterogeneity. Single-cell RNA sequencing studies have characterized a tumor microenvironment marked by abundant M2-like macrophages, exhausted CD8+GZMK+ T cells, and active TGF-β signaling between macrophages and fibroblasts[99], with recurrent tumors showing further immunosuppressive remodeling characterized by reduced antigen-presenting cell activity, fewer plasma cells, and lower BCR clonotype diversity[104]. PD-L1 expression is predominantly localized to tumor-infiltrating macrophages and lymphocytes rather than chordoma cells themselves[106], and a specific inflammatory cancer-associated fibroblast subset — spatially distant from tumor cells — has been functionally linked to promotion of chordoma progression[103], collectively suggesting that immunosuppression in chordoma is orchestrated largely through stromal and myeloid compartments rather than intrinsic tumor cell mechanisms. At the epigenomic level, unsupervised methylation clustering consistently identifies an immune-infiltrated subtype characterized by higher cytotoxic T lymphocyte, B cell, and macrophage infiltration, and lower tumor purity[8,81,82], with two studies reporting worse prognosis in this subtype[81,82]. Transcriptomic subtyping of skull base chordomas similarly identified two immune subtypes differing in macrophage and T cell infiltration, with higher CD68 and CD163 expression — markers of M2 macrophage polarization — correlating with shorter progression-free and overall survival [96]. Regarding link with the other major theme, CDKN2A/B deletions have been reported predominantly outside the immune-infiltrated subtype in one cohort[8,9], though this association is not consistent across studies[82].

This review has several limitations that must be acknowledged. Although a structured and reproducible search strategy was employed, the synthesis of included studies remains subjective. The field itself carries structural limitations that are difficult to mitigate: the rarity of chordoma means that the vast majority of published studies are retrospective in nature, and most cohorts comprise fewer than 100 patients, limiting statistical power and increasing the risk that reported associations will not replicate in independent datasets. Furthermore, the studies were conducted across anatomically distinct disease sites — skull base, spinal, and sacral — and in heterogeneous patient populations, making cross-cohort comparisons inherently difficult. Inconsistencies in molecular subtype nomenclature, differing platforms for the analyses, and variable clinical endpoint definitions compound this challenge. Finally, the rapid pace of discovery in single-cell sequencing, spatial transcriptomics, and liquid biopsy methodologies means that the evidence base continues to evolve quickly, and some conclusions drawn here will require revision as larger datasets mature. Addressing these limitations will require coordinated international efforts to establish biobanks and prospective registries capable of capturing sufficient patient numbers to validate the classification frameworks described in this review and link them rigorously to therapeutic outcomes.