Chordomas are rare, heterogeneous, notochord-derived malignancies of bone. These originate within the skull base (30%), vertebral spine (20%), and sacrum/coccyx (50%) (1–5). Chordomas comprise 1-4% of bone malignancies and possess a low annual tumor incidence of approximately 1 per 1,000,000 individuals (3, 4, 6, 7). Although they display patterns of indolent growth, chordomas expand in a locally destructive manner (2, 6), for which treatment options remain limited. Standard of care currently consists of surgical resection when feasible and radiotherapy (2). However, radical resection, our most effective treatment (8), is difficult to achieve, particularly when extricating skull base tumors due to a risk of damaging important, nearby neurovascular structures including cranial nerves and the brainstem (1, 2). Residual tumor cells render disease relapse likely (1). Furthermore, chordomas possess well-described characteristics of chemoradiation resistance and a high recurrence propensity (2, 6). Recurrence is deemed the most important factor in patient mortality following diagnosis (6, 9). Metastasis results in 3-30% of cases, most often in the context of a recurrent tumor (2, 10, 11). Cancer stem cells have been implicated in resistance to common therapies, recurrence, and metastasis (12–14), and thus, further investigation into this subpopulation may generate new angles from which to mitigate chordoma treatment failure.

The cancer stem cell (CSC) population is a group of tumor-initiating and tumor-maintaining cells with properties of self-renewal, de novo cancer formation following orthotopic implantation, and multi-lineage differentiation (12–17). To be accurately classified as such, CSCs should exhibit functional tumorigenic behavior, including colony- and tumorosphere- formation, increased migration, and enhanced invasion. CD15⁺CD133⁺ chordoma cells were recently were shown to meet these criteria in a 2019 study by Tuysuz et al (18). Prior to this, chordoma cells with enhanced CD15⁺CD133⁺ expression had been already been shown to grow anchorage-independent colonies (19). Importantly, despite the low prevalence of this cancer subpopulation (17), CSCs are implicated in radioresistance, chemoresistance, and metastasis (20, 21). CSCs have been shown to promote tumor formation in a host of other cancers including glioblastoma, myeloid malignancies, melanoma, and cancers of the breast, lung, colon, and pancreas (15, 17, 21–25). However, the expression of surface markers that define CSCs are tissue-type specific (17). Prior chordoma studies have illustrated that CSCs may be defined by surface expression of various combinations of markers, including CD15, CD24, CD133, and ALDH (15, 19, 20, 25). Investigation into the targeting of CSCs has been ongoing for years, using a range of strategies from focusing on stem cell associated pathways, such as the Wnt/β-catenin and Notch pathways, to immunologic techniques like adoptive cell transfer and checkpoint blockade, and even chimeric antigen receptor (CAR) T cell therapy directed at CD133⁺ CSCs (13, 16). However, much progress remains to be seen in the study of chordoma CSCs.

Herein we build upon work previously published by our group that identified chordoma CSCs with flow cytometry in four cell lines and showed that CSCs could be eliminated by avelumab (PD-L1 inhibitor)-mediated antibody dependent cellular cytotoxicity (ADCC) (20). We also demonstrated significantly increased surface expression of NK-activating receptor ligand B7H6 and PD-L1 in chordoma CSCs as compared to non-CSCs within a single cell line, UM-Chor1 (15). Our latter study went as far as to show preferential CSC vulnerability to a combinatorial treatment approach of natural killer cells, an anti-PD-L1 agent, and an IL-15 superagonist intended to induce NK- and T-cell compartment killing effects (15). These investigations provide initial evidence for the ability to enhance elimination of tumor-initiating cells in chordoma. To prime the development of efficacious molecular targeted therapy for decreasing CSC burden in chordoma, a better characterization of CSC behavior and phenotypic expression, or of viable immunotherapeutic targets found on CSCs, must be achieved.

With this study, we characterize chordoma CSCs by phenotypic surface marker expression and spatial distribution in two contexts, in vitro using flow cytometry and in vivo using multispectral immunofluorescence. We hypothesized that all chordoma cell lines examined in vitro would preferentially express immunotherapeutic target surface ligands in a subpopulation of CSC marker-enriched cells. Multispectral immunofluorescence (mIF) is an imaging method by which we may individually identify cells of interest within the tumor microenvironment, detect protein coexpression on these cells, and understand the spatial relationship among these (26). We proposed that using mIF, we could detect the presence and relative location of CSCs within operative chordoma samples. Such efforts to better characterize and later develop targeted therapy against chordoma CSCs may pave a path worth pursuing for improved control against recurrent and lethal disease.

Cancer stem cells (CSCs) have been associated with tumor recurrence and resistance to chemoradiation (20, 21) in a number of malignancies. Prior chordoma studies have demonstrated that CD15, CD24, CD133, and ALDH are markers for tumorigenic CSCs (18–20, 25). First, chordoma cell lines were investigated in vitro using flow cytometry for CSC density and characteristics. CSCs were defined using dual positive criteria for CD24 and CD133 expression based on a prior in vitro designation of the residential CSC population using this signature (20). Except for CH22, all chordoma cell lines had less than 5% CSC burden (SD 0.002-0.016). CH22, a recurrent, metastatic chordoma cell line of sacral origin, had the highest proportion of CSCs, with CSCs representing 16.98% (SD 0.014) of the cell population ( Table 1 ).

B7H6, UL16 binding protein 1 (ULBP1), and MHC class I chain-related protein A and B (MICA-B) are ligands for activating receptors on natural killer cells (27–30). B7H6 was overexpressed in CSCs as compared to non-CSCs in all cell lines, evidenced both by measures of percent positivity and mean fluorescence intensity (MFI) ( Figure 1 , Supplementary Table 2 ). B7H6 expression was increased by >20% in the CSC group in cell lines CH22 and UCH17M ( Supplementary Table 3 ). The mean B7H6 percent positivity of chordoma cells across cell lines was 13.2% in the CSC group and 0% in the non-CSC group (p <0.05) ( Figure 1 , Supplementary Table 2 ). The B7H6 antibody MFI detected across chordoma cell lines was 4,606 in the CSC group and 180 in the nonCSC group (p <0.0001) ( Supplementary Table 2 ). Similarly, ULBP1 was overexpressed in CSCs at a mean of 8.2% compared to 0% in non-CSCs (p <0.05) and at a MFI of 5691 in CSCs compared to 111 in non-CSCs (p <0.001) ( Figure 1 , Supplementary Table 2 ). This trend also held true for the MICA-B antigen, which had increased expression in CSCs, mean of 11.3% in CSCs and 0.1% in non-CSCs (p <0.01) and MFI of 20,146 in CSCs and 5,167 in non-CSCs (p <0.05) ( Figure 1 , Supplementary Table 2 ).

Human epidermal growth factor receptor 2 (HER2) and epidermal growth factor receptor (EGFR) are tyrosine kinases of the epidermal growth factor receptor family, and mutations of these are implicated in the development of multiple cancers (31). Both tyrosine kinases were uniformly highly expressed as measured by percent positivity in CSCs (HER2 mean 99.0%, EGFR mean 99.0%) and non-CSCs (HER2 mean 96.5%, EGFR mean 96.5%). However, the MFI of these tyrosine kinases was increased in the CSC group (HER2 mean 4,775, EGFR mean 5,209) as compared to the non-CSC group (HER2 mean 708, EGFR mean 1,277). The only statistically significant comparison in this series was HER2 MFI (p <0.05) ( Figure 1 , Supplementary Table 2 ).

Programmed cell death ligand 1 (PD-L1) forms part of the PD-1/PD-L1 pathway exploited by neoplasms to evade immune surveillance, and its inhibition has been shown to be clinically efficacious against a number of cancers (26, 32, 33). CSCs, as compared to non-CSCs, overexpressed PD-L1 in all cell lines except CH22 ( Supplementary Table 3 ). PD-L1 was identified in 65.2% of CSCs versus in 28.6% of non-CSCs on average (p <0.05) and at a MFI of 8,716 versus 1,545 ( Figure 1 , Supplementary Table 2 ). An increase of >20% PD-L1 expression within the CSC group was appreciated in nearly every chordoma cell line evaluated ( Supplementary Table 3 ). In CH22, the percent positivity of PD-L1 was comparable between groups (CSCs 16.2%, non-CSCs 20.1%). However, even in this cell line, the MFI of PD-L1 was notably greater in CSCs (3,777) as compared to non-CSCs (653) ( Supplementary Table 3 ). All values included above are means of technical triplicates and representative of three independent experiments. Overall, the amount of CSCs analyzed in vitro within chordoma cell samples ranged from 352-11,100. The average number of CSCs analyzed per experiment using flow cytometry was 2533.

In vivo, 18 resected chordoma samples were evaluated using multispectral immunofluorescence for CSCs, defined as cytokeratin positive chordoma cells that triple stained for CD15, CD24 and ALDH ( Figure 2 ). Of all pooled chordoma cells across these tumor samples, CSCs comprised 1.39% of the population on average ( Table 2 ). 75.18% of these CSCs were found to be PD-L1 positive ( Table 2 ), a percentage notably higher than the 18.94% of general chordoma cells identified with PD-L1 positivity. No differences in chordoma cell PD-L1 positivity, CSC burden, or CSC PD-L1 positivity were identified with clinical correlations of patient age, tumor anatomic site of origin, disease stage, or radiotherapy treatment exposure ( Supplementary Figure 2 ).

An infiltration spatial analysis of chordoma CSCs revealed an average distance of 71.51 μm (SD 73.40 μm) between CSCs and stroma. No difference in CSC distance to stroma was identified by patient age, tumor anatomic site of origin, disease stage, or radiotherapy treatment exposure ( Supplementary Figure 3 ). A series of density heat maps created of each individual chordoma sample suggested a tendency of chordoma CSCs to cluster. This cellular subpopulation tended to be in groups, near in proximity to other chordoma CSCs. A series of four density heat maps with representative chordoma CSC clustering behavior is provided in Figure 3 .

Chordoma is a devastating, locally invasive and destructive malignancy of the skull base, spine, and sacrum (2, 3, 5). Barriers to successful treatment lie in the tumor’s characteristic ability to resist chemoradiation and recur following surgical resection (2, 6). Chordoma cancer stem cells, a minority cellular subpopulation responsible for giving rise to and maintaining a proliferative tumor, are implicated in these aggressive attributes (12–14, 20, 21). This data demonstrates that NK activating ligand, tyrosine kinase, and PD-L1 markers, representative of potential immunotherapeutic targets, are favorably expressed by CSCs in six chordoma cell lines ( Figure 1 , Table 2 ). Furthermore, this study illustrates, for the first time, the in vivo presence and clustered spatial distribution of CSCs within the chordoma tumor microenvironment ( Figures 2 , 3 ). No statistically significant differences in CSCs were appreciated by clinical correlates of patient age, tumor anatomic site of origin, disease stage, or radiotherapy in this patient cohort ( Supplementary Figure 2 ). Chordoma CSCs quantified from cell culture using a CD24⁺CD133⁺ signature comprised <5% in all cell lines but one ( Table 1 ), whereas CSCs appraised within imaged chordoma tumor samples using the CD15⁺CD24⁺ALDH⁺ signature comprised 1.39% of tumor cells ( Table 2 ). Results confirm the sparse nature of this progenitor subpopulation and highlight the logistical challenge of its study in chordoma, a slow-growing tumor.

In this study, NK ligands B7H6, ULBP1, and MICA-B are expressed on the surface of a greater quantity of cells and at a significantly increased concentration within the CSC group as compared to the non-CSC group across six chordoma cell lines ( Figure 1 ). Engaging NK cells against chordoma CSCs is important as NK cells possess a unique ability to exert cytolysis in the absence of antigen presentation, while also releasing cytokines and precipitating a coordinated organization of T cells (15, 34). B7H6 is part of the B7 family and binds natural cytotoxicity receptor NKp30 (27, 35). The NKp30-B7H6 complex precipitates a crucial NK cell-driven immune response against a tumor (27, 36). A phase I dose-escalation trial of BI 765049, a B7H6/CD3 T cell engager, with or without ezabenlimab (PD-1 inhibitor) for advanced solid tumors is ongoing (NCT04752215). UL16 binding protein 1 (ULBP1) and major histocompatibility complex (MHC) class I-related chain A and B polypeptides (MICA-B) are ligands of natural killer group 2 member D (NKG2D), an activating receptor of NK and cytotoxic T cells (29, 30, 37). MICA-B may prove useful as an immunotherapeutic target given its described localization to tumor cells while mostly sparing surrounding non-cancerous tissue (30). MICA-B is typically expressed on cell surfaces at low levels but is triggered by viral illness or malignant transformation (38). As the presence of MICA-B on malignant cells promotes an NK cell and cytotoxic T cell antitumor immune response, several strategies to augment this effect have been proposed, including employing MICA-B targeted antibodies to facilitate antibody dependent cellular cytotoxicity (ADCC) or to prevent the immune complex’s proteolytic shedding (39). One additional way to augment the NK and T cell coordinated response against CSCs, particularly in the setting of increased CSC NK ligand expression, could be to employ N803, an interleukin (IL)-15/IL-15-Rα superagonist (N803) previously shown by our group to enhance the killing of chordoma CSCs¹⁵. Of note, the authors only measured surface NK markers using both flow cytometry and mIF methods in the CSC and nonCSC groups rather than tumor-derived soluble NKG2D ligands, which are known to encourage immune suppression. Given the negative effect of soluble NKG2D ligands on anti-tumor NK cell function and on immune checkpoint blockade efficacy (40), the future study of these in chordoma may prove useful.

This data demonstrates that EGFR and HER2, both receptor tyrosine kinases (31), are expressed on the surface of nearly all chordoma CSCs and non-CSCs alike, but are identified at higher concentrations within the CSC group ( Figure 1 ). Prior studies have described the amplification, transcriptional upregulation, and over-expression of EGFR in cancerous states and delineated this a biomarker of tumor resistance (41). In 2021, a review article described 14 globally approved EGFR reversible and irreversible inhibitors for anticancer treatment, including gefitinib and erlotinib, some of which may be considered in future studies of combinatorial immunotherapy against chordoma with a special focus on CSCs (42). In fact, a 2013 study has already demonstrated that erlotinib, a small molecule EGFR inhibitor, impeded the growth of a patient-derived chordoma in a xenograft model (43). Currently, three listed clinical trials are investigating EGFR inhibitors, including cetuximab (NCT05041127), afatinib (NCT03083678), and anlotinib hydrochloride versus imatinib (NCT04042597) for advanced, unresectable, or metastatic chordoma. Similarly, overexpression of human epidermal growth factor receptor 2 (HER2) in cancer has been linked to worse prognoses in breast, gastric, esophageal, ovarian, and endometrial malignancies (44). Monoclonal antibodies, trastuzumab and pertuzumab, and small molecule inhibitors, lapatinib, neratinib, and tucatinib developed to thwart HER2 (45) may be worth investigating in the context of chordoma CSCs.

PD-L1 was expressed by 65% of CSCs isolated from chordoma cell lines ( Figure 1 ) and by 75% of CSCs recognized in situ within the chordoma tumor microenvironment ( Table 2 ). Programmed cell death ligand 1 (PD-L1), expressed on many tumor cells, binds to programmed cell death 1 (PD-1) on T cells, B cells, NK cells, and myeloid derived suppressor cells (MDSCs) to drive an inhibitory response that suppresses T cell activity and fosters self-antigen tolerance. However, cancer cells exploit this pathway to circumvent the anti-tumor host immune response (33, 46). Preclinical evidence for the use of PD-L1 inhibitors suggests that chordoma cells express PD-L1, expression of which is further inducible by interferon gamma, and that chordoma CSCs are sensitive to PD-L1 inhibitor-mediated ADCC (15, 20). Current FDA approved PD-L1 inhibitors include avelumab, durvalumab, and atezolizumab for specific uses in advanced urothelial carcinoma, renal cell carcinoma, and non-small cell lung cancer (NSCLC), among others (46). Although pembrolizumab and nivolumab (PD-1 inhibitors) have been the more commonly studied immune checkpoint inhibitors in chordoma, durvalumab and FAZ053 (monoclonal antibodies with PD-L1 targets) were applied to limited chordoma treatment regimens with some clinical benefit (47, 48). To date, few ongoing clinical trials for PD-1/PD-L1 inhibitors include chordoma patients (NCT02936102, NCT02834013, NCT03190174). A neoadjuvant trial approach for PD-L1 blockade in chordoma would be interesting to determine whether targeting the chordoma PD-L1⁺ CSC subpopulation prior to standard of care surgical resection would help reduce disease recurrence.

Chordoma CSCs were found to take on a clustered pattern across in vivo tumor samples analyzed with multispectral immunofluorescence ( Figure 3 ). Prior studies have similarly described a CSC niche, or specialized microenvironment within a tissue that is home to CSCs. It has been reported that small tumor cell clusters, surrounded by a favorable regional environment, may be present prior to the onset of metastasis (49). The average distance between chordoma CSCs and stroma across samples was found to be 71.51 μm. For reference, the cancer cell diameter has been reported at about 20 μm (50, 51), from which we can deduce that the mean CSC in our sample was seen at a distance equivalent to 3-4 cancer cell lengths from the tumor parenchyma periphery. This describes a target cell bundle within the chordoma tumor microenvironment to pursue using immunotherapeutic tools.

Therapeutics that minimize tumor bulk without effectively extinguishing progenitor CSCs, such as surgery or cytotoxic chemoradiation, encouragingly provide lesion regression but may fail to confer a long-term solution to halt chordoma regrowth and spread (12, 52). This underscores the importance of eliminating the CSC subpopulation for longitudinal treatment results. Ideally, a combination of therapy that quickly reduces tumor burden while ensuring enduring relapse prevention (12) would maximize patient survival. Reevaluating NK cell and T-cell immunotherapies against antigen targets on CSCs may be an avenue from which to explore this. Based on the study results presented here, the authors propose that the design of future immunotherapeutics for chordoma hone in on extinguishing clusters of CSCs, a strategy to be later combined with standard of care treatment. These efforts may begin with anti-PD-L1, anti-EGFR, or B7H6-enhancing techniques in the setting of data that supports CSC overexpression of these.

To our knowledge, this study offers the first detailed assessment of differential marker expression by CSC subpopulation status in chordoma. It also provides new insight into the in situ presence and arrangement of chordoma CSCs. The complementation of two methods, flow cytometry for which a precedent has been well-established, and mIF, a method much newer to chordoma, yields different lenses through which to view what have thus far been poorly understood chordoma CSCs. Despite the heterogeneous nature of chordoma and the differences in technique employed, the proportion of chordoma CSCs described in cell culture and tumor samples here were analogous, supporting our study results’ validity. The mIF technique used conferred strengths of single cell identification of rare cells within a tumor microenvironment, such as a CSC. Standardization of tumor staining using the PerkinElmer platform also minimized human error that accompanies staining samples by hand. Lastly, a 2012 paper by Yu et al. on CSCs proposed a need for future investigation into the link between treatment resistant CSCs and clinical outcomes, and it suggested starting with identification of CSC-specific surface markers (14). One particular strength of this study is the knowledge it supplies to begin answering this key question.

The use of dissimilar CSC definitions is a limitation of this study. Fujii et al. indicated increased CD15 and ALDH biomarkers in CD24^high/CD133^high chordoma cells, termed the residential CSC population (20), substantiating the authors’ rationale for a CD24⁺CD133⁺ signature to discern CSCs in vitro. Furthermore, CD15 expression across all cell lines here was greater in CD24^high/CD133^high chordoma cells than in the general chordoma cell population. However, significant staining challenges were encountered while validating the CD133 antibody for multispectral immunofluorescence (mIF), despite having tested multiple antibody versions from a variety of makers at a wide range of dilutions. Thus, the authors deemed a triplicate CD15⁺CD24⁺ALDH⁺ signature to be most specific and reliable in the quantification of CSCs within chordoma tumors. An inherent constraint of the mIF technique includes a maximum of six markers per antibody panel, four of which were reserved for CSC markers (CD15, CD24, ALDH) and cytokeratin, restraining the authors’ ability to efficiently corroborate flow cytometry findings of CSC surface biomarker expression in an in vivo setting. The use of emerging spatial transcriptomics, single cell sequencing, and other techniques providing increased simultaneous marker evaluation may aid in the further study of this tumor sub-population. While the mIF technique offers a useful in situ illustration of a tumor’s individual cells and compartments, it provides a still snapshot and cannot measure growth, division, or differentiation of CSCs over time. The authors were also unable to report average distances between CSCs due a spatial analysis software limitation. This study described no statistical differences in CSC surface marker expression or spatial distribution by tumor characteristic or treatment regimen. A small sample size (n= 18) of tumors with retrospectively-collected clinical data, comprised only of two recurrent tumors and with no metastatic tumors, restricted the authors’ ability to draw meaningful conclusions regarding potentially informative correlations between CSC properties and patient status. The future study of CSCs in recurrent and metastatic chordomas, for which a more robust set of clinical data is required, represents a principle area of interest given the discussed associations between tumor-sustaining cells and recurrence and survival outcomes. Finally, a prior study of this same patient cohort characterized the myeloid cell, T cell, and natural killer cell compartments of the chordoma tumor immune microenvironment (TIME) using mIF (26). This study’s tumor images and those of the TIME analysis were unable to be superimposed to explore relationships between immune cells implicated in an anti-tumor response and the chordoma CSC subpopulation. However, this represents a future direction for this work.

This study provides a deeper understanding of the chordoma cancer stem cell surface phenotype and in situ organization. The sparing of CSCs with currently available antitumor therapies may account for continued high recurrence and metastasis rates. Thus, developing therapeutic regimens against tumor-initiating and sustaining CSCs may improve longitudinal chordoma control. Potential immunotherapeutic targets, such as PD-L1, NK binding ligands, and tyrosine kinases, are elucidated here for the elimination of chordoma CSC clusters.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the National Institutes of Health Institutional Review and the Johns Hopkins Institutional Review Board. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin because the National Institutes of Health Institutional Review Board provided an exemption and all Johns Hopkins participants had written consent for the use of tissue for research.

DL: Validation, Writing – review & editing, Writing – original draft, Visualization, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. KF: Writing – review & editing, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. MP: Writing – review & editing, Supervision, Resources. YR: Formal analysis, Writing – review & editing, Investigation, Data curation. JK: Writing – review & editing, Conceptualization. WL: Writing – review & editing, Investigation, Data curation. DP: Conceptualization, Writing – review & editing. CA: Writing – review & editing, Visualization, Resources, Data curation, Conceptualization. GG: Resources, Writing – review & editing, Conceptualization. JG: Writing – review & editing, Conceptualization. JH: Writing – review & editing, Visualization, Supervision, Resources, Data curation, Conceptualization. NL: Writing – review & editing, Visualization, Supervision, Resources, Funding acquisition, Data curation, Conceptualization.