Human pituitary adenomas express stemness genes and contain a distinct cell population capable of sphere formation
Multiple reports have consistently shown genes that regulate key stem cell properties such as self-renewal and differentiation, termed stemness genes, to contribute to intra- and intertumoral heterogeneity [23, 28], while predicting clinical variables such as treatment response and overall survivorship [13]. However, no such investigations have been performed in human PAs. The successful detection of stemness genes in pituitary adenomas has previously been hampered by limitations in robust, cost-effective gene expression platforms capable of detecting genes expressed at the resolution of a single transcript, as might be the case with stemness genes, which are exclusively expressed in rare TICs [15]. Using a nanoString-based 80-gene custom codeset, we determined the differential stemness gene expression profile across 14 human PAs (Table 1, PA 1–14) representing the most common hormonal subtypes (Fig. 1a). Interestingly, our analyses revealed significant intertumoral heterogeneity across PAs, indicating a spectrum of stemness in which some PAs retained a primitive gene expression network, while others displayed a less primitive genotype.
In order to functionally validate our observed spectrum of genetic stemness, we minimally cultured primary human PAs in tumor sphere media conditions as previously described [25]. Sphere formation, a function of clonally derived TICs, was observed in all primary adenoma subtypes to varying degrees (Fig. 1b), which supported our observed genetic heterogeneity in stemness. The most compelling evidence for cancer cell heterogeneity within a given tumor is obtained by using cell surface markers to prospectively isolate distinct subpopulations of malignant cells with corresponding phenotypic diversity. However, the utility of surface markers in dissecting intratumoral heterogeneity is complicated by the reported ability of certain cancer cells to reversibly transition between phenotypic states [8]. Regardless of the presence of a rigid hierarchy, determinants of stemness have been shown to contribute to treatment failure, irrespective of whether these determinants are present within a dynamic or static population. We therefore performed flow cytometric characterization of 22 human PAs for CD15, CD133, and Sox2 (Table 1, Additional file 1: Figure S1). Although, both CD15 [22, 27, 32] and CD133 [25, 26] have previously been shown to identify TICs in highly aggressive and malignant brain tumors, their prospective flow cytometric characterization of putative PAICs has not been described. Our analyses across multiple hormonal subtypes consistently characterized a distinct CD15+ population of cells within human PAs, while CD133 marked a negligible cell fraction (Fig. 1b). Given the presence of a differential stemness gene expression profile, coupled with the capacity for sphere formation and expression of putative TIC markers, our data supports the application of the TIC model for investigating the biological heterogeneity observed in human PAs.
Developmental genes of the normal pituitary are enriched in human pituitary adenomas, especially those adenomas with increased sphere formation
Since several brain tumors are comprised of a functionally heterogeneous population of cells that perpetuate tumor growth through unregulated self-renewal [13, 23], we performed functional assays to assess the subtype-dependent sphere formation rate within PAs. The rate of secondary tumor sphere formation was much lower in all PA subtypes when compared to highly aggressive and malignant tumors such as glioblastoma, medulloblastoma, and brain metastasis (data not shown). However, we did observe a wide distribution in the rate of secondary tumor sphere formation across multiple subtypes, particularly, GH-secreting and null cell adenomas (Fig. 2a). While the number of adenomas within each subgroup in our analyses was reflective of their prevalence in the general population, we were limited to a single tumor for a small number of subtypes (ACTH, FSH/LH, FSH/PRL, LH), making it difficult to comprehensively interrogate the spectrum of secondary sphere formation rates among these tumors. In order to supplement our sphere formation assays, we investigated the gene expression profile of Pax7 and Sox2 within our PA cohort. Both, Pax7 [10] and Sox2 [1] have been shown to identify unique primitive cell populations in the developing murine pituitary with the potential for initiating and maintaining tumorigenesis following transformation events. Both genes contained an assorted expression profile across subtypes (Fig. 2b, c, Additional file 1: Figure S1), further highlighting the significant inter- and intrasubtype heterogeneity within PAs. Given the wide distribution in sphere formation and gene expression profile of putative PA-specific stemness genes, Pax7 and Sox2, we sought to assess the secondary sphere formation rate of Pax7high and Sox2high PAs compared to Pax7low and Sox2low PAs, respectively. Using a median cutpoint based on the gene expression values of Pax7 and Sox2, we plotted PAs according to the high or low expression of these genes relative to the rate of secondary tumor sphere formation (Fig. 2d, e). An overall trend in keeping with an increased sphere formation among Pax7high and Sox2high PAs was detected, which fostered the development of our framework in which a distinct population of cells within PAs across all subtypes have an enhanced ability for sphere formation, which may be regulated by the re-expression of genes present during pituitary development.
CD15 enriches for pituitary adenoma-initiating cells
While the ability of human PA cells to be cultured as tumor spheres and the expression of genes associated with ontology have previously been described [3, 19, 35, 36], the prospective flow cytometric characterization and isolation of putative PAICs based on cell surface marker expression has yet to be performed. Such assays provide valuable evidence in support of heterogeneous cell-cell interactions that may be critical for the evolution of tumors from ontology to oncology and throughout the course of therapy. We performed flow cytometric characterization of CD133 [25, 26], Sox2 [30], and CD15 [22, 27, 32], three markers that have reproducibly been shown to mark TICs in a number of pediatric and adult brain tumors (Fig. 1b, Table 1). Since our work consistently showed CD15 to identify a unique fraction of PA cells, we assessed the genetic and functional cellular differences between CD15high and CD15low adenomas. Using a median cutpoint to distinguish between CD15high (n = 11, PA 3, 4, 8, 10, 11, 13, 16, 17, 19, 21, 22) and CD15low (n = 11, PA 1, 2, 5, 6, 7, 9, 12, 14, 15, 18, 20) adenomas, we investigated the expression of developmental genes Pax7 and Sox2, which we found to correlate with an increased rate of secondary sphere formation (Fig. 2d, e). Both, Pax7 (Fig. 3a, P < 0.005) and Sox2 (Fig. 3b, P < 0.001) were significantly enriched in CD15high adenomas when compared to CD15low tumors. Therefore, we hypothesized that CD15 may mark a unique fraction of cells with an enhanced potential for self-renewal and tumor initiation. Given the technical challenges in culturing primary human PAs combined with flow cytometric cell-sorting, a limited number of PAs (n = 2, PA 3, 22) were sorted for CD15+ and CD15- cells (Fig. 3c). Sorted cells were plated for secondary sphere formation assays, which yielded a significant difference in the frequency of spheres in CD15+ PAICs compared to CD15- cells (Fig. 3d, e, P < 0.01). Similarly, CD15high (n = 5) adenomas maintained a much higher sphere formation capacity when compared to CD15low (n = 5) tumors (Fig. 3f, P < 0.01), further demonstrating the presence of a highly tumorigenic regulatory network within CD15+ PAICs. Collectively, our data illustrates the prospective isolation and characterization of a distinct cell population in human PAs, which may contribute to intratumoral heterogeneity but also distinguish tumors based on their relative expression of CD15.
CD15high adenomas and CD15+ pituitary adenoma-initiating cells are capable of tumor initiation in human-mouse xenografts
The most definitive method to assess the presence and function of PAICs is to carry out in vivo xenotransplantation assays and thereby quantify the frequency of tumor initiation [26]. Given the limited proliferative potential of human PA primary cultures, we initially compared the tumor-initiating capacity across the unsorted bulk population of cells isolated from CD15high and CD15low adenomas. Three limiting dilutions (1×105, 5×104, 1×104) were used for intracranial injections of two biological replicates, each for CD15high (PA 19, 21) and CD15low adenomas (PA 18, 20). Interestingly, no tumors were formed from CD15low adenomas from all three dilutions (Fig. 4b). In contrast, CD15high adenomas generated small lesions at the single dilution of 1×105 cells (Fig. 4a), in keeping with the rare frequency of PAICs in benign brain tumors. While this data informs us of an increased capacity for tumorigenesis within CD15high adenomas, the gold standard for identifying and thoroughly characterizing TICs is through the prospective flow cytometric cell sorting of a putative TIC marker. We therefore set out to perform the first xenotransplantation assay in sorted human PA cells. Since cell-sorting for in vivo xenotransplantation assays requires a significant number of viable cells to ensure adequate engraftment, we were limited to a single biological replicate (PA 22) for this proof of principle experiment. In order to assess the frequency of PAICs in CD15+ and CD15- cells, we performed a limiting dilution consisting of 5×104 (n = 3) and 1×105 (n = 2) cells. We observed robust, invasive, multi-focal tumors with 5×104 CD15+ cells (Fig. 4c). In contrast, no tumors were generated with 5×104 and 1×104 CD15- cells (Fig. 4d) and two additional mice injected with 5×104 and 1×104 CD15- cells were still alive at the time of writing this manuscript. Further characterization of tumors initiated by CD15+ PAICs demonstrated significant mitoses and nuclear atypia (Fig. 4e, f), in support of CD15 functioning as a highly tumorigenic cell population in human PAs. Overall, our data support the enrichment of PAICs in CD15high adenomas, which is made apparent following cell sorting for CD15+ PAICs.
Xenografts generated from CD15+ pituitary adenoma-initiating cells maintain multi-lineage specification in vivo
An exhaustive immunohistochemical analysis was performed to further characterize the tumors generated from CD15+ PAICs. Given that the histological subtype of the original patient sample was a null cell adenoma (PA22), the lack of immunoreactivity within the xenograft for our pituitary hormone panel was expected (data not shown). Both samples were also immunonegative for GFAP, alpha-smooth muscle actin, EMA, and TTF1 (data not shown). In support of their neuroendocrine origin the primary patient sample and xenograft were both immunopositive for synaptophysin (Fig. 5a) and CD56 (Fig. 5b). However, desmin (Fig. 5c) and myogenin (Fig. 5d) were exclusively and unexpectedly expressed in the xenograft, highlighting the potential for multi-lineage differentiation within CD15+ PAICs. Interestingly, cytokeratin maker (AE1/AE3) expression was present in the original patient sample but was not detected in the xenograft (Fig. 5e), indicating the possibility of clonal divergence between the primary patient tumor and corresponding xenograft. Niche-specific effects may also contribute to these surprising results as our intracranial xenografts were derived from injecting cells into the frontal lobe and not the mouse pituitary. Nevertheless, previous reports of murine pituitary folliculostellate cells having the ability to differentiate into multinucleated tubular cells with immunoreactivity for myogenin [11, 17], raises the possibility of CD15+ PAICs representing a putative folliculostellate cell despite the lack of S100 expression in our xenografts.
CD15 is highly expressed in recurrent adenomas when compared to matched primary samples
While the identification of TICs has enhanced our conceptual understanding of the complex biological networks that initiate and maintain brain tumorigenesis, the clinical utility in identifying of these cells is largely dependent on their therapeutic targeting and use in prognostication as a measure of overall outcome. Since prior work has demonstrated an increasing TIC frequency to be associated with tumor aggressiveness [25, 26] and poor patient outcome [14, 20, 21], we investigated the enrichment of CD15 in recurrent/residual PAs. Given that recurrence was defined as reonset of clinical symptoms or radiological tumor growth >25%, we combined tumor recurrence and progression. We specifically assessed two primary and matched recurrent samples in which there was a significant period of time following surgery for the primary PA and the presence of a recurrent lesion (Fig. 6a, Table 2). Using our nanoString-based 80-gene custom codeset, we identified a subgroup of stemness genes that were enriched in recurrent adenomas compared to their matched primary tumors (Fig. 6b, n = 4, PA 23, 24, 25, 26). Of particular interest was the increased expression of components of developmental signaling pathway such as Ptch1 and Axin2, implicating the activation of these targets in driving PA recurrence. In order to validate CD15 as a clinically applicable marker of PAICs, we performed in silico analyses on a non-overlapping cohort of PAs for which gene expression profiles were curated for non-malignant pituitary glands, primary adenomas, and their recurrent tumors [16] (Fig. 6c). CD15 expression was able to discriminate between adenomas and normal pituitary specimens and was elevated in recurrent adenomas when compared to primary tumors (Fig. 6d).
Despite the enhanced expression of CD15 within recurrent adenomas on gene expression profiling, routine clinical neuropathology is primarily based on immunohistochemical staining. Recent work with molecular subgroups for pediatric brain tumors have attempted to distill the diagnosis of these clinically-relevant subgroups into cost-effective methods that may be readily introduced in both academic and non-academic settings [18, 34]. We therefore, performed CD15 immunostaining on a primary and matched recurrent PA as a proof of principle in describing the clinical utility of a putative PAIC marker. There was a substantial increase in CD15 expression within the recurrent tumor relative to the primary (Fig. 5e, f), illustrating the importance of combining functional in vitro and in vivo studies with novel technological platforms so that high resolution gene expression profiling may be easily translated to routine clinical immunostaining. Although, it is important to recognize CD15 enrichment in recurrent adenomas may also be due to the survival of proliferating cells or invasion of CD15+ non-resected adenoma cells. Collectively, CD15 identifies a unique population of PAICs that may drive tumor maintenance and eventual recurrence in response to multiple developmental signaling pathways.