Transcriptomic landscapes of functioning and non-functioning PitNET

We performed scRNA-seq of postoperative tissue from primary tumors of seven untreated GoPN and three untreated LaPN patients (see Methods for details; Suppl. Table 1 for clinical data). Subsequent UMAP embedding of 17,254 single-cell transcriptomes for the integrated GoPN samples revealed 21 clusters and that of the 20,369 single-cell transcriptomes of the LaPN samples revealed 25 clusters (Fig. 1a, b; Suppl. Figure 1a, b; Suppl. Table 2). Then, using specific marker genes and matching with reference datasets, we first annotated the definite TME clusters and then confirmed the potential tumor cell clusters by detecting deviant profiles in inferred copy number variation (CNV) (Fig. 1c, d; Suppl. Figure 2, 3).

Fig. 1

The single-cell transcriptomic landscapes of GoPN and LaPN. a UMAP plot showing integration and clustering of seven GoPN samples and b of three LaPN samples, as determined by scRNA-seq. c Cell type annotation of GoPN, and d of LaPN based on known marker genes and inferred CNV analysis (see Suppl. Figure 2 and Suppl. Figure 3). Clusters highlighted in red indicate selected tumor cell clusters (cluster 9 and 14 in GoPN, cluster 14 and 22 in LaPN) with specific gene signatures. Heatmaps of the relative expression levels of these and other tumor cell cluster-specific gene signatures are shown in (e), together with their functional gene ontology association and representative marker genes. ccImm cycling immune cells, ccTum cycling tumor cells, EC endothelial cells, epiDiff epithelial differentiation and proliferation, GoPN gonadotroph PitNET, hormSec hormone secretion, OxPhos oxidative phosphorylation, LaPN lactotroph PitNET, PC pericytes, UMAP uniform manifold approximation and projection

Intratumoral heterogeneity in GoPN and LaPN

Next, we focused on the characterization of the intratumoral heterogeneity (ITH) in GoPN and LaPN. Using differentially expressed gene (DEG) analysis, specific gene sets were identified for each individual cluster (Suppl. Table 3). These were then functionally annotated via Gene Ontology (GO) and presented in a heatmap to visualize the differences and similarities of different tumor cell subpopulations within GoPN and LaPN (Fig. 1e). In both subtypes, there are cell populations that behave relatively inertly concerning specific processes. These are processes where basic housekeeping or metabolic processes such as oxidative phosphorylation and mRNA processing are down-regulated (GoPN: cluster 1, 6, 15, 17; LaPN: cluster 0, 2, 7). In addition to these rather inconspicuous cell populations, there are also those whose gene expression signature indicate the participation in highly specific biological processes such as epithelial differentiation, hormone secretion or immune effector processes. However, it is essential to note that the respective signature genes for GoPN and LaPN are distinct despite these functional similarities. For example, the signature “epithelial differentiation and proliferation” (epiDiff) contains different representatives of the keratin gene family (GoPN: KRT8, KRT18; LaPN: KRT5, KRT7), the claudin gene family (GoPN: CLDN4; LaPN: CLDN5) or the growth factor receptor gene family (GoPN: EGFR; LaPN: FGFR1) in the two subtypes. Another signature is functionally associated with "hormone secretion" (hormSec), which indicates these tumor cells' endocrine origin. In this case, there is more overlap between the respective signatures: Both that of GoPN and LaPN contain the hormone gene CGA and the neuroendocrine marker genes chromogranin A (CHGA), secretogranin V (SCG5), and carboxypeptidase E (CPE). Further, a signature associated with “immune effector processes” can be detected in both GoPN and LaPN. Multiple genes related to immune cell activation and chemotaxis are contained in both signatures, e.g., in GoPN HLA-DMA, HLA-DQB1, and CD68, or LaPN HLA-DRA, S100A9, IL7R, CD14, and CD44. In addition, the LaPN signature contains genes related to antiviral processes, such as IFIT3 and RSAD2 (Fig. 1e).

Finally, this analysis revealed a small LaPN-restricted tumor cell population in cluster 22 (Fig. 1d, e). Its signature is characterized by a significant upregulation of numerous cell cycle and proliferation-associated genes such as cyclin B1 (CCNB1), centromere protein F (CENPF), or the proliferation marker Ki-67 (MKI67). Also included is the human securin gene PTTG1, which was initially identified as an oncogene in malignant pituitary tumors but has been associated with tumor aggressiveness and invasiveness in various other cancer entities [29]. In the following, this LaPN-specific signature is referred to as "cycling tumor cells" or ccTum. Notably, a related tumor cell population is missing in the non-functioning GoPN subtype.

In summary, both GoPN and LaPN exhibit high intratumoral heterogeneity. In both subtypes, there are similarities in functional networks within epithelial, endocrine, or immunological gene expression programs, but also a tumor cell population with highly proliferative characteristics that is exclusively present in LaPN.

Clinicopathological relevance of GoPN and LaPN single-cell signatures

Next, we investigated whether and how these signatures correlate with the clinicopathological properties of these tumors. To this end, we first checked whether these signatures can also be found in a large validation cohort, in which a total of 134 tumor transcriptomes from eight WHO-classified PitNET subtypes—including GoPN, LaPN, somatotroph (SomPN), thyrotroph (ThyPN), and corticotroph (CorPN)—have been analyzed using bulk RNA-seq [28]. For this comparison, we developed a scoring procedure in which the reference data of the validation cohort received a so-called signature score between 0 (no agreement) and 1 (complete agreement) (see Methods for details; Suppl. Tables 4, 5).

Essentially all signatures derived from our single-cell analysis were found in the bulk RNA-seq data of the different PitNET subtypes, underlining the validity and relevance of our approach (Fig. 2). However, very different scores were sometimes obtained for individual signatures within the respective subtypes. For example, while the two epithelial signatures epiDiff_GoPN and epiDiff_LaPN showed an essentially uniform distribution of their score (Fig. 2a), the values for the endocrine signature hormSec_GoPN and the proliferation-associated signature ccTum_LaPN were distributed much more specifically. It is noticeable that, in addition to GoPN itself, hormSec_GoPN has the relatively highest score in null-cell PitNET (NCPN), i.e. another representative of non-functioning PitNET (Fig. 2b). Even more striking was the specificity of ccTum_LaPN, which showed an average expression level in lactotroph PitNET that was about twice as high as in all other subtypes (Fig. 2b).

Fig. 2

Tumor-specific gene signatures correlate with clinicopathological features of PitNET. a Boxplots showing the normalized expression level of the condensed epithelial gene signatures derived from GoPN or LaPN single-cell data (epiDiff_GoPN, epiDiff_LaPN) in a bulk RNA-seq cohort comprising 134 PitNET tumors (29 gonadotroph (GoPN), 16 lactotroph (LaPN), 8 null-cell (NCPN), 35 corticotroph (CorPN), 6 thyrotroph (ThyPN), 23 somatotroph (SomPN), 9 plurihormonal (PluriPN) and 8 mixed (MixedPN) PitNET) [28]. b Boxplots of the normalized expression levels of the condensed hormone secretion signature (hormSec) from GoPN (left) and the cycling tumor cell signature (ccTum) from LaPN (right). For a statistical analysis of (a, b) see Suppl. Table 5. c, d Correlation of GoPN or LaPN gene signatures with aggressiveness level and/or clinical behavior of the bulk RNA-seq PitNET cohort (Rem, remission, n = 60; Pers, persistent, n = 48; Res, resistant, n = 12; Agg, aggressive, n = 14). Statistical analysis was performed using an unpaired two-sample Wilcoxon test, with */**/***/**** indicating p ≤ 0.05/0.01/0.001/0.0001, respectively. Non-significant results remain unmarked

In addition to their WHO classification and global gene expression profiles, other relevant data, such as the patient's response to treatment, are available for the samples of the reference cohort. Based on the clinical course, each sample is classified into one of four categories, in ascending order of severity from „remission” (Rem) to „persistent” (Pers) to „resistant” (Res) to „aggressive” (Agg) [28]. The definitions are as follows: (i) Classified as Agg are tumors that required unusual treatments (e.g., multiple surgeries, radiation therapy, or chemotherapy) and/or tumors that showed resistance to first-line treatments. (ii) Res refers to resistant tumors that grew slowly but did not require radiation or chemotherapy. (iii) Pers refers to those defined by stable disease or slow tumor growth after incomplete surgery, and (iv) Rem includes all tumors in remission after first-line therapy. Looking more closely at the PitNET subtypes we examined, the lactotroph PitNET samples stand out with an increased aggressive phenotype, along with thyrotroph, corticotroph, and plurihormonal PitNET. One quarter of the 16 lactotroph PitNET samples fall into the category Agg. In contrast, the gonadotrophic PitNET samples (n = 29) show much less aggressive features and are found in either one of the two less aggressive categories Rem or Pers (Suppl. Figure 4).

When comparing our sc-derived signatures with these clinical data, it is noticeable that the scores of all four signatures gradually increase from category Pers to Res (Fig. 2c, d). However, while in the case of the epithelial and endocrine signatures, the scores in aggressive tumors drop again, this value increases significantly in the case of the proliferation signature ccTum_LaPN (Fig. 2d).

In summary, we used scRNA-seq to identify a tumor cell population mainly present in the F-PitNET subtype LaPN and characterized by a gene expression signature associated with a more resilient and aggressive clinical presentation.

The tumor microenvironment of GoPN and LaPN

Previous single-cell studies of PitNET have not, or only sparsely, addressed the composition of the TME and its potential impact on tumor growth or other pathological processes [18, 20]. However, our study shows that GoPN and LaPN represent two entities harboring a comparatively rich TME (Fig. 1c, d) compared to PitNET of the TPIT lineage [17]. As the TME is assumed to impact pituitary tumor behavior significantly [12, 30], we have focused on a closer examination of this aspect in the following.

An initial mapping of the cellular TME landscape is shown in Fig. 3a for GoPN and Fig. 3b for LaPN. With few exceptions, both subtypes show similar distributions of the individual cell types between the respective samples (Suppl. Figure 5). TME cells represented more than half of all cells analyzed, namely 61.7% in GoPN and 51.2% in LaPN. Within the TME, immune cells dominated with proportions of approximately 95% and 98%, respectively (Fig. 3c, d). Although we were also able to detect pericytes (PC) and endothelial cells (EC) (Suppl. Figure 2), we refrained from further analyzing them due to their small number (GoPN: 349 PC, 142 EC; LaPN: 149 PC, 71 EC) and low proportion (GoPN: 4.6%; LaPN: 2.1%; Fig. 3c). This first overview of the immune cell landscape already illustrates a striking difference between GoPN and LaPN, namely the TME dominated by myeloid cell types in GoPN (64% myeloid vs. 31% lymphoid cell types) and, vice versa, dominance of lymphoid cells in LaPN (78% vs. 20% myeloid cell types) (Fig. 3c; Suppl. Table 6).

Fig. 3

Tumor microenvironment of GoPN and LaPN. a UMAP plot with cell type annotation of GoPN, and b of LaPN tumors. c Pie charts showing the relative proportions of lymphoid, myeloid, and other cell types in the TME of GoPN and LaPN. d Relative distribution of all cell types, including tumor cells in GoPN (left) and LaPN (right), based on the number of cells per annotated cluster. e Quantification of immunostaining of the myeloid marker CD68 (top) and the T cell marker CD3 (bottom) in tissue sections of GoPN (n = 20) and LaPN (n = 19) tumors. Relative tumor cell infiltration (y-axis) was assessed by expert judgment, and the differences in mean score values of GoPN (red) versus LaPN (blue) were tested for significance using a Mann–Whitney U-test. ns, not specific. B B-cells, ccLy cycling lymphoid cells, ccImm cycling immune cells, CD4−/CD8−DN CD3+CD4/CD8-double negative T cells, cDC CD1C-expressing conventional dendritic cells, EC endothelial cells, act.Macs activated macrophages (act-Macs), MC mast cells, MiMe high-level expression of mitochondrial and metabolic genes, NK natural killer cells, PC pericytes, Plasma plasma B-cells

A closer look revealed other distinct features of the immune cell landscape. We found small subpopulations of proliferating immune cells (ccImm) in GoPN and proliferating lymphoid cells (ccLy) in LaPN. In the latter, we were able to detect small populations of mast cells (MC) and plasma B cells (Plasma), both of which were absent in GoPN. Moreover, we also observed a cell group that—although clustering together with other lymphoid cell types—did not express any corresponding marker genes but was characterized by a high expression of mitotic and metabolic genes and is referred to here as "MiMe" (Fig. 3b, d).

Upon closer examination of the myeloid compartment, we find activated and M2-polarized macrophages (Macs) in the TME of subtypes. Expression of the M2-specific markers CD163 and MRC1/CD206 [31, 32] is found in cluster 9 (LaPN) as well as clusters 0, 11 and 13 (GoPN) (Fig. 3a, b). M2-Macs often possess pro-tumorigenic effects and have been associated with higher invasion in PitNET [15, 33].

Next, we observed myeloid cells with increased VCAN, S100A8/S100A9 and IL1B expression indicating that these cells are active in inflammatory processes (Suppl. Figure 2). In the following, we refer to them as activated macrophages (act-Macs). Interleukin 1 beta (IL1B) plays an essential role in the induction of inflammatory and anti-tumoral activities [34]. In contrast, elevated VCAN and S100A8/S100A9 expression has been associated with the recruitment of myeloid-derived suppressor cells (MDSC) and, thus, the creation of an immunosuppressive TME in some tumors [35,36,37]. While act-Macs express CD44 and CCR2, M2-polarized and other macrophages (GoPN: 21; LaPN: cl. 18 and 23) show high-level expression of CX3CR1. This expression pattern suggests that act-Macs represent tumor-infiltrating blood-derived macrophages, while the latter are tissue-resident cells [38, 39].

Also, there is a striking difference between the two subtypes in the distribution of T cells and natural killer (NK) cells. For instance, at 11.1%, the latter cell type is almost 2.5 times more abundant in LaPN than in GoPN, where NK cells account for only 4.5% of all cells (Fig. 3d). Moreover, while GoPN is characterized by a comparatively high number of double CD3/CD4-positive (CD4+) T helper cells, this cell type is absent in LaPN. In contrast, CD3+CD4/CD8-double negative (CD4−/CD8−DN) T cells, which normally comprise a rare subset of peripheral T cells [40], forms the largest fraction of all non-tumor cells in LaPN at almost 16%, while it is missing in GoPN (Fig. 3d).

We validated this differential immune cell by performing immunostaining on tissue sections from GoPN and LaPN tumors which confirmed significantly stronger lymphoid tumor infiltration in LaPN vs. GoPN (Fig. 3e).

Differences in the distribution of specific TAM populations

To gain more detailed insights into the different cellular composition of the myeloid compartment of the two subtypes, we isolated the respective myeloid cell clusters from GoPN and LaPN, merged them and re-clustered the integrated myeloid subset of 8,076 cells (Fig. 4a; Suppl. Figure 6a).

Fig. 4

Distinct cell type expressions in myeloid cells in GoPN and LaPN. a Cluster arrangement of the myeloid cells from both GoPN and LaPN tumors after integration. b Dot plot showing the average expression levels (avg. exp.) and occurrence (pct. exp.) of myeloid cell type-specific marker genes in the individual clusters. c UMAP plot with annotated myeloid cells from GoPN and LaPN. d Relative distribution of identified cell types within GoPN and LaPN cells. cDC CD1C-expressing conventional dendritic cells, MDSC myeloid-derived suppressor cells, Mono monocytes, TAMs tumor-associated macrophages

This higher resolution of the myeloid compartment and the subsequent comparison with a reference data set of so-called tumor-associated macrophages (TAMs) [41] led to the identification of ISG15+, C1Q+ and SPP1+ TAM populations (Fig. 4b, c; Suppl. Figure 6b). ISG15+ and C1Q+ TAMs have been previously detected in all three PitNET lineages, while this is notably not the case for SPP1+ population [42]. Furthermore, it is striking that the relative frequencies of SPP1+ TAMs differ significantly: GoPN (4.9%) contains proportionally about 4.5 times more of these cells than LaPN (1%). This is remarkable as the abundance of SPP1+ TAMs is associated with poor prognosis in lung adenocarcinoma and cervical cancer patients [43, 44] as well as the promotion of cancer stemness in pancreatic cancer [45]. Nevertheless, in PitNET we find a notably higher number of SPP1+ TAMs in the GoPN cohort, which is associated with significantly better clinical outcomes compared to LaPN.

The influence of lymphoid cell subsets on the clinical features of PitNET

As shown above the striking difference between the GoPN and LaPN TME lies in the abundance of T cells and natural killer (NK) cells (Fig. 3e). To decipher these differences more precisely, we proceeded as before and generated a subset of all lymphoid cells.

Single-cell transcriptomes of all cells of lymphoid origin from both entities (a total of 11,512 cells) were integrated into a single UMAP and re-annotated (Fig. 5a; Suppl. Figure 7b; Suppl. Table 7). The relative GoPN/ LaPN proportions of each cell type are shown in Fig. 5b, and those of individual samples in Suppl. Figure 7a. As previously observed with the myeloid subset, the higher resolution allows the identification of previously obscured cell types. For example, in addition to a large cluster of CD79B+- B cells, we identified a smaller population of CD79B−—B cells in cluster 13. Also, a small number (n = 55) of innate lymphoid cells (ILC) now appear in cluster 12. The same applies to natural killer T cells (NKT), which are identified in cluster 6.

Fig. 5

The lymphoid compartment and its relation to clinicopathological features of PitNET. a Cluster arrangement of the lymphoid compartment from both GoPN and LaPN tumors after integration. b Relative proportions of GoPN and LaPN cells in the identified clusters. c–e Expression of cycling lymphoid cell (ccLy), CD8+ T cell (CD8+), and natural killer cell (NK) signatures in PitNET tumors [28], categorized by degree of aggressiveness in (c), invasion into the sphenoid sinus in (d), or by proliferation index (percentage of MIB1-stained Ki67-positive cells) in (e). Statistical analysis was done as in Fig. 2 (Suppl. Table 9). ILC innate lymphoid cells

After cell type annotation, we derived specific gene signatures for individual, defined subgroups of lymphoid cells as described above for the tumor cells. In the next step, these lymphoid-specific signatures were then correlated with the clinicopathological properties of PitNET by comparison with the RNA-seq validation cohort (Suppl. Tables 8, 9). Starting with the levels of tumor aggressiveness, it was noticed that increasing expression of the proliferative ccLy signature was associated with increased tumor aggressiveness and resistance to treatment, as it was already the case with the proliferative ccTum_LaPN signature. Furthermore, the analysis also shows reduced resistance to treatment coinciding with increased CD8+ T cell infiltration as well as a reduced aggressiveness in correlation to a higher amount of NK cells (Fig. 5c). All other lymphoid signatures showed insignificant changes in tumor aggressiveness (Suppl. Figure 8a).

A similar picture is seen in the correlation of expression levels of ccLy and CD8+ signatures to tumor invasion data in the sphenoid sinus. Increased ccLy expression is associated with increased invasion, whereas increased CD8+ expression is associated with decreased invasion (Fig. 5d). This result complements a publication on somatotroph PitNET, in which both tumors without cavernous sinus invasion and first-generation somatostatin analogs responder tumors showed higher numbers of CD8+ lymphocytes compared to invasive or resistant tumors [46].

Finally, the expression levels of ccLy increase significantly with a higher MIB1/KI67 proliferation index of the tumor (Fig. 5e).

The correlation of tumor and lymphoid cell type signatures regarding their association with aggressive phenotypes

In both the TME and the tumor cells of gonadotroph and lactotroph PitNET, we found clear signatures with expression patterns correlating with either a better outcome (less aggressiveness, less resistance to treatment and/or lower invasiveness), namely epiDiff_GoPN, epiDiff_LaPN, CD8+ T cells and NK cells, or a worse outcome (higher aggressiveness, higher resistance and/ or more invasiveness), namely ccTum_LaPN and ccLy. Here, the question arises whether there is a linear relationship between the expression of these distinct signatures. Therefore, we computed a Pearson correlation coefficient between the tumor signatures ccTum_LAPN, epiDiff_GoPN, and epiDiff_LaPN and the lymphoid cell signatures ccLy, CD8+, and NK.

Interestingly, there is a strong positive correlation (R = 0.57) between the expression of the ccLy signature of the TME and the ccTum_LaPN signature of the LaPN tumor cells (Fig. 6a). Both signatures are associated with higher levels of aggressiveness and invasion in the sphenoid sinus (Figs. 2d and Fig. 5c, d).

Fig. 6

Correlation of distinct tumor and lymphoid cell signatures associated with better or worse tumor progression outcome. Linear correlation of expression values of a CD8+ T cell (CD8+) signature with tumor signature epiDiff_GoPN, b cycling lymphoid cell (ccLy) signature with cycling tumor cell (ccTum) signature and natural killer cell (NK) signature with both tumor signatures epiDiff_GoPN (c) and epiDiff_LaPN (d). Signatures associated with a less aggressive outcome are marked in green, while those associated with a more aggressive outcome are marked in red. Pearson correlation coefficient (R) and p-value are indicated (p)

The expression of both the CD8+ T cell signature and the NK cell signature correlate positively with the expression of the epiDiff signatures of GoPN and LaPN (Fig. 6b–d). All of these signatures are associated with a better outcome. In one case, there is a positive correlation between the expression of a signature associated with a better outcome and one with a worse outcome, NK cells with ccTum_LaPN (Fig. 6e).