This retrospective study enrolled 72 adult patients with NF-PitNET from 26 tertiary centers from all over Spain who underwent pituitary surgery and had tissue availability from 2014 to 2020. NF-PitNET was diagnosed by magnetic resonance image (MRI) of the sella turcica with a tumor visualized, in the absence of symptoms suggesting hormone hypersecretion and a biochemical confirmation of normal or hyposecretion of pituitary hormones. All tumors underwent surgical treatment. Furthermore, the pathologist selected a surplus of tumor to be embedded in RNAlater (Invitrogen, Carlsbad, CA, USA) for research purposes. All the tumors were naïve to medical treatment and radiotherapy. NF-PitNET invasiveness and size were established according to the preoperative MRI following the conceptual classification by Raverot (31). Tumor recurrence was considered when a new tumor image was detected in patients with no apparent postsurgical remnants or when a regrowth of a known remnant tumor was of sufficient clinical entity that required a second surgical intervention. The minimum surveillance time for considering a recurrence was 2 years, with a median of 5.32 ± 2.05 years for a total number of 58 patients. The clinical and neuropathogical characteristics of the cohort are described in Table 1 . Sixteen non-tumoral pituitary samples from autopsies (8 samples) and organ donors (8 samples) were also analyzed as a non-pathological condition reference (mean age: 62.4 years ± 10.9; 37.5% females).

The molecular data of the GH-secreting adenomas used was obtained from our previous paper where the cohort is fully described (23). Briefly, a total of 57 (32 women) acromegaly patients from the REMAH cohort who underwent pituitary surgery and had tissue availability was used. Mean age was 45.74 ± 12.35, 82% of the tumors were macroadenomas, 68% presented extrasellar extension, with a mean tumor diameter of 19.49 ± 10.03.

The study was conducted in accordance with the ethical principles of the Declaration of Helsinki and implemented and reported in accordance with the International Conference on Harmonized Tripartite Guideline for Good Clinical Practice. The study was approved by the Germans Trias i Pujol Hospital Ethical Committee for Clinical Research (Ref.: EO-11-080 http://www.ceicgermanstrias.cat/index.html). The protocol and informed consent forms were approved by the institutional review board of the participating centers, independent ethics committee, and/or research ethics board of each study site. All patients provided written informed consent to participate in the study.

Formalin-fixed paraffin-embedded tumor samples were used for anterior pituitary hormone expression assessment. Growth hormone (GH), prolactin (PRL), thyroid-stimulating hormone (TSH), adrenocorticotropic hormone (ACTH), luteinizing hormone (LH) and follicle-stimulating hormone (FSH) expression were all tested with the corresponding antibodies, following local protocols for diagnostic assessment. Hematoxylin-eosin and reticulin staining, as well as cytokeratin, Ki-67, α-subunit and p53 immunolabeling, were also carried out in most of the cases as part of the standard pathological evaluation. Plurihormonal tumors were defined as tumors that showed immunoreactivity for more than one hormone that cannot be explained by normal cytophysiology or developmental mechanisms. We considered negative tumors those that did not express positivity for any of the hormones.

Representative fragments of the fresh tumor were selected by a pathologist and embedded in RNAlater (Invitrogen, Carlsbad, CA, USA) for 24 h, after which the samples were frozen at −80°C. Total RNA was isolated from pituitary adenomas using AllPrep DNA/RNA/miRNA Universal Kit (Qiagen, Hilden, Germany). We removed contaminating genomic DNA from RNA by treating samples with RNAse-free DNAse twice: the first time during the extraction of RNA following the manufacturer’s protocol and the second time, before the retrotranscription, for which ezDNase Enzyme (Invitrogen, Carlsbad, CA, USA) was used. The quantity and purity of DNA and RNA were measured using a NanoDrop™ 1000 spectrophotometer (RRID : SCR_016517, Thermo Fisher Scientific, Waltham, MA, USA). Integrity of the RNA was checked by agarose gel electrophoresis.

Five hundred nanograms of total RNA were reverse transcribed using SuperScript IV reverse transcriptase (Invitrogen, Carlsbad, California, USA), and random hexamers in a final volume of 20 uL according to the manufacturer’s protocol.

Gene expression was quantified using Taqman assays (Applied Biosystems, Fosters City, California, USA). The genes analyzed were Somatostatin Receptor Subtype 2 (SSTR2, Hs00990356_m1), Somatostatin Receptor Subtype 3 (SSTR3, Hs00265633_s1), Somatostatin Receptor subtype 5 (SSTR5, Hs00990408_s1), short Dopamine Receptor D2 (DRD2) Isoform (Hs01014210_m1), long DRD2 Isoform (Hs01024460_m1), Arrestin Beta 1 (ARRB1, Hs00930516_m1), Pleiomorphic Adenoma Gene-Like 1 (PLAGL1, Hs00414677_m1), Raf Kinase Inhibitory Protein (RKIP/PEBP1, Hs01110783_g1), E-cadherin (CDH1, Hs01023894_m1), Ki-67 (MKI67, Hs01032443_m1), Ghrelin and Obestatin Prepropeptide (GHRL, Hs01074053_m1), Aryl Hydrocarbon Receptor Interacting Protein (AIP, Hs00610222_m1), Snail Family Transcriptional Repressor 1 (SNAI1, Hs00195591_m1), Snail Family Transcriptional Repressor 2 (SNAI2, Hs00950344_m1), Epithelial Splicing Regulatory Protein 1 (ESRP1, Hs00214472_m1), RAR Related Orphan Receptor C (RORC, Hs01076112_m1), N-Cadherin (CDH2, Hs00983056_m1), Twist Family bHLH Transcription Factor 1 (TWIST1, Hs00361186_m1) and Vimentin (VIM, Hs00958111_m1); furthermore, a custom assay was ordered for Intron 1 Ghrelin In1-GHRL (AJ89KWC). We tested six reference genes to normalize gene expression: Hypoxanthine Phosphoribosyltransferase 1 (HPRT1, Hs99999909_m1), Proteasome 26S Subunit ATPase 4 (PSMC4, Hs00197826_m1), Glucuronidase Beta (GUSB, Hs00939627_m1), TATA-Box Binding Protein (TBP, Hs00427621_m1), Mitochondrial Ribosomal Protein L19 (MRPL19, Hs01040217_m1) and Phosphoglycerate Kinase 1 (PGK1, Hs00943178_g1), and selected the last three reference genes based on their stability in our samples according to Chainy software (available on: http://maplab.imppc.org/chainy/) (32).

Quantitative polymerase chain reactions (qPCR) were carried out in a 7900HT Fast Real-Time PCR System (RRID : SCR_018060; Applied Biosystems, Fosters City, California, USA). We used TaqMan Gene Expression Master Mix (Applied Biosystems, Fosters City, California, USA), and the amplification reactions were performed in triplicate for each sample in a final volume of 10 uL in 384-well plates. To minimize the inter-assay variation, all the genes, including the reference genes, for each sample were analyzed in the same plate. To quantify relative gene expression we calculated a normalization factor for each sample based on the geometric mean of the selected reference genes, according to geNorm (RRID : SCR_006763, https://genorm.cmgg.be/) algorithms (33).

Descriptive results were expressed as mean ± standard deviation. Unsupervised hierarchical clustering was used to investigate the potential identification of patient’s response subgroups, based on their molecular expression profile. Differences between groups were compared using analysis of variance (Student’s t-test, Wilcoxon signed-rank test and Kruskal-Wallis analysis of variance as appropriate). Samples from all groups within an experiment were processed at the same time. The P values were two-sided, and statistical significance was considered when P < 0.05. All statistical analyses were performed using R version 4.2.1 (R Project for Statistical Computing, RRID : SCR_001905). Unsupervised hierarchical clustering was performed using the R package pheatmap (Pretty Heatmaps, https://CRAN.R-project.org/package=pheatmap). The graphical representation was done using package ggplot 2 (RRID : SCR_014601, Whickham https://CRAN.R-project.org/package=ggplot2) and the P values were added using the ggpubr package (‘ggplot2’ Based Publication Ready Plots, https://CRAN.R-project.org/package=ggpubr).

We analyzed 72 patients with NF-PitNETs, 37 men and 35 women. The mean age was 60.3 years ± 13.5 and the mean follow-up was 915.7 ± 696.4 days. During follow-up, 6 patients with tumor remnants presented a recurrence after a mean of 759.8 days and 3 patients without visible remnants after surgery presented a recurrence after a mean of 783 days. Based on MRI, maximal tumor diameter did not show significant differences in tumors with or without extrasellar extension (p=0.22). However, maximal tumor diameter showed differences among tumors with sinus invasion (mean: 28.3 mm) and tumors without (mean: 23.3 mm) (p=0.02). The proportion of patients presenting headache and visual alterations fell after surgery (43.1% to 11.3%, and 40.3% to 22.6%, respectively in intrasellar and extrasellar tumors) while the proportion of patients presenting hypopituitarism increased after surgery (from 47.8% to 67.9%) ( Table 1 ).

We analyzed the gene expression of different EMT-related markers and their relationship with clinical features of NF-PitNETs ( Table 2 ). We found that SNAI1 and vimentin expression was associated with invasive tumors (p=0.049 and p=0.039, respectively) while E-cadherin expression was associated with non-invasive intrasellar tumors (p=0.047). Moreover, vimentin and SNAI1 overexpression showed an association with tumors presenting, respectively, headache and hypopituitarism before surgery (p=0.050 and p=0.048).

We recently reported that most GH-secreting pituitary tumors showed a hybrid and variable EMT expression profile rather than a clear binarial epithelial or mesenchymal phenotype (23). Taking these data into account, we compared the expression of EMT-related markers in GH-secreting tumors, NF-PitNETs and normal pituitaries. Unsupervised hierarchical clustering analysis revealed multiple EMT transition states, with normal pituitaries forming a subcluster associated with high levels of the epithelial marker E-cadherin while NF-PitNETs and GH-secreting tumors were mixed in several different subclusters. Interestingly, a cluster associated with a mesenchymal expression profile was found in 1 GH-secreting adenoma and 6 NF-PitNETs ( Figure 1 ). This result showed that 12.5% of NF-PitNETs harbored an expression profile compatible with an advanced EMT transformation compared to 3.6% of GH-secreting tumors.

We also analyzed the expression of SRLs response biomarkers in NF-PitNETs. In those patients in which clinical information was available, gene expression was correlated with clinical parameters ( Table 3 ). Tumor size showed a negative association with the DRD2 long isoform (Pearson’s r = -0.25, p = 0.05). Extrasellar extension was related to SSTR2, PEBP1, Ki-67 and KLK10 (p = 0.031, p=0.033, p =0.042 and p = 0.008, respectively).

Differences were found also in SRLs biomarkers between NF-PitNETs, GH-secreting pituitary adenomas and non-tumoral pituitaries. GH-secreting tumors showed higher levels of SSTR2 and SSTR5, whereas SSTR3 was more expressed in NF-PitNETs ( Figures 2A–C ). ARRB1 levels were higher in normal tissue ( Figure 2D ). On the other hand, NF-PitNETs seem to be characterized by high levels of KLK10 and PLAGL1 ( Figures 2E, F ). Finally, non-tumoral tissue presented high levels of E-cadherin and DRD2 ( Figures 2G–I ).

Furthermore, we wanted to investigate if some of these genes could be used to predict patient’s outcome. We investigated the ability of these markers in predicting recurrence in NF-PitNETs ( Table 4 ), and found that only PEBP1 showed a difference between tumors that recurred and tumors that did not (p=0.036) ( Figure 3A ). When performing a binomial logistic regression, PEBP1 showed an AUC of 69.9%, for a cut-off of 14.0, with a sensitivity of 100% and a specificity of 42.6% ( Figure 3B ).

We wanted to compare relative quantities of drug receptor expression in both NF-PitNETs and GH-secreting tumors. In NF-PitNETs the more prevalent receptor was DRD2, followed by SSTR3, while SSTR2 and SSTR5 presented a low expression ( Figure 4A ). In the case of GH-secreting tumors DRD2 was also the most quantitatively expressed, SSTR2 and SSTR5 were expressed at the same level as the SSTR3 in NF-PitNETs, while SSTR3 showed a very low expression level ( Figure 4B ).

We correlated levels of EMT and SRL response gene expression with immunohistochemical characteristics of NF-PitNETs ( Table 5 ). We found differences in SSTR3 comparing ACTH-expressing NF-PitNETs versus FSH/LH-positive, plurihormonal NF-PitNETs and negative tumors (p = 0.001, p = 0.005 and p = 0.004, respectively) ( Figure 5A ). We also found differences in ARRB1 comparing ACTH-expressing NF-PitNETs versus FSH/LH-positive and negative cell tumors (p=0.005 and p=0.012, respectively) ( Figure 5B ).