Pituitary neuroendocrine tumors with PIT1/SF1 co-expression show distinct clinicopathological and molecular features

Pituitary neuroendocrine tumors (PitNETs) are classified according to cell lineage, which requires immunohistochemistry for adenohypophyseal hormones and the transcription factors (TFs) PIT1, SF1, and TPIT. According to the current WHO 2022 classification, PitNETs with co-expression of multiple TFs are termed “plurihormonal”. Previously, PIT1/SF1 co-expression was prevailingly reported in PitNETs, which otherwise correspond to the somatotroph lineage. However, little is known about such tumors and the WHO classification has not recognized their significance. We compiled an in-house case series of 100 tumors, previously diagnosed as somatotroph PitNETs. Following TF staining, histopathological features associated with PIT1/SF1 co-expression were assessed. Integration of in-house and publicly available sample data allowed for a meta-analysis of SF1-associated clinicopathological and molecular features across a total of 270 somatotroph PitNETs. The majority (74%, 52/70) of our densely granulated somatotroph PitNETs (DGST) unequivocally co-expressed PIT1 and SF1 (DGST-PIT1/SF1). None (0%, 0/30) of our sparsely granulated somatotroph PitNETs (SGST) stained positive for SF1 (SGST-PIT1). Among DGST, PIT1/SF1 co-expression was significantly associated with scarce FSH/LH expression and fewer fibrous bodies compared to DGST-PIT1. Integrated molecular analyses including publicly available samples confirmed that DGST-PIT1/SF1, DGST-PIT1 and SGST-PIT1 represent distinct tumor subtypes. Clinicopathological meta-analyses indicated that DGST-PIT1 respond more favorably towards treatment with somatostatin analogs compared to DGST-PIT1/SF1, while both these subtypes show an overall less aggressive clinical course than SGST-PIT1. In this study, we spotlight that DGST with co-expression of PIT1 and SF1 represent a common, yet underrecognized, distinct PitNET subtype. Our study questions the rationale of generally classifying such tumors as “plurihormonal”, and calls for a refinement of the WHO classification. We propose the term “somatogonadotroph PitNET”. Supplementary Information The online version contains supplementary material available at 10.1007/s00401-024-02686-1.


Introduction
Pituitary neuroendocrine tumors (PitNETs), formerly termed pituitary adenomas, are common intracranial neoplasms, which originate from adenohypophyseal cells of the anterior pituitary lobe.PitNET classification is based on cell lineage, determined by immunohistochemistry for adenohypophyseal hormones and the pituitary transcription factors (TFs) PIT1, SF1, and TPIT [2,28].According to the current WHO classification of 2022, immunopositivity for one of these TFs denotes the PIT1-, gonadotroph, or corticotroph lineage, respectively.PitNETs of the PIT1-lineage can further exhibit somatotroph, mammosomatotroph, lactotroph and/ or thyrotroph differentiation.Some PitNETs may express multiple adenohypophyseal hormones and/or TFs, belonging to more than one cell lineage.Currently, the WHO classification subsumes all such tumors under the term "plurihormonal PitNET/adenoma" [31].
First reports on PitNETs demonstrating simultaneous expression of multiple hormones date back to over 40 years ago [9,22].In particular, expression of gonadotropins was recurrently described in growth hormone (GH)-secreting PitNET/adenomas causing acromegaly [9,12,14,22], suggesting affiliations with both the somatotroph and gonadotroph lineage in a subset of such tumors.
Through the routine implementation of TFs in pituitary diagnostics, multilineage PitNETs regained increased attention over the last few years.Several studies have reported concurrent PIT1 and SF1 immunopositivity in unusual PitNETs, which otherwise reflected somatotroph tumors [3,16,24,29].Previous transcriptome-based studies confirmed occasional SF1 expression in somatotroph PitNETs, predominantly in the densely granulated subtype [17,20].Moreover, SF1 expression was recently linked to distinct transcriptomic signatures and methylation profiles among somatotroph PitNETs [11,20].In summary, the current literature suggests that PIT1/SF1 co-expression associates with the somatotroph lineage.This finding is, however, yet underappreciated by the WHO classification and challenges the precept of declaring any and all PitNETs with co-expression of multiple TFs as "plurihormonal".Currently, little is known about the clinical, histopathological, and molecular features linked to PIT1/SF1 co-expression in somatotroph PitNETs and such tumors are considered rare.
In this study, we aimed to demonstrate the prevalence of PIT1/SF1 co-expression in PitNETs, which otherwise correspond to somatotroph PitNETs, as defined by the WHO 2022 classification.We further explored the clinical, histopathological, epigenomic and genomic features associated with PIT1/SF1 co-expression in somatotroph tumors in a large meta-analysis by integrating molecular in-house data and publicly available data deposits.

Case series and tissue assembly
The archive of the University Medical Center Hamburg-Eppendorf was searched for PitNET/adenomas diagnosed between 2017 and 2023 as either sparsely or densely granulated somatotroph PitNET/adenomas or plurihormonal Pit-NET/adenomas.PitNETs fulfilling the essential diagnostic criteria for sparsely or densely granulated somatotroph PitNETs, as defined by the WHO 2022 Classification of Endocrine and Neuroendocrine Tumours (5th ed.) [31] were included in this study.In detail, inclusion criteria for DGST were adopted as follows: (i) diffuse GH expression (arbitrarily required in at least 30% of tumor cells), (ii) absence of other pituitary cell differentiation was disregarded owing to the purpose of this study, (iii) perinuclear LMWCK staining (perinuclear cytoplasmic CAM5.2 immunopositivity arbitrarily required in at least 30% of tumor cells), and (iv) acidophilic cytoplasm on H&E-stained sections (evaluated as either moderate or strong cytoplasmic eosinophilia).Six PitNETs were entirely immunonegative for CAM5.2 and thus did not fulfill the third WHO criterion.They were nevertheless included in the case series, due to a lack of reasonable differential diagnoses other than DGST and with the aim to further explore these unusual tumors.
Inclusion criteria for SGST were adopted as follows: (i) variable GH expression (any immunostaining pattern was accepted), (ii) absence of other pituitary cell differentiation was disregarded owing to the purpose of this study, (iii) abundant fibrous bodies (in more than 70% of cells), and (iv) absence of acidophilic cytoplasm on H&E-stained sections (evaluated as either none or weak cytoplasmic eosinophilia).
Cases of insufficient formalin-fixed paraffin-embedded (FFPE) material quantity or quality were excluded.To avoid inclusion of mammosomatotroph, mixed GH-PRL and plurihormonal PIT1-lineage tumors, cases with marked expression of estrogen receptor, prolactin or TSH were excluded from this study.Scattered intratumoral PRL expression up to 5% was tolerated for study inclusion, owing to the high prevalence of this finding, the possibility that PRL expression may stem from intratumorally entrapped residual adenohypophyseal cells and the aim to further explore this feature.
The use of all tissue specimens for research upon anonymization was in accordance with local and national ethical standards and with the 1964 Helsinki declaration and its later amendments.

DNA methylation analysis
DNA was isolated from FFPE tissue using the Reli-aPrep™ FFPE gDNA Miniprep System (Promega).Around 100-500 ng of DNA was bisulfite-converted using the EZ DNA Methylation Kit (Zymo Research).The DNA Clean & Concentrator-5 kit (Zymo Research) and the Infinium HD FFPE DNA Restore Kit (Illumina) were used to clean and restore the converted DNA.Finally, the Illumina Infinium Methylation EPIC BeadChip Kit was used to quantify the methylation status of 850,000 CpG sites on an iScan device (Illumina).
Raw methylation array data (idat files) from this study and publicly available deposits [Capper et [11], and Silva-Júnior et al. (GSE207937) [23]] were processed using the minfi package [1] in R [19].Probes on sex chromosomes, probes with a detection p value of or above 0.01, probes with SNPs at the CpG site, and cross-reactive probes were excluded.When combining different types of arrays, probes, which were not represented in both the EPIC and the 450 k array were excluded.
Consensus cluster analyses were performed using the ConsensusClusterPlus package [32].Uniform Manifold Approximation and Projection (UMAP) transformation and plotting were performed using the umap package [15].
Cumulative copy number profiles (CNPs) were calculated using the GenVisR package [26].To reduce CNP noise, three consecutive segments were required to surpass the cutoffs set to − 0.35 and 0.35.
Methylation-based classification was performed using the R package of the MNP brain tumor methylation classifier v12.5 [6].

Data integration and annotation of external samples
For evaluation of clinicopathological parameters in DGST-PIT1/SF1, DGST-PIT1, and SGST-PIT1 across studies, we compiled an extended case series consisting of our in-house samples, and publicly available data of somatotroph PitNETs derived from Capper et al. [6] (GSE109381), Neou et al. [17] (E-MTAB-7762, E-MTAB-7768), Silva-Júnior et al. [23] (GSE207937, GSE209903), Kober et al. [11] (GSE226764) and Rymuza et al. [20] (E-MTAB-11889).Extended sample data of the latter study were kindly provided by Mateusz Bujko (Dept. of Molecular and Translational Oncology, Maria Sklodowska-Curie National Research Institute of Oncology, Warsaw, Poland).For preparation of the integrated dataset, in-house samples were classified as either DGST-PIT1/SF1, DGST-PIT1, or SGST-PIT1, based on histomorphology and SF1-IHC.External samples were reclassified according to NR5A1 (SF1) RNA expression levels and/or methylation classifier results.Normalized NR5A1 counts above or below 160 were considered high, or low, respectively.External samples of Rymuza et al. lacked both DNA methylation and RNA expression data and were reclassified according to the previously described qPCRbased transcriptomic subgroups [20].To maintain reclassification accuracy, samples of Rymuza et al. were excluded if histomorphology was not in line with the qPCR-based transcriptomic subgroup.
Due to the purpose of this study, all external sample with a classifier match or histomorphology other than somatotroph PitNET (e.g., mixed GH/PRL) were excluded.Furthermore, samples demonstrating an inconclusive mismatch between SF1 status and classifier result, as well as insufficient data for accurate reclassification were excluded.The final extended case series comprised a total of 270 tumors (99 in-house, 38

Statistical analyses
Statistical analyses of clinicopathological parameters were performed using the stats R package (v4.1.3)[19].Continuous scale data were analyzed using the student's t test.Ranked/ordinal data were analyzed using the Wilcoxon rank sum test.Binary categorical data were analyzed using the chi-square test.For detection of significant recurrent chromosomal alterations, the Genomic Identification of Significant Targets in Cancer (GISTIC) procedure [4] was applied using the CNVRanger R package [25].Population ranges were computed based on reciprocal overlap between genomic regions with a cutoff of 0.51.In order to increase specificity for significant recurrent chromosomal alterations, at least 5 Mbp of genomic regions were required to show p values < 0.05 for each chromosome.

Results
We searched the archive of the University Medical Center Hamburg-Eppendorf for PitNETs, which fulfilled the essential diagnostic criteria for densely (DGST) or sparsely granulated somatotroph PitNET (SGST) as defined by the WHO 2022.Owing to the purpose of this study, hormone, and TF expression outside of the PIT1-lineage were overlooked as exclusion criteria.

Discussion
In this study, we aimed to explore the histopathological, molecular, and clinical features associated with the emerging role of PIT1/SF1 co-expression in PitNETs.Our study design was centered around the current WHO classification and based on findings of previous publications, which triggered the search for PIT1/SF1 co-expression among PitNETs, which otherwise correspond to the somatotroph PIT1-lineage.

PIT1/SF1 co-expression is highly prevalent among DGST
We were surprised to find that the vast majority (nearly 3/4) of previously diagnosed DGST co-expressed PIT1 and SF1.This unexpectedly high rate can be explained by two circumstances.Firstly, the WHO has only recently dictated necessity for staining all three TFs in routine pituitary diagnostics.Thus, data on unusual TF expression constellations are currently scarce and the prevalence of multilineage PitNETs may be much higher than generally assumed.Secondly, the strict implementation of WHO-based inclusion criteria may have inflated the prevalence of PIT1/SF1 co-expression within the case series.Thus, the frequency of PIT1/SF1 coexpression among PitNETs, which do not fit into the WHO class of somatotroph tumors has yet to be determined.This pertains to pure somatotroph tumors, which do not fulfill the essential criteria for either SGST or DGST, to somatotroph tumors exhibiting elements of mammosomatotroph, mixed somatotroph-lactotroph, or PIT1-plurihormonal differentiation and to further non-somatotroph PitNETs of the PIT1-lineage.

DGST-PIT1/SF1 exhibit distinct molecular, histopathological, and clinical features
Global DNA methylation patterns are considered to reflect the cell of origin, making epigenomic analyses useful for classifying tumors based on their lineage [6,7,21].We compared methylation profiles of DGSTs with and without PIT1/ SF1 co-expression and showed that they are epigenomically distinct.This result is in line with a recent publication by Kober et al. [11] and compatible with the notion that transcription factors play an early role in pituitary lineage development.The existence of two epigenomically distinct groups of tumors among DGST has also been described before by Capper et al. [6], and was implemented into the Brain tumor methylation classifier.The two groups had been termed mc "DNS-A" and mc "DNS-B" (v11b4, v12.5, v12.8), while their significance remained unknown.In this report, we clarify that the mc "DNS-A" affiliates with PIT1/SF1 coexpression, whereas the mc "DNS-B" mainly comprises pure PIT1-lineage tumors.
Moreover, we show that histopathological features of most DGST-PIT1/SF1 reflect those of prototypical DGST.In contrast, DGST-PIT1 display fibrous bodies in significantly higher amounts than DGST-PIT1/SF1.We conclude that DGST-PIT1 predominantly pertain to the previously proposed class of "intermediate type" granulated somatotroph PitNET, which were considered within the histomorphological spectrum of DGST [18].Although "intermediate type" granulation was not exclusively encountered among DGST-PIT1, our results show that this granulation type associates with molecular distinctness from somatotroph PitNETs with densely granulated morphology.
Tumor sizes in DGST-PIT1 were significantly smaller than in DGST-PIT1/SF1 and SGST-PIT1 after SSA treatment, indicating that tumor growth of DGST-PIT1 is more efficiently impeded by medical management compared to the other subtypes.In line with this, GNAS mutations and dense granulation patterns were previously linked to favorable SSA responses in somatotroph PitNETs [5,8,13,34].We find that both these features affiliate with DGST-PIT1.

Somatotroph PitNETs frequently demonstrate intratumoral PRL expression and may be CAM5.2 immunonegative
Handling apparent intratumoral PRL expression in somatotroph PitNETs poses a diagnostic difficulty.Firstly, it can be challenging to histomorphologically differentiate if scarce immunosignal stems from entrapped non-neoplastic pituitary cells or scattered tumor cells.Secondly, in contrast  [10,27].The epigenomic data presented in this study suggest that lack of CAM5.2 immunoreactivity among DGST does not accompany epigenomic distinctness.In addition, the highly variant extent of cytoplasmic CAM5.2 immunoreactivity (ranging from 30 to 100% of tumor cells) in the CAM5.2-positiveDGST of our case series also did not associate with epigenomic distinctness.Taken together, our data suggest that prominent perinuclear cytoplasmic CAM5.2 staining is not a crucial histopathological feature of DGST.Further investigations are needed to clarify if CAM5.2 immunoreactivity may associate with separate clinical features in DGST and whether CAM5.2-negativePitNETs with epigenomic profiles of SGST also exist.

Are DGST-PIT1/SF1 truly plurihormonal tumors?
The question arises, how DGST-PIT1/SF1 should be meaningfully termed and classified.As previously mentioned, the WHO 2022 dictates to bundle all PitNETs with immunopositivity for more than one TF as "plurihormonal PitNET/ adenoma".Consequently, the WHO distinguishes DGST-PIT1/SF1 from pure PIT1-lineage somatotroph PitNETs and formally classifies these tumors together with various PitNETs exhibiting unusual combinations of TF and hormone expression patterns.Our data, however, suggests that DGST-PIT1/SF1 represents a distinct somatotroph PitNET subtype.Moreover, since DGST-PIT1/SF1 rarely expressed FSH or LH in our study, questioning the true plurihormonal identity of these tumors stands to reason.We propose the term "somatogonadotroph PitNET" for DGST-PIT1/SF1.
In conclusion, a substantial proportion of previously diagnosed somatotroph PitNETs co-express PIT1 and SF1 and exhibit clinical, histopathological, and molecular distinctness from other pure PIT1-lineage somatotroph PitNETs.We present a comprehensive meta-analysis of the three emerging molecular subtypes of somatotroph PitNETs, which call for a refinement of the current WHO 2022 classification.

Fig. 3
Fig. 3 Comparison of molecular features in DGST-PIT1/SF1, DGST-PIT1 and SGST-PIT1.a Consensus clustering of global DNA methylation data demonstrated three epigenetically distinct subgroups of somatotroph tumors.The first subtype (left branch) mostly harbored tumors with densely granulated histology and evident SF1-expression via RNA or immunostaining, showed an affiliation with the transcriptomic subgroup 1 (defined by Rymuza et al. [20]), and matched with the methylation class (mc) "Pituitary adenoma, subtype STH-producing, subclass densely granulated A" (PA STH DENSE A) (Brain classifier version 12.5).The second subtype (middle branch) mostly harbored tumors with densely granulated histology and insignificant SF1-expression via RNA or immunostaining, showed an affiliation with the transcriptomic subgroup 2, and matched with the mc PA STH DENSE B. The third subtype (right branch) mostly harbored tumors with sparsely granulated histology and insignificant SF1expression via RNA or immunostaining, showed an affiliation with the transcriptomic subgroup 3, and matched with the mc PA STH SPARSE.Taken together, the three epigenetic subgroups correspond to DGST-PIT1/SF1 (left), DGST-PIT1 (middle), and SGST-PIT1
2017, the WHO 2022 states that somatotroph tumors are negative for PRL, raising the question how to classify bona fide somatotroph tumors with little, yet obvious intratumoral PRL expression.In study, intratumoral PRL expression up to 5% was tolerated for somatotroph Pit-NET diagnosis.We found that intratumoral PRL expression did not relate to epigenomic distinctness in somatotroph Pit-NETs.This result clearly showcases that detection of limited PRL expression does not justify exclusion of somatotroph PitNET diagnosis.This should be considered by the WHO classification, which currently states that absence of hormone expression other than GH is an essential diagnostic criterion for somatotroph PitNETs.To further clarify this topic, extended and comprehensive epigenomic investigations on the various histopathological subtypes of PIT1lineage PitNETs are needed.Moreover, we epigenomically analyzed four CAM5.2 immunonegative somatotroph PitNETs in our case series.Because CAM5.2-negativity poses a diagnostic dilemma, previous studies refrained from classifying such tumors as sparsely or densely granulated