bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; this version posted April 27, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
1 DNA Methylation-based Signatures Classify Sporadic Pituitary Tumors According to
2 Clinicopathological Features.
3 Maritza S. Mosella 1,2,3, Thais S. Sabedot 1,2, Tiago C. Silva 3,#, Tathiane M. Malta 1,2,&,
4 Felipe D. Segato 3,*, Karam P. Asmaro 1,2, Michael Wells 1,2, Abir Mukherjee 4, Laila M.
5 Poisson 5, James Snyder 2, Ana C. deCarvalho 1, Tobias Walbert T 2, Todd Aho 6, Steven
6 Kalkanis 1, Paula C. Elias 7; Sonir R. Antonini 8, Jack Rock 1, Houtan Noushmehr 1,2,3,↟ ,
7 Margaret Castro 7↟ ; Ana Valeria Castro1,2,↟ ,^
8 1Department of Neurosurgery, Hermelin Brain Tumor Center, Henry Ford Health System,
9 Detroit, MI, USA. 2Department of Neurosurgery, Omics Laboratory, Henry Ford Health
10 System, Detroit, MI, USA. 3Department of Genetics, Ribeirao Preto Medical School,
11 University of São Paulo, Ribeirao Preto, Brazil. 4Department of Pathology, Henry Ford
12 Health System, Detroit, MI, USA. 5Department of Biostatistics, Henry Ford Health System,
13 Detroit, MI, USA; 6Department of Radiology, Henry Ford Health System, Detroit, MI, USA.
14 7Internal Medicine Department, Ribeirao Preto Medical School, University of São Paulo,
15 Ribeirao Preto, Brazil. 8Department of Pediatrics, Ribeirao Preto Medical School, University
16 of São Paulo, Ribeirao Preto, Brazil.
17 ^Corresponding author:
18 Ana Valeria Castro
19 Scientist/Professor
20 Department of Neurosurgery
21 Hermelin Brain Tumor Center
22 Henry Ford Health System
23 2799 West Grand Blvd, E&R 3096
24 Detroit, MI, 48202
1 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
26 Running title
27 DNA-methylation markers classify pituitary tumors
28 Funding
29 This work was supported by the Henry Ford Health System, Department of Neurosurgery,
30 and the Hermelin Brain Tumor Center. Additionally, MSM and MC are supported by The
31 São Paulo Research Foundation (FAPES), Brazil ( #16/11039-3; #17/10357-4,#14/03989-6);
32 AC and KPA by Henry Ford Hospital (A30935, A30957; GME 202199); LMP, HN, AD,
33 MW, and AM are supported by the National Institutes of Health (R01CA222146), HN, TSS,
34 TMM, LMP, and AD are supported by the Department of Defense (CA170278).
35 Conflict of interest
36 The authors declare to have no competing interests.
37 Authorship
38 Overall concept and coordination of the study: AVC, MC, HN, SA, PE; retrieval of publicly
39 available molecular and clinical data: MSM; Bioinformatic and statistical analyses: MSM,
40 TSS, TCS, TMM, FSD, HN and input from LMP; HFHF cohort: pathology review MF, AM;
41 clinicopathological and radiological data were generation and collection: AVC, MSM, KPA,
42 AM, MF, TA; molecular data generation: AC, TMM; manuscript was written by MSM,
43 AVC, HN, intellectual contribution: JS, SK, TW. All authors contributed to the revision of
44 the manuscript. # Present address: University of Miami, FL, USA & Present address:
45 University of São Paulo, Ribeirao Preto, Brazil * Present address: Cedars-Sinai Medical
46 Center, Los Angeles, CA, USA. ↟ Contributed equally
47 Word count
48 Abstract: 230
49 Article text: 3,182
50 Reference list: 1,187
2 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
51 Figures and Tables captions: 782
52 Manuscript text: 5,095 (Title: 414; abstract: 230; article text: 3,182; Acknowledgements: 82;
53 reference list: 1,187)
54
3 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
55 ABSTRACT
56 Background: Distinct genome-wide methylation patterns have consistently clustered pituitary
57 neuroendocrine tumors (PT) into molecular groups associated with specific clinicopathological
58 features. Here we aim to identify, characterize and validate the methylation signatures that
59 objectively classify PT into those molecular groups.
60 Methods: Combining in-house and publicly available data, we conducted an analysis of the
61 methylome profile of a comprehensive cohort of 177 tumor and 20 non-tumor specimens from
62 the pituitary gland. We also retrieved methylome data from an independent pituitary tumor
63 (PT) cohort (N=86) to validate our findings.
64 Results: We identified three methylation clusters associated with functional status and
65 adenohypophyseal cell lineages using an unsupervised approach. We also identified signatures
66 based on differentially methylated CpG probes (DMP), some of which overlapped with
67 pituitary-specific transcription factors genes (SF1 and Tpit), that significantly distinguished
68 pairs of clusters related to functional status and adenohypophyseal cell lineage. These findings
69 were reproduced in an independent cohort, validating these methylation signatures. The DMPs
70 were mainly annotated in enhancer regions associated with pathways and genes involved in
71 cell identity and tumorigenesis.
72 Conclusions: We identified and validated methylation signatures that distinguished PT by
73 distinct functional status and adenohypophyseal cell lineages. These signatures, annotated in
74 enhancer regions, indicate the importance of these elements in pituitary tumorigenesis. They
75 also provide an unbiased approach to classify pituitary tumors according to the most recent
76 classification recommended by the WHO 2017 using methylation profiling.
77 Key-words: DNA methylation; classification; regulatory elements; enhancers; pituitary
78 tumors.
4 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
79 Key-points
80 ● Distinct methylation landscapes define PT groups with specific functional
81 status/subtypes and adenohypophyseal lineages subtypes.
82 ● Methylation abnormalities in each cluster mainly occur in CpG annotated in distal
83 regions overlapping predicted enhancers regions associated with pathways and genes
84 involved in cell identity and tumorigenesis.
85 ● DNA methylation signatures provide an unbiased approach to classify PT.
86 Importance of the study
87 This study harnessed the largest methylome data to date from a comprehensive cohort
88 of pituitary specimens obtained from four different institutions. We identified and validated
89 methylation signatures that distinguished pituitary tumors into molecular groups that reflect the
90 functionality and adenohypophyseal cell lineages of these tumors. These signatures, mainly
91 located in enhancers, are associated with pathways and genes involved in cell identity and
92 tumorigenesis. Our results show that methylome profiling provides an objective approach to
93 classify PT according to the most recent classification of PT recommended by the 2017 WHO.
5 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
94 1 Introduction
95 Pituitary tumors (PT), a type of neuroendocrine tumors, comprise the second most
96 common neoplasm of the central nervous system (CNS) (~17%) 1–3 with an annual average
97 age-adjusted incidence of 4.08 per 100,000 population 3. Stratified by endocrine status, PT are
98 classified as functioning (FPT) (46-64%) or nonfunctioning (NFPT) (36%–54%) subtypes, the
99 latter also referred to as silent adenomas 1,4. In relation to the histopathological classification
100 the most recent recommendation by the World Health Organization (WHO) in 2017 is based
101 on immunostaining for adenohypophyseal hormones and cell-lineage 5. Despite its importance
102 and extensive use for tumor classification, the immunopathology method is limited by its
103 subjectivity and inaccuracy in pituitary and several other tumors 6,7. To circumvent these
104 limitations and/or to complement pathology results, inclusion of genomic-based signatures
105 such as recurrent somatic mutations and chromosome number alterations has advanced and
106 refined the classification of many tumors, including CNS neoplasms 8,9. However, in contrast
107 to other CNS tumors, genomic alterations are rare in PT or are restricted to some subtypes (i.e.
108 GNAS mutation in somatotroph adenomas) 10–12. Contrariwise, distinct genome-wide DNA
109 methylation patterns are consistently associated with their clinicopathological features such as
110 function and histological variants. In contrast to other CNS tumors, the importance of
111 methylation markers in the taxonomy of PT has not been explored to date 7,10,13–17. A recent
112 study reported that the combination of genome-wide molecular profiling of multiple platforms,
113 including DNA methylation, provided an objective approach to the classification of PT 10.
114 However, the identification of specific methylation signatures to classify PT into these
115 methylome-based groups has not been exploited.
116 To address this knowledge gap, we aimed to identify, characterize and validate
117 methylation-based markers that define PT according to clinicopathological features. For this
118 purpose, we compiled and performed a comprehensive analysis of the genome-wide DNA
6 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
119 methylation data on 177 adult sporadic tumors and 20 non-tumorous tissue from the pituitary
120 gland obtained from a multicenter source and validated our findings in an independent cohort.
121 We identified methylation signatures that classified PT according to methylation groups that
122 reflected the functionality and the adenohypophyseal lineages of these tumors.
123 2 Subjects and methods
124 2.1 Patients and tissue specimens
125 DNA methylation data were retrieved from 3 publicly available cohort datasets (data
126 retrieval freeze on August 2018) 7,13,14 and from our cohort at the Hermelin Brain Tumor Center
127 (HBTC) of the Henry Ford Health System (Detroit, MI, USA) (n=23; HBTC cohort,
128 unpublished). The final cohort consists of 20 non-neoplastic pituitary glands (non-tumor/NT)
129 and 177 PT from patients over 18 years (Pan-pit), composed of 87 FPT and 90 NFPT. Clinical
130 information about this study cohort is summarized in Table 1 and detailed in Supplementary
131 Table S1. The PT clinicopathological classification criteria of features such as invasion from
132 our internal cohort were matched with the criteria provided by the authors of the external
133 cohorts 7,13,14. Specific transcription factor-based adenohypophyseal lineages were assigned to
134 each PT subtype in accordance with the 2017 WHO classification 5. All the studies were
135 approved by their respective Institutional Review Boards and ethics committees and written
136 consent obtained from each patient at the attending institutions 7,10,13,14.
137 2.2 Data retrieval and preprocessing
138 DNA from pituitary tissue was obtained from fresh-frozen (FF) 13,14 or formalin-fixed-
139 paraffin-embedded specimens (FFPE) 7. The method of DNA extraction, processing, and
140 assays for each cohort included in this study is reported in the respective manuscripts 7,13,14 and
141 for the HBTC cohort is available in the Supplementary Methods. The IDAT files for the DNA
142 methylation bead arrays (450K or 850K/EPIC) from the Ling, Kober, and Capper cohorts 7,13,14,
143 were retrieved through the GEOquery package 18 and preprocessed simultaneously using minfi
7 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
144 package19. The EPIC array data from Capper and HBTC were processed separately using the
145 same approach. Detailed information about preprocessing of IDAT files is depicted in
146 Supplementary methods. Principal component analysis (PCA) revealed that there were no
147 batch effects related to the cohort, tissue sources (FFPE or FF) and array platform (450K/EPIC)
148 (Supplementary Figure S1A-C). The HBTC DNA methylation data was submitted to Gene
149 Expression Omnibus (GSEXXXX).
150 To test the reproducibility of our findings, we retrieved the methylome data from 86
151 patients with PT obtained from an independent study published after our data freeze (EMBL-
152 EBI accession number E-MTAB-7762) 10.
153 2.3 Unsupervised and supervised analysis of the PT methylome
154 To reduce noise, the most variable probes were sorted according to standard deviation
155 across the 177 PT specimens. Next, a consensus clustering analysis20 was applied to the top
156 1% most variant probes (n=4,526). Agglomerative hierarchical clustering with 1000
157 resampling steps, upon euclidean distance was used, as described previously 8,20. Optimal
158 clusters were selected based on statistical parameters: Calinski-Harabasz curve and consensus
159 cluster delta area and overlapped with clinicopathological annotations.
160 Two-tailed Wilcoxon rank sum tests were conducted for each CpG in order to identify
161 pairwise differentially methylated probes (DMP) between the methylation clusters and
162 invasion status. DMP were defined based on differences in mean methylation (diffmean) and
163 adjusted p-value significance. For those comparisons with the highest number of significant
164 probes (e.g. for NFPT-e vs. FPT-e), we applied a more stringent adjusted p-value cutoff, in
165 order to retrieve DMP sets with similar number of probes. An absolute diffmean of 0.3 was
166 used as a threshold for all methylation clusters comparisons. Between ACTH-e vs. NFPT-e
167 clusters comparison adjusted p-values <1e-17 were obtained for both hypo- and
168 hypermethylated probes; for ACTH-e vs. FPT-e and NFPT-e vs. FPT-e comparisons, the
8 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
169 following adjusted p-values for the respective hypo and hypermethylated probes were obtained:
170 p< 1e-16 and p<1e-19 and p<1e-21 p<1e-38, respectively. No significant adjusted p-values
171 were obtained from the invasion status comparative analysis, therefore, the results are related
172 to non-adjusted p-value <0.001 along with delta diffmean of 0.15, similar to the approach
173 reported by other authors 13,14.
174 CpG probes were mapped to their CpG genomic location as CpG islands (CGI), shores,
175 shelves, and open sea regions as previously defined 21–24. DMPs were integrated with predicted
176 enhancers listed in the GeneHancer database 25 through the Genomic Ranges package
177 (https://bioconductor.org/packages/release/bioc/html/GenomicRanges.html). Enrichment or
178 depletion frequencies of DMP sets by genomic location and enhancer overlap were calculated
179 as Odds Ratio using the 485,000 probes from HM450K platform as the reference.
180 Determination of the number of overlapping and distinct DMPs (hyper or
181 hypomethylated) between pairwise comparisons were calculated using the UpsetR package 26.
182 2.4 DNA motif analysis
183 To predict transcription factor binding sites, we surveyed the DNA motifs within
184 enhancer regions which overlapped our DMPs (namely enhancer-related DMP or eDMP).
185 Sequence motifs were identified within ±200 bp from the core of the hyper- or hypomethylated
186 enhancer regions using the bioinformatic tools HOMER 27 . To increase the sensitivity of motif
187 detection, up to two mismatches were allowed in each oligonucleotide sequence. The
188 distributions of motif content in ‘target’ (i.e., the enhancer regions) and random ‘background’
189 sequences were assessed for fold change values.
190 2.5 Integrative analysis between methylation and enhancer prediction databank
191 We identified CpG probes that overlapped with their predicted regulatory regions listed
192 in the GeneHancer database 25. Based on the GeneHancer’s annotation confidence score for
193 functional enhancers 25, the highest scored enhancers for each DMP set were selected. The
9 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
194 selected enhancers were connected to their putative target genes as defined by four
195 experimental metrics 25. The genes with the highest enhancer-gene pairing association score
196 were selected (GeneHancer interaction score)25 and their expressions per methylation cluster
197 are depicted as boxplots .
198 2.6 Methylation status of pituitary-specific transcription factors-related probes
199 We assessed the DNA methylation profile of CpG probes related to pituitary-specific
200 transcription factor (TF) coding genes, known to determine the development of various
201 adenohypophyseal cell types in the pituitary gland: Pit-1 (POU1F1), Tpit (TBX19), ER-alpha
202 (ESR1), GATA2 and SF1 (NR5A1) 1,22. Next, we overlapped the pituitary-specific TF-related
203 probes with the three DMP sets obtained from the pairwise comparison between the clusters.
204 For each significant TF-related probe, we established methylation degree cutoffs that
205 distinguished between clusters. The accuracy of these cutoffs in defining PT subtypes was
206 investigated in the validation cohort 10.
207 3 Results
208 3.1 Genome-wide DNA methylation patterns define three groups of pituitary tumors
209 related to specific functional subtypes and adenohypophyseal cell-lineages.
210 Consensus clustering of the most variable DNA methylation probes in the pan-pit
211 dataset, revealed three main clusters. The clusters were named according to their enrichment
212 for specific functional subtypes (Figure 1A-B and Supplementary Figure S1C). The
213 adrenocorticotropic hormone-enriched cluster (ACTH-e; n=29) is composed mainly of
214 corticotroph adenomas (n=21) and 8 NFPT, including 1 silent corticotroph adenoma, 1
215 gonadotroph, 1 null cell adenoma and 5 of unknown histological NFPT subtypes. The NFPT-
216 e cluster (n=82) consists mainly of NFPT (n=69), including 60 gonadotrophs, 7 null cell
217 tumors, and 2 mixed non-functioning adenomas (LH/FSH/TSH), 1 FPT (lactotroph adenoma)
218 and 12 tumors with unknown histological subtypes. The FPT-e cluster (n=66) is composed
10 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
219 mainly of FPT (n=65), i.e., 41 somatotrophs, 11 thyrotrophs, 8 lactotrophs, 5
220 plurihormonal/mixed adenomas (2 GH/PRL, 1 GH/TSH, 2 GH/PRL/TSH) and 1 NFPT
221 (gonadotroph). The top most variant probes (according to standard deviation) across tumor
222 specimens (n=4,526) are hypomethylated in the FPT-e cluster, hypermethylated in the NFPT-
223 e cluster, and present an intermediate methylation pattern in the ACTH-e cluster (Figure 1B
224 and 1C). ACTH-e, NFPT-e and FPT-e clusters are composed mainly by Tpit- (22/24; 92%);
225 SF1- (60/70; 86%) and Pit-1 (65/66; 98%) cell lineages-derived subtypes, respectively.
226 3.2 Differential methylation between the three methylation-based PT clusters is observed
227 in distal regions overlapping enhancer elements.
228 The supervised comparisons between the clusters retrieved three DMP sets totaling
229 3,316 differentially methylated probes (Figure 2A discovery heatmap) that distinguished
230 among the groups which were validated in an external cohort 10 (Figure 2A, validation
231 heatmap). We observed that 1,973 (59%) out of 3,316 DMPs were located in enhancer regions
232 (Figure 2A, left row label track). DMPs were enriched in open sea regions (n=2,257, 68%),
233 while depleted in CGI, for the three pairwise methylation cluster comparisons (Figure 2B).
234 Detailed information about the distribution of the number of DMPs according to methylation
235 direction and CpG annotation is depicted in Table S2 (Supplemental material). The survey of
236 common or exclusive DMPs among the methylation cluster comparisons are depicted in Figure
237 2C and D, based on the methylation direction (hypo or hypermethylated).
238 When we surveyed for DNA motifs in differentially methylated enhancer candidates
239 that overlapped our DMPs, we identified 31 TF predicted to bind to DNA motifs in the 295
240 hypomethylated eDMP from the comparison between ACTH-e vs. NFPT-e (adjusted p-
241 value<0.05). The top binding site was predicted to bind to Fra1/Fra2 subunits of the Activator
242 protein (AP-1) (Figure 3A). Hypermethylated eDMP (n=88) predicted to bind to NR5A2
243 presented a trend to be significant (adjusted p-value= 0.06). Hypermethylated DMP from
11 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
244 ACTH-e vs. FPT-e comprised cluster-related hypermethylated eDMP (n=360) presented Basic
245 -Leucine Zipper domain (bZIP) motifs predicted to bind to 17 TFs, which are also known to
246 interact with AP-1 subunits (Figure 3A). NFPT-e vs. FPT-e comparison retrieved
247 hypermethylated eDMP-related DNA motifs (n=398) predicted to bind to 12 TF. The top most
248 enriched TF was Jun, another subunit of AP-1 and related to the bZIP DNA binding proteins
249 (Figure 3A). No adjusted significant DNA motifs were retrieved from the hypomethylated
250 eDMP regions from ACTH-e vs. FPT-e or from NFPT-e vs. FPT-e comparisons (n=15 or 48
251 hypo-DMP, respectively). The eDMPs related to the top three high-scored enhancers are
252 depicted in Figure 3B. Their respective associated target gene (TEAD3, PTCH1 and BAHCC1)
253 expression levels are depicted in Figure 3C.
254 We observed a tendency of a negative association between the methylation degree of
255 the eDMP and the expression of some of their putative target genes (p>0.05). For instance, in
256 association with the correspondent hypomethylation in the eDMP, the expression of TEAD3
257 was higher in ACTH-e than in the NFPT-e cluster (Figure 3C). Hypomethylated eDMP in the
258 FPT-e cluster (compared to ACTH-e) was related to upregulation of PTCH1 (Figure 3B-C) and
259 ERCC6L2 target genes (not shown) in the FPT-e cluster. In the comparison between NFPT-e
260 and FPT-e, we identified a cluster of 3 CpG probes overlapping a large enhancer
261 (GH171061384) that presented a heterogeneous methylation pattern. Two out of three probes
262 in this eDMP were hypomethylated in NFPT-e in relation to FPT-e. Their putative highest
263 scored target genes, BAHCC1 and a lncRNA (AC110285.7), were upregulated in NFPT-e. A
264 Genome Browser overview is used to highlight chromatin interactions between the top
265 enhancer and putative genes pairs (Figure 3D).
266 3.3 Pituitary-related transcription factors Tpit (TBX19) and SF1 (NR5A1)-related probes
267 differentiate specific methylation clusters.
12 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
268 We investigated whether pituitary-specific TF probes were differentially methylated
269 among the clusters. We found 125 probes-related to these TF (data not shown). Among the
270 pituitary-specific TF investigated in our analysis, SF1 (NR5A1) and Tpit (TBX19) were
271 significantly differentially methylated between two or three clusters (Figure 4A). SF1-related
272 DMPs were all located within enhancer intronic regions (n=2), while Tpit-related DMPs mainly
273 overlapped promoter (n=3) and intronic regions (n=2). SF1 DMPs were significantly
274 hypermethylated in ACTH-e and FPT-e when compared to the NFPT-e (Figure 4A). Tpit-
275 DMPs were entirely hypomethylated in ACTH-e in comparison with NFPT-e
276 (promoter/intronic) and significantly distinguished from FPT-e-DMPs by hypomethylation in
277 the promoter region (Figure 4B). The intronic Tpit region also distinguished the
278 hypermethylated NFPT-e from the hypomethylated FPT-e cluster (Figure 4A). Taken together,
279 T-pit methylation significantly distinguished all three methylation clusters; SF1 intronic
280 methylation distinguished ACTH-e and NFPT-e and, less significantly, NFPT-e from FPT-e
281 cluster. We established SF-1-related cutoffs for the two probes that significantly distinguished
282 ACTH-e and NFPT-e. Specimens that presented SF1-related probes with methylation values
283 below 0.5 (cg14143574) and 0.3 (cg02853418) were assigned to belong to NFPT-e, while beta-
284 values above those cutoffs were considered to belong to ACTH-e cluster (Figure 4B). The
285 application of these cutoffs to the Neou cohort correctly assigned 95% to either ACTH/T-pit
286 or NFPT/ SF1 specimens (χ2-squared p-value = 1.68e-08). Likewise, three out of five T-pit-
287 related DMP accurately distinguished ACTH-e from non-ACTH-e specimens (NFPT-e and
288 FPT-e). Specimens with CpG methylation below the thresholds for at least two out of those 3
289 probes, i.e. 0.5 (cg26160839), 0.7 (cg01732037) and 0.5 (cg15140722), assigned samples to
290 the ACTH-e group (Figure 4B). Applying this criteria to the validation cohort 10, 75% of the
291 ACTH tumors were classified accordingly while 100% of the other functioning and
292 nonfunctioning PT were assigned as the nonACTH category (χ2-squared p-value = 7.56 e-14).
13 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
293 4 Discussion
294 We conducted a comprehensive and integrative analysis between the molecular and
295 clinicopathological features of the largest multicentric methylome cohort of pituitary gland
296 specimens to date. An unsupervised analysis of the methylome cohort defined three groups of
297 PT characterized by hypo-, hyper- or intermediately methylated profiles mainly segregated by
298 functionality (functioning and nonfunctioning) as well as adenohypophyseal cell lineages,
299 namely FPT-e, NFPT-e and ACTH-e, respectively, similar to the results reported in recent
300 studies 10,15. These methylation clusters contain subtypes that recall and provide an epigenetic
301 support for the pituitary cell lineages-based criteria to classify PT recommended by the 2017
302 WHO classification, particularly to the Pit-1-derived tumors 5.
303 Notably, in the NFPT-e cluster, the methylome similarities between gonadotrophs and
304 null cell adenomas reported by us and others 10,15,28 may indicate that hypermethylation
305 underlies the pathogenesis of the lack of production and/or secretion of adenohypophyseal
306 hormones. In addition, we and others showed that silent and functioning corticotroph adenomas
307 (i.e. overt Cushing’s disease and subclinical) shared a more similar methylation profile
308 compared to each other and with NFPT. These results are concordant with a report that silent
309 ACTH and gonadotroph tumors share cytopathological features and clinical behavior 29.
310 Taking into account that cell-specific methylation patterns are preserved during development
311 and tumorigenesis, it is also possible that null cell, a subset of gonadotrophs and of corticotroph
312 adenomas share a common precursor 30,31 10,32,33 (Figure 1B). In addition, a study showed that
313 a subset of functioning corticotroph adenoma displayed a more intermediate methylation
314 profile between FPT and NFPT 10,32,33. Specific gene mutations such as GATA3 and USP8
315 accounted for the differences among these corticotroph adenomas variants 10,28. Altogether,
316 these results show that corticotroph adenomas consist of a molecularly heterogeneous tumor
14 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
317 10,15,28. Whether the association between methylation patterns and mutations in these variants
318 are drivers or passengers remains to be elucidated.
319 Pairwise comparisons between the methylation clusters yielded the identification of a
320 significant number of DMP (Figure 2A) that were mainly located in enhancer regions and
321 enriched for open sea CpGs (Figure 3B). Validating our findings, the overlap of either the DMP
322 or the top eDMP with the methylome, assigned the specimens of an independent cohort 10 into
323 similar methylation and clinicopathological groups as we observed in the pan-pit cohort. These
324 novel results indicate that these regulatory regions may play key roles in pituitary
325 tumorigenesis as well as in cell-specific development and differentiation as described in other
326 tumors 34. The biological significance of the differentially methylated enhancers across the
327 methylation clusters are endorsed by their association with target genes involved in pathways
328 related to developmental cytodifferentiation and in the tumorigenic process (e.g. TEAD3,
329 TAF11, lnRNA/AC110285.7, ERCC6L2, PTCH1) (Figures 3B,C)35–42.
330 Recently, the integration of multiple platforms including genomic and epigenomic
331 features has been proposed as an unbiased molecular approach to classify PT 10. Despite its
332 thoroughness, the complexity and the current cost of profiling multiple platforms to classify a
333 sample may limit its application in the clinics. Interestingly, their methylation profiles showed
334 similar clusters of PT as we described here. In addition the methylation-based groups were
335 similar to the ones obtained using the pangenomic approach 10. Altogether, these findings
336 provide evidence that methylome profiling per se may be a reliable and an unbiased approach
337 to classify PT 10. In our analysis, we were able to identify specific methylation thresholds
338 related to SF-1 and T-pit genes that significantly distinguished ACTH-e/Tpit from NFPT-e/SF1
339 methylation groups in our cohort that correctly assigned ACTH and NFPT specimens in the
340 independent cohort 10. These results, if validated by other methods (e.g. , pyrosequencing),
15 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
341 suggest that the methylation degree of these probes could be used as an objective approach to
342 classify PT specimens according to adenohypophyseal cell lineage.
343 The prognostic and predictive significance of methylation-based stratification in PT is
344 not clear, in contrast to the findings reported in other CNS tumors such as gliomas and
345 meningiomas 8,43. We and others showed that PT methylation clusters were not associated with
346 features used as surrogates of aggressive behavior in PT (e.g. invasion, resistance to treatment
347 etc) 5,10,28. However, the association of PT methylation clusters with a prognostic grading which
348 has been validated as predictors of tumor aggressive behavior remains to be investigated
349 8,43,44,45.
350 Besides refining tumor classification and stratification, methylome profiling can also
351 be applied to differentiate tumors originated from different tissues 7,46. Indeed, one study
352 showed that distinct methylome patterns segregated PT apart from other CNS tumors7.
353 Potentially, PT-specific signatures, as the ones herein described, could be helpful to distinguish
354 PT from other primary or secondary sellar tumors whose diagnosis by morphologic and
355 immunohistochemical approaches may be challenging and inconclusive 7,47,48.
356 In summary, we identified and validated methylation signatures that distinguished PT
357 by distinct functional status/subtype and adenohypophyseal cell lineages. These signatures,
358 annotated in enhancer regions, indicate the importance of these elements in the pathogenesis
359 of these tumors. They also provide an unbiased approach to classify pituitary tumors according
360 to the most recent classification recommended by the 2017 WHO using methylation profiling.
361
362 Acknowledgements
363 The authors are grateful to the HFHS patients who consented to the usage of PT for research
364 purposes. We thank Nancy Takacs and Heather Mengel for their administrative support;
365 Kevin Nelson for the collection, handling and maintenance of the tumor bank at the Hermelin
16 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
366 Brain Tumor Center; Andrea Transou for tumor pathology processing; Laura A. Hasselbach
367 for DNA extraction; Daniel Weisenberger and team at USC Epigenome Center for assistance
368 with DNA methylation profiling (HFHS support);Susan MacPhee for proofreading the
369 manuscript.
370 Bibliography
371 1. Mete O, Cintosun A, Pressman I, Asa SL. Epidemiology and biomarker profile of
372 pituitary adenohypophysial tumors. Mod Pathol. 2018;31(6):900-909.
373 doi:10.1038/s41379-018-0016-8
374 2. Asa SL, Casar-Borota O, Chanson P, et al. From pituitary adenoma to pituitary
375 neuroendocrine tumor (PitNET): an International Pituitary Pathology Club proposal.
376 Endocr Relat Cancer. 2017;24(4):C5-C8. doi:10.1530/ERC-17-0004
377 3. Ostrom QT, Cioffi G, Gittleman H, et al. CBTRUS Statistical Report: Primary Brain
378 and Other Central Nervous System Tumors Diagnosed in the United States in 2012-
379 2016. Neuro Oncol. 2019;21(Suppl 5):v1-v100. doi:10.1093/neuonc/noz150
380 4. Molitch ME. Diagnosis and treatment of pituitary adenomas: A review. JAMA.
381 2017;317(5):516-524. doi:10.1001/jama.2016.19699
382 5. Mete O, Lopes MB. Overview of the 2017 WHO classification of pituitary tumors.
383 Endocr Pathol. 2017;28(3):228-243. doi:10.1007/s12022-017-9498-z
384 6. Torregrosa-Quesada ME, García-Martínez A, Silva-Ortega S, et al. How Valuable Is
385 the RT-qPCR of Pituitary-Specific Transcription Factors for Identifying Pituitary
386 Neuroendocrine Tumor Subtypes According to the New WHO 2017 Criteria? Cancers
387 (Basel). 2019;11(12). doi:10.3390/cancers11121990
388 7. Capper D, Jones DTW, Sill M, et al. DNA methylation-based classification of central
389 nervous system tumours. Nature. 2018;555(7697):469-474. doi:10.1038/nature26000
17 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
390 8. Ceccarelli M, Barthel FP, Malta TM, et al. Molecular profiling reveals biologically
391 discrete subsets and pathways of progression in diffuse glioma. Cell. 2016;164(3):550-
392 563. doi:10.1016/j.cell.2015.12.028
393 9. Collord G, Tarpey P, Kurbatova N, et al. An integrated genomic analysis of anaplastic
394 meningioma identifies prognostic molecular signatures. Sci Rep. 2018;8(1):13537.
395 doi:10.1038/s41598-018-31659-0
396 10. Neou M, Villa C, Armignacco R, et al. Pangenomic classification of pituitary
397 neuroendocrine tumors. Cancer Cell. December 2019. doi:10.1016/j.ccell.2019.11.002
398 11. Bi WL, Horowitz P, Greenwald NF, et al. Landscape of genomic alterations in
399 pituitary adenomas. Clin Cancer Res. 2017;23(7):1841-1851. doi:10.1158/1078-
400 0432.CCR-16-0790
401 12. Song Z-J, Reitman ZJ, Ma Z-Y, et al. The genome-wide mutational landscape of
402 pituitary adenomas. Cell Res. 2016;26(11):1255-1259. doi:10.1038/cr.2016.114
403 13. Ling C, Pease M, Shi L, et al. A pilot genome-scale profiling of DNA methylation in
404 sporadic pituitary macroadenomas: association with tumor invasion and
405 histopathological subtype. PLoS ONE. 2014;9(4):e96178.
406 doi:10.1371/journal.pone.0096178
407 14. Kober P, Boresowicz J, Rusetska N, et al. DNA methylation profiling in
408 nonfunctioning pituitary adenomas. Mol Cell Endocrinol. 2018;473:194-204.
409 doi:10.1016/j.mce.2018.01.020
410 15. Salomon MP, Wang X, Marzese DM, et al. The Epigenomic Landscape of Pituitary
411 Adenomas Reveals Specific Alterations and Differentiates Among Acromegaly,
412 Cushing’s Disease and Endocrine-Inactive Subtypes. Clin Cancer Res.
413 2018;24(17):4126-4136. doi:10.1158/1078-0432.CCR-17-2206
18 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
414 16. Duong CV, Emes RD, Wessely F, Yacqub-Usman K, Clayton RN, Farrell WE.
415 Quantitative, genome-wide analysis of the DNA methylome in sporadic pituitary
416 adenomas. Endocr Relat Cancer. 2012;19(6):805-816. doi:10.1530/ERC-12-0251
417 17. Gu Y, Zhou X, Hu F, et al. Differential DNA methylome profiling of nonfunctioning
418 pituitary adenomas suggesting tumour invasion is correlated with cell adhesion. J
419 Neurooncol. 2016;129(1):23-31. doi:10.1007/s11060-016-2139-4
420 18. Davis S, Meltzer PS. GEOquery: a bridge between the Gene Expression Omnibus
421 (GEO) and BioConductor. Bioinformatics. 2007;23(14):1846-1847.
422 doi:10.1093/bioinformatics/btm254
423 19. Fortin J-P, Triche TJ, Hansen KD. Preprocessing, normalization and integration of the
424 Illumina HumanMethylationEPIC array with minfi. Bioinformatics. 2017;33(4):558-
425 560. doi:10.1093/bioinformatics/btw691
426 20. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with
427 confidence assessments and item tracking. Bioinformatics. 2010;26(12):1572-1573.
428 doi:10.1093/bioinformatics/btq170
429 21. Gardiner-Garden M, Frommer M. CpG islands in vertebrate genomes. J Mol Biol.
430 1987;196(2):261-282. doi:10.1016/0022-2836(87)90689-9
431 22. Zhou W, Laird PW, Shen H. Comprehensive characterization, annotation and
432 innovative use of Infinium DNA methylation BeadChip probes. Nucleic Acids Res.
433 2017;45(4):e22. doi:10.1093/nar/gkw967
434 23. Sandoval J, Heyn H, Moran S, et al. Validation of a DNA methylation microarray for
435 450,000 CpG sites in the human genome. Epigenetics. 2011;6(6):692-702.
436 doi:10.4161/epi.6.6.16196
437 24. Takai D, Jones PA. The CpG island searcher: a new WWW resource. In Silico Biol
19 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
438 (Gedrukt). 2003;3(3):235-240.
439 25. Fishilevich S, Nudel R, Rappaport N, et al. GeneHancer: genome-wide integration of
440 enhancers and target genes in GeneCards. Database (Oxford). 2017;2017.
441 doi:10.1093/database/bax028
442 26. Conway JR, Lex A, Gehlenborg N. UpSetR: an R package for the visualization of
443 intersecting sets and their properties. Bioinformatics. 2017;33(18):2938-2940.
444 doi:10.1093/bioinformatics/btx364
445 27. Boeva V. Analysis of genomic sequence motifs for deciphering transcription factor
446 binding and transcriptional regulation in eukaryotic cells. Front Genet. 2016;7:24.
447 doi:10.3389/fgene.2016.00024
448 28. Ricklefs FL, Fita KD, Rotermund R, et al. Genome-wide DNA methylation profiles
449 distinguish silent from non-silent ACTH adenomas. Acta Neuropathol. March 2020.
450 doi:10.1007/s00401-020-02149-3
451 29. Cooper O, Ben-Shlomo A, Bonert V, Bannykh S, Mirocha J, Melmed S. Silent
452 corticogonadotroph adenomas: clinical and cellular characteristics and long-term
453 outcomes. Horm Cancer. 2010;1(2):80-92. doi:10.1007/s12672-010-0014-x
454 30. Pulichino A-M, Vallette-Kasic S, Tsai JP-Y, Couture C, Gauthier Y, Drouin J. Tpit
455 determines alternate fates during pituitary cell differentiation. Genes Dev.
456 2003;17(6):738-747. doi:10.1101/gad.1065703
457 31. Drouin J. Epigenetic mechanisms of pituitary cell fate specification. In: Pfaff D,
458 Christen Y, eds. Stem Cells in Neuroendocrinology. Cham (CH): Springer; 2016.
459 doi:10.1007/978-3-319-41603-8_9
460 32. Manojlovic-Gacic E, Bollerslev J, Casar-Borota O. Invited Review: Pathology of
461 pituitary neuroendocrine tumours: present status, modern diagnostic approach,
20 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
462 controversies and future perspectives from a neuropathological and clinical standpoint.
463 Neuropathol Appl Neurobiol. May 2019. doi:10.1111/nan.12568
464 33. Kontogeorgos G, Thodou E. The gonadotroph origin of null cell adenomas. Hormones
465 (Athens). 2016;15(2):243-247. doi:10.14310/horm.2002.1652
466 34. Yao L, Berman BP, Farnham PJ. Demystifying the secret mission of enhancers:
467 linking distal regulatory elements to target genes. Crit Rev Biochem Mol Biol.
468 2015;50(6):550-573. doi:10.3109/10409238.2015.1087961
469 35. Zanconato F, Forcato M, Battilana G, et al. Genome-wide association between
470 YAP/TAZ/TEAD and AP-1 at enhancers drives oncogenic growth. Nat Cell Biol.
471 2015;17(9):1218-1227. doi:10.1038/ncb3216
472 36. Vierbuchen T, Ling E, Cowley CJ, et al. AP-1 Transcription Factors and the BAF
473 Complex Mediate Signal-Dependent Enhancer Selection. Mol Cell. 2017;68(6):1067-
474 1082.e12. doi:10.1016/j.molcel.2017.11.026
475 37. Atsaves V, Leventaki V, Rassidakis GZ, Claret FX. AP-1 Transcription Factors as
476 Regulators of Immune Responses in Cancer. Cancers (Basel). 2019;11(7).
477 doi:10.3390/cancers11071037
478 38. Carreno G, Apps JR, Lodge EJ, et al. Hypothalamic sonic hedgehog is required for cell
479 specification and proliferation of LHX3/LHX4 pituitary embryonic precursors.
480 Development. 2017;144(18):3289-3302. doi:10.1242/dev.153387
481 39. Qi C, Di Minin G, Vercellino I, Wutz A, Korkhov VM. Structural basis of sterol
482 recognition by human hedgehog receptor PTCH1. Sci Adv. 2019;5(9):eaaw6490.
483 doi:10.1126/sciadv.aaw6490
484 40. Liu X, Li H, Rajurkar M, et al. Tead and AP1 coordinate transcription and motility.
485 Cell Rep. 2016;14(5):1169-1180. doi:10.1016/j.celrep.2015.12.104
21 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
486 41. Gupta K, Watson AA, Baptista T, et al. Architecture of TAF11/TAF13/TBP complex
487 suggests novel regulation properties of general transcription factor TFIID. elife.
488 2017;6. doi:10.7554/eLife.30395
489 42. Tummala H, Dokal AD, Walne A, et al. Genome instability is a consequence of
490 transcription deficiency in patients with bone marrow failure harboring biallelic
491 ERCC6L2 variants. Proc Natl Acad Sci USA. 2018;115(30):7777-7782.
492 doi:10.1073/pnas.1803275115
493 43. Sahm F, Schrimpf D, Stichel D, et al. DNA methylation-based classification and
494 grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol.
495 2017;18(5):682-694. doi:10.1016/S1470-2045(17)30155-9
496 44. Asioli S, Righi A, Iommi M, et al. Validation of a clinicopathological score for the
497 prediction of post-surgical evolution of pituitary adenoma: retrospective analysis on
498 566 patients from a tertiary care centre. Eur J Endocrinol. 2019;180(2):127-134.
499 doi:10.1530/EJE-18-0749
500 45. Trouillas J, Roy P, Sturm N, et al. A new prognostic clinicopathological classification
501 of pituitary adenomas: a multicentric case-control study of 410 patients with 8 years
502 post-operative follow-up. Acta Neuropathol. 2013;126(1):123-135.
503 doi:10.1007/s00401-013-1084-y
504 46. Hoadley KA, Yau C, Hinoue T, et al. Cell-of-Origin Patterns Dominate the Molecular
505 Classification of 10,000 Tumors from 33 Types of Cancer. Cell. 2018;173(2):291-
506 304.e6. doi:10.1016/j.cell.2018.03.022
507 47. Bormann F, Rodríguez-Paredes M, Lasitschka F, et al. Cell-of-Origin DNA
508 Methylation Signatures Are Maintained during Colorectal Carcinogenesis. Cell Rep.
509 2018;23(11):3407-3418. doi:10.1016/j.celrep.2018.05.045
22 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
510 48. Moss J, Magenheim J, Neiman D, et al. Comprehensive human cell-type methylation
511 atlas reveals origins of circulating cell-free DNA in health and disease. Nat Commun.
512 2018;9(1):5068. doi:10.1038/s41467-018-07466-6.
513
23 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
514 Figure captions
515 Figure 1 - Methylome landscape segregates pituitary tumors into three distinct groups in
516 a multicenter cohort.
517 (A) Consensus Cluster correlation heatmap depicting three identified methylation-based
518 groups among pituitary tumors.
519 (B) Heatmap of the most variant CpG methylated probes (n=4,526) among 177 pituitary tumors
520 according to their methylation-based cluster and for 20 non-tumor pituitary specimens.
521 Columns represent specimens; rows represent the CpG probes.
522 (C) Boxplots depicting the mean DNA methylation of the top most variant probes according
523 to non-tumor and tumor pituitary specimens methylation groups.
524 Figure 2 - Differentially methylated distal probes define three clusters.
525 A) Heatmap of the differentially methylated CpG probes (DMPs) from pairwise comparisons
526 among clusters. Heatmap on the left displays the methylation landscape of the pan-pit cohort
527 (discovery, n=177) and the heatmap on the right depicts the methylation patterns of the
528 corresponding CpG applied to the Neou cohort (validation, n=86).
529 (B) Odds Ratio for the frequencies of DMPs for each pairwise comparison, depicting genomic
530 location and enhancer overlap relative to the expected genome-wide distribution of 450K
531 probes. Relative to the first cluster of the pairwise comparison label, red- and blue-shaded
532 boxes represent hypermethylated and hypomethylated probes respectively.
533 (C) Barplots showing the number of hypermethylated DMP that are unique or common in the
534 intersection among the three differentially methylated probe sets.
535 (D) Barplots showing the number of hypomethylated DMP that are unique or common in the
536 intersection among the three differentially methylated probe sets.
537 Figure 3 - Differentially methylated candidate enhancers predicted AP-1 and NR5A2
538 transcription factors and respective putative enhancer-target genes.
24 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
539 (A) Representation of DNA motifs significantly enriched among differentially methylated
540 enhancer regions obtained from the pairwise comparison between ACTH-e, NFPT-e and FPT-
541 e.
542 (B) Methylation of probes in the top three highest-scored enhancers.
543 (C) Expression of the high scored enhancer-target genes expression, per methylation cluster.
544 (D) Genome browser view of top enhancers and the highest scored target genes, depicting the
545 GENCODE transcripts, the predicted chromatin interactions and the conservation among
546 vertebrates.
547 Figure 4 - Pituitary-specific transcription factors found differentially methylated among
548 the three methylation clusters.
549 (A) Heatmap indicating significant differentially methylated probes associated with SF-1
550 (NR5A1) and T-pit (TBX19) among the three methylation clusters: ACTH-e, NFPT-e and
551 FPT-e. Heatmap on the left displays the pan-pit cohort (discovery, n=177) and the heatmap on
552 the right depicts the methylation patterns of the corresponding CpG applied to the Neou cohort
553 (validation, n=86).
554 (B) Density plots on the distribution of beta value and tentative thresholds (vertical dashed
555 lines) for SF-1- and T-pit-related differentially methylated probes that distinguish the ACTH-
556 e from NFPTs and ACTH from non-ACTHs, respectively.
557 Table caption
558 Table 1 - Clinicopathological classification features of pituitary tumors from a multicenter
559 cohort. NFPT: Nonfunctioning pituitary tumor; Supplementary Figures
560 Figure S1 - Principal component analysis (PCA) colored by (A) cohort source; (B) by PT
561 functional status; (C) by methylation cluster; (D) Cumulative distribution function of consensus
562 clustering classification and (E) Consensus cluster delta area plot of consensus clustering
563 classification.
25 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
564 Figure S2 - DNA methylation-based stemness index (mDNAsi) of pan-pituitary cohort
565 specimens, according to methylation clusters and clinicopathological features.
566 (A) By methylation clusters.
567 (B) By functional status.
568 (C) By the high risk subtypes according to the 2017 WHO classification.
569 (D) By gender.
570 (E) By age.
571 *=p<0.05;**=p<0.01;***=p<0.001;****=p<0.0001
572 Figure S3 - Supervised analysis of the methylome according to invasion status.
573 (A) Heatmap of the 13 differentially methylated probes between invasive vs. noninvasive PT
574 (n=34 vs. n=46, respectively; p< 0.001 and ∆β=0.15).
575 (B) Venn diagram depicting common and distinct differentially methylated genes from our
576 study compared to several studies.
577 (C) Differentially methylated probes overlapping enhancer regions (eDMPs) and the respective
578 candidate enhancer-target genes correlation according to genomic context.
579 (D) Differentially methylated probes overlapping enhancer regions (eDMPs) and the respective
580 candidate enhancer-target genes correlation according to invasion status.
581 Figure S4 - DNA methylation-based stemness index and RNA-based stemness index of
582 Pan-pit cohort and integrative subset cohort based on the invasion status.
583 (A) DNA methylation-based stemness index (mDNAsi) of the pan-pituitary specimens cohort
584 colored by invasion status (n=197).
585 (B) DNA methylation-based stemness index (mDNAsi) of a subset of the pan-pituitary cohort
586 colored by invasion status (n=23)
587 (C) RNA expression-based stemness index (mRNAsi) of a PT cohort subset, colored by
588 invasion status (n=23).
26 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; thisDNA version Methylation-based posted April 27, 2020. Clusters The copyright in Pituitary holder for Tumors this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
589 Supplementary tables
590 Table S1 - Compiled clinical information for each patient pituitary specimen.
591 Table S2 - Differentially methylated probes from supervised comparisons. 1-3) differentially
592 methylated probes (DMP) among the three methylation clusters comparisons; 4) DMPs
593 between tumorous and non-tumorous pituitary specimens. 5) DMPs between invasive and
594 noninvasive PT.
595
27 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; this version posted April 27, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
A B PITUITARY ADENOMAS (n=177) NT (n=20) PITUITARY ADENOMAS (n=177) Methylation cluster samples (n=197) Cell lineage TF samples (n=197) Functional status Cohort Hormone staining WHO 2017_High risk Gender Age
Correlation Methylation 1 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0
PROBE LEGEND CpG Location CpG.island Shore Shelf Open.sea Top 1% most variant probes (n= 4,526) 1% Top (n= probes variant most
C Top 1% most variant probes (n= 4,526)
0.8 lation
y 0.6 Methylation cluster
th Non-tumor ACTH−e NFPT−e SAMPLE LEGEND A me FPT−e Methylation cluster Functional status Hormone staining WHO 2017.High risk 0.4 ACTH−e Functioning ACTH Yes NFPT−e Nonfunctioning GH No FPT−e GH/PRL Undetermined GH/TSH Cell lineage TF Cohort GH/PRL/TSH Gender Female mean DN Pit−1 Nontumor-Kober GH densely T−pit Nontumor-Capper GH sparsely Male 0.2 Pit−1, ER-α Nontumor-HFHS LH and/or FSH Age Pit−1, GATA2 Tumor-Ling LH/FSH/TSH 18−20 SF−1, GATA2, ER-α Tumor-Kober TSH 20−40 SF−1, GATA2, ER-α, Pit-1 Tumor-Capper Non-tumor ACTH-e NFPT-e FPT-e PRL 40−60 All Tumor-HFHS PRL densely 60−80 None Null cell >80 Methylation cluster Unknown Unknown bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; this version posted April 27, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
A DISCOVERYDISCOVERYDISCOVERY VALIDATION NT (n=20) TUMOR (n=177) TUMOR (n=86) MethMethylationyMethylationlation clust cluster clusterer Functional group CellCell Celllineag lineage lineagee TF TF TF TF cell lineage WHOWHOWHO 2017 2017 2017 WHO 2017
SAMPLE LEGEND
Methylation cluster Functional group
(n=1,067) ACTH−e ACTH NFPT-e NFPT
ACTH vs NFPT FPT-e FPT
NFPT-e vs ACTH-e silent ACTH subclinical ACTH
Cell lineage TF TF cell lineage Pit−1 Pit−1 T−pit T−pit Pit−1, ERalpha SF−1 Methylation Pit−1, GATA2 None 1 SF−1, GATA2, ERalpha 0.8 0.6 SF−1, GATA2, ERalpha, Pit−1
(n=1,039) 0.4 All 0.2 None
FPT vs ACTH 0 Unknown
FPT-e vs ACTH-e PROBE LEGEND
WHO 2017 CpG Location Corticotroph CpG island Silent corticotroph Shore Somatotroph Shelf Mammosomatotroph Open sea Plurihormonal Densely somatotroph Enhancer Sparsely somatotroph No Gonadotroph Yes Thyrotroph (n=1,210) Lactotroph Null cell FPT NFPT vs Undetermined FPT-e vs NFPT-e n er o i c at an c h o L En G Cp
B ACTH-e vs. FPT-e C D ACTH-e vs. NFPT-e ACTH-e vs. FPT-e NFPT-e vs. FPT-e Enhancer 800 733 793 750 630 Not enhancer 600
500 ion Size Open.sea ion Size 400 e 345 r sect sect u t er er
a Shelf 220 250 e Int 200 Int F 131 Shore 17 9 37 0 0 1
CpG.island
0 1 2 3 0 1 2 3 0 1 2 3 900 600 300 0 800 600 400 200 0 Odds−ratio Set Size Set Size
LEGEND Significance Probe direction DMP sets Depleted ACTH-e (n=29) vs. NFPT-e (n=82) Hypermethylated probes Enriched ACTH-e (n=29) vs. FPT-e (n=66) Hypomethylated probes Non-significant NFPT-e (n=82) vs. FPT-e (n=66) bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; this version posted April 27, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
A Hypo enhancers in ACTH-e (n=295) from ACTH-e vs. NFPT-e Hyper enhancers in ACTH-e (n=88) from ACTH-e vs. NFPT-e Hyper enhancers in ACTH-e (n=360) from ACTH-e vs. FPT-e p.value =1e-3; 3.9 Fold p.value =1e-11; 3.2 Fold
bZIP motif AP-1(Fra1/Fra2) NR motif Nr5a2
2 2
1 1 mation content mation content or or 0 Inf Inf 0 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 Position Position B C
NT (n=20) PITUITARY ADENOMAS (n=177)
Methylation cluster Top enhancer target-genes expression Methylation Cell lineage TF 1 GH06I035483 GH09I095503 GH17I081384 0.8 Functional status PTCH1 BAHCC1 0.6 TEAD3 Hormone staining 0.4 0.2 WHO 2017_High risk 0 Gender SAMPLE LEGEND Methylation cluster WHO 2017_High risk 0.6 Age ACTH−e Yes NFPT−e No FPT−e Undetermined Functional status cg02333187_GH06I035483 Functioning Gender Nonfunctioning Female Male Cell lineage TF Pit−1 Age 0.4 T−pit 18−20
=5) Pit−1, ER-α 20−40 cg13754720_GH09I095503 Pit−1, GATA2
(n 40−60 2(TPM + 1)
n SF−1, GATA2, ER-α
60−80 g o o i SF−1, GATA2, ER-α, Pit-1 >80 l
at All l
y None h Unknown et PTCH1 0.2 cg06636541_GH17I081384 Hormone staining m ACTH er GH c GH/PRL an GH/TSH h cg11031567_GH17I081384 GH/PRL/TSH En GH densely GH sparsely LH and/or FSH 0.0 LH/FSH/TSH TSH ACTH−e NFPT−e ACTH−e FPT−e NFPT−e FPT−e cg26218110_GH17I081384 PRL PRL densely Null cell Unknown
D chr6 (p21.31) 6p22.3 6q126q13 q14.1 q15 16.1 6q21 22.31 26 6q27
10 kb hg38 35,485,000 35,490,000 35,495,000 35,500,000 35,505,000 GENCODE Transcript Set TEAD3 TEAD3
483 Predicted chromatin interactions D3 GH06I035483 TEAD3 TEA 4.88 _ 100 vertebrates Basewise Conservation by PhyloP GH06I035 0 - -4.5 _ 21.1 21.3 9p23 (q22.32) chr9
chr9 (q22.32) 9p23 21.3 21.1 11 9q12 13 31.1 32 33.1
200 kb hg38
D3 95,600,000 95,700,000 95,800,000 95,900,000 96,000,000
PTCH1 PTCH1 GENCODE Transcript Set ERCC6L2 PTCH1 PTCH1 PTCH1 PTCH1 TEA PTCH1 9IPTCH1095503 PTCH1 PTCH1 ERCC6L2 PTCH1 ERCC6L2 PTCH1
503 ERCC6L2 ERCC6L2 H1 Predicted chromatin interactions C GH09I095503 ERCC6L2 PT PTCH1 GH09I095
100 vertebrates Basewise Conservation by PhyloP 4.88 _
0 - -4.5 _
21.1 21.3 9p23 (q22.32) chr9 chr17 (q25.3) 13.3 13.2 p13.1 17p12 17p11.2 17q11.2 17q12 21.31 17q22 23.2 q24.3 q25.1 17q25.3
10 kb hg38 81,385,000 81,390,000 81,395,000 81,400,000 81,405,000 81,410,000 GENCODE Transcript Set
BAHCC1 AC110285.6 BAHCC1 384
C1 Predicted chromatin interactions
BAHCC1 GH17I081384 AHC B
GH17I081 100 vertebrates Basewise Conservation by PhyloP 4.88 _
0 - -4.5 _ bioRxiv preprint doi: https://doi.org/10.1101/2020.04.25.061903; this version posted April 27, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
A DISCOVERY VALIDATION
NT (n=20) PITUITARY ADENOMAS (n=177) PITUITARY ADENOMAS (n=86) Methylation cluster Functional group Cell lineage TF TF cell lineage WHO 2017 WHO 2017
cg14143574 SF-1 cg02853418
Methylation cg15140722 1 0.8 0.6 cg01732037 0.4 0.2 0 T-pit cg26160839
cg06941869
cg17095936
SAMPLE LEGEND Methylation cluster Cell lineage TF WHO 2017 ACTH−e Pit−1 Corticotroph NFPT-e T−pit Silent corticotroph FPT-e Pit−1, ERalpha Somatotroph Pit−1, GATA2 Mammosomatotroph SF−1, GATA2, ERalpha Plurihormonal Functional group SF−1, GATA2, ERalpha, Pit−1 Densely somatotroph ACTH All Sparsely somatotroph NFPT None Gonadotroph FPT Unknown Thyrotroph silent ACTH Lactotroph subclinical ACTH TF cell lineage Null cell Pit−1 Undetermined T−pit SF−1 None
B SF-1 (NR5A1) cg14143574 cg02853418 1.00 Methylation cluster 0.75 ACTH−e 0.50 NFPT−e FPT−e 0.25
0.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
NFPT-e ACTH-e NFPT-e ACTH-e
y / Count T-pit (TBX19) cg26160839 cg01732037 cg15140722 1.00 Densit
0.75
0.50
0.25
0.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
ACTH-e Non-ACTHs ACTH-e Non-ACTHs ACTH-e Non-ACTHs
cg26160839 DNA methcg26160839ylation level cg26160839