Genome-wide methylation profiling of Peripheral T-cell lymphomas identifies TRIP13 as a critical driver of tumor proliferation and survival

Cytosine methylation of genomic DNA contributes to the regulation of gene expression and is involved in normal development including hematopoiesis in mammals. It is catalyzed by the family of DNA methyltransferases (DNMTs) that include DNMT1, DNMT3A, and DNMT3B. Peripheral T-cell lymphomas (PTCLs) represent a diverse group of aggressive mature T-cell malignancies accounting for approximately 10–15% of non-Hodgkin lymphoma cases in the US. PTCLs exhibit a broad spectrum of clinical, histological, and immunophenotypic features with poor prognosis and inadequately understood molecular pathobiology. To better understand the molecular landscape and identify candidate genes involved in disease maintenance, we used high-resolution Whole Genome Bisulfite Sequencing (WGBS) and RNA-seq to profile DNA methylation and gene expression of PTCLs and normal T-cells. We found that the methylation patterns in PTCLs are deregulated and heterogeneous but share 767 hypo- and 567 hypermethylated differentially methylated regions (DMRs) along with 231 genes up- and 91 genes downregulated in all samples suggesting a potential association with tumor development. We further identified 39 hypomethylated promoters associated with increased gene expression in the majority of PTCLs. This putative oncogenic signature included the TRIP13 (thyroid hormone receptor interactor 13) gene whose both genetic and pharmacologic inactivation, inhibited cellular growth of PTCL cell lines by inducing G2-M arrest accompanied by apoptosis suggesting that such an approach might be beneficial in human lymphoma treatment. Altogether we show that human PTCLs are characterized by a large number of recurrent methylation alterations, and demonstrated that TRIP13 is critical for PTCL maintenance in vitro.


Introduction
Non-Hodgkin's lymphomas (NHLs) are a heterogeneous group of lymphoid malignancies that arise from the transformation of B-, T-, and NK cells.Although the majority of NHLs are B-cell lymphomas (BCLs), ~ 15% of patients in Western countries suffer from more aggressive Peripheral T-cell lymphomas (PTCLs) with poor survival rates [1][2][3] .PTCLs exhibit a broad spectrum of clinical, histological, and immunophenotypic features, which present signi cant therapeutic challenges and an overall unfavorable prognosis.Based on their manifestations -disseminated, nodal disease, extranodal, or cutaneous -and molecular features, at least 29 discrete types of PTCLs have been identi ed to date 4 .
ALCL are CD30-positive aggressive lymphoproliferative disorders affecting lymph nodes and extranodal sites and characterized by the expression of the Anaplastic lymphoma kinase (ALK) protein as either ALKpositive and ALK-negative subtypes 5 .AITLs arise from the follicular T helper cells and manifest by liver and spleen enlargement, lymphadenopathy, and weight loss 4,5 .
value of such information, cancer-speci c DNA methylation patterns of several human PTCL types have been partially characterized to date.In subtypes of human ALCL, DNA methylation signatures affect TCR and other genes involved in T-cell differentiation, as well as transcription factor induction and occupancy thereby contributing to ALCL oncogenic signaling 24 .In hepatosplenic TCL, several hyper-or hypomethylated loci were identi ed in gene regulatory elements including enhancers with eight-gene signatures of recurrent changes 25 .
To better understand the recurrent molecular changes, and their possible contribution to disease maintenance, here we used high-resolution methylation and gene expression pro ling of up to 10 PTCLs along with normal T-cells.
We found that whereas normal T-cell methylation patterns are quite similar across several samples, methylation patterns in PTCL are heterogeneous and characterized by both hypo-and hypermethylated regions relative to normal controls with the majority of changes not shared across tumors.In addition, we identi ed methylation (767 hypo -and 567 hypermethylated DMRs) and gene expression signatures (31 genes up-and 91 genes down) present in all tested tumors suggesting a potential association with tumor development.We also found that putative tumor suppressors, such as FOXP1, STK4, and ATM were frequently hypermethylated and silenced whereas oncogenes, such as UHRF1, MPZL1, CDK14, RACGAP1, and RAB13, were hypomethylated and overexpressed suggesting that DNA methylation changes contribute to PTCL development.Among putative oncogenes was also TRIP13 (thyroid hormone receptor interactor 13) hypomethylated and overexpressed in 9/10 PTCLs on average of ~ 10-fold.We show that both TRIP13 knockdown and its pharmacological inhibition by a small-molecule inhibitor DCZ0415 impaired cellular growth of T-cell lymphoma cell line T8ML-1 26 , as well as T-cell leukemia cell lines by inducing G2-M arrest and apoptosis suggesting the role of this gene in maintenance of T-cell malignancies.
Altogether we show that human PTCL are characterized by a large number of recurrent methylation alterations affecting the expression of genes involved in tumorigenesis and functionally demonstrate that TRIP13 plays a role in PTCL lymphoma maintenance in vitro.Therefore, targeting TRIP13 might be bene cial in human PTCL treatment.

DNA methylome and transcriptome of normal human CD4 + and CD8 + T-cells
To identify molecular changes associated with PTCL development, we rst sought to determine methylation and gene expression patterns in normal T-cells.To achieve that we obtained CD4 + and CD8 + normal human T-cell samples isolated from peripheral blood (Precision Medicine) and subjected them to Whole Genome Bisul te Sequencing (WGBS) and RNA-seq analysis.WGBS yielded more than 28M reads for each sample (Fig. S1).We next analyzed methylation data along with data obtained from puri ed CD4 + T-cells (sample CD4_1) for which both WGBS and RNA-seq were publicly available.The methylation analysis revealed that 21,936,712 individual CpG dinucleotides (CGs) were covered ≥ 5x in all three samples and were used further.A majority of CGs had a high level of methylation with more than 1.4 x 10 7 CGs methylated to at least 75% in all three samples, whereas only 1 x 10 6 CGs were methylated ≤ 25% (Fig. 1A).Further analysis revealed that more than 15,000 out of 34,858 promoters (de ned as − 1500 bp to + 500 bp relative to the transcription start site -TSS) were heavily methylated (at least 75%) and only ~ 4,000 promoters were methylated at lower levels (less than 25%) in each sample (Figs.1B, 1C and Supporting Information 1).The positional variability in promoter methylation was low with very similar patterns across samples as determined by Pearson's correlation with R values close to 1 in all pairwise comparisons (Fig. 1D).We next utilized RNA-seq data to determine if methylation may impact gene expression.This analysis revealed that in general, the degree of promoter methylation is inversely correlated with gene expression (Figs. 1E, 1F and Supporting Information 2).Pairwise comparisons of gene expression and methylation revealed that FPKM values for genes with different percentages of promoter methylation were signi cantly different (p < 0.05, two-tailed Student's t-test) in all comparisons except 0-25% vs 26-50% and 26-50% vs 51-75% in the rst CD4 + T-cells sample (Fig. 1F).For example, genes with promoters that were less than 25% methylated were expressed at higher levels than genes with highly methylated promoters (> 75%; Fig. 1F).Three analyzed T-cell samples were more variable in gene expression patterns as R values in pairwise comparisons were lower than those seen based on promoter methylation (Fig. 1G).This result suggests that promoter methylome is more stable than transcriptome in normal T-cells across individuals and perhaps less prone to polymorphic changes in the human population.
Pathway enrichment analysis of 3,001 highly expressed genes (FPKM ≥ 5) by EnrichR revealed, not surprisingly, signi cant enrichment in genes related to T-cell development.The most signi cant in Biocarta 2016 were T-cell receptor, IL2 -, IL7-signaling, and T-cell apoptosis (Fig. 1H).
To further investigate methylation variability in human T-cells, we analyzed the similarity of methylation patterns among an additional three samples for which WGBS data are available.A pairwise comparison of the methylation status of 7,628,019 cytosines (covered 5x) that overlapped in all controls revealed that samples showed a high degree of similarity irrespective of their immunophenotypes (R = 0.73-0.85)for all pairwise comparisons (Fig. 1I).
Our analysis thus demonstrates that a majority of CpG dinucleotides and promoters are mostly methylated in normal T-cells and that methylation polymorphism is limited resulting in relatively infrequent differences in promoter methylation between CD4 + and CD8 + T-cells.

DNA methylome of Peripheral T-cell lymphomas
To analyze DNA methylation in human PTCL, we rst obtained a set of seven primary human lymphoma samples from both the Cooperative Human Tissue Network (CHTN) and commercial sources.This resulted in a collection of four ALCL (T1-T4), one AITL sample (T5), and two PTCL-NOS (T6, T8) (Fig. S2).
After DNA isolation, we subjected the obtained samples to global methylation pro ling using WGBS.
Analysis of the data revealed that we were able to obtain sequencing depths ranging from 16.5 to 24M CGs that were covered by at least ve sequence reads in each sample (Fig. S3).For analysis of tumorspeci c methylation, we used control sample consisting of methylation data obtained from ve normal human T cells -CD3, CD4_1, CD4_2, CD4_3 and CD8 -averaged out by Metilene.A pairwise comparison of the methylation status of 7,628,019 cytosines (covered 5x, averaged out by Metilene) that overlapped in all samples revealed that tumors were severely hypomethylated relative to the controls (Figs.2A, S1 and S3).This was further con rmed by the analysis of differentially methylated cytosines (DMCs) which were those that had at least 10% methylation change relative to controls (p < 0.05 (MWU)).Our pairwise analysis on CGs covered in all samples revealed that most of the DMCs were hypomethylated relative to normal T-cells in all tumors (Fig. 2B).Hypomethylation was most pronounced in T1 and T8 lymphomas and the least in T2 and T4 that had almost equal numbers of hypo-and hypermethylated DMCs (Fig. 2B).
More stringent analysis of methylation changes using differentially methylated regions (DMRs; de ned as ≥ 10% methylation change in the same direction in three consecutive cytosines in ≤ 100 bp, p < 0.05 (MWU)) con rmed the large-scale deregulation of methylation in all tumors.Notably, thousands of hyperand especially hypomethylated DMRs were observed in all tumors in various genomic elements including promoters, enhancers, introns, exons, and repeats (Figs.2C-G).Like in mouse hematologic malignancies [17][18][19] , the presence of hypo-rather than hypermethylated DMRs in promoters was a more frequent event in all tumors ranging from ~ 6,000 in T2 to ~ 20,000 in T8 (Fig. 2C).However, promoter hypermethylation was seen in all tumors and ranged from ~ 2,000 (T1) to 8,000 (T3).In some tumors -T2 and T3promoter hypo and hypermethylation were almost equal in frequency (Fig. 2C).Patterns of hypo-and hypermethylation changes didn't appear to depend on tumor type since the most hypomethylated tumors -T1 and T8 -belonged to different PTCL subtypes (ALCL and PTCL-NOS, respectively).
While tumors varied in the frequency of hypo and hypermethylated DMRs in genomic elements, the trend remained similar within individual tumors.For example, the most hypomethylated tumors were T1 and T8, and hypomethylation was manifested across all genomic elements.Similarly, the ratio between hypo and hypermethylated DMRs was the smallest in T2, and that trend was seen across all genomic elements (Figs.2C-G and data not shown).
Further dissection of this signature may reveal methylation markers of PTCL associated with the disease initiation and progression, as well as genes driving the disease development.

Analysis of gene expression in ALCL and PTCL-NOS
To determine the extent to which methylation changes may contribute to deregulated transcriptome in PTCL, we next performed global gene expression pro ling of PTCL samples using RNA-seq analysis.In addition to PTCL samples used for methylation analysis, we also included RNA isolated from three additional PTCL-NOS samples (T7, T9, T10) along with RNA from normal T-cells used for methylation analysis (Figs. Receptor Signaling, and HIPPO signaling (Fig. 3B).Additional frequently suppressed pathways included RHOGDI Signaling, Cell Cycle G1/S Checkpoint Regulation, tumor suppressive PTEN Signaling, and others (Fig. 3B).In contrast, pathways frequently activated in all tested samples included Estrogen-mediated Sphase Entry, Cell Cycle Control of Chromosomal Replication, Signaling by Rho Family GTPases, and the STAT3 pathway.Additionally, there was a high frequency of activation in Interferon and RHOA Signaling pathways, along with other cancer-related pathways (Fig. 3B).
The expression of 231 genes was increased and the expression of 91 genes was decreased at least 2-fold in all ten PTCLs.We term these molecular changes as 'Core Gene Expression Signature' (Figs. 3C and Supporting Information 5).The TOP ve genes overexpressed in all PTCL samples encode the following: phospholipase PLA2G2D, organic anion transporting polypeptide SLCO2B1, metalloprotease ADAMDEC1, complement protein C1QB, and receptor activity modifying protein RAMP3, with the average increase in expression over 500-fold.(Supplementary Figs.S7 and Supporting Information 5).None of these genes seem to have an established role in T-cell transformation and their deregulation can be a consequence of changes in signaling and epigenetic alterations in PTCL.GO enrichment analysis identi ed pathways linked to cell cycle regulation and in particular to Cell Cycle G2/M Phase Transition, Mitotic Nuclear Division, and Sister Chromatid Segregation (Fig. 3D).Analysis of genes not directly linked to the cycle revealed pathways related to Complement Activation, Angiogenesis, Blood Coagulation, and Notch Pathway Signaling (Fig. S8).ChIP enrichment analysis revealed the possible involvement of FOXM1, E2F1 and E2F4, SOX2, KLF4, and MYBL2 transcription factors in the pathogenesis of PTCL as they may play a role in the deregulation of the 'Core PTCL Expression Signature' (Fig. S9).
In all 10 PTCL samples, 91 genes exhibited a consistent decrease in expression, with at least a 2-fold reduction (Fig. 3C).Notably, the most frequently downregulated genes included FGFBP2 (also known as KSP37), teneurin TENM1, and a group of Y-linked genes such as USP9Y, TTTY15, and UTY (Figs. 3C and  S7).Their role in lymphomagenesis remains unclear.
The GO enrichment analysis identi ed pathways linked to the Cell cycle, Histone demethylation, epigenetic dysregulation, and B cell differentiation likely occurring with the contribution of transcription factors FOXO1, FOXOM1, MYB, and others (Figs.3E, S10 and S11).
Up-regulation of TBX21 or GATA3 and their target genes (EOMES, CXCR3, IL2RB, CCL3, IFNγ, and CCR4, IL18RA, CXCR7, IK respectively) was previously shown to distinguish two subclasses of PTCL-NOS with distinct clinical outcomes 6,27 .To determine if any of the ve samples included in our analysis belongs to a speci c PTCL-NOS subtype, we analyzed RNA-seq expression patterns further.However, we did not see up-regulation of TBX21 or GATA3 when compared to normal controls in any of the ve cases of PTCL-NOS perhaps because our analysis was limited by the small sample size (Fig. S12).

DNA methylation modi ers are deregulated in human PTCL
To determine if levels of DNA methylation modi ers are deregulated in human PTCL, we further analyzed RNA-seq data and found that DNMT3A was signi cantly reduced in all tumors when compared to either CD4 + or CD8 + normal T-cells from peripheral blood (Fig. 4A).In contrast, DNMT1 and DNMT3B were unchanged with only one ALCL tumor, T3, showing a signi cant increase in transcript levels (Fig. 4A).Analysis of DNA demethylases showed that TET1 transcripts were low in all samples (FPKM < 1) but still signi cantly downregulated in the majority of tested tumors, whereas levels of TET2 and TET3 were unchanged relative to normal thymocytes (Figs.4A and S13).Another protein that can affect DNA methylation is the TCL1A protein which was shown to inhibit DNA methyltransferases biochemically and affect global methylation in a mouse model of hematological malignancies 15 .We, therefore, analyzed its expression and found six out of ten analyzed tumors had signi cantly increased levels of TCL1A RNA relative to normal T-cells (Fig. 4A).In contrast, the expression of another member of the TCL family -TCL1B -that does not inhibit Dnmts 15 was unchanged in PTCL (Fig. S13).
To analyze protein levels, we next performed immunoblot analysis of DNMTs in a subset of PTCLs for which we had frozen tissues available (Tumors 1-7).Consistent with RNA-seq data analysis, DNMT3A was downregulated in most tumors relative to normal peripheral blood lymphocytes (Fig. 4B).Interestingly, we detected low protein levels of DNMT3B in T2, T4, T5, and T6 suggesting that despite unchanged mRNA levels, protein down-regulation may occur in primary PTCL (Fig. 4C).Whether DNMT3B levels are affected by TCL1A overexpression or the protein is down-regulated by other mechanisms, remains unclear.Unlike DNMT3A/B, DNMT1 levels did not seem to be changed in lymphomas (data not shown).
To determine whether deregulated expression of methylation modi ers is more broadly observed in PTCL subtypes, we next utilized publicly available data generated by RNA-seq analysis of 15 primary Natural killer/T-cell lymphomas (NKTCL), 21 ALCL (ALCL-2), eight Adult T-cell leukemia/lymphoma (ATLL), and eight T-lymphoblastic lymphomas (TLBL), and our ve NOS and four ALCL (ALCL-1).While the expression of DNMT1 was mostly unchanged, DNMT3A levels were signi cantly reduced across all PTCL subtypes except TLBL (Fig. 4D).Like DNMT1, DNMT3B expression was unchanged across most PTCL except for a signi cant increase in TLBL (Fig. 4D).The levels of TET1 RNA showed a tendency for downregulation in a majority of analyzed PTCLs although it did not reach statistical signi cance, while TET3 was mostly unchanged irrespective of tumor type (Fig. S14).
TCL1A was upregulated in a majority of PTCLs, whereas TCL1B was not signi cantly increased in any of the tested tumor subtypes (Figs.4D, S14).
Altogether, this analysis revealed that several DNA methylation modi ers are deregulated in PTCL, with the downregulation of DNMT3A and TET1 and up-regulation of TCL1A in most tested samples.These molecular changes, along with the downregulation of DNMT3B protein and genetic alterations found in these modi ers in a subset of PTCL as reported by others [28][29][30] , are the most likely reasons for the largescale deregulation of PTCL methylomes we observed.

Deregulated promoter methylation is associated with changes in gene expression
To determine the extent to which methylation changes in PTCL may affect transcription we further compared the data sets generated by WGBS and RNA-seq.First, we found that methylation changes also affected the expression of various repetitive elements, including DNA transposons, LTR and Non-LTR retrotransposons, and satellite repeats (Fig. S15).In a subset of repeat elements, such as UCON80, LTR39, and LTR180, hypomethylation mostly correlated with an increased expression.In contrast, hypomethylation of UCON34 correlated with decreased, rather than increased expression (Fig. S15).However, we didn't detect any consistent pattern of methylation changes, gene expression, or their correlation in examined tumors suggesting that repeats may not serve as good markers of disease development or progression.
Next, we examined gene expression changes that correlated with methylation changes in gene promoters at high frequency and found that 2,810 genes contained hypomethylated DMRs in 5 out of 7 tumors (Supporting Information 6).Out of those, 153 genes had increased expression in 5 out of 7 tumors (Fig. S16).EnrichR pathway enrichment analysis coupled with Wiki Pathway datasets revealed pathways regulating multiple cellular processes that are commonly upregulated in cancers, such as VEGFA-VEGFR2 Signaling, PI3K-Akt signaling, and EGF/EGFR signaling (Fig. S17).Interestingly, several pathways related to Hippo signaling such as Hippo-Merlin Signaling Dysregulation, Overview of leukocyte-intrinsic Hippo pathway, Pathways Regulating Hippo and Hippo-Yap signaling pathway raising the possibility that the deregulation of this pathway in T-cells could play a functional role in transformation.Promoters of 1,281 genes contained hypomethylated DMRs in at least 6 out of 7 tumors.Out of these, 39 genes were associated with increased expression (≥ 1.5) in PTCL suggesting that loss of methylation may have contributed to their deregulation (Fig. 5A).
We identi ed 1,220 hypermethylated genes with the frequency of 5 out of 7 tumors out of which 74 are also downregulated in 5 out of 7 tumors (Supporting Information 6 and Fig. S18).Decreased expression of these genes was associated with the deregulation of various pathways including the Wnt Signaling Pathway and Pluripotency and Pathways Regulating Hippo Signaling.(Fig. S19).Promoters of 578 genes showed hypermethylated DMRs with a high frequency of 6 out of 7 PTCL.Out of these, the expression of 56 genes decreased by at least 1.5-fold.(Fig. 5A).Among these were genes with putative tumor suppressor function in different cancer types including FOXP1, STK4, ATM, and others [36][37][38] .
Interestingly, several genes with potential oncogenic functions, such as SF3B1, FYN, and LCK are also frequently downregulated but the physiological relevance of these changes for PTCL development remains unclear.
Our data show a relatively high number of recurrent methylation events observed in PTCL correlate with changes in gene transcription.Given that changes in expression affect many genes involved in various aspects of tumor biology, DNA methylation changes likely contribute to PTCL development in a causative way.

Loss of DNA methylation correlates with up-regulation of genes critical for cancer cell proliferation
To further explore whether promoter hypomethylation may affect aspects of T-cell lymphomagenesis, we next focused on the analysis of 39 genes hypomethylated and overexpressed with high frequency in PTCL (Fig. 5A).To determine if any of these genes may play a role in PTCL maintenance, we utilized data from CRISPR knockout screens in the DepMap database (Broad Institute, (39, 40)) and investigated whether any of the genes affected the growth of hematologic cell lines.This search revealed that knockouts of most genes did not signi cantly impact the proliferation of hematological cell lines, as indicated by the gene effect scores ranging from − 0.75 to + 0.75 (Figs.5B and S20).In contrast, the knockout of RACGAP1 and RCC1 genes was lethal to the majority of cell lines, and they were therefore classi ed as Common essential required for proliferation of cells in general.Interestingly, TRIP13 (thyroid hormone receptor interactor 13) characterized by DepMap as Strongly Selective was critical for the maintenance of several cancer cell lines including B-cell lymphoma OCILY19 and Human eosinophilic leukemia EOL1.Because TRIP13 is not a Common essential and therefore its targeting may be less toxic, we, therefore, focused on the analysis of this gene and sought to further explore its role in lymphomagenesis by determining its expression in primary TCL.Our analysis revealed that the gene was overexpressed in 9/10 PTCLs on average by ~ 10-fold relative to controls (Fig. 6A).Analysis of publicly available RNA-seq data revealed overexpression of TRIP13 in ALCL, ATLL, NKTCL, TLBL and NOS (Fig. 6B).
We next performed immunoblot analysis of various cell lines to determine if the TRIP13 protein is present in hematologic cell lines.This analysis revealed that TRIP13 was expressed in T-cell lymphoma (T8ML-1, MJ, HH), T-cell leukemia (JURKAT, MOLT4, MOT, SUPT1, LOUCY, CCRF-CEM, DND-41) B-cell lymphoma (RAJI), B-cell leukemia (MEC-1, MEC-2) and myeloid leukemia (K562, CTV-1) cell lines (Fig. 6C).Altogether, our data show that TRIP13 is overexpressed in primary TCL and expressed in various hematologic cell lines and may be involved in their maintenance.

TRIP13 downregulation inhibits the proliferation of malignant T-cells
The AAA ATPase TRIP13 (thyroid hormone receptor interactor 13) is known to participate in various regulatory steps related to the cell cycle, such as the mitotic spindle assembly checkpoint and meiotic recombination, as well as the DNA repair by immediate-early DNA damage sensing and ATM signaling activation 39 .To determine if TRIP13 is required for the proliferation of PTCL, we transduced a PTCL-NOS cell line, T8ML-1, with lentiviruses expressing shRNAs either targeting scrambled or TRIP13 and coexpressing mCherry.Cells were cultured over twenty days and the percentage of mCherry positive cells was measured at different time points by FACS.The percentage of cells expressing scrambled shRNA remained relatively stable suggesting that expression of scrambled shRNA or mCherry did not affect the proliferation of cells (Figs. 7A and 7B).In contrast, the percentage of TRIP13 shRNA-1 expressing cells was gradually reduced over time suggesting that the proliferation of T8ML-1 cells is impaired by TRIP13 downregulation.(Figs.7A and 7B).The proliferative defect was reproducible and was also seen using independent TRIP13 shRNA-2 (Figs.7C and S21).We next asked whether TRIP13 knockdown can affect the proliferation of T-cell leukemia JURKAT cells.These cells could be routinely transduced to more than 99% e ciency thus allowing for direct cellular and molecular analysis (Fig. S22).As expected, both TRIP13-speci c shRNAs e ciently decreased RNA and protein levels of TRIP13 (Figs. 7D and S23).Similar to T8ML-1 cells, TRIP13 knockdown resulted in impaired proliferation and reduced viability of JURKAT cells as determined by the decreased percentage of cells in "live gate" in FACS analysis and cell counts upon culturing over time (Figs.7E and data not shown).TRIP13 reduction resulted in decreased BrdU incorporation, G2-M arrest, increased Annexin V expression, and apoptosis (Figs.7F-I).In contrast to downregulation, lentivirally mediated overexpression of C-terminally FLAG-tagged TRIP13 CDS from lentiviral backbone co-expressing uorescent protein in T8ML-1 did not affect cell growth of T8ML-1 or JURKAT cells as the percentage of EGFP + cells remained relatively stable over time in both TRIP13 and control cells (data not shown).Altogether, these data suggest that the downregulation of TRIP13 has antiproliferative effects by inducing G2-M arrest accompanied by apoptosis.

Treatment of T8ML-1 cells with TRIP13 inhibitor DCZ0415 impairs proliferation and induces cell death
To test if targeting TRIP13 may have therapeutic potential, we next treated lymphoma cell lines with DCZ0415, a TRIP13-speci c inhibitor 40 .The treatment of T8ML-1 cells with various concentrations of DCZ0415 (10-25 µM) showed dose-dependent cell death with EC50 values of 10.0 µM and 5.5 µM upon 2-and 3-day treatment, respectively (Figs. 8A and S24).The cell treatment was accompanied by the downregulation of endogenous TRIP13 protein suggesting not only that DCZ0415 inhibits this protein but also contributes to its downregulation accompanied by impaired proliferation and cell death (Fig. 8B).When T8ML-1 cells were treated longer, even lower DCZ0415 concentration effectively impaired cellular proliferation.For example, a 14-day treatment of T8ML-1 cells severely reduced viability at 5 µM and 10 µM DCZ0415 concentrations (Fig. 8C).Even at concentrations as low as 1 µM, DCZ0415 reduced cellular viability by 50%.Like with shRNA-mediated TRIP13 downregulation, the treatment with DCZ0415 inhibitor impaired cell proliferation by reducing BrdU incorporation, inducing G2-M arrest, and cell death (Figs.8D  and 8E).Consistently with G2-M arrest, the levels of G2-M proteins -Cdc25A and Cyclin B1 -were elevated upon the drug treatment, whereas G1-S Cyclin D1 was downregulated.
Altogether, our data show that targeting TRIP13 may be bene cial in treating PTCL.

Discussion
The goal of our studies was to better understand the molecular landscape of PTCL with a focus on recurrent molecular changes and identi cation of putative genes driving lymphomagenesis.Using highresolution methylation of seven PTCLs and gene expression pro ling of ten PTCLs coupled with bioinformatic analysis of previously published data and functional studies, we made several interesting observations in this study.
First, hypomethylation of the PTCL genome is more frequent than hypermethylation regardless of the genomic elements or tumor types.These data are consistent with the numerous studies in mouse models that demonstrated that hypomethylation in T-cell malignancies is more frequent than hypermethylation 17,20,41 .Furthermore, methylation patterns in tumors are heterogonous with large differences and only a few common features.Some studies in hematological malignancies found tumor methylation patterns to have little to no cancer speci city.For example, based on the pro ling of Chronic Lymphocytic Leukemia (CLL) it was proposed that tumor patterns are re ections of the methylation status of the cell of origin, rather than being truly cancer-speci c as most previously reported tumor-speci c methylation events are normally present in non-malignant B-cells 42 .Similarly, recently identi ed DNA methylation patterns of ALCL were found to be similar to thymic progenitor cells 24 .Due to the tremendous methylation heterogeneity we observed across PTCLs, the possibility that PTCLs re ect methylation patterns of the cell of origin appears unlikely.Rather, methylation differences between normal and malignant T-cells appeared to be a consequence of at least three molecular events we observed with high-frequencydownregulation of DNMT3A on the transcript level, downregulation of DNMT3B protein, and up-regulation of TCL-1 protein that may biochemically inhibit DNMT functions.These molecular changes are likely responsible for the 'Core PTCL Methylation Signature' consisting of 767 hypo-and 567 hypermethylated DMRs that were observed across various genomic elements in all seven analyzed PTCLs.This signature had no overlap with the 12 gene signature (hyper -BCL11B, CD5, CXCR6, GIMAP7, LTA, SEPT9, UBAC2, UXS1); hypo -ADARB1, NFIC, NR1H3, ST3GAL3) observed in Hepatosplenic T-cell lymphoma 25 .Interestingly, we observed hypermethylation of LCK, previously shown to be hypermethylated and repressed in ALCL and PTCL samples 24 .Although not part of the 'Core PTCL Methylation Signature', we also observed hypermethylation of LEF1 and TCF7, but not BCL11B reported previously in PTCLs 24 .Thus, LCK promoter hypermethylation and to a lesser degree LEF1 and TCF7 belong to the most frequent molecular changes characterizing PTCLs.In addition to methylation signatures, we also identi ed the 'Core PTCL Expression Signature' consisting of 231 genes that were up-and 91 genes that were downregulated in all 10 tested tumors.Not surprisingly, at least 48 genes in this signature were associated with the cell cycle.Additional genes in the signature were related to complement activation, blood coagulation, and Notch pathway signaling.Several genes of the 'Core PTCL Expression Signature' are less characterized but their recurrent deregulation suggests a potential association with tumor development and have the potential to serve as biomarkers.Among them is TEDC2 (also known as C16orf59) whose increased expression is associated with poor prognosis of lung adenocarcinoma 43 .Overexpression of ZWINT predicts poor prognosis and promotes the proliferation of hepatocellular carcinoma 44 .Increased C10orf10 levels (DEPP Autophagy Regulator 1, DEPP1) correlated with the shorter survival time of patients with primary gliomas 45 .
Second, we identi ed recurrent methylation events in promoters that were associated with the changes in gene expression.Namely, 39 genes upregulated in PTCL whose promoters were hypomethylated, and 56 genes repressed in tumors whose promoters were hypermethylated.Importantly, several oncogenes and tumor suppressor genes were identi ed in these signatures.For example, the hypomethylated and overexpressed gene UHRF1 has oncogenic functions promoting tumorigenesis through the silencing of DNA repair genes and inhibiting apoptosis in various cancers 32 .UHRF1 is overexpressed in T-cell ALL and its knockdown reduces c-MYC expression and viability in these malignancies 33 .Other genes hypomethylated and overexpressed in PTCL -RAB13, MPZL1, CDK14 -were implicated in tumor progression of several different tumors including B-cell lymphomas, and glioblastomas, lung and ovarian 34,35,46,47 .Among genes, we detected to be hypermethylated and silenced in PTCL are major tumor suppressors, e.g.ATM, which plays a critical role in DNA repair and whose functions are often compromised in PTCL through acquired genetic alterations 48 .Another hypermethylated and silenced gene in PTCL -serine/threonine-protein kinase 4 (STK4) -regulates cell differentiation and apoptosis and is the tumor suppressor in hepatocellular carcinoma, breast cancer, and lymphoma 49 .We also detected somewhat puzzling hypermethylation and silencing of genes that would be predicted to be oncogenes, rather than tumor suppressors, such as SF3B1, LCK, FOXP1, FYN 50,51 .Given the complexities of gene functions and the large-scale-deregulation of signaling networks in cancer, it is not surprising that some genes normally promoting tumorigenesis would be silenced.Alternatively, their effects on tumorigenesis would be context-dependent.For example, loss of proto-oncogene LCK accelerates CLL development in mice suggesting the tumor suppressor role for this gene 52 .The speci c roles of genes whose expression changes coincided with methylation alterations remains to be worked out.However, our data strongly suggests that both DNA hypo-and hypermethylation affect the expression of genes that are important in various aspects of T-cell tumorigenesis and at least some of them are likely to contribute to PTCL development.
Third, we found that genetic and pharmacological inhibition of TRIP13 impaired the proliferation of T-cell malignant cell lines by inducing G2-M arrest and apoptosis.Interestingly, we identi ed mouse TRIP13 to be hypomethylated and overexpressed in mouse T-cell lymphomas 21 .Our ndings collectively underscore the critical role of DNA methylation in regulating TRIP13, as well as its signi cant involvement in the development and/or maintenance of TCL.This positions TRIP13 as a potential key oncogene in PTCL.Reinforcing this hypothesis, similar conclusions have been drawn from research conducted on various solid tumors, where the in uence of TRIP13 on cancer progression has been similarly observed.
For example, TRIP13 is highly expressed in Glioblastoma and its inhibition impaired the proliferation, migration, and invasion of tumor cells by regulating c-MYC stability 53 .TRIP13 promotes metastasis of colorectal cancer and its inhibition decreased cell proliferation in vitro and tumor formation in vivo of colorectal carcinoma cell lines 40,54 .
Our functional assays demonstrate that promoter hypomethylation is important even in human lymphomagenesis, at least in its maintenance.Whether it plays a role in initiation and progression remains to be tested.
It's important to note certain limitations in our study.Although our use of WGBS allowed for a more comprehensive analysis of the methylation landscape of PTCLs than previous studies 24,25 , our research was limited to seven samples.This limitation might impact the generalizability of our ndings in the broader context of PTCL research.Furthermore, while we performed RNA-seq gene-expression analysis on 10 PTCL samples, only seven of those lymphomas were pro led for DNA methylation alterations due to the unavailability of DNA.Altogether, this limited our ability to address tumor-speci c differences between ALCL (T1-4), AITL (T5), and PTCL-NOS (T5-10).Thus, by and large, we did not attempt to dissect the molecular differences within tumor subtypes as such analysis would result in a large number of falsepositive and negative alterations and therefore provide substantially skewed views of molecular landscapes.Rather, our focus was on methylation and gene expression landscapes of PTCLs analyzed as one single tumor entity in an attempt to identify potential molecular events involved in the pathogenesis of these malignancies.In one analysis attempting to classify PTCL-NOS into TBX21 or GATA3 subgroups, we were unable to do so either using their gene expression values or readout of their target genes.While this might be due to the limited sample size, it is also possible that such classi cation of PTCL-NOS, which was primarily based on microarray results, may not be easily seen by RNA-seq analysis or be less pronounced than previously thought.
Our future study will also address the relationships of methylation and gene expression changes observed in this study.However, given the large number of genes with demonstrated biological effects in T-cell biology, our study clearly points to a causative relationship of deregulated methylation in PTCL development.

Clinical samples and data sources
The study included in total two normal human puri ed CD4 + T-cells (84200-1.0/13122)and normal human puri ed CD8 + T-cells (84300-1.0/13083)obtained from Precision Medicine.These samples were used for methylation analysis by WGBS and gene expression analysis by RNA-seq.We also used seven frozen PTCL tissue samples and one normal lymph node sample obtained from the NCI Cooperative Human Tissue Network (CHTN).Other investigators may have received samples from the same tissue specimens.PTCL samples were used for the isolation of DNA (for WGBS), RNA (RNA-seq and real-time qRT-PCR), and proteins (immunoblots).The frozen lymph node was used for protein isolation and served as a control for immunoblots.Additional three total RNA samples isolated from PTCL tumors and one paired DNA sample were acquired from Origene (Rockville, MD, USA).Methylation and gene expression analyses were done on paired DNA-RNA samples except for two RNA samples for which DNA or tissue was not available.The Institutional Review Board of the University of Nebraska Medical Center approved the use of these samples.The sample sets are described in this section, in Figures S1 and S3.For data analysis, we utilized several data sets available online.For the methylation analysis -in addition to normal human T-cell controls generated in this study -we also used publicly available WGBS data on CD3 + and CD4 + T-cells obtained from GEO [GSM1186660, GSM2103005, GSM2103006, GSM2103007] 55 .For gene expression analysis publicly available data for human ALCL (SRA accession numbers Cells were maintained in DMEM or RPMI 1640 (Invitrogen) containing 10% fetal bovine serum.Cell lines were cultured at 37°C in a humidi ed 5% CO2 atmosphere and were passaged according to recommendations.
To generate lentiviruses, 293T cells were seeded in a 100 mm tissue culture plate to obtain ~ 80-90% con uence and transfected with plasmid DNA and packaging plasmids psPAX2 and pMD2.G at a ratio 1:0.65:0.35using 70 µg of polyethylenimine (PEI) (Polysciences).Viruses were collected 48-96 h posttransfection.Transduction was performed as described previously 17 using Polybrene Infection / Transfection Reagent (Sigma).The e ciency of transduction was determined by measuring the percentage of mCherry-positive cells by FACS.

FACS, BrdU incorporation, and apoptosis assays
For shRNA experiments, the growth of T8ML-1 or JURKAT cells transduced with lentiviruses was monitored by FACS over 2-20 days.The maximum percentage of mCherry-expressing cells was typically observed 72 hours post-transduction.In some experiments, all subsequent mCherry data points were normalized to this time point and expressed as a relative percentage of the initial time point.EGFP was measured periodically by ow cytometry on the LSRII available at the University of Nebraska ow cytometry core facility.In JURKAT cells, where we achieved + 99% transduction e ciency, the growth of cells was evaluated by Trypan blue staining and cell counting at different time points.Viability was also evaluated by scoring percentages of cells in the "Live gate" of FACS diagrams obtained using cells at different timepoints.For in vitro BrdU labeling, BrdU at a concentration of 0.01 mM was added 110 min before harvests of cells.BrdU-positive cells were quanti ed using anti-BrdU-FITC (BrdU-Flow Kit, BD Biosciences PharMingen) as described by the manufacturer.For cell cycle analysis, 7-ADD was added to the samples.For analysis of apoptosis, cells were stained with allophycocyanin-conjugated annexin V antibody and analyzed by FACS according to the manufacturer's recommendations (eBioscience).FACS analysis was done using the BD Accuri C6 Plus Flow Cytometer.

WGBS and bioinformatics analysis
Genomic DNA from PTCL samples was isolated using QIAamp DNA Mini Kit (Qiagen).
The WGBS libraries using DNA from normal human CD4 + and CD8 + puri ed T-cells and seven PTCLs were prepared and sequenced on an Illumina NovaSeq6000 sequencer using 150 bp long paired-end reads by Novogene, USA.Quality checks, trimming, ltering, alignment of reads to the UCSC hg38 reference genome, and methylation calling were performed with Bismark software 58 .Only CpG sites with a minimum sequencing depth 5x were included in the analysis.Differential methylation was calculated by comparison of individual tumors to the control consisting of methylation data obtained from ve normal human T-cells -CD3, CD4_1, CD4_2, CD4_3, and CD8 -averaged out by Metilene.
Differentially Methylated Regions (DMRs) were determined by Metilene 59 and de ned based on the average of a minimum of three consecutive DMCs with methylation change of ≥ 10% in the same direction with p values < 0.05 (as determined by MWU test).The maximal base pair cut-off for a distance between consecutive DMCs in DMR was set to 100 bp.Annotation of methylated CpGs and DMRs to promoters, gene bodies, enhancers, and repeats was performed using bedtools intersect.Chromosomal coordinates of TSS, gene bodies, and repeats were acquired from the USCS Table browser.Enhancer coordinates identi ed in CD4 + CD8 + cells and thymus cells were obtained from Enhancer atlas 60 .
Promoter was de ned as 1,500 bp upstream to 500 bp downstream of the TSS.
Methylation scores were visualized with the Integrated Genome Browser (IGB) 61 .Scatter plots of methylation score were generated in RStudio v1.1.4.6 using package gplots.Genome-wide Pearson correlation analysis of CpG sites was performed using deepTools package multi bigwig summary and plot Correlation 62 .

Immunoblotting
Frozen tissue was cut on a glass plate on dry ice and 20-50 mg of tissue was placed into a round bottom Eppendorf tube.20 µL/mg of ice-cold RIPA buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% NP-40, 1% Na-deoxycholate, 0.1% SDS, sterile-ltered) along with protease inhibitor cocktail and sodium orthovanadate was added and tissue was homogenized with an electric homogenizer on ice.
RNA isolation, RNA-seq, and bioinformatics analysis Total RNA from frozen tumor tissues was isolated using the TRIzol reagent (Invitrogen) and repuri ed using an RNAeasy kit (Qiagen).Library generation and sequencing on NovaSeq 6000 platform using paired-end 150 bp runs was performed by Novogene, USA.Tissue samples from seven Peripheral T-cell lymphomas were provided by the NCI Cooperative Human Tissue Network (CHTN).Other investigators may have received samples from the same tissue specimens.Three total RNA samples isolated from PTCL tumors were acquired from Origene (Rockville, MD, USA).The Institutional Review Board of the University of Nebraska Medical Center approved the use of these samples.The sample description is summarized in Supplementary Fig. 2.
Publicly available data for human ALCL, ATLL, NKTCL, and TLBL were obtained from GEO 55 and SRR (accession numbers listed in Supporting Information 7).Trimmed sequencing data were rst aligned to Homo sapiens UCSC hg38 reference genome using STAR aligner.RNA-seq data with minimum mapped quality 50 were quanti ed using the RNA-seq quanti cation pipeline in SeqMonk software (http://www.bioinformatics.babraham.ac.uk/projects/seqmonk/).DESeq2 was used to calculate differential expression.For differentially expressed genes, only genes with a fold change ≥ 1.5 and a p value < 0.05 were considered signi cant.Ingenuity pathway analysis (Qiagen) 63  TRIP13 drug treatment, cell counting, and molecular assays DCZ0415, an inhibitor of TRIP13, was obtained from (MedChem, Cat.#: HY-130603, Monmouth Junction, NJ, USA) and dissolved in DMSO.T8ML-1 cells were subjected to treatments with either DMSO (serving as the vehicle control) or the DCZ0415 at concentrations ranging from 1 to 25 µM per 1 mL of culture.
The cell viability was assessed using Trypan Blue exclusion dye cell counts at different time points after the drug addition.EC50 values were determined by assessing the drug concentration that corresponds to 50% cell viability and were analyzed by AAT Bioquest Software.For the analysis of TRIP13 protein levels upon the treatment, 1 x 10 8 T8ML-1 cells were treated with DMSO or 20 µM DCZ0415 per 1 mL of culture, and protein lysates were made similarly as described above.BrdU assays and apoptosis were done as described above.

Real-time qRT-PCR
Two micrograms of RNA were reverse transcribed with the SuperScript III Reverse transcriptase (Thermo Fisher) using oligo (dT) primers.Real-time qRT-PCR was performed with the iQ™ SYBR® Green Supermix (Bio-Rad) on a CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad).Fast PCR cycling conditions were used (95°C for 3 min, 40 cycles (95°C for 10 s, 58-63.5°Cfor 30 s)), followed by a dissociation curve analysis.All qPCR measurements were performed in duplicate reactions and normalized to the expression of the housekeeping gene (β-actin).In parallel, no-RT controls were ampli ed to rule out the presence of contaminating genomic DNA (Supporting Information 8).

Statistical analysis
The statistical signi cance of means ± SEM was evaluated using the two-tailed Student's t-test.For all statistical analyses p values < 0.05 were considered signi cant.
Abbreviations and nomenclature A scale is shown below the heat map, in which yellow and blue correspond to a lower and higher methylation status, respectively.The corresponding expression is shown as a heat map with highly expressed genes denoted in red and lowly expressed genes denoted in green.
(F) Analysis of promoter methylation and gene expression in normal human T-cells for 19,253 genes.
Genes were divided into four groups based on the percentage of promoter methylation (0%-25%, 26%-50%, 51%-75%, and 76%-100% The treatment of T8ML-1 cells with DCZ0415 inhibits cellular growth by inducing G2-M arrest and cell death was used to analyze activated and decreased core signaling pathways for differentially expressed genes.Activated and inhibited pathways (Z-score > 1.5, p < 0.05) identi ed in individual PTCLs are shown in Fig. 3B.The top subcategories obtained in Physiological System, Development, and Functions are displayed (p < 0.05, for all subcategories).

Figure 2 DNA
Figure 2

Figure 7 TRIP13
Figure 7 Breakdown of promoter methylation for 34,858 genes in CD4+ and CD8+ normal human T-cell samples.Methylation percentages for all CpGs across the 2,000 bp region (−1,500 bp to +500 bp relative to TSS) were averaged to give a mean methylation value for each gene promoter.Promoters were placed into four categories based on percent methylation (0-25%, 26-50%, 51-75%, and 76-100%).(C)Methylation status of 34,858 promoters in CD4+ and CD8+ samples determined by WGBS.Mean promoter methylation was determined as described in B. A scale is shown below the heat map, in which yellow and blue correspond to a lower and higher methylation status, respectively.