Graphical abstract

   graphic file with name fx1.jpg
   [45]Open in a new tab

Highlights

     * •
       PTCL molecular subtypes reveal distinct genetic and transcriptional
       features
     * •
       Four molecular subtypes exhibit potential vulnerabilities to
       targeted agents
     * •
       Four microenvironment subtypes reflect different immune communities
     __________________________________________________________________

   Huang et al. integrate the genomic and transcriptomic data from 221
   patients of peripheral T cell lymphoma to establish 4 molecular and
   microenvironment subtypes with distinct features and therapeutic
   responses. These findings suggest potential precision medicine
   approaches that target both tumors and the microenvironment in this
   hard-to-treat disease.

Introduction

   Peripheral T cell lymphoma (PTCL), comprising ∼15%–20% of all
   aggressive non-Hodgkin’s lymphomas, is characterized by malignant
   proliferation of post-thymic lymphocytes and heterogeneous
   clinicopathological manifestations.[46]^1 The most frequent subtypes
   include PTCL not otherwise specified (PTCL-NOS; ∼35%),
   angioimmunoblastic T cell lymphoma (AITL; ∼15%–42%), and anaplastic
   large cell lymphoma (ALCL; ∼15%), which can be further divided into
   anaplastic lymphoma kinase (ALK) fusion-positive ALCL (ALK+ ALCL) and
   ALK fusion-negative ALCL (ALK− ALCL).[47]^2^,[48]^3 Except for ALK+
   ALCL, PTCL exhibits a poor response to standard anthracycline-based
   cyclophosphamide, doxorubicin, vincristine, and prednisolone (CHOP)
   chemotherapy, resulting in a 5-year survival of less than 30% for most
   pathological subtypes.[49]^4 Consequently, effective targeted therapies
   are an urgent need for this lethal disease, which has long been impeded
   by the dearth of knowledge of the genetic landscape of PTCL.

   Previous studies have suggested a complex pathobiology underlying the
   heterogeneity of PTCL.[50]^3 Various intrinsic culprits dysregulate
   essential pathways in mature T cells, thus promoting malignant
   transformation.[51]^5 Frequent gene mutations associated with epigenome
   (e.g., TET2, DNMT3A, and IDH2) and T cell receptor (TCR) signaling
   (e.g., RHOA) have been detected in AITL, accompanied by activation of
   the TCR and phosphatidylinositol 3-kinase (PI3K)-AKT
   pathways.[52]^6^,[53]^7^,[54]^8^,[55]^9 In PTCL-NOS, histone-modifying
   genes, mainly on methylation and acetylation, are commonly
   observed.[56]^10 ALK+ ALCL is defined by the NPM1-ALK fusion protein,
   along with an increased expression of genes related to
   hypoxia-inducible factor alpha (HIF1-α) and interleukin-10
   (IL-10),[57]^11 while ALK− ALCL represents a pathologically distinct
   but genetically heterogeneous group of disease, all expressing three
   genes: TNFRSF8, BATF3, and TMOD1.[58]^12 As for the cell of origin
   (COO), gene expression profiling disclosed a significant enrichment of
   the T follicular helper (Tfh) cell signature in AITL.[59]^13 In
   PTCL-NOS, two major molecular subgroups have been delineated,
   PTCL-TBX21 and PTCL-GATA3, characterized by high expression of Th1 and
   Th2 cell differentiation regulators, respectively.[60]^11 More
   recently, a Tfh-originating PTCL-NOS subtype has been identified that
   is highly similar to AITL[61]^5 and harbors frequent Tfh-relevant
   mutations.[62]^14^,[63]^15

   In addition to intrinsic defects, growing evidence suggests that
   extrinsic stimuli by cellular and non-cellular components in the
   lymphoma microenvironment (LME) are highly paramount for tumor cell
   behaviors and clinical manifestations in PTCL.[64]^3 In AITL, gene
   expression profiling unveiled an enrichment of B cells and plasma cells
   as well as a complex cytokine milieu involved in various immune
   processes, indicating a complicated immunologic
   network.[65]^11^,[66]^16 High expression of IL-4, IL-6, and IL-21 is
   associated with the expansion of B cells and plasma cells, likely
   leading to hypergammaglobulinemia and autoimmune phenomena.
   Accumulation of tumor-associated macrophages and overexpression
   ofvascular endothelial growth factor (VEGF), transforming growth factor
   β (TGF-β), and IL-10 are associated with immunosuppression and frequent
   Epstein-Barr virus (EBV) infection. Clinically, B cell association is
   significantly correlated with favorable outcomes, while high VEGF
   expression is relevant to tumor progression in AITL.[67]^17 In
   PTCL-NOS, TBX21 expression presents a Th1-driven subset, which features
   abundant immune cell infiltration, especially macrophage polarization,
   thereby accounting for unfavorable outcomes.[68]^18^,[69]^19^,[70]^20
   In contrast, a minimally inflamed microenvironment was detected in the
   Th2-originating PTCL-GATA3 subset, which has an even worse
   prognosis.[71]^20 Another study revealed a microenvironment
   classification in PTCL-NOS based on gene expression signatures of
   tumor-infiltrating immune cells, in which the subtype exhibiting both B
   cell and dentritic cell (DC) signatures shows better prognosis, while
   the subtype with neither signature represents the worst.[72]^21
   Elevated expression of immune checkpoint genes, including PD-L1 and
   IDO1, was observed in macrophage-enriched cases, implying potential use
   of immune checkpoint inhibitors (ICIs) in this subtype.

   Despite the previous efforts, a lack of large-scale RNA sequencing
   (RNA-seq) in PTCL put limitations on the discovery of potential
   biomarkers for genetic subtyping and targeted therapies. In this study,
   by integrating whole-exome sequencing (WES), targeted sequencing, and
   RNA-seq on 221 PTCL patients, we aimed to portray a comprehensive
   genomic landscape of mutations as well as establish a molecular
   stratification based on genetic signatures, biological alterations, and
   therapeutic responses, proposing four distinct molecular subgroups of
   PTCL. Additionally, we defined a microenvironment stratification
   derived from gene expression footprints of infiltrating immune cells
   and stromal components, which also reflected distinct biological
   features and clinical properties. This is a multiomics study exploiting
   RNA-seq data on a large cohort of PTCL, which revealed a series of
   cancerous as well as microenvironment biomarkers, providing a rationale
   for personalized treatment of PTCL patients.

Results

Unique molecular subtypes of PTCL defined by genetic landscape

   Based on WES data of 101 PTCL patients, we detected a total of 36,625
   somatic mutations. Among all mutation genes, 82 candidate genes (SNVs
   and insertions or deletions [indels]) with predicted functional
   alterations in tumorigenesis were selected to initiate a gene panel for
   deep targeted capture sequencing in an additional cohort of 120 PTCL
   patients. With a combined overview of the results from WES and targeted
   sequencing, 1,412 non-silent somatic mutations were discovered in 82
   genes, which included 1,065 missense, 96 nonsense, 229 indel, and 22
   splice-site mutations, showing a significantly high frequency of
   missense mutations ([73]Figure S1A). Among these 82 genes, an average
   of 6 mutations were detected in each sample, ranging from 1 to
   20 ([74]Figure S1B). SNVs in PTCL were dominated by C>T and T>C
   transitions, analogous to the somatic SNV spectrum in other cancers
   ([75]Figure S1C; [76]Table S1). Besides, copy number aberrations (CNAs)
   analysis was conducted in WES samples, and the results showed that copy
   number gains/amplifications or losses/deletions were relatively common
   ([77]Figures S1D and S1E). Consistent with previous studies, the loss
   of significant genes, including TP53, CDKN2A, and PTEN, was frequently
   observed ([78]Figure S1F).[79]^5^,[80]^16

   With the comprehensive list of mutations in PTCL, 40 genes were
   designated as recurrently mutated, as they were detected in more than
   5% of all cases ([81]Figure 1A). The most frequently mutated genes
   included TET2 (49%), RHOA (27%), KMT2C (20%), DNMT3A (16%), and KMT2D
   (12%). Involved pathways included DNA methylation (e.g., TET2, DNMT3A,
   and IDH2), TCR signaling (e.g., RHOA, CARD11, and FYN), PI3K-AKT
   signaling (e.g., VAV1, PI3KR1, and ITPR3), histone modification (e.g.,
   KMT2C, KMT2D, CREBBP, and EP300), p53 signaling (e.g., TP53, ATM, and
   JMY), Janus kinase-signal transducer and activator of transcription
   (JAK-STAT) signaling (e.g., PTPN13 and JAK3), immune surveillance
   (e.g., HLA-A and HLA-B), chromatin remodeling (e.g., ARID1A and
   ARID1B), NOTCH signaling (e.g., NOTCH1 and NOTCH3), and tumor
   suppression (e.g., MGA and CIC). Of note, all patients were covered by
   mutations of these 10 categories, indicating a complicated pathogenic
   role of intrinsic defects in PTCL ([82]Table S2). Analysis of pairwise
   relationships across the recurrent genetic lesions uncovered a series
   of co-occurring and mutually exclusive mutations ([83]Figure S1G). We
   discovered strong co-alteration tendencies in Tfh-related genes,
   including TET2, RHOA, DNMT3A, and IDH2, as well as in the
   immune-related genes PTPN13 and HLA-B. KMT2C and KMT2D mutations were
   found mutually exclusive of the aforementioned alterations, providing a
   hint about subtype-defining genes for categorization. We then performed
   principal-component analysis (PCA) of the mutation data and identified
   3 major molecular subtypes in PTCL associated with genetic alterations,
   including T1 (60 cases, characterized by TET2 and RHOA mutations), T2
   (49 cases, characterized by TET2 without RHOA mutations), and T3 (112
   cases, characterized by other mutations with wild-type TET2 and RHOA)
   ([84]Figure 1B). Moreover, AITL and PTCL-NOS patients could be
   separated into 3 subtypes by PCA ([85]Figures S2A and S2B). The number
   of ALK+ ALCL and ALK− ALCL patients was limited to be further
   subclassified.

Figure 1.

   [86]Figure 1
   [87]Open in a new tab

   Genomic landscape and mutational features of PTCL (n = 221)

   (A) Landscape of somatic mutations detected by WES (n = 101) and
   targeted sequencing (n = 120) in 221 patients with PTCL. Mutations in
   82 candidate genes (SNVs and indels) with predicted functional
   alterations in tumorigenesis were ranked by prevalence in all cases.
   The numbers of mutational burdens in each patient are indicated at the
   top, and the prevalence of each mutation is depicted on the right.
   Pathological subtypes and IHC makers (BCL6, CD10, PD-1, and CXCL13) are
   also shown.

   (B) PCA of the genetic data and projection of representative clusters.
   Projection of mutational status along with the first two collective
   principal components (PC1 and PC2) divided 221 patients into 3
   molecular subtypes. The individual cases are colored by subtype, and a
   95% confidence ellipse is depicted to define each subtype.

   (C) Evolutionary tree plot of the unsupervised hierarchical clustering
   within the T3 subtype, revealing 2 subtypes: T3.1 and T3.2.

   (D) Bar chart overview of the defining alterations significantly
   different between T3.1 and T3.2.

   (E) Frequencies of common genetic mutations across 4 molecular
   subtypes.

   The T1 subtype has been extensively investigated due to its distinctive
   co-existence of TET2 and RHOA mutations. In addition, we explored an
   increased frequency of other mutations regarding DNA methylation and
   the TCR signaling pathway, such as DNMT3A, IDH2, and CARD11. On the
   other hand, the T2 subtype was found to bear frequent mutations in the
   PI3K-AKT signaling pathway, such as TSC2 and VAV1. The largest PTCL
   subtype T3, accounting for 51% of our cohort, featured heterogeneous
   genetic alterations that participated in a complex of biological
   functions. Subsequently, further stratification was carried out,
   utilizing unsupervised hierarchical clustering, and 2 subtypes, T3.1
   (71 cases) and T3.2 (41 cases), were discovered according to their
   discrete genetic traits and gene functions ([88]Figure 1C). To explore
   the defining alterations that distinguished T3.1 and T3.2, we compared
   the mutational loads between the 2 subtypes ([89]Figure 1D). T3.1
   presented a significant enrichment in mutations involving histone
   modification, including methylation (KMT2C, 31%; KMT2D, 21%; and KDM6B,
   17%) and acetylation (YEATS2, 13%, and CREBBP, 7%). Similarly, compared
   with T1 and T2, T3.1 also showed distinctively high frequencies in
   these mutations, thereby determining its mutational feature as histone
   modification alterations. T3.2 favored immune-related mutations, such
   as PTPN13 (17%), HLA-B (12%), and HLA-A (7%), within both T3 and the
   whole cohort ([90]Figure 1E). In addition, some relatively common gene
   mutations were evenly distributed in each subtype, especially ARID1A
   and ARID1B (switch defective/sucrose non-fermentable chromatin
   remodeling complex, SWI/SNF). It has been reported that ARID1A and
   ARID1B participated in multiple biological processes, involving
   methyltransferase activity, the PI3K-AKT pathway, p53 signaling, and
   immune responsiveness, which may play a general role in the
   pathogenesis of different PTCL subtypes.[91]^22

   In summary, PTCL was classified into 4 subgroups according to the
   mutational status of RHOA, TET2, histone-modifying, and immune-related
   genes ([92]Figures 1E and [93]S2C).

PTCL molecular subtypes associated with distinct clinicopathologic
manifestations

   Molecular subtypes differed significantly in pathological distribution
   ([94]Figure 2A). It is well known that TET2, RHOA, DNMT3A, and IDH2,
   which are recognized as Tfh-associated genetic alterations, are
   frequently altered in Tfh-originating AITL and PTCL-Tfh entities, and
   affect 73%, 50%, 24%, and 18% of AITL in our cohort, respectively
   ([95]Figure S3A).[96]^1^,[97]^23 Accordingly, T1 and T2 were mostly
   composed of AITL, taking up 85% and 49% of the subtype, respectively,
   followed by PTCL-NOS (13% in T1 and 37% in T2). T3.1 was heterogeneous
   and included AITL, PTCL-NOS, ALK+ ALCL, and ALK− ALCL cases, accounting
   for 23%, 34%, 23%, and 21%, respectively. That is, histone-modifying
   gene alterations were frequently discovered regardless of pathological
   classification. T3.2 also represented a subgroup with pathological
   heterogeneity, thus indicating an essential role of immune-related
   mutations across different pathological subtypes.

Figure 2.

   [98]Figure 2
   [99]Open in a new tab

   Clinical features of the molecular subtypes in PTCL (n = 221)

   (A) Sankey plot of the distribution of pathological categories within
   each molecular subtype (left) and correlations between molecular
   subtypes and pathological categories (right).

   (B) IPI (0–2 vs. 3–5) according to molecular subtypes.

   (C) EBER (positive vs. negative) by in situ hybridization according to
   molecular subtypes (n = 171).

   (D) Kaplan-Meier curves of progression-free survival (PFS) and overall
   survival (OS) according to molecular subtypes.

   The clinicopathological characteristics of the patients are listed in
   [100]Table S3. Among 4 pathological subtypes, ALK+ ALCL patients showed
   the highest percentage of low-risk international prognostic index (IPI)
   ([101]Figure S3B). It is worth noting that T1 presented the lowest
   percentage of low-risk IPI patients ([102]Figure 2B). As for
   Epstein-Barr virus-encoded RNA (EBER) expression, EBV infection was
   detected in 67% of AITL and 38% of PTCL-NOS, consistent with previous
   studies ([103]Figure S3B).[104]^24 Interestingly, the T1 subtype
   exhibited a remarkably higher proportion of EBER positivity, implying a
   potential association of TET2 and RHOA mutations with EBV infection in
   PTCL ([105]Figure 2C). With a median follow-up of 43.5 months (range,
   0.4–123.0 months), ALK+ ALCL patients had better progression-free
   survival (PFS) and overall survival (OS) than those with PTCL-NOS,
   AITL, and ALK− ALCL, while the other 3 subtypes shared a similar
   unfavorable prognosis ([106]Figure S3C). Besides, we segregated the 71
   PTCL-NOS patients into PTCL-Tfh, PTCL-GATA3, PTCL-TBX21, and
   unclassified[107]^15 ([108]Figure S3D). Most patients of T1 and T2
   subtypes were defined as PTCL-Tfh. Consistent with previous
   studies,[109]^11^,[110]^15 PTCL-GATA3 patients present with the worst
   prognosis compared with others ([111]Figure S3E). According to
   molecular subtypes, all 4 molecular subtypes showed inferior PFS and
   OS. The predicted 3-year OS rate for the T1, T2, T3.1, and T3.2
   subtypes was 41.3%, 55.5%, 54.8%, and 41.8%, respectively
   ([112]Figure 2D).

PTCL molecular subtypes associated with divergent transcriptional signatures

   Next, we investigated RNA-seq data to profile biological features in
   different molecular subtypes. We performed differential gene expression
   (DGE) analysis, contrasting each subtype to all others, and revealed
   that the 4 molecular subtypes differed in gene expression pattern
   ([113]Figure 3A). As revealed by gene set enrichment analysis (GSEA)
   ([114]Figure 3B), the TCR signaling pathway was the most enriched gene
   set in T1, followed by the PI3K-AKT, chemokine, JAK-STAT, ERBB, Hippo,
   VEGF, mitogen-activated protein kinase (MAPK), and mammalian target of
   rapamycin (MTOR) signaling pathways. In T2, the PI3K-AKT signaling
   pathway was most significantly activated, with the TCR and MAPK
   signaling pathways slightly enriched. Conversely, T3.1 was involved in
   other cellular processes, including cell cycle checkpoints, DNA
   replication, and translation, while T3.2 featured NOTCH-, interferon
   (IFN)-α/β-, and VEGF-related genes. The signaling pathways enriched in
   each subtype and the corresponding genomic alterations are summarized
   in [115]Figures 3C and 3D. T1 was characterized by enrichment of the
   TCR signaling pathway, in accordance with RHOA mutation. In comparison,
   T2 exhibited a peculiar enrichment in the PI3K-AKT signaling pathway
   both transcriptionally and genetically, with minor enrichment in the
   TCR signaling pathway, probably due to the absence of RHOA mutation. On
   the other hand, T3.1 harbored genetic alterations involving the histone
   modification, chromatin modeling, tumor suppressor, and p53 pathways,
   consistent with enrichment of cognate pathways, including translation
   and cell cycle checkpoints. Finally, T3.2 exhibited a genetic profile
   of immune surveillance defects, accompanied by enrichment of the
   IFN-α/β and VEGF signaling pathways.

Figure 3.

   [116]Figure 3
   [117]Open in a new tab

   PTCL molecular subtypes associated with divergent transcriptomic
   signatures

   (A) Heatmap showing gene expression profiles across 4 molecular
   subtypes.

   (B) Dot plot representing hallmarks of enrichment scores for each
   molecular subtype compared with all other patients by gene set
   enrichment analysis (GSEA). Yellow represents upregulated pathways and
   blue downregulated pathways.

   (C) Density distribution of oncogenic pathways according to molecular
   subtypes. The x axis shows the rank metric score, and the y axis shows
   the running enrichment score by GSEA.

   (D) Driver mutations affecting functional pathways of PTCL are selected
   and summarized for comparison between T1, T2, T3.1, and T3.2. Each gene
   box includes the mutation frequency and regulatory status of activation
   or disruption.

Synergy of targeted agents and exploration of potential monotherapy in T1 and
T2 subtypes

   To better understand the features of the T1 and T2 subtypes, both
   harboring TET2 mutations, we compared gene expression with respect to
   oncogenic pathways between the 2 subtypes. As expected, the T1 subtype
   showed enrichment of the TCR signaling pathway as well as chemokines,
   NOTCH, VEGF, and negative regulation of PI3K-AKT compared with the T2
   subtype ([118]Figure 4A). As the TCR and PI3K-AKT signaling pathways
   may be potential therapeutic targets for the T1 and T2 subtypes,
   respectively ([119]Figure 4B), we explored the combination of the
   hypomethylating agent 5-azacytidine, which can target TET2
   mutation,[120]^25 with the multikinase inhibitor dasatinib, which can
   inhibit TCR signaling,[121]^26 in the T1 subtype and with the PI3K
   inhibitor duvelisib in the T2 subtype. In vitro, TET2^mutRHOA^mut
   Jurkat and H9 cells were treated with different concentrations of
   5-azacytidine and dasatinib for 48 h. The majority of the data points
   were found in the area <1 on the combination index (CI) curve, denoting
   synergistic interactions ([122]Figures 4C and [123]S4A). To clarify the
   molecular basis for synergy, RNA-seq of TET2^mutRHOA^mut Jurkat cells,
   treated with 5-azacytidine and dasatinib alone or in combination, was
   performed. The Venn diagram identified a specific pattern of 1,551
   genes altered by the combination group but not by the untreated or the
   single-agent group ([124]Figure S4B). Combined treatment led to
   significant downregulation of multiple signaling pathways associated
   with cancer, especially the TCR signaling pathway ([125]Figures 4D and
   [126]S4C). Furthermore, preclinical zebrafish patient-derived xenograft
   (zPDX) models bearing TET2 and RHOA mutations present longer survival
   in the combination group than the untreated and the single-agent group
   ([127]Figure 4E). Meanwhile, TET2^mutRHOA^wt Jurkat and H9 cells were
   treated with different concentrations of 5-azacytidine and duvelisib
   for 48 h. A synergistic inhibitory effect was found in the combination
   group ([128]Figures 4F and [129]S4D), accompanied by a specific pattern
   of 1,487 genes, as well as downregulation of PI3K-AKT and inactivation
   of IRF7 through the TLR signaling pathways ([130]Figures 4G, [131]S4E,
   and S4F). In zPDX models, combined treatment also resulted in increased
   survival of those carrying TET2 mutations ([132]Figure 4H).

Figure 4.

   [133]Figure 4
   [134]Open in a new tab

   Synergy of targeted agents and exploration of potential monotherapy in
   T1 and T2 subtypes

   (A) Pathway enrichment analysis in T1 and T2 according to the Kyoto
   Encyclopedia of Genes and Genomes (KEGG) and Reactome databases.

   (B) GSEA enriched differentially expressed genes of TCR signaling
   pathway and negative regulation of the PI3K-AKT network in T1 compared
   with T2.

   (C) Combination index (CI) curve calculated by Compusyn software in
   TET2^mutRHOA^mut Jurkat cells treated with 5-azacytidine (Aza) and
   dasatinib (Dasa) for 48 h.

   (D) Pathway analysis of differentially downregulated genes in each
   group upon treatment with Aza and/or Dasa.

   (E) Survival of zebrafish patient-derived xenografts (zPDXs) upon Aza
   and/or Dasa treatment alone or in combination (n = 11 for each group).

   (F) CI curve calculated by Compusyn software in TET2^mut Jurkat cells
   treated with Aza and duvelisib (Duv).

   (G) Pathway analysis of differentially downregulated genes in each
   group upon treatment with Aza and/or Duv.

   (H) Survival of zPDX upon Aza and/or Duv treatment alone or in
   combination (n = 11 for each group).

   (I) The ratios of cell viability of TET2^mutRHOA^mut Jurkat cells and
   TET2^WTRHOA^WT cells after treatment with 7,316 compounds at 0.4 μM for
   72 h (left). Compounds that specially suppressed the growth of
   TET2^mutRHOA^mut Jurkat cells are listed on the right panel.

   (J) Cell viability of TET2^mutRHOA^mut Jurkat cells after treatment
   with different concentrations of KPT-9274 for 48 h. ∗p < 0.05 compared
   with TET2^WTRHOA^WT cells. Data are presented as the mean ± SD (n = 3).

   (K) Cell viability ratio of TET2^mutRHOA^WT Jurkat cells compared with
   TET2^WTRHOA^WT cells after treatment with 7,316 compounds at 0.4 μM for
   72 h (left). Compounds that specially suppressed the growth of
   TET2^WTRHOA^WT Jurkat cells are listed on the right.

   (L) Cell viability of TET2^mutRHOA^wt Jurkat cells after treatment with
   different concentrations of LY-364947 for 48 h. ∗p < 0.05 compared with
   TET2^WTRHOA^WT cells. Data are presented as the mean ± SD (n = 3).

   In search of reagents selectively targeting TET2^mutRHOA^mut and
   TET2^mutRHOA^wt cells, we examined cell growth of Jurkat cells exposed
   to 7,316 targeted agents ([135]Table S4). A number of potent inhibitors
   of the growth of TET2^mutRHOA^mut Jurkat cells, but not TET2^WTRHOA^WT
   cells, were identified by a drug sensitivity screen. The most effective
   agents are depicted in [136]Figure 4I. The selective suppressive
   function of KPT-9274, a dual inhibitor of PAK4 and nicotinamide
   phosphoribosyltransferase (NAMPT), decreased the growth of both
   TET2^mutRHOA^mut Jurkat and H9 cells ([137]Figures 4J and [138]S4G).
   Similar results were obtained when screening TET2^mutRHOA^WT Jurkat
   cells ([139]Figure 4K). The efficacy of LY-364947, a TGF-β1 receptor I
   (TβRI) inhibitor, was observed and validated in TET2^mutRHOA^WT Jurkat
   and H9 cells ([140]Figures 4L and [141]S4H). These findings suggest the
   potential of KPT-9274 and LY-364947 as a monotherapy for the T1 and T2
   subtypes, respectively.

Synergy of targeted agents and exploration of potential monotherapy in T3.1
and T3.2 subtypes

   Next, we evaluated differentially expressed genes between the T3.1 and
   T3.2 subtypes. As demonstrated by GSEA, higher expression of genes
   associated with DNA replication, translation, cell cycle checkpoints,
   and p53 signaling pathways were found in the T3.1 subtype, while VEGF,
   TGF-β, NOTCH, WNT, and MAPK signaling pathways were in the T3.2 subtype
   ([142]Figure 5A). Significant activation of DNA replication and VEGF
   signaling pathways is illustrated in [143]Figure 5B. As the T3.1
   subtype was characterized by histone-modifying gene mutations, we
   examined the combination of the histone deacetylase inhibitor chidamide
   with decitabine in KMT2C^mut Jurkat and H9 cells and observed synergy
   function compared with KMT2C^WT cells ([144]Figures 5C and [145]S4I).
   The Venn diagram identified a specific pattern of 2,168 genes altered
   by the combination group, but not by the untreated or the single-agent
   group ([146]Figure S4J). Genes associated with cytokine-cytokine
   receptor interaction were enriched in the combination group as well as
   the MAPK signaling pathway, in agreement with our previous study of
   KMT2D mutations ([147]Figures 5D and [148]S4K).[149]^10 zPDX models of
   the T3.1 subtype showed longer survival in the combination group than
   in the untreated or the single-agent group ([150]Figure 5E). Meanwhile,
   Ki-67 expression was evaluated to test the inhibitory activity of the
   anti-angiogenic agent apatinib and PD-1 inhibitor nivolumab in
   PTPN13^mut Jurkat and H9 cells when co-cultured with peripheral blood
   mononuclear cells (PBMCs) at a ratio of 1:5. Apatinib and nivolumab
   exhibited an elevated anti-tumor activity in PTPN13^mut Jurkat and H9
   cells ([151]Figures 5F and [152]S4L). As revealed by multicolor flow
   cytometry for immunophenotyping in the co-culture system of Jurkat
   cells and PBMCs, the combination group significantly decreased GATA3
   expression of CD3+/CD4+ Th2 cells in the co-culture system of
   PTPN13^mut Jurkat cells compared with the untreated group, while the
   inhibitory effect was not observed in PTPN13^WT cells ([153]Figure 5G).
   Accordingly, upregulation of GATA3 was observed in the T3.2 subtype
   using RNA-seq data of PTCL patients ([154]Figure 5H).

Figure 5.

   [155]Figure 5
   [156]Open in a new tab

   Synergy of targeted agents and exploration of potential monotherapy in
   T3.1 and T3.2 subtypes

   (A) Pathway enrichment analysis in T3.1 and T3.2 according to the KEGG
   and Reactome databases.

   (B) GSEA enriched differentially expressed genes of DNA replication in
   T3.1 and the VEGF signaling pathway in T3.2.

   (C) CI curve calculated by Compusyn software in KMT2C^mut Jurkat cells
   treated with chidamide (Chid) and decitabine (Deci) for 48 h.

   (D) Pathway analysis of differentially downregulated genes in each
   group upon treatment with Chid and/or Deci.

   (E) Survival of zPDX upon Chid and/or Deci treatment alone or in
   combination (n = 11 for each group).

   (F) Ki-67 positivity of Jurkat cells transfected with PTPN13^WT and
   PTPN13^mut upon nivolumab (Nivo) and apatinib (Apa) treatment after 72
   h. Peripheral blood mononuclear cells (PBMCs) and Jurkat cells were
   co-cultured at a ratio of 5:1. Nivo was at a concentration of 10 μg/mL,
   and Apa was at 10 μM. Data are presented as the mean ± SD (n = 3).

   (G) Multiflow cytometry analysis of Th2 cell markers (CD4 and GATA3) in
   PBMCs, co-cultured with PTPN13^mut Jurkat cells with or without Nivo
   (10 μg/mL) and Apa treatment for 72 h. Data are presented as the mean ±
   SD (n = 3).

   (H) Normalized mRNA expression of GATA3 across 4 molecular subtypes as
   revealed by RNA-seq.

   (I) Cell viability ratio of KMT2C^mut Jurkat cells and KMT2C^WT cells
   after treatment with 7,316 compounds at 0.4 μM for 72 h (left).
   Compounds that specially suppressed the growth of KMT2C^mut Jurkat
   cells are listed on the right.

   (J) Cell viability of KMT2C^mut Jurkat cells after treatment with
   different concentrations of KDM4D-IN-1 for 48 h. ∗p < 0.05 compared
   with KMT2C^WT cells. Data are presented as the mean ± SD (n = 3).

   (K) Cell viability ratio of PTPN13^mut Jurkat cells and PTPN13^WT cells
   after treatment with 7,316 compounds at 0.4 μM for 72 h (left).
   Compounds that specially suppressed the growth of PTPN13^mut Jurkat
   cells are listed on the right.

   (L) Cell viability of PTPN13^mut Jurkat cells after treatment with
   different concentrations of anlotinib for 48 h. ∗p < 0.05 compared with
   PTPN13^WT cells. Data are presented as the mean ± SD (n = 3).

   We also investigated cellular vulnerabilities to monotherapy induced by
   KMT2C and PTPN13 mutations, respectively, using a drug sensitivity
   screen. The selective sensitivity of KMT2C mutation to the KDM4D
   inhibitor KDM4D-IN-1 was validated by KMT2C^mut Jurkat and H9 cells
   ([157]Figures 5I, 5J, and [158]S4M), and that of PTPN13 mutation to the
   VEGFR2 inhibitor anlotinib was validated by PTPN13^mut Jurkat and H9
   cells ([159]Figures 5K, 5L, and [160]S4N).

Unique LME subtypes of PTCL defined by gene expression profiling

   Besides the molecular alterations of tumor cells in PTCL, their
   interactions with LME play a pivotal role in tumor biology and clinical
   behavior as well.[161]^3^,[162]^21 To define the diversity in the
   composition of LME in PTCL, we conducted unsupervised hierarchical
   clustering of RNA-seq data based on a list of essential immune
   profiling genes ([163]Table S5) and identified 4 distinct subtypes
   associated with immune infiltration and stromal components: LME1
   (39.2%), LME2 (19.9%), LME3 (11.8%), and LME4 (29.0%)
   ([164]Figure S5A). To validate our classification scheme, we utilized
   the statistical power conferred by the external gene expression
   profiles of 396 PTCL patients provided in published
   literature (Published Data:
   [165]https://github.com/emacgene/PTCL/blob/master/Supplementary_Data.Rm
   d)[166]^16 ([167]Table S6), and 4 LME clusters were identified with
   similar proportions: LME1 (36.6%), LME2 (15.7%), LME3 (12.1%), and LME4
   (35.6%) ([168]Figure S5B).

   To further explore specific characteristics of different LME subtypes,
   we developed 25 functional gene expression signatures covering various
   aspects of the PTCL microenvironment based on previous studies,
   including tumor cells of origin (e.g., Tfh, Th1, Th2, and Th17 cells),
   immune cells (e.g., B cells, plasma cells, regulatory T [Treg] cells,
   and M1 and M2 macrophages), stroma-related cellular and non-cellular
   components (e.g., lymphatic endothelial cells [LECs], vascular
   endothelial cells [VECs], cancer-associated fibroblasts [CAFs],
   fibroblastic reticular cells [FRCs], and extracellular matrix [ECM]),
   and LME-dependent signaling and transcriptional pathways (e.g., B cell
   traffic, TNF, immune checkpoints, IFN-α/β, VEGF, and HIF)
   ([169]Table S7). To reconstruct the general composition and biological
   features, the values of single-sample GSEA (ssGSEA) were compared among
   4 LME subtypes ([170]Figure 6A). According to the combinations of
   signatures within each subtype, they were termed “Tfh like” for the
   unique enrichment of Tfh cells as its tumor COO in LME1; “inflammatory”
   due to the aggregation of immune cells (e.g., Treg cells and M1
   macrophages) and the extensive activation of various
   immune-infiltration signaling pathways (e.g., VEGF, IFN-α/β, and immune
   checkpoints) in LME2; “mesenchymal,” considering the abundance of M2
   macrophages, stromal cells, and ECM in LME3; and finally “depleted” as
   LME4 was featured by an overall low expression of most microenvironment
   signatures compared with other subtypes. Moreover, all samples were
   examined for tumor content, and the results were comparable with
   previous studies ([171]Figure S5C).[172]^5^,[173]^15 No significant
   difference was detected between different subtypes based on
   pathological and molecular classification. However, when comparing the
   tumor content of the 4 LME subtypes, we observed remarkable
   differences. The depleted subtype showed significantly higher purity,
   consistent with the scarcity of immune- or stroma-related signatures
   and strong cell proliferation in this subtype. The inflammatory and
   mesenchymal subtypes revealed lower tumor content since they both
   featured abundant LME components. LME subtypes revealed an uneven
   pathological distribution ([174]Figure 6B). The Tfh-like subtype was
   mainly composed of AITL (53.4%) and PTCL-NOS (37.0%), corresponding to
   AITL and Tfh-originating PTCL-Tfh entities.[175]^5^,[176]^15 The major
   components of the inflammatory subtype were also AITL (56.8%) and
   PTCL-NOS (29.7%). Of note, most of mesenchymal subtype were ALK+ ALCL
   with a Th17-cell and stromal enrichment signature. The depleted subtype
   showed no specific inclination. We also performed ssGSEA analysis on
   the published gene expression data, and all immune-related signatures
   of each LME subtype were validated by the external cohort
   ([177]Figure 6C). In addition, despite the different proportions of
   pathological subtypes in the external cohort, LME subtypes shared
   similar pathologic distributions with our data ([178]Figure S5D). The
   Tfh-like and inflammatory subtypes mainly contained AITL (61.4% and
   25.8%) and PTCL-NOS (32.4% and 59.7%); most ALK+ ALCL were classified
   into the mesenchymal (42.9%) and depleted (44.6%) subtypes. As for the
   association of LME subtypes with molecular subtypes, we observed a
   correlation between Tfh-like and T2 subtypes, in accordance with the
   association of TET2 mutations with Tfh and B cell markers in AITL and
   PTCL-NOS.[179]^27^,[180]^28^,[181]^29^,[182]^30 The inflammatory
   subtype showed the closest association with the T1 subtype, which was
   characterized by RHOA mutations. The mesenchymal subtype, enriched in
   M2 macrophages and Th17-originating tumor cells, both of which are
   associated with histone-modifying mutations,[183]^31^,[184]^32 was
   mainly composed of the T3.1 subtype ([185]Figure 6D).

Figure 6.

   [186]Figure 6
   [187]Open in a new tab

   Immune clusters of PTCL revealed distinct gene expression-based
   lymphoma microenvironment (LME) subtypes

   (A) Heatmap of the activity scores of 25 functional gene expression
   signatures denoting four major LME clusters termed Tfh like,
   inflammatory, mesenchymal, and depleted. Tfh, follicular T helper; LEC,
   lymphatic endothelial cell; VEC, vascular endothelial cell; CAF,
   cancer-associated fibroblast; FRC, fibroblastic reticular cell; ECM,
   extracellular matrix; TNF, TNF signaling pathway; IS cytokines,
   immunosuppressive cytokines; IFNab, IFN-α/β signaling pathway; VEGF,
   VEGF signaling pathway; HIF, HIF signaling pathway; TGFb, TGF-β
   signaling pathway.

   (B) The proportion of pathological category distributions of four LME
   subtypes. The number of cases in each subtype is shown.

   (C) Heatmap of the activity scores of 25 functional gene expression
   signatures from external GEP of 396 PTCL patients, denoting four major
   LME clusters termed Tfh like, inflammatory, mesenchymal, and depleted.

   (D) Relationships and phi coefficients between LME subtypes and
   molecular subtypes.

   (E) PFS and OS curves of 186 PTCL patients according to LME subtype.

   Both PFS and OS in our cohort indicated a better prognosis of the
   mesenchymal subtype compared with other subtypes ([188]Figure 6E).
   Similarly, OS of the mesenchymal subtype in the external cohort also
   presented a significantly better prognosis, confirming its association
   with favorable outcomes ([189]Figure S5E). Moreover, among PTCL-NOS
   patients, PTCL-GATA3 mostly belonged to the depleted subtype with a
   worse prognosis, and PTCL-Tfh was associated with the Tfh-like subtype
   with a better prognosis ([190]Figure S5F).

Distinct immune communities dissected by LME subtypes

   To better profile the immune nature of each LME subtype, a precise
   overview of cellular abundance and relative population within each LME
   subtype was calculated using Cibersort. Moreover, the reproducibility
   of the LME subtypes was validated in the external cohort, showing
   comparable community enrichment profiles in each LME subtype
   ([191]Figures 7 and [192]S6). The Tfh-like subtype revealed a Tfh
   origin of lymphoma cells ([193]Figures 7A and [194]S6A), along with
   remarkably higher proportions of B cells and plasma cells
   ([195]Figures 7B, 7C, and [196]S6B). The inflammatory subtype mainly
   originated from Th1 or Th2 cells ([197]Figures 7D and [198]S6C) with
   enrichment of Treg cells and M1 macrophages ([199]Figures 7E, 7F, and
   [200]S6D). The mesenchymal subtype mainly originated from Th17 cells
   ([201]Figures 7G and [202]S6E) with enrichment of M2 macrophages and
   extensive stromal components, encompassing LECs, VECs, CAFs, FRCs, and
   ECM ([203]Figures 7H, 7I, and [204]S6F). As for the depleted subtype,
   it was characterized by the minimal presence of most LME components.

Figure 7.

   [205]Figure 7
   [206]Open in a new tab

   Dissection of LME subtypes uncovered unique immune communities

   (A) Significant enrichment of Tfh cells in Tfh-like LME, obtained by
   Cibersort.

   (B) Significant enrichment of B cells and plasma cells in Tfh-like LME,
   obtained by Cibersort and single-sample GSEA (ssGSEA), respectively.

   (C) Schematic of selected features of Tfh-like LME.

   (D) Significant enrichment of Th1 and Th2 cells in inflammatory LME
   obtained by ssGSEA.

   (E) Significant enrichment of Treg cells and M1 macrophages in
   inflammatory LME, obtained by ssGSEA and Cibersort, respectively.

   (F) Schematic of selected features of inflammatory LME.

   (G) Significant enrichment of Th17 cells in mesenchymal LME obtained by
   ssGSEA.

   (H) Significant enrichment of M2 macrophages, LECs, VECs, CAFs, FRCs,
   and ECM in mesenchymal LME, obtained by Cibersort or ssGSEA.

   (I) Schematic of selected features of mesenchymal LME.

   (J) Normalized mRNA expression of CRBN, TIGIT, and CD52 in Tfh-like LME
   vs. other LMEs as revealed by RNA-seq.

   (K) Normalized mRNA expression of PD-L1, LAG3, and IDO1 in inflammatory
   LME vs. other LMEs as revealed by RNA-seq and normalized mRNA
   expression of TIM-3 and CD30 in inflammatory LME vs. other LMEs as
   revealed by RNA-seq.

   As for possible immunotherapeutic targets, expression levels of
   Cereblon (CRBN), T-cell immunoglobulin and ITIM domain (TIGIT), and
   CD52 were higher in the Tfh-like subtype than in other LME subtypes,
   suggesting potential therapeutic vulnerabilities ([207]Figures 7J and
   [208]S6G). Increased expression of various immune checkpoint genes was
   observed in the inflammatory subtype, such as PD-L1, LAG3, and IDO1
   ([209]Figures 7K and [210]S6H), and TIM-3 in the mesenchymal subtype
   ([211]Figure 7L). Moreover, enrichment of CD30 expression was
   discovered in the mesenchymal subtype, indicating potential targeted
   therapy with an anti-CD30 antibody ([212]Figures 7L and [213]S6I).
   Therefore, the immunosuppressive condition may effectively contribute
   to the reconstruction of anti-tumor immunity and subsequent tumor
   growth inhibition in each LME subtype.

Discussion

   Despite the revolutionary introduction of targeted therapy
   and immunotherapy into lymphoma treatment, CHOP-based chemotherapy
   protocols remain an “unsatisfactory” standard of care in PTCL. Here we
   reported an integrated genomic and transcriptomic analysis of 221 PTCL
   patients, representing a large set of DNA- and RNA-seq profiling to
   understand the molecular and microenvironment characterization of PTCL
   and, more importantly, to provide a basis for the selection of more
   effective targeted and immunotherapeutic approaches in PTCL.

   Our study identified distinct molecular subtypes varying from genomic
   alterations and transcriptional signatures to clinical features. The T1
   subtype was defined as the coexistence of TET2 and RHOA mutations. TET2
   is a dioxygenase that functions within the DNA demethylation
   process.[214]^33 RHOA is a GTPase essential for multiple T cell
   processes, including cell proliferation and modulation of TCR
   signaling.[215]^30^,[216]^34 It has been reported that RHOA mutation
   together with TET2 loss induces Tfh cell specification and
   Tfh-originating lymphomagenesis, including AITL and PTCL-Tfh.[217]^35
   As RHOA functions as an essential GTPase in TCR signaling, we
   identified an enrichment of TCR-related gene expression in the T1
   subtype. Due to the coexistence of TCR-related (RHOA, CARD11, and FYN)
   or DNA methylation-related (TET2, IDH2, and DNMT3A) mutations, we
   evaluated co-targeting therapy with the hypomethylating agent
   5-azacytidine and the multikinase inhibitor dasatinib and exploited
   their synergistic function in TET2^mutRHOA^mut cells, which has been
   reported recently in a clinical trial for relapsed/refractory AITL
   patients.[218]^26 Interestingly, the drug sensitivity screen showed a
   potential monotherapy, KPT-9274, a dual inhibitor of PAK4 and NAMPT, to
   selectively suppress the growth of TET2^mutRHOA^mut cells. The T2
   subtype was characterized by TET2 mutation with wild-type RHOA and
   mainly consisted of AITL and PTCL-NOS patients. Mutations in
   PI3K-AKT-associated genes were enriched in T2, such as TSC2, ITPR3, and
   PIK3R1, in accordance with activation of the PI3K-AKT pathway. The
   dysregulation of PI3K-AKT has been reported to drive the pathogenesis
   of AITL and PTCL-NOS and is linked to poor clinical
   outcomes.[219]^36^,[220]^37 Subsequently, we proposed and confirmed
   that a combination of agents co-targeting DNA methylation and PI3K,
   5-azacytidine and duvelisib, exerted a synergistic impact on
   TET2^mutRHOA^WT cells. Besides, the TβRI inhibitor LY-364947 was
   identified as a potential monotherapy for the T2 subtype. The T3.1
   subtype mainly presented alterations in histone modification,
   particularly KMT2C and KMT2D. So far, inactivating histone
   methyltransferase KMT2C and KMT2D mutations have been identified in
   PTCL, including PTCL-NOS, PTCL-Tfh, and
   ALCL.[221]^38^,[222]^39^,[223]^40 Pathologically, the T3.1 subtype was
   enriched in ALCL, which was also associated with epigenetic modifier
   mutations.[224]^31 The dysfunction in histone modifiers led to
   enrichment in DNA replication and cell cycle checkpoint pathways.
   Combined treatment with two epigenetic agents, chidamide and decitabine
   as well as the monotherapy KDM4D-IN-1 both exerted potential inhibition
   of KMT2C^mut tumor cells. The T3.2 subtype was enriched for mutations
   in immune-related genes, especially HLA-A, HLA-B, and PTPN13, sharing
   similar features of a PTCL-NOS subtype with immune surveillance
   alterations.[225]^5 The VEGF pathway was upregulated in T3.2, together
   with immune-related mutations, explaining why an anti-PD-1 antibody and
   VEGF inhibitor as well as anlotinib, a tyrosine kinase inhibitor
   targeting the VEGF receptor,[226]^41 may contribute to favorable
   efficacy in the T3.2 subtype.

   In the second approach, we obtained microenvironment evidence and
   established distinct immune-related LME subtypes in PTCL,
   simultaneously integrating malignant T cell populations and
   microenvironment components into the prognosis and treatment of the
   disease.

   The Tfh-like subtype resembled a Tfh-originated lymphoma and was mainly
   composed of AITL and PTCL-NOS patients.[227]^13^,[228]^42 The high
   frequency of RHOA, TET2, DNMT3A, and IDH2 mutations in this subtype
   explained the differentiation of Tfh cells and Tfh-related
   lymphomagenesis. Cereblon, the direct binding target for
   immunomodulatory agents, was significantly overexpressed in this
   subtype, which may indicate a better response to
   lenalidomide-containing regimens. Moreover, increased expression of
   immune checkpoint TIGIT and the cell marker CD52 renders the Tfh-like
   subtype sensitive to TIGIT inhibition and the CD52-antibody
   alemtuzumab, respectively. The inflammatory subtype represented a
   microenvironment mediated by M1 macrophages and Treg cells.
   Pro-inflammatory M1 macrophages may contribute to the differentiation
   of neoplastic T cells, Th1 and Th2, and suggest a certain degree of
   anti-lymphoma immunity. However, Treg cell-associated immunosuppressive
   LME has been reported in different PTCL pathological entities and is
   associated with angiogenesis and tumor aggressiveness.[229]^3^,[230]^43
   Gene expression profiling unveiled significant enrichment of immune
   therapy-associated and angiogenesis pathways, including immune
   checkpoints, immunosuppressive cytokines, IFN-α/β, and the VEGF
   signaling pathway. Therapeutically, PD-L1, IDO1, and LAG3 were
   increasingly expressed, implying more potential efficacy of ICIs for
   these targets, while VEGF pathway activation suggested a favorable
   response to anti-angiogenesis agents. The mesenchymal subtype featured
   M2 macrophage infiltration as well as abundant cellular and
   non-cellular stromal components, including LECs, VECs, CAFs, FRCs, and
   ECM, together shaping an immunosuppressive microenvironment. The
   transcriptional signature was identified as upregulation of ECM
   remodeling, VEGF, HIF, and TGF-β pathways. As essential tumor-promoting
   cells, M2 macrophages and CAFs play immunosuppressive roles by directly
   modulating immune cell activities with cytokines and chemokines (e.g.,
   VEGF, IL-10, and TGF-β) and indirectly mediating ECM
   accumulation.[231]^44 In addition, TGF-β represents the master
   chemokine regulator in mesenchymal LME and, notably, induces Th17 cell
   differentiation.[232]^45 This subtype showed special preference for
   ALK+ ALCL, resulting in a dominant COO of Th17 and favorable clinical
   outcomes. Interestingly, recent advances have attributed the
   dynamic process of epithelial-mesenchymal transition to epigenetic
   modification, including histone methylation and acetylation,[233]^46
   consistent with the high frequency of epigenetic modification-related
   mutations in mesenchymal subtypes. TIM-3 overexpression provides an
   opportunity for ICI treatment in the mesenchymal subtype. This LME
   subtype was also characterized by a high CD30 positivity rate. Thus,
   the therapeutic activity of CD30-targeted agents is potentiated in
   relevant patients and has already been recommended as first-line
   therapy in CD30+ PTCL.[234]^47 The depleted subtype presented low
   microenvironment-derived signatures, featuring heterogenic COO and a
   lack of typical pathological preference. Consistent with the high tumor
   content, this subtype showed enrichment in cell proliferation
   signatures. A similar LME subtype with non-significant immune
   components has been reported in diffuse large B cell lymphoma and
   Burkitt lymphoma.[235]^48 Notably, despite the overexpression of GATA3,
   immunohistochemistry (IHC) and gene expression profile (GEP) approaches
   have characterized the microenvironment of the PTCL-GATA3 subgroup as
   “minimally inflamed” due to the limited amount of various immune cells,
   comparable with our depleted LME signature.[236]^20 Enrichment of
   available immunotherapy markers is not detected in this subtype; thus,
   effective therapies warrant further investigation.

   In conclusion, assisted by large-scale tumor biopsies and system
   biology studies, genomic and transcriptomic characterization were
   accomplished for categorizing PTCL into distinct molecular and
   microenvironment subtypes. Future study is necessary to elucidate the
   potential synergy of targeted therapy and immunotherapy, realizing
   curable PTCL in the era of precision medicine.

Limitations of the study

   Although we did not identify novel gene mutations in PTCL, we provided
   a simple and practical subtyping method suitable for clinical
   application. Meanwhile, due to the long-term lack of suitable cell and
   animal models in the basic research field of PTCL, we failed to obtain
   the most appropriate preclinical research model. We tried to overcome
   this limitation by maximizing the clinical utility of the scheme. A
   multicenter, randomized, phase 2 trial within each molecular subtype is
   currently ongoing (ClinicalTrials.gov: [237]NCT05675813). Finally, our
   LME subtyping is based on transcriptome sequencing, which requires
   further validation by pathological immunohistochemistry.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Antibodies
     __________________________________________________________________

   Alexa Fluor® 647 Mouse anti-Ki-67 BD Cat# 561126; RRID:
   [238]AB_10611874
   BUV395 Mouse Anti-Human CD45 BD Cat# 563792; RRID: [239]AB_2869519
   High Parameter Custom BV570 Conjugate Anti-Human CD4 BD Cat# 624298;
   special order
   High Parameter Custom BB660-P2 Conjugate Anti-Human T-Bet BD Cat#
   624295; special order
   High Parameter Custom BV605 Conjugate Anti-Human RORγt BD Cat# 624290;
   special order
   BUV496 Mouse Anti-Human CD3 BD Cat# 612940; RRID: [240]AB_2870222
   BV510 Mouse Anti-Human CD8 BD Cat# 563919; RRID:[241]AB_2722546
   PE-Cy™7 Mouse anti-GATA3 BD Cat# 560405; RRID:[242]AB_1645544
     __________________________________________________________________

   Biological samples
     __________________________________________________________________

   Human PTCL tumor tissues/peripheral blood This paper N/A
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   5′-azacytidine (Aza) Selleck S1782
   Dasatinib (Dasa) Selleck S1021
   Duvelisib (Duv) Selleck S7028
   Chidamide (Chid) Shenzhen Chipscreen N/A
   Decitabine (Deci) Selleck S1200
   Nivolumab (anti-PD-1) Selleck A2002
   Apatinib (Apa) Hengrui N/A
   KPT-9274 Selleck S8444
   LY-364947 Selleck S2805
   KDM4D-IN-1 Selleck S1059
   Anlotinib Selleck S8726
     __________________________________________________________________

   Critical commercial assays
     __________________________________________________________________

   Wizard® Genomic DNA Purification Kit Promega A1120
   DNeasy Blood & Tissue Kit Qiagen 69504
   RNeasy Mini Kit Qiagen 74106
   Qubit dsDNA HS Assay Kit Invitrogen [243]Q32854
   Qubit RNA HS Assay Kit Invitrogen [244]Q32852
   Qubit Assay Tube Invitrogen [245]Q32856
   Covaris microTube Covaris 520045
   Agilent High Sensitivity DNA Kit Agilent 5067–4626
   Agilent DNA 1000 Kit Agilent 5067–1504
   Agencourt® AMPure® XP Kit Beckman Coulter A63881
   Dynabeads MyOne Streptavidin T1 Invitrogen 65602
   SureSelect XT Human All Exon v6 Agilent 5190–8863
   SureSelect XT Reagent Kit, Illumina (ILM) platforms Agilent G9611A
   Herculase II Fusion DNA Polymerase Agilent 600677
   TargetSeq One Kit iGeneTech TPB1-i96
   Enzyme Plus Library Prep Kit iGeneTech TLP13-i96
   VAHTS Total RNA-seq (HMR) Library Prep Kit for Illumina vazyme NR603
   NovaSeq 6000 S4 Reagent Kit v1.5 (300 cycles) Illumina 20028312
   100% Ethanol, molecular biology grade Sigma E7023
   1X Low TE Buffer (10 mM Tris-HCl, pH 8.0, 0.1 mM EDTA) Life
   Technologies 12090–015
   KAPA RNA HyperPrep Kit with RiboErase KAPA KK8560
   Nuclease-free Water Ambion AM9930
   RosetteSep Human T Cells StemCell 15021
   lipofectamine 2000 Invitrogen 11668019
   Transcription Factor Buffer Set BD 652574
   CellTiter-Glo Luminescent Cell Viability Assay Promega G7572
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   Raw sequencing data This paper GSA-Human database (HRA004246)
     __________________________________________________________________

   Experimental models: Cell lines
     __________________________________________________________________

   Jurkat ATCC TIB-152
   H9 ATCC HTB-176
   HEK-293T ATCC CRL-11268
     __________________________________________________________________

   Experimental models: Organisms/strains
     __________________________________________________________________

   Zebrafish: Tuebingen (wild type) China Zebrafish Resource Center CZ3
     __________________________________________________________________

   Recombinant DNA
     __________________________________________________________________

   plasmids pCV186/Cherry/Puro-RHOA^wt This paper NA
   Plasmids pCV186/Cherry/Puro- RHOA^G17V This paper NA
   Plasmids pGV655/GFP/Puro-TET2^wt This paper NA
   Plasmids pGV655/GFP/Puro-TET2^H1380Y This paper NA
   Plasmids pGV655/GFP/Puro-KMT2C^wt This paper NA
   Plasmids pGV655/GFP/Puro- KMT2C^N4686S This paper NA
   Plasmids pGV655/GFP/Puro-PTPN13^wt This paper NA
   Plasmids pGV655/GFP/Puro- PTPN13^D2176N This paper NA
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   FastQC NA
   [246]http://www.bioinformatics.babraham.ac.uk/projects/fastqc/
   Trimmomatic Bolger et al., 2014[247]^49
   [248]https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
   BWA Li and Durbin, 2009[249]^50 [250]http://bio-bwa.sourceforge.net/
   Genome Analysis Toolkit McKenna et al., 2010[251]^51
   [252]https://software.broadinstitute.org/gatk/
   ANNOVAR Wang et al., 2010[253]^52
   [254]http://annovar.openbioinformatics.org/en/latest/
   STAR NA [255]https://github.com/alexdobin/STAR
   Integrative Genomics Viewer Thorvaldsdóttir et al., 2013[256]^53
   [257]http://software.broadinstitute.org/software/igv/
   Picard Broad Institute [258]http://broadinstitute.github.io/picard/
   BEDtools Quinlan, 2014[259]^54
   [260]https://bedtools.readthedocs.io/en/latest/
   HTSeq Anders et al., 2015[261]^55 [262]https://htseq.readthedocs.io/
   DESeq2 Love et al., 2014[263]^56
   [264]https://bioconductor.org/packages/DESeq2/
   Samtools Li et al., 2009[265]^57 [266]http://samtools.sourceforge.net/
   CNVkit Talevich et al., 2016[267]^58
   [268]https://cnvkit.readthedocs.io/en/stable/pipeline.html
   Genomic Identification of Significant Targets in Cancer (GISTIC 2.0)
   Mermel et al., 2011[269]^59 NA
   Maftools Mayakonda et al., 2018[270]^60
   [271]https://bioconductor.org/packages/release/bioc/html/maftools.html
   Limma (v3⋅38⋅3) NA
   [272]http://bioconductor.riken.jp/packages/3.0/bioc/vignettes/limma/ins
   t/doc/intro.pdf
   ggplot2 (v3.5.5) NA [273]https://ggplot2-book.org/
   ggbiplot NA [274]http://github.com/vqv/ggbiplot
   Pheatmap v1.0.12 NA
   [275]https://bioconductor.org/packages/release/bioc/manuals/heatmaps/ma
   n/heatmaps.pdf
   Gene set enrichment analysis (GSEA) Subramanian et al., 2005[276]^61
   [277]http://www.broad.mit.edu/gsea/
   Cibersort Newman et al., 2015[278]^62
   [279]https://cibersort.stanford.edu
   GSVA package v1.40.1 Hänzelmann et al., 2013[280]^63
   [281]http://www.sagebase.org
   HISAT2 software (v2.0.4) Kim et al., 2015[282]^64
   [283]https://daehwankimlab.github.io/hisat2/
   GFOLD (generalized fold change) algorithm Feng et al., 2012[284]^65
   [285]https://zhanglab.tongji.edu.cn/softwares/GFOLD/index.html
   Database for Annotation, Visualization and Integrated Discovery (DAVID)
   v6.8 NA [286]https://david.ncifcrf.gov
   Calcusyn software program Chou, 2006[287]^66 NA
   [288]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources and reagents should be
   directed to and will be fulfilled by the Lead Contact, Wei-Li Zhao
   (zhao.weili@yahoo.com).

Materials availability

   All unique/stable reagents generated in this study are available from
   the [289]lead contact with a completed Materials Transfer Agreement.

Data and code availability

     * •
       All sequencing data (including WES and RNA-seq) have been deposited
       in the GSA-Human database and can be accessed through the following
       link: [290]https://bigd.big.ac.cn/gsa-human/browse/HRA004246. It is
       available by contacting the [291]lead contact, Wei-Li Zhao
       (zhao.weili@yahoo.com), upon request.
     * •
       The paper does not generate original code. We utilized publicly
       available software in all the analyses. These are listed with
       appropriate citations in the methods.
     * •
       Any additional information required to reanalyze the data reported
       in this work paper is available from the [292]lead contact upon
       request.

Experimental model and study participant details

Patient information

   We collected 221 frozen tumor samples of PTCL patients. Histological
   diagnoses were established according to the World Health Organization
   (WHO) classification and reviewed by two independent pathologists
   (C.-F. Wang and B.-SH. Ou-Yang). All the patients received 6 cycles of
   CHOP/CHOP-like regimen, consisting of 103 AITL (47%), 71 PTCL-NOS
   (32%), 24 ALK+ALCL (11%), and 23 ALK-ALCL (10%) patients. Two or more
   TFH makers (including BCL6, CD10, PD-1, and CXCL13) were evaluated by
   IHC in most of AITL and PTCL-NOS ([293]Figure 1A). These samples were
   subjected to WES (n = 101, including 46 AITL, 39 PTCL-NOS, 12 ALK+ALCL,
   and 4 ALK-ALCL) or targeted sequencing (n = 120, including 57 AITL, 32
   PTCL-NOS, 12 ALK+ALCL, and 19 ALK-ALCL), and RNA-seq (n = 186). The
   study was approved by the Institutional Review Boards of all centers,
   and informed consent was obtained in accordance with the Declaration of
   Helsinki.

Zebrafish patient-derived xenografts (zPDX)

   Tumor tissues of three PTCL patients from T1, T2, and T3.1 subtypes
   were used to establish zPDX models, which were obtained from Shanghai
   Ruijin Hospital with written informed consent. Tumor cells were stained
   with Dil (Vybrant; Molecular Probes, Invitrogen) and then carefully
   injected into the perivitelline space of each anesthetized larva from
   wild-type Tuebingen zebrafish. In each subtype, successful injected
   xenografts were chosen and were randomly distributed in four groups:
   the untreated, the single-agent, and the combination group (3μM
   5-azacytidine with 0.1 μM dasatinib for T1 subtype, 3μM 5-azacytidine
   with 5μM duvelisib for T2 subtype, and 3μM chidamide with 3μM
   decitabine for T3.1 subtype for three consecutive days). In the
   following 10 days, the death event of zebrafish was recorded. The
   zebrafish study was approved by the Ethics Committee of Rui Jin
   Hospital Affiliated to Shanghai Jiao Tong University School of Medicine
   and the methods were carried out in accordance with the approved
   guidelines.

Method details

Whole exome sequencing

   Genomic DNA was extracted from frozen tumor samples of 101 patients and
   peripheral blood of 21 patients randomly selected from 101 patients,
   using Wizard Genomic DNA Purification Kit (Promega, Wisconsin-Madison,
   USA) according to the manufacturer’s instructions. A SureSelect XT
   Human All Exon v6 was used for capturing exome regions of these
   samples, which were sequenced on NovaSeq 6000 platform with 150bp
   paired-end strategy. Raw sequencing data were assessed by FastQC
   (version 1.11.4), and then processed by Trimmomatic (version 3.6) to
   remove sequencing adapters and low-quality reads that the joint
   sequence fragments of the 3′ end and low-quality fragments with Q value
   <25 and fragments with <35 bp. All sequence data was processed through
   the Broad Institute’s data processing pipeline to prepare read
   alignments for analysis. Next, reads were aligned to the Human Genome
   Reference Consortium build 37 (GRCh37) using BWA (version
   0.7.17-r1188), and the quality scores were recalibrated using the
   TableRecalibration tool from the Genome Analysis Toolkit (GATK)
   (version 4.1.4.0). Variant detection and analysis were performed by the
   Broad Institute’s Cancer Genome Analysis infrastructure program Mutect2
   and mutations were annotated using the Annovar software (version
   2017-07-17). The mean depth of each sample was 97.9× (range
   66.6–159.1×). A total of 82 genes were selected for further study based
   on two criteria: (i) genes with recurrent mutations (>3%) and/or (ii)
   genes with relevance to oncogenesis of PTCL ([294]Table S2). Sanger
   sequencing was used to confirm these somatic mutations in all WES
   samples. CNAs analysis for WES samples were inferred by CNVkit using
   the batch pipeline recommended, and significant focal CNAs across all
   samples were identified by Genomic Identification of Significant
   Targets in Cancer (GISTIC, version 2.0) with default parameters.

Targeted sequencing

   Genomic DNA was extracted from frozen or paraffin tumor samples of 120
   PTCL patients using a DNeasy Blood & Tissue Kit (Qiagen, Hilden,
   Germany) according to the manufacturer’s instructions. Targeted capture
   sequencing for 82 genes selected from WES data was performed using
   established Illumina protocols on novaseq6000 platform. Multiplexed
   libraries of tagged amplicons from PTCL tumor samples were generated by
   Shanghai Yuanqi Bio-Pharmaceutical Multiplex-PCR Amplification System.

RNA sequencing of tumor samples

   Total RNA was extracted from frozen tumor samples of 186 PTCL patients
   using RNeasy Mini Kit (Qiagen, Hilden, Germany). RNA size,
   concentration, and integrity were verified using Agilent 2100 System
   (Agilent). RNA libraries were constructed by using VAHTS total RNA-seq
   (HMR) library prep kit according to the manufacturer’s instructions,
   and sequenced on Illumina NovaSeq 6000 System. Paired-end reads were
   harvested from Illumina NovaSeq 6000 System, and the quality were
   controlled by Q30. After 3′ adaptor-trimming and low-quality reads
   removing by Trimmomatic software (v0.36), the high-quality trimmed
   reads were aligned to the reference genome (UCSC hg19) guided by the
   Ensembl GFF gene annotation file with STAR software (v2.5.3a).

   R package limma (v3⋅38⋅3) was used to normalize raw reads and obtain
   differentially expressed genes. GSEA was performed using the BROAD
   Institute GSEA software ([295]http://www.broad.mit.edu/gsea/) referring
   to Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome
   databases.[296]^48^,[297]^67

LME heterogeneity analysis

   In our RNA-seq dataset, we first calculated the log2(TPM+1) matrix
   based on the expression of 152 immune profiling genes extracted from
   LM22 signature gene list ([298]Table S5). The selected genes
   represented characteristic and exclusive markers in various immune
   components, covering all 22 major leukocyte subsets and activation
   state.[299]^68 Hierarchical clustering of the matrix was performed
   using the R package pheatmap (v1.0.12) using ward.D2 as linkage. To
   determine the optimal number of clusters, we utilized the silhouette
   analysis of KMeans by the cluster R package. The number of 4 clusters
   was identified as a better pick than more numerous clusters, and the 4
   clusters were identified by the cutree function. In the validation
   cohort, we clustered the gene expression data using 149 genes that were
   presented in all clustered datasets ([300]Table S5). Cibersort is a
   deconvolution algorithm that predict the proportion of different cell
   types from bulk RNA-seq data based on a list of 547 reference gene
   expression values, using support vector regression.[301]^62 We used the
   Cibersort algorithm to calculate percentages of 22 immune cell types
   from the normalized RNA-seq data. Next, we curated 25 gene expression
   signatures, reflecting various immune cells, stromal cells, noncellular
   substances based on published gene sets, as well as pathways based on
   KEGG and Reactome databases ([302]Table S7). Gene set analysis was
   carried out using the GSVA Bioconductor package v1.40.1. We then
   obtained a score of each sample, representing the enrichment of a gene
   set using gene expression data, including RNA-seq data and gene
   expression profile.

Cell transfection

   For RHOA^wt, RHOA^mut, TET2^wt, TET2^mut, KMT2C^wt, KMT2C^mut,
   PTPN13^wt, and PTPN13^mut transfection, purified plasmids
   pCV186/Cherry/Puro-RHOA^wt, pCV186/Cherry/Puro-RHOA^G17V,
   pGV655/GFP/Puro-TET2^wt, pGV655/GFP/Puro-TET2^H1380Y,
   pGV655/GFP/Puro-KMT2C^wt, pGV655/GFP/Puro-KMT2C^N4686S,
   pGV655/GFP/Puro-PTPN13^wt, and pGV655/GFP/Puro-PTPN13^D2176N were
   transfected into package HEK-293T cells using lipofectamine 2000
   (11668019, Invitrogen, Carlsbad, CA, USA) according to the
   manufacturer’s protocol. The supernatant fraction of HEK-293T cell
   cultures was condensed to a viral concentration of approximately
   2 × 10^8 units/ml. Then, the lentiviral particles were incubated with
   Jurkat and H9 cells for 72h with addition of polybrene (8 μg/mL). The
   stably transduced clones were selected by green and/or red fluorescence
   protein using flow cytometry or puromycin treatment for two weeks.

RNA sequencing of cell lines

   TET2^mutRHOA^mut, TET2^mutRHOA^wt, and KMT2C^mut Jurkat cells were
   treated with different agents. RNA size, concentration, and integrity
   were verified using Agilent 4200 TapeStation System (Agilent). RNA
   libraries were constructed by using KAPA RNA HyperPrep Kit with
   RiboErase (HMR, KAPA) according to the manufacturer’s instructions.
   Libraries were sequenced on Illumina NovaSeq 6000 System. Paired-end
   reads were harvested from Illumina NovaSeq 6000 System, and were
   quality controlled by Q30. After 3′adaptor-trimming and low-quality
   reads removing by cutadapt software (v1.9), the high-quality trimmed
   reads were aligned to the reference genome (UCSC hg19) guided by the
   Ensembl GFF gene annotation file with hisat2 software (v2.0.4). The
   GFOLD (generalized fold change) algorithm was used to find biologically
   meaningful rankings of differentially expressed genes from RNA-seq data
   of cell lines.[303]^65 DEGs were then analyzed by the Database for
   Annotation, Visualization and Integrated Discovery (DAVID) and were
   enriched in KEGG pathways.

Isobolographic analysis

   To determine the synergistic effect of targeted agent combination,
   including 5-azacytidine with dasatinib, 5-azacytidine with duvelisib,
   and chidamide with decitabine, combination index (CI) method described
   by Chou and Talalay[304]^66 was applied for automated analysis by the
   Calcusyn software program (Biosoft, Cambridge, UK). When at least 80%
   of CI values for a combination were <1, the drug combination was
   considered to be synergistic, while CI values = 1, and >1, the drug
   combination was considered to be additive effect, and antagonism,
   respectively.

T-lymphoma cell co-culture and multi-color flow cytometry

   Co-culture of PTPN13^wt or PTPN13^mut cells with PBMCs was conducted at
   1:5 ratio, treated with anti-PD-1 antibody nivolumab and/or apatinib
   for 72h. All cells were maintained in RPMI-1640 medium supplemented
   with 10% heat-inactivated fetal bovine serum and 1%
   penicillin/streptomycin. Then multi-color flow cytometry was carried
   out to assess the immune receptor expression of immune cells. CD45,
   CD3, CD4, CD8, T-Bet, RORγt, FoxP3, and GATA3 antibodies were used for
   cell labeling. Ki-67 were used to assess tumor cell proliferation. Flow
   cytometry data were collected by a FACS Calibur cytometer (BD
   Biosciences) and analyzed by FlowJo software.

Drug library screen

   TET2^wtRHOA^wt, TET2^mutRHOA^mut, TET2^wt, TET2^mut, KMT2C^wt,
   KMT2C^mut, PTPN13^wt, and PTPN13^mut Jurkat cells were used for
   screening assays. Cells were seeded in 384-well plates and then treated
   with the drug at a concentration of 0.4 μM (SCADS Inhibitor Kit,
   including 7315 compounds, [305]Table S4). Cell viability was assessed
   after 72h using the CellTiter-Glo Luminescent Cell Viability Assay
   (Promega, G7572). Luminescence was measured using an Envision
   Multi-label plate reader (PerkinElmer), and the luminescence reading
   was used to determine the cell viability relative to that of cells
   treated with solvent (DMSO). Candidate compounds were considered if
   their viability in TET2^mutRHOA^mut, TET2^mut, KMT2C^mut, or PTPN13^mut
   Jurkat cells was less than 50%, and that in TET2^wtRHOA^wt, TET2^wt,
   KMT2C^wt, or PTPN13^wt Jurkat cells was more than 80%.[306]^69

Quantification and statistical analysis

   Data were calculated as the mean ± standard deviation (SD) from three
   separate experiments. The Student’s t test was used to compare two
   normally distributed groups and the Mann-Whitney U test to compare two
   groups which did not conform to normal distribution. PFS was calculated
   from the date when treatment began to the date when the disease
   progression was recognized or the date of the last follow-up. OS was
   measured from the date of diagnosis to the date of death or the last
   follow-up. The Kaplan-Meier method was used to estimate survival
   functions were using and the log rank test to assess differences in PFS
   and OS between different subtypes. The data analysis for this
   manuscript was conducted using Statistical Package for the Social
   Sciences (SPSS, v24.0), GraphPad Prism 8 software or R software.
   p < 0.05 was considered statistically significant.

Acknowledgments