Genome-Wide miRNA Expression Profiling of Molecular Subgroups of Peripheral T-cell Lymphoma

Peripheral T-cell lymphoma (PTCL) is a group of mature T-cell malignancies with several entities recognized in the current World Health Organization (WHO) classification (1). PTCLs represent 10% to 15% of non–Hodgkin lymphomas (NHL), with angioimmunoblastic T-cell lymphoma (AITL), anaplastic large cell lymphoma (ALCL), and an unclassified group of PTCLs, called PTCL-NOS, encompassing approximately 70% of all PTCL cases (2–5).The clinical outcome of PTCLs have not improved in the last four decades, but introduction of targeted therapies has shown remarkable promise, especially in ALCL subtypes. Using gene expression profiling (GEP), we have defined robust molecular classifiers for major subtypes of PTCL, and more importantly, we identified two biological subtypes within PTCL-NOS, characterized by high expression of distinct T-cell transcriptional programs, called PTCL-GATA3 and PTCL-TBX21 subtypes (6). In addition, we have shown that tumor microenvironment signatures are prognostic in AITL (6).

We have also identified frequent mutations in genes that regulate the epigenome (TET2, DNMT3A, and IDH2; refs. 7–9) and in genes encoding proximal TCR/CD28 signaling molecules (RHOA^G17V, CD28, FYN, PLCγ, VAV1; refs. 7, 10–13) in AITL and other PTCL subtypes. Several studies have uncovered that AITL is derived from TFH cells, a subset of CD4⁺ T cells (14, 15). The transformation of the TFH cells to a malignant state in AITL is not well understood. However, genetic analysis revealed two major cooperative oncogenic mechanisms, initial epigenomic dysregulation followed by aberrant activation of T-cell receptor (TCR) signaling (reviewed in Iqbal and colleagues; ref. 5). Similarly, the integration of genome-wide DNA copy-number analysis in PTCL-NOS led to identification of specific pathogenetic aberrations in the novel subtypes driving distinct oncogenic pathways and possible therapeutic targets (e.g., constitutive PI3K–mTOR activation) in the PTCL–GATA3 subtype (16). Many of the GEP or genetic studies in PTCL were carried out using fresh frozen specimens (7, 10–13). The application of these findings to routine clinical practice has been challenging due to the use of formalin-fixed paraffin-embedded (FFPE) tissues for routine diagnosis (16, 17). Formalin fixation leads to fragmentation, cross-linking, and chemical modifications of RNA or DNA (18). Although we have used IHC algorithms for dissecting molecular subgroups (19), it is challenging for high-throughput analysis due to subjective quantitation methods for antibody staining. In contrast, miRNAs are well preserved in FFPE tissue due to their smaller size, and therefore, can be studied in archival materials with high reproducibility and accuracy (20). The use of miRNA expression profiling for tumor classification due to their stage-specific expression pattern, including in hematologic malignancies, has been explored (21–25). Studies have demonstrated the role of miR17∼92 in STAT3 activation and miR-135b in maintaining an IL17-like immunophenotype, and the association of miR-101 expression with mTOR activation in ALCL (26–31), an observation also consistent with GEP findings (6). In contrast, few genome-wide miRNA studies have been performed in common nodal PTCLs [e.g., AITL (32) or PTCL-NOS (33, 34)] and cutaneous T-cell lymphoma (CTCL; ref. 35). We performed miRNA expression profiling (n = 99) of AITL and PTCL-NOS molecular subtypes (PTCL-GATA3 and PTCL-TBX21) to identify reliable miRNA signatures, and compared miRNA profiles with normal human CD4⁺ Th-cell effector subsets (e.g., IL21-, Th1-, and Th2-polarized cells) to identify normal and oncogenic pathways regulated by miRNA expression. Integrative analysis of miRNA and gene expression was performed to delineate the oncogenic pathways perturbed due to miRNA expression and correponding role in PTCL pathogenesis.

The majority of PTCL specimens have GEP data obtained (Supplementary Table S1) using fresh frozen RNA (6, 16, 36). The corresponding FFPE samples were used for miRNA profiling using two platforms, including the TaqMan Array Human MicroRNA Card Set v3.0T, (ABI, Inc.) and the nCounter miRNA panels (NanoString, Inc). As detailed below, the miRNA data from the nCounter platform were validated in TaqMan Array quantitative real-time (qRT-PCR)-based platform (ABI, Inc.) in a different cohort. These cases were diagnosed in accordance with the 2016 WHO classification, and PTCL cases with a TFH immunophenotype were excluded. The study was approved by the Institutional Review Board of the University of Nebraska Medical Center (Omaha, NE). Two CD4⁺ T-cell lines representative of lymphoblastic leukemia (Jurkat) and Sezary Syndrome (HuT78) were used in the study. These T-cell lines were cultured in 2.05 mmol/L l-glutamine supplemental RPMI1640 (Jurkat) or IMDM (HuT78), supplemented with 10% or 20% FBS, respectively, and penicillin G (199 U/mL) and streptomycin (100 μg/mL). Cell cultures were maintained at 37°C in 5% CO₂. All cell lines underwent routine Mycoplasma testing using the Universal Mycoplasma Detection Kit (ATCC), with the most recent testing completed on November 20, 2019. Approximately four passages of the cells occurred between thawing and experimental use. Both Jurkat and HuT78 were previously purchased from ATCC.

Normal CD4⁺ T cells were isolated from peripheral blood mononuclear cells (PBMC) using the EasySep (human) negative selection CD4⁺ T-cell Isolation Kit (StemCell Technologies, Inc #17951). The CD4⁺ T-cell purity was analyzed by flow cytometry on an ACEA NovoCyte flow cytometer. CD4⁺ T cells were maintained in culture at 1 × 10⁶ cells/mL in RPMI1640 containing glutamine, 20% FBS, 1% penicillin–streptomycin, 1× 2-mercaptoethanol, and recombinant human IL2 (Cell Signaling Technology; # 78145; 20 ng/mL). The cells were polarized in IL21 (Cell signaling Technology; #45095; 20 ng/mL), and the polarization of CD4⁺ T cells to Th1 and Th2 was carried out using ImmunoCult human Th1 or Th2 [Cell Signaling Technology; Th1 (#10975), Th2 (#10973)] differentiation supplement combined with T-cell expansion medium (Cell Signaling Technology; #10981), as per the manufacturer's recommendations. The cells were analyzed on days 7 and 14 for Th1 and Th2 expression patterns, respectively.

FFPE tissue scrolls of 10–20-μm thickness with a surface area of 1 cm² were cut and stored at −20°C, until RNA extraction. Total RNA was extracted using the RNAStorm or Qiagen DNA/RNA Isolation Kit, and quality assessed using an Agilent Bioanalyzer or TapeStation2200 to estimate the RNA integrity analysis (RIN) value (37). The RNA was quantified using a Qubit fluorometer and the quality measure was calculated by the percentage of RNA fragments with a length of >200 nucleotides (i.e., DV₂₀₀)_. The total RNA from cell lines and normal cells was isolated with miRNeasy Mini Kit (Qiagen, Inc; #217004).

Total RNA (100 ng) was used for miRNA profiling utilizing the nCounter platform with the 799 miRNA probe panel (NanoString Human v3, Inc), following the manufacturer's instructions. The NanoString Digital Analyzer was used for data extraction and imported to n-Solver followed by normalization to the geometric mean of the top 100 miRNAs. To select miRNAs for analysis, >50 counts were considered as the threshold for the minimum level of expression and the normalized data were uploaded into BRB-Array Tool version 3.9.0 for analysis (http://linus.nci.nih.gov/BRB-ArrayTools.html). To select miRNAs for analysis, we used two approaches: (i) miRNA showing minimal variation across the arrays were excluded from the analysis by including only miRNAs whose expression differed by >2-fold from the median in at least 10% of the cases, and (ii) miRNAs with counts < 50 (i.e., threshold for the minimum level of expression) were excluded. In our series, differentially expressed miRNAs were selected at a significance of P < 0.05 by performing pair-wise comparisons. Specific miRNA signatures were derived for AITL, PTCL-TBX21, and PTCL-GATA3 from the nCounter platform and subsequently validated on Taqman qRT-PCR based (Applied Biosystems, Inc) platform (26, 33). The miRNA data from fresh-frozen cell lines or normal Th cells was used for comparative analysis only.

We integrated the GEP from the previous studies (6, 16, 36) and miRNA data from this study in a subset of PTCL cases (AITL = 14, PTCL-TBX21 = 14, PTCL-GATA3 = 12). The upregulated miRNA in AITL, PTCL-TBX21, and PTCL-GATA3 were queried against the TargetScan database (Release 7.1) to retrieve all predicted gene targets. These predicted miRNA targets were used to perform two types of analysis: correlation analysis and target gene repression analysis. The correlation analysis included integration of GEP-miRNA expression wherein negatively correlated genes were selected for each miRNA in each subgroup (r < −0.3 and P < 0.05).

Furthermore, subtype-specific analysis was done by performing pairwise differential gene expression analysis (e.g., AITL vs. PTCL-TBX21/PTCL-GATA3; PTCL-TBX21 vs. AITL/PTCL-GATA3; and PTCL-GATA3 vs. AITL/PTCL-TBX21) on targets predicted by TargetScan. The downregulated gene targets were selected (log fold change < −0.1, P < 0.05) from the three comparisons. To validate the derived target signature, we examined the expression in an extended cohort of AITL (n = 50) and PTCL molecular subgroups (n = 106) from our earlier studies (6, 16, 36).

The functional annotation analysis was performed on genes that were significantly negatively correlated with the miRNA expression using ConsensusPathDB (http://cpdb.molgen.mpg.de/), Ingenuity Pathway analysis [QIAGEN Ingenuity Pathway Analysis (IPA)], and DAVID (https://david.ncifcrf.gov/). Enriched pathways that overlapped between ConcensusPathDB and/or DAVID and IPA were identified and were further assessed.

The overall survival (OS; death from any cause) was estimated using the Kaplan–Meier method, and differences assessed using the log-rank test. Statistical analyses were performed with the R-language survival package. Differences among groups were considered significant at P values below 0.05.

The mature sequence of miR-126-3p and miR-145-5p was subcloned in TET-PLKO1 lentiviral vector for ectopic expression of miR-126 or miR-145. The lentiviruses were generated using 293T cells and the particles were transduced in cell lines using polybrene (8 μg/mL, Chemicon-Millipore) and puromycin (1 μg/mL) or GFP⁺ cells were sorted on a BD FACSAria II to select the transduced cells. Specific antibodies used for Western blots and flow cytometry are listed in Supplementary Tables S2 and S3, respectively. Primers and probes used in this study are listed in Supplementary Tables S4 and S5, respectively.

Apoptosis was quantified using a FACSCalibur flow cytometer (BD Biosciences) after staining the cells with Annexin V-PE (Apoptosis Detection Kit; BD Pharmingen # 4227986) according to the manufacturer's instructions.

The cell-cycle profile was analyzed using propidium iodide (#11868600). The cells were washed once with PBS and resuspended in buffer containing 1× PBS + 0.1% TritonX at 0.5 × 10⁶ cells/mL. RNAase was added to the cells at a final concentration of 10 μg/mL. The cells were incubated at 37°C for 15 minutes and then analyzed using a ACEA NovoCyte flow cytometer.

Cell viability was performed in the 96-well plates by using PrestoBlue Cell Viability Reagent (Invitrogen, A13261) as per the manufacturer's protocol and reading was taken on a fluorescence plate reader (Tecan Infinite M200Pro).

The cells were lysed in RIPA buffer (50 mmol/L Tris, pH 7.4, 150 mmol/L NaCl, 1% Triton X-100, 0.1% SDS, 1% sodium deoxycholate, 10 mmol/L sodium 2-glycerolphosphate, 1 mmol/L phenylmethylsulfonyl fluoride, 0.4 U/mL aprotinin, 1 mmol/L sodium fluoride, and 0.1 mmol/L sodium vanadate). Twenty to 100 μg of whole-cell extracts were separated by SDS-PAGE and transferred to nitrocellulose (Bio-Rad) or polyvinylidene difluoride (Bio-Rad) membranes. Membranes were blocked in Odyssey Blocking Buffer (LI-COR) or Tris-buffered saline with 0.1% Tween and 5% milk at room temperature for 30 minutes to 1 hour and then incubated with primary antibody (Supplementary Table S2) at 4°C overnight, followed by treatment with secondary antibodies. The immunoblots were visualized in Odyssey CLX (LI-COR) or Bio-Rad, Chemi Doc MP.

Transwell migration assay was performed in a 12-well plate (#07–200–156; Corning Transwell Multiple Well Plate) after induction of miR-126-3p in T-cell lines and were placed on the top layer of a cell culture insert with permeable membrane and B cells (Raji or DHL16) were placed below the cell-permeable membrane. Following an incubation period (3–18 hours), the migration of T cells with or without miR-126-3p expression toward B-cell lines (RAJI or DHL16) was assessed by the presence of CD4⁺ marker (flow antibody) in the bottom chamber using a NovoCyte 2060R.

Cells (10⁶) were washed twice in PBS and suspended in 100 μL 10% BSA staining buffer and antibody to CD4, CXCR5, PD-1, and CD19 (Supplementary Table S3) for 1 hour at 4°C for surface marker staining. Cells were then washed twice in staining buffer and run on a NovoCyte 2060R with data analysis using the NovoExpress software. Intracellular staining for GATA3 and TBX21 expression (Supplementary Table S3) was achieved using the Tru-Nuclear kit (BioLegend #424401) following the manufacturer's instructions. For phospho-flow antibodies (Supplementary Table S3), cells were fixed with BD phosflow fix buffer (#557870) and permeabilized with BD phosflow buffer III (#558050).

The endogenous levels of GTP-bound (active) Rho were detected using the GTPase pull down kit (Cell Signaling Technology, # 8820) as per the manufacturer's recommendation. Briefly, spin cups were treated with glutathione resin followed by the treatment of GST-Rhotekin-RBD. Whole lysates (empty or miRNA-transduced cells) containing 500 μg of proteins were added onto the spin cup; GTP binding protein and glutathione resin were incubated for 1 hour at 4°C. The GST-Rhotekin-RBD fusion protein binds the activated form of GTP-bound Rho, which can then be immunoprecipitated with glutathione resin. The unbound proteins were removed by centrifugation followed by elution of glutathione resin-bound GTPase with SDS buffer. The eluted sample was analyzed by Western blot by using the RhoA antibody.

The basic clinicopathologic characteristics of the patients with PTCL are summarized in Supplementary Table S1, with the majority of cases classified using GEP in earlier studies (6, 16, 36). We used classifier signatures in cases with GEP data to delineate PTCL-NOS into PTCL-GATA3 and PTCL-TBX21 (ref. 6; Supplementary Fig. S1A). In five cases, where GEP was not available, an IHC algorithm was used for PTCL-GATA3 versus PTCL-TBX21 subclassification (19). In agreement with previous findings, the PTCL-GATA3 subtype had an inferior OS compared with PTCL-TBX21 (Supplementary Fig. S1B; refs. 6, 19). Genome-wide miRNA expression profiling was performed in AITL (n = 39), PTCL-GATA3 (n = 24), and PTCL-TBX21 (n = 36) cases using hybridization based nCounter (NanoString Inc) or Taqman qRT-PCR based (Applied Biosystems, Inc) platforms.

Using human PMBCs from three healthy donors, CD4⁺ T cells were isolated and polarized in vitro into Th effector subsets (i.e., TFH-like, Th1-polarized, and Th2-polarized cells) with two activation modes. One included CD4⁺ T-cell activation with αCD3/αCD28 and IL2 for three days prior to polarization (donor A); the other started with direct polarization of naïve CD4⁺ T-cells into Th effector subsets (donors B and C; Fig. 1A; Supplementary Fig. S2A and S2B), using either IL21 cytokine stimulation (TFH) or commercial polarization kits (STEMCELL Technologies) for Th1 or Th2. We observed the expected changes in the expression of Th differentiation markers (BCL6, PRDM1, TBX21, GATA3) for each subset after polarization of naïve CD4⁺ T cells using published differentiation methods (38) and noted that IL21 stimulation generated a TFH-like immunophenotype, assessed by TFH markers (Fig. 1A–C; Supplementary Fig. S2A and S2B). The direct polarization of naïve CD4⁺ T cells resulted in a greater percentage of cells expressing effector Th subset differentiation markers compared with pre-IL2–activated CD4⁺ T cells (range: 38%–84% vs. 22%–49%), and thus were prioritized for miRNA profiling (Fig. 1A–D; Supplementary Fig. S2A and S2B). Stimulation with IL2 alone did not result in a major change in expression of most Th subset markers, but did result in higher CD4⁺/PD-1⁺ T cells.

Unsupervised hierarchical clustering (HC) and three-dimensional Principal Component Analysis (PCA) of the miRNA profiles showed distinct clusters of Th subsets, comprised either of cells that were activated with IL2 prior to polarization (first experiment with donor A), or cells that were directly polarized into different effector Th-cell subsets. The biological replicates of Th effector cells showed high correlations (r > 0.9; Fig. 2A and B), however, IL2-stimulated cells showed a wide range in their clustering and tended to cocluster with IL21 or with Th2-polarized cells, which illustrates the pleiotropic nature of IL2.

To generate specific miRNA signatures for Th effector subsets, we performed differential miRNA expression analysis by applying the pairwise analysis method (one-way ANOVA, P < 0.05; fold-change ≥2-fold) using the directly polarized Th subset data (Fig. 2C). Several known miRNAs associated with TFH differentiation were identified in the IL21-polarized cells [e.g., upregulated: miR21–5p (P = 0.0013), miR19b3-p (P = 0.03; ref. 39); miR-106b (P = 0.01), miR-146 (P = 0.04; ref. 40); and downregulated: miR-155 (P = 0.04; ref. 41)] as well as novel ones [miR-26a-5p (P = 0.03), miR-26b-5p (P = 0.03), miR-140 (P = 0.01)]. Because miRNA function is highly variable and cell dependent, we used the recently developed computational tool (miEAA; ref. 42) to identify signaling pathways associated with up- or downregulated miRNA signatures in each Th subset. Of the upregulated miRNAs in IL21-polarized cells, a subset regulates transcriptional repression of genes involved in inhibition of sphingosine-1-phosphate (S1PR1) receptor signaling (P < 0.001) or activation of PI3K–AKT signaling (P < 0.0001), whereas downregulated miRNAs promote activation of TCR signaling (P = 0.03) or Ras/RhoA inhibition (P = 0.003; Fig. 2D), suggesting a role of these miRNAs in regulating the signaling circuit of TFH differentiation. The upregulated miRNAs [miR-625-5p (P = 0.01), miR-181a-5p (P = 0.01), miR-150-5p (P = 0.001)] in Th1-polarized cells were enriched in transcriptional repression of the TGFβ signaling cascade (P = 4.43E-08), consistent with the negative role of TGFβ signaling in Th1 differentiation (43), whereas the downregulated miRNAs suggested transcriptional activation of ERBB2–MAPK–ERK-2 signaling (P = 0.03; Fig. 2D). In contrast, the Th2-polarized cells showed significant upregulation of known proinflammatory miRNAs [e.g., miR-221 (P = 0.02) and miR-155 (P = 0.04; ref. 44)] or miRNAs that lead to the upregulation of the p53 signaling response (P = 0.02) or activation of the PTEN–AKT–mTOR pathway [miR-337–3p (P = 0.008), miR-19a-3p (P = 0.02), miR-92a-3p (P = 0.01); Fig. 2D]. The latter pathway is known to promote Th2 immune response (45). Comparison of Th1- versus Th2-polarized cells (Fig. 2C, right, smaller heatmap) delineates the similar trend of miRNA signatures as mentioned above, but apparent signature differences is observed in both Th1- and Th2-polarized cells. T cells activated with IL2 prior to polarization (experiment 1) showed a large proportion of miRNAs with a similar expression pattern (Supplementary Fig. S3). Th1- and IL21-polarized cells had an overlap in miRNA expression (Fig. 2C), supporting a recent finding showing a TFH-like cell transition phase in Th1 differentiation (46) and potentially could be due to plasticity of Th effector cells (47). To support the in silico findings, we further evaluated corresponding key protein expression using flow cytometry or Western blot and demonstrated that IL21-polarized cells showed higher TCR activation as assessed by flow (p-LCK, p-Zap70, or p-SLP70) or by p-LCK protein expression compared with other subsets or naïve T cells. In addition, upregulation of p-AKT (ser473) in TFH cells, ERK-2 (p42MAPK) in Th1 cells as reported elsewhere (48), and p-S6 ribosomal protein [RPS6/(ser235/236)] indicative of increase in translation of mRNA transcripts or mTOR activity was confirmed in Th2 cells, thus validated the functional attributes of miRNA signatures (Fig. 2E and F).

We also identified miRNAs that were induced or repressed upon Th differentiation by comparing each subset with naïve T cells (P = 0.05; Fig. 3A and B; Supplementary Table S6). Other than miRNAs identified in the above analysis, miRNAs regulating JAK–STAT (P = 0.03) were induced in IL21-polarized cells, whereas miRNAs that negatively regulate IL4 production were induced in Th1-polarized cells, and miRNAs regulating activation of the p53 signaling pathway were induced in Th2-polarized cells (P = 0.03; Fig. 3A and B; Supplementary Table S6). Irrespective of the differentiation stage, the miR17∼92 cluster and paralog miR-106a∼363 were induced in all Th-polarized subsets. A miRNA with known involvement in inhibition of T-cell differentiation program [ref. 49; e.g., miR-125b-5p (P = 0.0001)] and other novel miRNAs [miR-574–5p (P = 0.005), miR-1285–5p (P = 0.008), miR-1972 (P = 0.01), miR-549a (P = 0.02)] were downregulated upon activation.

Ours and others' GEP studies suggested that TFH, Th1, and Th2 are the putative cells of origin (COO) for AITL, PTCL-GATA3, and PTCL-TBX21, respectively (6, 14, 36). Therefore, we correlated miRNA profiles of these subsets with their corresponding PTCL subtypes. Unsurprisingly, miRNAs from the nonmalignant polarized Th cells showed nonsignificant overlap with PTCLs, likely due to the plasticity of T cells and the presence of nonmalignant cells from the tumor microenvironment of the tumor sample biopsies. However, several miRNAs showed concordant expression identified by rank analysis (i.e., upper fifth percentile) between normal counterpart Th subsets and corresponding PTCL entities (Fig. 3C). Seven upregulated miRNAs were shared between AITL and nonmalignant IL21-polarized cells including miRNAs associated with RhoA protein signaling (P = 0.03) or NF-κB activation (P = 0.02). Similarly, 12 miRNAs were shared between PTCL-TBX21 and Th1-polarized cells, including miRNAs involved in regulating type I IFNγ signaling, whereas miRNAs negatively regulating IL6R signaling antagonists of Th2 differentiation, were shared between the PTCL-GATA3 subtype and Th2-polarized cells (Fig. 3D). In contrast, differential expression analysis of miRNAs in AITL versus IL21-polarized cells identified upregulation of miRNA associated with T-cell differentiation program (P = 0.008), and downregulation of miRNAs predicted to target regulators of Ras signaling (P = 0.03), potentially activating Ras pathway in AITL (Fig. 3E and F; Supplementary Table S7). Similarly, upregulated miRNAs in PTCL-TBX21 compared with Th1-polarized cells, were associated with inhibition of Notch and MAPK signaling and downregulated miRNAs were associated with activation of VEGF (P = 0.016) pathway and positive regulation of IL12 production (P = 0.02). In contrast, the PTCL-GATA3 subtype compared with Th2-polarized cells showed upregulation of miRNAs associated with inhibition of p53 signaling (P = 0.030) and the activation of PI3K pathway (P = 0.01) and Th2 immune response (P = 0.03; Fig. 3E and F; Supplementary Table S7).

Unsupervised HC and PCA analysis of the entire PTCL cohort (n = 58), which was profiled using the nCounter platform, showed that molecular PTCL subtypes were nonsignificantly grouped within subclusters, with cases largely interspersed (Fig. 4A and B). Similar findings were obtained using an independent PTCL cohort (n = 41) profiled on a TaqMan based qRT-PCR platform (ABI Inc; Supplementary Fig. S4A and S4B) using overlapping miRNAs between the platforms. We also used the qRT-PCR cohort for validation, as four cases profiled on both platforms showed a good correlation of miRNA profiles (Spearman r ≥ 0.63, P < 0.001) between the platforms (Supplementary Fig. S4C).

Differential miRNA expression analysis (ANOVA, P < 0.05, fold change >2-fold) identified 12 miRNAs significantly upregulated in AITL compared with PTCL subsets (PTCL-TBX21 and PTCL-GATA3; Fig. 4C), and upregulated miRNAs were associated with activation of TCR signaling as their predicted target genes negatively regulate TCR signaling (P = 0.013) or are involved in epigenetic maintenance (Fig. 4D; Supplementary Table S8). Conversely, targets of the downregulated miRNAs were associated with PI3K activation (P = 0.001) or mediating inhibition of Ras/RhoA signaling (P = 0.01). Of the 12 upregulated miRNAs, 8 were included on the qRT-PCR–based ABI platform using a different AITL cohort (n = 20) and had a concordant elevated expression pattern (Supplementary Fig. S5) compared with other PTCL subtypes (n = 24).

PTCL-TBX21 showed distinct expression of five miRNAs (Fig. 4C), associated with negative regulation of ERK signaling, a finding that substantiates the role of the ERK-1 signaling in Th2, but not in Th1 differentiation (50–52), whereas the downregulated signatures showed inhibition of JNK cascade (P = 0.002) or activation of NF-κB (P = 0.05; Fig. 4D; Supplementary Table S8). In contrast, the PTCL-GATA3 subtype showed upregulated miRNAs likely resulting in activation of AKT–mTOR pathway, p53 signaling response, MAPK activation, or negative STAT3 activation (Fig. 4D; Supplementary Table S8), which has a known role in regulating the Th2 differentiation (53). Interestingly, the downregulated miRNAs were predicted to activate the PI3K pathway. Supervised HC of these differentially expressed miRNAs in the ABI cohort demonstrated that these subtypes formed different subclusters, although a few cases were interspersed, likely due to fewer miRNAs available on ABI platform (Supplementary Fig. S4D).

We examined the tumor-associated miRNA signatures in normal T-cell subsets and expression of 5 of 12 miRNAs in the AITL signature were observed in naïve or IL2-activated CD4⁺ T cells (Fig. 4C). Overall, miRNAs expressed in PTCL-TBX21 or PTCL-GATA3 were largely represented in all normal Th-cell subsets.

We integrated the miRNA data with corresponding GEP in PTCL cases (n = 40) using the workflow shown in Fig. 5A. The upregulated miRNA specific to each PTCL subtype (Fig. 4C) were queried against the TargetScan database to determine predicted targets. These predicted miRNA targets were used for correlation analysis and target gene repression analysis. The correlation analysis included integration of GEP-miRNA expression wherein negatively correlated genes were selected for each miRNA in the subtypes (r < −0.3 and P < 0.05). We rationalized that correlation analysis would identify the specific predicted miRNA target genes whose expression is negatively correlated with target mRNA expression and target gene expression analysis would identify the repressed target specific to each PTCL subtype. Furthermore, we performed binary or pairwise differential gene expression analysis (logFC < −0.1) of the target genes between PTCL subsets (AITL vs. PTCL-TBX21/GATA3; PTCL-TBX21 vs. AITL/PTCL-GATA3; and PTCL-GATA3 vs. PTCL-TBX21/AITL). The resulting targets gene repressed in PTCL subtypes due to high miRNA expression (Fig. 5B) were further validated in an extended cohort of AITL (n = 50) and PTCL molecular subtypes (n = 106; Fig. 5C). We observed that the downregulated target gene signature for the PTCL subtypes had similar expression trends in the validation cohort as observed in training cohort, suggesting that miRNA mediates biological function in lymphomagenesis. The functional annotation of the significant target genes using in silico computational algorithms [i.e., ConsensusPathDB and/or enriched in other programs (IPA, DAVID)] identified target genes leading to inhibition of sphingolipid signaling (P = 0.01) and Wnt signaling (P = 0.02) in AITL, as shown in Fig. 5D. These findings are concordant with the TFH-like polarized cells mentioned above and demonstrated in another study where S1PR1 inhibition led to an increase in TFH cells (54). On a similar note, deregulated Wnt signaling has been reported in AITLs with RHOA^G17V and TET2 mutations (13). We also observed an inverse correlation (r = −0.65) between miRNA-603 and its target DNMT3A (54) in AITL (P = 0.03; Fig. 5D, scatter plot, top), concordant with findings of DNMT3A dysregulation in AITL from other GEP or genomic studies (7, 10–13).

PTCL-TBX21 miRNA target gene analysis showed target genes regulating cell “stemness” and inhibition of MAPK signaling pathway (P = 0.007; Fig. 5D), while PTCL-GATA3 were characterized by gene signatures either regulating cell-cycle regulators or RAP1/ERK5 signaling genes known to regulate cell proliferation and PI3K activation (55). The expression of DUSP10, critical negative regulator of ERK signaling (56), was negatively correlated with miR-181a-5p (r = −0.63) in the PTCL-GATA3 subtype (Fig. 5D, scatter plot, bottom) and miR-181a-5p as an essential modulator for ERK/MAPK signaling leading to proliferation has been reported (57, 58). Overall, this integrative analysis identified several deregulated signaling pathways in PTCL subtypes potentially mediated by deregulated miRNA expression.

PTCL-TBX21 and PTCL-GATA3 subtypes of PTCL-NOS have prognostic significance of the distinctive miRNAs identified between two subtypes, miR-29c (favorable) and miR-106 (unfavorable), also showed association with prognosis in PTCL-NOS cases (Supplementary Fig. S6A and S6B).

Of the miRNAs significantly associated with AITL, miR-126-3p and miR-145-5p expression showed marginal (P = 0.23) or significant (P < 0.01) association with OS (Fig. 6A and B). However, correlation of miR-126-3p and miR-145-5p (r = 0.1) illustrate their independent association with the prognosis (Supplementary Fig. S6C). miR-126-3p expression was validated by qRT-PCR (Fig. 6C), and high expression was noted in AITL cases profiled by the qRT-PCR–based platform (Supplementary Fig. S5A). In normal CD4⁺ T cells, miR-126-3p expression was upregulated in IL21 and Th2-polarized cells, but not in Th1-polarized or IL2-activated cells (Fig. 6D). Furthermore, it was noted that high expression of miR-126 was noted in CD4⁺ T cells, compared with B cells or stromal components of lymph nodes (Fig. 6E).

The ectopic miR-126-3p expression in T-cell lines (Jurkat, HuT78) or in normal CD4⁺ T cells resulted in elevated miR-126-3p expression (2–3-fold) and approximately 90% repression of a target recognition sequence in 293T cells using a luciferase reporter assay (Fig. 6F and G). This also resulted in induced TFH immunophenotype, as assessed by PD-1⁺/CXCR5⁺ in normal CD4⁺ T cells with miR-126-3p ectopic expression (Fig. 6H). Of the several targets identified through in silico analysis, we selected S1PR2, which has been shown to be important for CXCR5⁺ T- and B-cell interaction in germinal center (GC) and TFH differentiation (59) for experimental validation. As shown, ectopic miR-126-3p expression led to the repression of S1PR2 protein in T-cell lines (Jurkat; Fig. 6L), HUT78, and normal CD4⁺ T (Supplementary Fig. S7A and S7B) at protein or mRNA levels (Supplementary Fig. S7B). It is worth noting that low expression of S1PR2 showed association with bad prognosis, replicating likely miR-126-3p association with prognosis in AITL cases (Fig. 6I).

S1PR2 regulates the migration of CXCR5⁺ T cells toward CXCR5⁺ B cells within the GC follicle (59, 60); thus, we investigated the effect of miR-126-3p on T- and B-cell interaction using a transwell migration assay. Jurkat T cells (CXCR5⁺/S1PR2⁺) stably transduced with miR-126-3p or empty vector were plated in the top chamber, while a GC B-cell line with high CXCR5 expression (Raji) or with low CXCR5 expression (DHL16; Supplementary Fig. S7C) were plated in the bottom chamber; migration of T cells into the bottom chamber was assessed by flow cytometry by using CD4 and CD19 flow panel for T and B cells, respectively (Fig. 6J). As a control, we treated Jurkat T cells with a pharmacologic inhibitor of S1PR2 (JTE013, 10 μmol). We observed an approximately 50% reduction in the number of migrated Jurkat cells transduced with miR-126-3p cells toward Raji, whereas approximately 75% reduction in the number that migrated toward DHL16, indicating an inhibitory role of miR-126-3p in T- and B-cell interaction (Fig. 6K) and migration results were consistent with inhibitor JTE013 (Supplementary Fig. S7D).

S1PR2 signaling activates RhoA-GTPase activity (61), and we observed reduced active RhoA-GTP in miR-126-3p transduced T cells or drug-treated cells using pull-down assay for RhoA-GTP (Fig. 6L; Supplementary Fig. S7E), suggesting reduced RHOA enzymatic activity in transduced cells. High miR-126-3p expression showed no proliferative or survival advantage (Supplementary Fig. S7F–S7I), indicating miR-126-3p alters T-cell motility by regulating S1PR2 expression, a critical step in TFH differentiation, rather than modulating proliferative signals. RHOA inactivation inhibits migration (62, 63) and this study showed that miR-126-3p is playing a vital role in inhibiting the migration through S1PR2–RHOA axis and might have an important role in the pathobiology of AITL subtypes.

miR-145-5p expression was also associated with poor clinical outcome in AITL patients and its expression further validated by qRT-PCR in a subset of cases (Fig. 6M). We showed that ectopic expression of miR-145-5p in T cells inhibited T–B-cell interaction in a T-cell migration assay (∼45%) and repressed ROCK1 expression leading to a marginal decrease in active RhoA, via a known negative feedback (ROCK1-TIAM1/2-RAC1-RHOA) axis (Fig. 6N and O). Like miR-126-3p, ectopic expression of miR-145-5p did not show a proliferative advantage in T cells (Supplementary Fig. S7J–S7M). Overall, these data suggested that the RhoA pathway is negatively regulated upstream by miR-126-3p (via S1PR2), whereas downstream by miR-145-5p (via ROCK1), thus these miRNAs may be critical in TFH differentiation and AITL pathogenesis.

There are few studies involving genome-wide miRNA profiling of CD4⁺ Th subsets, though some miRNAs associated with the Th differentiation process (e.g., miR-17∼92, miR-155, miR-181a, miR-146a, miR-125) have been characterized (49, 64, 65). While generating Th subsets in vitro, we observed that prior stimulation of naïve T cells with IL2 dampens the Th differentiation program, as we observed reduced levels of lineage-specific markers. IL2 is a pleiotropic cytokine, often used for in vitro T-cell activation studies, as it shapes the transcriptional and metabolic programs of CD4⁺ T cells and prevents T-cell anergy (66). IL2 inhibits TFH (67) or Th17 differentiation but is required for Th1 and Th2 effector response (67). These observations led us to generate miRNA profiles on cells directly polarized into different Th subsets. Although miRNAs may have modest repression on target genes, a single miRNA can affect >200 genes, thus deregulated miRNA expression can concomitantly perturb multiple pathways. Instead of designating single miRNA signatures for each subset, we used miEAA computational models (42) to catalog their functions in the context of pathways for easier interpretation of their effects. For example, the miRNA significantly upregulated in IL21-polarized cells were identified to mediate negative regulation of S1PR1 or RHOA signaling, whereas downregulated miRNAs may lead to activation of PI3K–AKT and TCR signaling compared with other subsets. Low levels of S1PR1 were also found to be essential to maintain in vivo TFH homing (54, 68). In contrast, Th1 (polarized cells) miRNA signature was predicted to negatively regulate TGFβ signaling and Th2 (polarized cells) miRNAs may lead to PTEN–AKT–mTOR pathway activation or enrichment of Tp53 signaling response. These distinct pathways have been identified in earlier studies (7–9, 14, 15); our findings substantiate the earlier observations and indicate that miRNA expression may indeed fine-tune these signaling pathways to regulate distinct Th differentiation programs.

miRNAs regulate a transcriptional repertoire affecting particular signaling pathways, and the repression of a positive or a negative regulator may be marginal, but a persistent effect may develop a significant physiologic change (69). This was evident when comparing the miRNA profile of AITLs with normal counterpart IL21-polarized cells, which indicated that constitutive activation of TCR and PI3K signaling may be due to miRNA-mediated repression of negative regulators (e.g., CTLA4 by miR-145 or PIK3R2 by miR-29c) in AITL. MiRNAs (miR-150-5p, let-7a-5p, miR-142-3p, miR-155-5p, miR-21-5p, let-7g-5p, miR-16-5p) that regulate Ras/RhoA protein signaling or NF-kB activation were shared between TFH cells/IL21-polarized and AITL cases, albeit at different expression levels. A similar comparison of the miRNA profiles of PTCL-TBX21 and their normal counterpart (Th1-polarized) showed activation of VEGF signaling and inhibition of the MAPK pathway, whereas miRNAs regulating the IFNγ signaling pathway (e.g., miR-155, let-7ip, miR-23a, miR-146a, miR-16-5p, miR-181a, miR-15b-5p) were shared between Th1 cells and PTCL-TBX21. In contrast, comparison of PTCL-GATA3 with normal counterpart Th2-polarized cells showed upregulation of miRNA associated with p53 signaling [e.g., miR-26b-5p (P = 0.009), miR-29a-3p (P = 0.009), miR-19b-3p (P = 0.003), miR-92a-3p (P = 0.0001), miR-93-5p (P = 0.01)] and PI3K–AKT pathway [(e.g., miR-26a-5p (P = 0.009), let-7a-5p (P = 0.02), miR-20a-5p (P = 0.04), miR-20b-5p (P = 0.0001), miR-155-5p (P = 0.0001), miR-93-5p (P = 0.01), miR-625-5p (P = 0.004)], whereas IL6/IL12 signaling were shared (70, 71). These findings indicate that miRNA expression is important for normal Th-cell activation, but perturbed expression is associated with oncogenic transformation.

We also demonstrated distinct miRNA expression signatures associated with PTCL subtypes, and integration with corresponding GEP led to the identification of miRNA-regulated lymphomagenesis mechanisms unique to PTCL subtypes. It has been demonstrated that miRNAs negatively regulate their targets translationally (72) as well as through degradation of the transcript and the relative importance of these two mechanisms varies with the miRNA and the target. We were not able to investigate the effects on translation but focused on the transcripts and performed correlation analysis of GEP from the earlier studies (6, 16, 36) and current miRNA data in PTCL cases.We observed that significant association of miRNA signatures that regulate sphingolipid and Wnt signaling and epigenetic maintenance were altered in AITL compared with other PTCL subtypes. This observation is concordant with genetic findings, which demonstrated epigenetic dysregulation in AITL (16), while inhibition of sphingolipid signaling might be important for the maintenance of TFH gene expression program in AITL. The PTCL-TBX21 miRNA profile was associated with negative regulation of MAPK/ERK signaling, which substantiates the role of the ERK signaling in Th2, but not Th1 differentiation (50). We also observed the inhibition of ERK-1 and activation of ERK-2 in Th1 cells. Therefore, ERK signaling may play a distinctive role of Th1/Th2 cells through ERK-2 and ERK-1, respectively. In contrast, the PTCL-GATA3 subtype miRNA profile was associated with regulation of cell-cycle genes and RAP-1/ERK-5 signaling, associated with proliferative advantage in the PTCL-GATA3 subtype. Taken together, GEP–miRNA integration analysis suggests cooperative oncogenic mechanisms resulting from deregulated SIP and Wnt and epigenomic dysregulation in AITL, while other subtypes are characterized by either stemness or ERK/MAPK cascade leading to deregulated cell-cycle regulation. Earlier GEP or genetic studies in these PTCL subtypes have also identified distinctive oncogenic pathways associated with each PTCL subtype (6, 16, 25, 36).

Of the several miRNAs included in these signatures, we validated the function of miR-126-3p and miR-145-5p, as differential miRNA expression was correlated with clinical outcome in this AITL cohort. However, the number of cases with clinical outcome was few and additional cases with complete clinical outcome would be needed to confirm the association of these miRNAs with prognosis in future studies. Of these, miR-145-5p, transcribed as part of the miR-143/miR-145 cluster located on chromosome 5q33, a loci gained or amplified in 40% of AITL, was among the top tenth percentile miRNAome by rank analysis. The target genes identified through a negative correlation between miRNA–mRNA expressions included genes associated with cell motility or migration. Consistent with previous findings, we also observed that miR-145-5p may regulate ROCK1 expression in T cells, leading to inhibition of RhoA kinase activity as demonstrated in earlier studies (59, 73, 74)

The role of miR-126-3p in AITL pathogenesis is interesting, as it has been shown to regulate DNA methylation (75), T-cell activation (76), differentiation (77), immunosuppression (78), and Th2 response (79). Our findings suggest that miR-126 is expressed in the top fifth percentile of the AITL miRNAome, and regulates the expression of key molecules (i.e., SIPR2) involved in T- and B-cell interaction in the GC, a fundamental step for TFH differentiation. S1PR2 and CXCR5 regulate TFH cell localization near B cells to support GC development (59, 60), and requires CXCL13 produced by stromal or FDC cells (80, 81) or B cells (82). We show that miR-126-3p represses the S1PR2 expression leading to a diminished RhoA kinase activity. In AITL, deregulated RhoA activity is a key pathologic event, as 50% of the AITL cases harbor dominant negative RHOA^G17V mutation. It is likely that miR-126-3p or miR-145-5p expression may also perturb the RhoA pathway by repressing the expression of the upstream receptor, S1PR2, via mir-126-3p in TFH cells or by regulating the expression of downstream ROCK1 via miR-145-5p, further modulating the RhoA signaling in AITL at two different junctions: receptor and effector level. However, it is also interesting for future investigation to identify the role of RHOA^G17V mutation status in cases with high expression of miR-126-3p and miR-145-5p.

In summary, our study identified miRNAs expressed in Th subsets and their association with corresponding PTCL signatures, and their potential role in PTCL pathogenesis and diagnosis. However, a larger series of cases would be required to refine the miRNA signatures and to elucidate the differences between the major biological subtypes of PTCL.

W. Lone reports grants from NIH/NCI during the conduct of the study. A. Bouska reports grants from NIH/NCI during the conduct of the study. G.W. Slack reports other support from Seagen outside the submitted work. K.J. Savage reports other support from BMS, Merck, Seattle Genetics, Gilead, Janssen, AstraZeneca, Kyowa, Novartis, and BeiGene outside the submitted work. J.R. Cook reports grants from NIH (P01 1P01CA229100–01) during the conduct of the study. A.L. Feldman reports grants from Seattle Genetics outside the submitted work; in addition, A.L. Feldman has a patent 8,679,743 issued, a patent 9,677,137 issued, and a patent for Intellectual property licensed and with royalties paid from Zeno Pharmaceuticals. L.M. Rimsza reports other support from NanoString outside the submitted work. T.W. McKeithan reports grants from U.S. government during the conduct of the study. T.C. Greiner reports other support from Invivoscribe outside the submitted work. S. Pileri reports grants from AIRC- Italian Association for Cancer Research during the conduct of the study, as well as other support from BeiGene, Roche, and NanoString outside the submitted work. W.C. Chan reports grants from NIH outside the submitted work; in addition, W.C. Chan has a patent for Methods and Assay Kits for Diagnosing and Prognosing Peripheral T-Cell Lymphoma pending to University of Nebraska Medical Center. No disclosures were reported by the other authors.

W. Lone: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A. Bouska: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. S. Sharma: Conceptualization, resources, data curation, software. C. Amador: Conceptualization, resources. M. Saumyaranjan: Resources. T.A. Herek: Conceptualization, resources, visualization. T.B. Heavican: Visualization. J. Yu: Visualization. S.T. Lim: Resources. C.K. Ong: Resources. G.W. Slack: Resources. K.J. Savage: Resources. A. Rosenwald: Visualization. G. Ott: Visualization. J.R. Cook: Resources. A.L. Feldman: Resources. L.M. Rimsza: Resources. T.W. McKeithan: Resources, data curation. T.C. Greiner: Visualization. D.D. Weisenburger: Conceptualization, visualization. F. Melle: Resources. G. Motta: Resources. S. Pileri: Resources, data curation, investigation, methodology. J.M. Vose: Resources, visualization. W.C. Chan: Conceptualization, resources, data curation, formal analysis, funding acquisition. J. Iqbal: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, project administration, writing–review and editing.

The authors thank the University of Nebraska Medical Center Human Genetics Laboratory at the Munroe-Meyer Institute for facilitating us to scan our nCounter-cartridges on the digital analyzer and the University of Nebraska Medical Center Flow Cytometry Core. This work would not be possible without support provided by NIH NCI grants UH2 CA206127 -02 (to J. Iqbal and W.C. Chan) and P01 CA229100 (to J. Iqbal), the Leukemia and Lymphoma Society (TRP-6129–04; to J. Iqbal), NIH NCI Eppley Cancer Center Support grant P30 CA036727 (to J. Iqbal), NIH NCI Strategic Partnering to Evaluate Cancer Signatures (SPECS) II 5 UO1 CA157581–01 (to L.M. Rimsza, J. Iqbal, W.C. Chan), NIH NCI Specialized Programs of Research Excellence 1P50 CA136411–01 01A1 PP-4, and City of Hope internal funds (W.C.C.), AIRC 5 × 1000 grant (no. 21198; to S. Pileri). This study is also supported by Singapore Ministry of Health's National Medical Research Council, NMRC OF-LCG (MOH-000205–00) and TANOTO and LING Foundation (NRDUKSN18101; to C.K. Ong, S.T. Lim).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.