T-Cell Receptor Dependent and Independent NF-kappa B Activation is a Prognostic Marker and a Therapeutic Target in Peripheral T-cell Lymphoma Not Otherwise Specified

Gene expression analysis

We studied 180 PTCLs from which GEP were previously generated using fresh/frozen tissues [9–11, 21] (GEO data sets: GSE6338 and GSE19069). All of the cases were reviewed by at least two expert hematopathologists and diagnosed according to the WHO Classification [2]; furthermore, the diagnosis was refined by applying a molecular classifier recently developed by our group [10]. The detailed clinicopathological characteristics of these cases were previously reported [10].

Specifically, to characterize the activation of NF–kB pathway, we studied the expression of RELA (201783_s_at and 209878_s_at), RELB (205205_at), and REL (206035_at, and 206036_s_at), as well as the one of a series of well-known NF–kB targets, previously used for analogue purposes [14] and validated biochemically (http://www.bu.edu/nf-kb/gene-resources/target-genes/; http://bioinfo.lifl.fr/NF-KB/) [25]. Further, we studied genes encoding for proteins involved in TCR signaling [19].

Raw data were normalized in GeneSpring GX 12.5 (Agilent, CA); details on sample normalization and analysis have been previously reported [9–11, 21, 26, 27]. Reverse engineering was carried on by ARACNe algorithm in GeWorkbench 2.6.0 as previously described [27, 28]. Possible relationships among genes were further investigated by Cognoscente (http://vanburenlab.tamhsc.edu/cognoscente).

Gene expression studies were conducted according to MIAME guidelines. Raw gene expression are available at http://www.ncbi.nlm.nih.gov/projects/geo/.

Cells culture

Four human PTCL cell lines (Fe–Pd, Mac1, Karpass-299, and Jurkat) served as in vitro PTCLs models for treatment with BMS-345541 (SigmaAldrich) and Bortezomib (kindly provided by Dr. Guarguaglini). BMS-345541 (4(2^′-aminoethyl) amino-1,8-dimethylimidazo (1,2-𝛼)-quinoxaline)-4,5-dihydro-1,8-dimethyli-midazo (1,2-𝛼)quinoxalin-4-one-2-carboxylic acid, was dissolved in DMSO to produce a 10 mM (mmol/L) stock solution while, Bortezomib was dissolved in water at 1 mM stock solution concentration. The FE-PD cell line was kindly provided by Annarosa del Mistro (Padova University, Italy). The NPM-ALK + ALCL Karpas 299 and CTLC MAC1 were kindly provided by Prof. Inghirami (Department of Pathology and CERMS, University of Torino, Turin, Italy). Jurkat cell line (T-ALL) was purchased from DSMZ cell collection (Braunschweig, Germany). In this study Jurkat cell line was considered as negative control of BMS-345541 treatment as already reported in other study [29]. We also analyzed Pheripheral blood mononuclear cells (PBMC) obtained from healthy donor. Briefly, PBMC were separated by ficoll gradient density centrifugation using Lympholyte-H (Cedarlane Laboratories).

All the four cell lines and PBMC were cultured in RPMI medium 1640 containing 10% FBS and 2 mM L-glutamine (Lonza).

Cell viability assessment

Cell viability was measured by CellTiter-Glo Luminescent Cell Viability Assay (Promega,Corporation Italy). Aliquots of 5 × 10⁴ cells per well were distributed in 96-well opaque microplates in 100 μl of medium and incubated at 37 ^°C in a 5% CO2 humidified incubator for 48 h. BMS-345541 and Bortezomib were added at progressively increasing concentrations ranging from 0.1 μM up to 9 μM and from 2 nM up to 15 nM, respectively, to determine growth inhibition curves for all cell lines. Following incubation, 100 μl of CellTiter-Glo Reagent was added to the volume of cell culture medium present in each well, according to the manufacturer’s instructions. The relative cell viability was determined at 490 nm using a Victor2 (PerkinElmer) 96-well plate reader instrument. Each experiment was performed in triplicate.

For each cell line the value of half maximal inhibitory concentration (IC₅₀) [30, 31] was computed using the GraphPad software.

Flow-cytometry

To evaluate apoptosis and cell cycle, the four cell lines were incubated for 48 h with and without BMS-345541 (3.5, 4, 5 and 6 μM) and Bortezomib (5 and 8 nM).

In order to detect and quantify apoptosis we used Annexin-V-Fluos staining kit (Roche, Italy). Cells were harvested, washed twice and re-suspended in 100 μl of Annexin-V-Fluos labeling solution for 15 min, according to the manufacturer’s instructions. Then, we added 500 μl of incubation Buffer and analyzed with NAVIOS cytometer (Beckman Coulter). Data were elaborated by Kaluza dedicated software. Annexin V + PI-cells represented the early apoptotic populations. Annexin V + PI + cells represented either late apoptotic or secondary necrotic populations.

We followed the progression of S-phase through the cell cycle by labeling cells with the thymidine analog 5-bromo-2^′-deoxyuridine (BrdU) (Roche). After treatments, cells were collected and fixed to analyze the embedding of Br-dU according to the manufacturer’s instructions. The progression of labeled cells through the cell cycle was quantified by measurement of the fraction of BrdU-labeled cells in S-phase,G2/M, and G1 by flow-cytometry NAVIOS cytometer (Beckman Coulter).

Cytochemistry

To assess the effective inhibition of Nf-kB pathway via IKK inhibitor, we used immunohistochemistry by staining the Fe–Pd cell line after 6 h of treatment with 5 μM of BMS-345541. Briefly, cells were collected, washed once with cold phosphate-buffered saline (PBS), and centrifuged at 1500 r.p.m. for 5 min. The cell pellets were fixed in PBS-buffered formalin at room temperature at least 2 h. The cells were centrifuged for 10 min, then the cell pellets were washed with physiological saline and centrifuged. Subsequently, we added to the cell pellets human plasma and Simplastin (Tromborel) in ratio of 1:2 to allow formation of the clot. Finally, the clots were then routinely embedded in paraffin wax, as described previously [32], and processed for the detection of RELA and P50 by IHC. Briefly, FFPE clots were investigated by antibodies raised against fixation resistant epitopes: RELA (mouse monoclonal; Santa Cruz Inc.) and p50 (rabbit polyclonal; Thermo Scientific). The antigen retrieval protocols, dilutions and revelation system are detailed in table 1. Immunohistochemical preparations were visualized and images were captured using Olympus Dot-slide microscope digital system equipped with the VS110 image analysis software.

Statistical analysis

Statistical analyses were carried on with the IBM SPSS Statistics 20.0 (IBM, USA). Anova, and unpaired T-Test were adopted for GEP data analyses. Survival data were analyzed by using the Kaplan-Meier method [34]. Specifically, overall survival (OS) was calculated from the time of diagnosis to death or last follow up; progression free survival (PFS) was calculated from the end of induction treatment until progression, death, or last contact. Clinical information and complete follow up were available for 43/48 cases.

The limit of significance for all analyses was defined as P < 0.05; two-sided tests were used in all calculations.

All cases were collected at diagnosis, before any treatment administration. The study was approved by the Local Ethical Committee and conduced according to the Helsinki Declaration principles.

NF–kB REL components are variably expressed in PTCL

First, we studied the expression of RELA, RELB and REL in a series of 180 PTCL cases. We documented variable expression levels across the panel (figure 1A). As high levels of NF–kB effectors (RELA, RELB, and REL) have been previously related to NF–kB pathway activation, we divided our series into quartiles (Q1-4), based on the expression of each molecule (figure 1A). We then used GSEA to assess which group(s) could present with functional signs of NF–kB activation. Interestingly, we found a significant enrichment in both NF–kB targets and NF–kB pathway (as defined by GeneOntology) in Q2, Q3, and Q4 for RELA and REL. Conversely, no enrichment was recorded for any RELB subgroup. Therefore, we classified samples as NF–kB positive (84%) if falling in Q2-4 for at least one out of RELA and REL, and NF–kB negative (16%) in any other circumstance (figure 1B). Interestingly, REL and RELA expression tended to be mutually exclusive as documented by the inverse relation (r2 = 0.29; p = 0.01; figure 1C), though cases with high or null expression of both were recorded.

To further test whether such division could be reliable, we performed a supervised analysis between the two groups (NF–kB positive vs. NF–kB negative) and identified 781 probe sets, corresponding to 668 unique genes, that were differentially expressed in the two groups (figure 2A; supplementary table 1). Based on the expression of these genes, cases were then clustered and roughly separated according to NF–kB activation (figure 2B). Of interest, when we sought for pathways and cellular programs enriched in these genes, we found, among others, the TCR signaling (the main upstream of NF–kB in T-cells), the NF–kB pathway cascade itself, and the NF–kB transcriptional targets (figure 3; supplementary tables 2–4).

Noteworthy, IHC did confirm GEP data, with nuclear RELA/REL expression in activated cases (figure 4). Specifically, 17/17 NF–kB positive cases showed nuclear staining for either RELA or RELB or REL. Conversely, none of the 24 negative cases presented nuclear staining of a REL family component (table 2).

Together, these data indicated that PTCL/NOS category includes two discrete subgroups apparently characterized by activated and non-activated NF–kB pathway. As RELA and REL but not RELB were involved, it appeared that the canonical rather than the alternative pathway was activated in PTCL/NOS.

RELA/NF–kB activation is related to TCR signaling activation in PTCL/NOS

As NF–kB pathway is a downstream of TCR signaling in T-lymphocytes, we then focused on the expression of genes representative of TCR signaling (Unigene) [19]. We found several genes differentially expressed between cases with or without NF–kB activation, including FAS (p = 0.007), CFLAR (p < 0.001), CASP1 (p < 0.001), LCK (p < 0.001), FOS (p = 0.01), NFKB1 (p < 0.001), NFKBIA (p = 0.01), and MAP2K4 (p = 0.002), PIK3CG (p < 0.001), LAT (p < 0.001), PRKCB (p < 0.001), and ZAP70 (p < 0.001). Consistently, the two groups were clearly discriminated when clustered based on the expression of the TCR-related signature (figure 5; supplementary table 5). Moreover, among genes differentially expressed based on NF–kB activation status, we found genes encoding for TCR components (CD3E, CD5, and CD7) as well as TCR costimulatory molecules (CTLA4, CD28 and ICOS) (supplementary figure 1).

We then investigated whether RELA and REL were related to TCR at the same extent. We found that RELA expression was directly correlated to that one of both CD3E and CD7 (p < 0.0001 and p < 0.0001, respectively; supplementary figure 2A–B). Conversely, we observed that REL presented an inverse correlation with CD3E and CD7 (p < 0.0001 and p = 0.0015, respectively; supplementary figure 2C–D). In line with this evidence, RELA expression was directly correlated with the expression of the downstream molecules STAT1 and STAT3 (supplementary figure 2E–F), while REL presented an inverse relation with both (supplementary figure 2G–H).

When the costimulatory molecules were considered, RELA expression turned out to be directly correlated with that of CTLA4, CD28 and ICOS, while REL presented with no significant relations (supplementary figure 3).

Together, these data indicate that NF–kB activation is mediated by TCR signaling and RELA in a fraction of PTCL/NOS cases, being at least partially TCR independent and REL associated in other instances.

Reverse engineering revealed similarities and differences between RELA and REL transcriptional networks

As RELA and REL appeared to be activated at least in part by different mechanisms (i.e. TCR dependent or independent), we used reverse engineering to characterize their functional networks aiming to identify possible similarities and differences. To do this, we applied ARACNe, an algorithm previously found to be very effective in recognizing transcription factors targets as well as other molecules functionally related to them. We identified 281 genes significantly related to REL and 849 genes significantly related to RELA. Within the two networks, 133 and 723 genes were univocally related to REL or RELA, respectively. By contrast, of note, 161 genes were in common, the overlap thus being highly significant (p < 0.0001; figure 6A; supplementary figure 4; supplementary table 6). Such 161 genes, when tested for possible functional connection were found to have 147 already proved interactions (supplementary table 7; figure 6B). Among others, we identified CD44, IER2, NR3C1, and PGK1 as known NF–kB targets as well as ARID1A, BRK1, BTG1, CCNL1, CXCR4, FYN, and RHOA all previously related to neoplastic phenotype in humans.

The value of the identified networks was then further validated demonstrating by GSEA their significant enrichment in genes already found to be REL/RELA transcriptional targets (obtained from http://www.bu.edu/nf-kb/gene-resources/target-genes/ and http://bioinfo.lifl.fr/NF-KB/) (p = 0.004 and p = 0.005 for RELA and REL, respectively). In addition, they largely corresponded to the gens found to be differentially expressed in NF–kB positive vs. NF–kB negative cases (p < 0.0001).

Together, these data documented that RELA and REL are involved in partially different functional networks in PTCLs, but do also share a significant number of first neighbor

NF–kB activation is associated with adverse outcomes in PTCL/NOS

As PTCL/NOS appeared to be composed by two populations according to NF-KB molecular pathway, we sought to investigate the possible prognostic significance of such distinction.

When we studied the overall survival of patients divided in the 4 quartiles, we observed that patients with lower levels of RELA and REL fared better (figure 7). Similarly, when cases were divided into two groups based on REL and RELA expression (positive vs. negative; positive being defined when at least one gene was highly expressed), we observed a clear–cut differences between cases with higher or lower RELA and REL levels (figure 7). Particularly, the difference appeared to be significant as far as RELA was concerned. Most interestingly, when cases were grouped as NF–kB positive vs. NF–kB negative, by considering REL and RELA together, we documented a significantly different OS. Specifically, the estimated mean OS for NF–kB positive cases was 28.67 (95% CI, 15.875–41.459) vs. 56.018 months (95% CI, 39.252–72.785) (p = 0.004; figure 7). By contrast, in line with its inactivity indicated by GEP, RELB expression was not associated with the clinical outcome. Survival data are summarized in table 3.

Canonical NF–kB pathway is activated in PTCL cell lines

We established whether NF–kB pathway was active in the four available cell lines. Based on GEP, Fe–Pd, Mac1, Karpass-299, and Jurkat cells were considered RELA-positive, REL-positive, and RELB-negative. Consistently, RELA (p65), REL and p50, representative of the canonical pathway, showed a nucleus-cytoplasmic localization at immunofluorescence, while RELB and p52 typical of the alternative pathway, appeared to be confined to the cytoplasm, indicating an active and inactive status for the canonical and alternative pathways, respectively (supplementary figure 5).

NF–kB inhibition through BMS-345541 and Bortezomib limits proliferation and induces apoptotic cell death in PTCL cell lines

We then sought to assess whether NF–kB pathway shut off could affect the cellular vitality of PTCL cells. We used BMS-345541, a specific IKK inhibitor (upstream NF–kB activator) as well as Bortezomib, a proteasome inhibitor. The latter, though less specific, is already available in the clinical practice and determines NF–kB blockage via proteasome inhibition [35].

To verify whether BMS-345541 and Bortezomib could inhibit PTCL cell growth we used increasing concentrations of the two drugs from 0.1 μM to 9 μM and 2 nM to 15 nM for 48 h, respectively. The cell growth of PTCL cells was inhibited in a dose-dependent manner. Specifically, the IC50 values for BMS-345541 (figure 8A) and Bortezomib (figure 8B) at 48 h for the four cells lines were as follows: Fe–Pd 5 μM and 8 nM; MAC1 6.5 μM and 10 nM; Karpas 6.97 μM and 9.63 nM; Jurkat >12 μM and >15 nM (table 4). Jurkat cells were resistant to BMS-345541 and Bortezomib treatment, as expected.

We then investigated whether NF–kB inactivation could be responsible for PTCL cell cycle progression arrest and induction of cell death. To evaluate the effect of NF–kB inhibition over cell cycle progression we considered 5 μM and 8 nM for BMS-345541 and Bortezomib, respectively, corresponding to their Ic50. After measurement of the fraction of BrdU-labeled cells by flow-cytometry, we observed an arrest of cell cycle in G0/G1 phase in Fe–Pd (68% for BMS-345541 and 73.6% for Bortezomib), Mac1 (33% for BMS-345541 and 47% for Bortezomib), Karpas (50% for BMS-345541 and 48% for Bortezomib). On the contrary, Jurkat cells were insensitive to both the treatments (16% for BMS-345541 and 19% for Bortezomib) (figure 8).

We subsequently evaluated the apoptosis rate by Annexin V assay in PTCL cells lines treated or not with BMS-345541 and Bortezomib as single agents at different concentrations (3.5–6 μM for BMS-345541; 5–8 nM for Bortezomib). After 48 h, we observed an increase in the apoptotic events up to 76% in Fe–Pd, 36% in MAC1, 47% in Karpas and 9% in Jurkat cells line upon BMS-345541 treatment (figure 8). Similar results we observed when we treated PTCLs cells lines with Bortezomib. Particularly, we documented 58% of apoptotic events in Fe–Pd, 22% in MAC1, 10% in Karpas, and 8% in Jurkat cells lines upon administration of Bortezomib 8 nM. To further assess the specificity of the effects exerted by NF–kB inhibitors on PTCL cells and evaluate the potential toxicity of such treatment, we investigated the induction of apoptosis in PBMCs from healthy donors. Healthy donor PBMCs were exposed to BMS-345541 at 5 μM and Bortezomib at 8 nM (IC50) for 48 h. Importantly, we did not record any significant cytotoxicity or apoptosis induction in non-neoplastic T-cells (supplementary figure 6).

Finally, to prove that the observed pharmacologic effects were associated with NF–kB inhibition, we performed immunocytochemical analysis of RELA and p50 expression in Fe–Pd cell line before and after 6 h of treatment with 5 μM BMS-345541. While untreated Fe–Pd cells showed a marked nucleo-cytoplasmic positivity for both RELA and p50 proteins, after incubation with BMS-345541 a dramatic reduction of the nuclear staining with RELA and p50 was observed, the two molecules expression being substantially confined to the cytoplasm (supplementary figure 7).

Taken together, these data indicate that NF–kB inhibition is effective against NF–kB positive PTCL cells ex vivo.