Cases from multiple institutions were reviewed by two hematopathologists to confirm the diagnosis of ENKTL (n=89), PTCL-EBV (n=25) and PTCL-NOS (n=36) based on the 2017 WHO lymphoma classification.⁵ PTCL-EBV shows a similar phenotype as ENKTL but, unlike ENKTL, (i) patients present primarily with nodal disease where the bulk of tumor is localized; (ii) nasal involvement is lacking; (iii) it often shows a CD8⁺/CD56^– phenotype; and (iv) it is often of T-cell origin (Online Supplementary Table S1). Systemic and cutaneous EBV-positive T/NK lymphoproliferative diseases occurring in children were excluded. The diagnosis of PTCL-NOS was established when other specific subtypes of PTCL had been excluded. Cytotoxicity was defined as the expression of at least one cytotoxic marker (TIA1, granzyme B). Clinical data including age, sex, disease type, stage, International Prognostic Index (IPI) score, expression of CD4, CD8 and CD56, T or NK lineage, treatment and overall survival were obtained (Online Supplementary Table S2A, B). This study was approved by the National Healthcare Group Domain Specific Review Board B (2009/00212).

An OncoScan® FFPE assay was performed on 34 ENKTL, 14 PTCL-EBV and 29 PTCL-NOS cases as previously published.⁹ Copy number aberrations were analyzed using OncoScan^® Console (v1.3) software (ThermoFisher Scientific, Waltham, MA, USA). Segmentation results were analyzed using GISTIC (v2.0.22)¹⁰ with the GENCODE hg19 build. GI and homologous recombination deficiency (HRD) scores were calculated according to published methods.¹¹

Gene expression profiling was performed on 35 ENKTL, 23 PTCL-EBV, and 26 PTCL-NOS cases using a GeneChip^® Clariom D Assay (Human) array and the data were analyzed as described previously.¹² Gene expression profiling (GSE160119) and copy number aberration (GSE160118) raw data were deposited in the Gene Expression Omnibus (GEO).

Quantitative reverse transcription polymerase chain reaction (RT-qPCR) analysis of miRNA expression was performed using IDEAL miRNA qPCR assays (MiRXES, Singapore) (n=24) according to the manufacturer’s instructions on a QuantStudio™ 5 System (ThermoFisher Scientific, Waltham, MA, USA). RT-qPCR of EBV genes was performed to confirm EBV latency in PTCL-EBV and ENKTL.

Fluorescence in situ hybridization (FISH), multiplex immunofluorescence and multispectral imaging were performed as previously described.⁴

Targeted mutation analysis was performed by next-generation sequencing (Ion GeneStudio S5 prime, Thermo Fisher Scientific, Waltham, MA, USA) using an AmpliSeq customized T/NK-lymphoid panel comprising 35 genes recurrently mutated in ENKTL and PTCL-NOS (Online Supplementary Table S3A) and the 484-gene NovoPMTM 2.0 panel (Online Supplementary Table S3B) (total 500 genes including 19 common genes).

Overall survival was investigated using Kaplan-Meier non-parametric survival analysis and log-rank tests. The effects of disease type, age, sex and disease stage were analyzed using unadjusted univariate and multivariate Cox proportional hazard models. Analyses were performed in R with the “survival” (v0.1.2) and “survminer” (v0.4.6) packages. Gene expression profiling analysis was performed in R with the “limma” package (v3.40.6). Unsupervised hierarchical clustering was performed using Spearman distance and Ward.D2 linkage. All additional statistical tests were performed in R.

The diagnostic criteria for the diseases, T/NK lineage assignment, detailed descriptions of study cohorts, experimental methods, data analysis and data accession are provided in the Online Supplementary Materials.

The salient clinicopathological and immunohistochemical features of ENKTL, PTCL-EBV and PTCL-NOS are summarized in Table 1A and Online Supplementary Table S2A, B. All 25 patients with PTCL-EBV, comprising 14 Japanese, eight Chinese, two Korean and one Bangladeshi person, presented with nodal disease and the bulk of disease involved lymph nodes. Phenotypically, PTCL-EBV was often positive for CD8 (17/25, 68%), cytotoxic markers (100%) and TCRB (12/25, 48%). CD56 was commonly negative (18/24, 75%) and none of the cases tested expressed TCRG. Four of 23 (17%) cases tested expressed CD4, of which one was positive for CXCL13 and CD10 (focal). Most of the cases of PTCL-EBV were of T-cell origin (20/23, 87%), in accordance with previous reports^3,13–15 (Figure 1A). With regards to treatment, the patients in all three groups from different institutions were treated with a heterogeneous combination of chemotherapy regimens (Online Supplementary Table S2A). This is understandable as these are rare lymphomas lacking standardized and effective treatment. Two out of 15 PTCL-EBV patients with known treatment data were treated with SMILE therapy. One died 3.5 months after diagnosis and the other died 12.6 months after diagnosis.

Despite the treatment heterogeneity, our results revealed that patients with PTCL-EBV had a significantly shorter median overall survival (4.6 months) compared to those with ENKTL (14.7 months, P=0.001) and PTCL-NOS (26 months, P=0.007). (Figure 1B; Table 1). Univariate analysis identified advanced disease stage (stage 4) (P<0.001) and older age (P=0.001) as significantly associated with poor prognosis. Compared to patients with PTCL-EBV, patients with ENKTL had a 61% lower risk of death (HR=0.39, 95% CI: 0.22-0.69, P=0.001), while those with PTCL-NOS had a 59% lower risk (HR=0.41, 95% CI: 0.21-0.80, P=0.009). After adjusting for disease stage, sex and age, PTCL-NOS remained significantly less aggressive compared to PTCLEBV (HR=0.3, 95% CI: 0.14-0.65, P=0.002) (Table 2). A Cox proportional-hazards model was also performed and after controlling for lineage, patients with PTCL-EBV showed significantly worse outcome than those with ENKTL (HR=2.34, 95% CI: 1.05-5.24, P=0.038), indicating that the worse outcome of PTCL-EBV patients compared to ENKTL patients is independent of lineage. Interestingly, there was no longer a significant difference in the survival outcome between ENKTL and PTCL-EBV patients, suggesting that the worse outcome of those with PTCL-EBV could be ascribed to older age and more advanced disease stage at diagnosis.

In order to identify differences in copy number profiles of the three diseases, we analyzed copy number aberrations of patients with ENKTL (n=34), PTCL-EBV (n=14) and PTCLNOS (n=29) and identified recurrent aberrations (q<0.25) with their putative target genes across all samples (Figure 2A). Previously,⁴ comparing ENKTL (n=29) and PTCL-EBV (n=12), we noted differences in 14q11.2 and were able to reproduce those here. However, with the addition of five and two new ENKTL and PTCL-EBV samples, respectively – as well as a cohort of PTCL-NOS – we were able to identify multiple unreported differences in focal copy number aberration rates across these groups.

Of the 11 recurrently gained regions, 3p14.1, 6p22.3, 6p22.1 and 17q21.33 occurred at significantly different frequencies across disease groups (P<0.05, χ² test) (Online Supplementary Table S4). Gain of 3p14.1 was found in 14.3% of PTCLEBV cases compared to 5.9% and 76.0% of ENKTL and PTCL-NOS cases, respectively (P<0.001). Gain of 6p22.3 was observed more frequently in ENKTL (20.6%) and PTCL-NOS (58.6%) than in PTCL-EBV (7.1%) (P=0.005). Gain of 6p22.1 was observed in 21.4% of PTCL-EBV cases compared to 8.8% of ENKTL and 58.6% of PTCL-NOS cases (P=0.001). Of the nine recurrent losses, only 14q11.2 occurred at significantly different frequencies among the disease groups (Online Supplementary Table S5). As expected, loss of 14q11.2 was observed in 100% of PTCL-EBV cases,⁴ whereas it was only found in 20.6% and 58.6% of ENKTL and PTCL-NOS, respectively (P=0.001). It is unclear whether the high frequency of 14q11.2 loss in PTCL-EBV is a result of additional critical driver events occurring at the TCRA locus, on top of monoclonal TCRA gene rearrangement, or related to preferential TCR usage in PTCL-EBV due to possible underlying immune perturbations and antigen selection that predispose to malignancy.¹⁶ The 14q11.2 loss in PTCL-NOS is slightly low, which may be a result of coverage bias related to the probe design of the Oncoscan assay or a lack of TCRA rearrangement in some PTCL-NOS cases.⁸ The top two recurrent copy-number gains (3p14.1 and 6p22.1) were subsequently validated by FISH (Online Supplementary Figure S1 and Online Supplementary Table S6).

We further compared the genome-wide levels of copy number aberrations across the three disease types (Figure 2B, Online Supplementary Figure S2). Analysis of copy number burden (segment counts) revealed that PTCL-EBV exhibited significantly fewer segments than ENKTL (P=0.016, Mann-Whitney U test) and PTCL-NOS (P<0.001) (Figure 2C, left panel). While copy number segment differences between ENKTL and PTCL-EBV appeared primarily driven by gains (Figure 2C, middle panel), PTCL-NOS demonstrated increased losses compared to other groups (Figure 2C, right panel). The sizes of gains and losses are known to vary across cancer types and these differences may be attributable to different mutagenic processes.¹⁷ Pairwise comparisons of copy number aberration size distributions indicated that all three diseases exhibited distinct patterns of gains and losses (P<0.05, Kolmogorov-Smirnov test) (Figure 2D, E) with PTCL-EBV showing a propensity to smaller gains (~300 kb in size) and a unimodal distribution for losses peaking around 300 kb. To further examine the genomic complexities of the three disease groups, GI and HRD scores were calculated; the former based on the ratio of total length of regions with aberrant copy number and the latter on the loss of heterozygosity, telomere allelic imbalance and large-scale state transitions.¹¹ We observed that PTCL-EBV had significantly lower GI and HRD scores compared to ENKTL (P<0.001 [GI], P=0.004 [HRD], Mann-Whitney U) and PTCLNOS (P=0.0012 [GI], P=0.025 [HRD]) (Figure 3A, B). Despite these lower scores, ploidy levels were similar across all three groups, indicating that these results were not attributable to differential rates of whole genome duplication (Figure 3C). The GI score remained significantly lower in PTCL-EBV T-lineage (n=13) than in ENKTL T-lineage (n=9) cases, indicating that the lower GI score in PTCL-EBV compared to ENKTL is not related to lineage. Nevertheless, future studies with larger sample sizes will be necessary for a comparison of lineage effects across the disease groups. In line with previous reports describing TP53 alterations co-occurring with increased GI or HRD,¹⁸ we also observed that TP53 losses were associated with higher GI score (Online Supplementary Figure S3) and were less frequent in PTCL-EBV (7.1%) than in ENKTL (26.5%; n.s., Fisher exact test) and PTCL-NOS (31.4%; P=0.03) (Online Supplementary Table S5).

Using 14 publicly available Oncoscan datasets (Online Supplementary Table S7), we compared GI and HRD scores of other hematolymphoid neoplasms and solid cancers with those of PTCL-EBV, ENKTL and PTCL-NOS. Oncoscan datasets for T-cell lymphomas were unavailable. GI and HRD scores varied widely across cancer types, with the highest scores in solid cancers and the lowest in chronic myeloid leukemia and pediatric-type follicular lymphoma (Figure 3D). Within hematolymphoid malignancies, high-grade lymphomas, such as Burkitt-like lymphomas, large B-cell lymphomas, ENKTL and PTCL-NOS, had higher GI than low-grade malignancies. Compared to other aggressive B-cell lymphomas, PTCL-EBV demonstrated a remarkably low GI score (all P<0.003).

To explore the biological pathways associated with PTCLEBV, we investigated genes that were differentially expressed between PTCL-EBV and ENKTL (EBVvsENKTL) or PTCL-NOS (EBVvsNOS). After filtering non-consensus coding sequence probes, 244 and 95 differentially expressed genes (P<0.01, adjusted P<0.05) were identified in EBVvsENKTL and EBVvsNOS, respectively (Online Supplementary Table S8). Interestingly, all 95 EBVvsNOS differentially expressed genes overlapped with the 244 from EBVvsENKTL. As expected, unsupervised hierarchical clustering of these 244 genes revealed three distinct clusters with PTCL-EBV separated from ENKTL and PTCL-NOS (Online Supplementary Figure S4). To assess differentially expressed genes for over-representation of gene ontology terms and protein-protein interactions, genes differentially expressed in EBVvsENKTL and EBVvsNOS were independently submitted to STRING.¹⁹ Both sets of differentially expressed genes were enriched for numerous immunity-related processes (Online Supplementary Tables S9 and S10) and known protein-protein interactions (P<0.005) (Online Supplementary Figure S5). Subsequent analyses of known gene-gene interactions identified NFκB-associated genes BIRC3, NFKB1, TLR8 and CD27 – which are upregulated in PTCL-EBV – as central nodes in the differentially expressed gene networks (Figure 4). To corroborate differential expression data at the protein level, multiplexed immunofluorescence was performed (Figure 5, Online Supplementary Figure S6A) and revealed significantly upregulated expression of BIRC3 and p50 (NFKB1) in tumor (Online Supplementary Figure S6B, C) and non-tumor (Online Supplementary Figure S6E, F) cells of PTCL-EBV compared to ENKTL and PTCL-NOS. CD27 expression was significantly higher in tumor and non-tumor cells (Online Supplementary Figure S6D, G) of PTCL-EBV compared with ENKTL (both P<0.001) but not with PTCLNOS. Interestingly, the proportions of non-tumor cells which were double positive for CD27/BIRC3, CD27/p50, and triple positive for CD27/p50/BIRC3 within a single non-tumor cell, were also significantly higher in PTCL-EBV than in ENKTL and PTCL-NOS (Online Supplementary Figure S6A, H-K).

To further explore NFκB across EBVvsENKTL and EBVvsNOS, we performed gene set enrichment analysis using five curated sets of NFκB target genes (Online Supplementary Methods). We observed consistent and significant upregulation of NFκB target gene expression in PTCL-EBV compared to ENKTL and PTCL-NOS (Figure 6A). In addition, we found no significant difference in tumor content (Online Supplementary Figure S7A) and observed a rather homogeneous composition of the main immune components of the TME according to computational transcriptome deconvolution across the disease groups (Online Supplementary Figure S7B) using CIBERSORTx. This suggests that the upregulation of immune-related pathways in PTCL-EBV is unlikely to be related to tumor content or TME composition difference between the disease groups. Overall, these results indicate prominent immune pathway upregulation and NFκB activation in PTCL-EBV, suggesting the potential role of persistent NFkB signaling in the disease pathogenesis.

To further elucidate the biological pathways associated with GI, we correlated GI score to the expression of each gene across all samples, which resulted in a list of genes ranked by the Spearman rho. This list was submitted to gene set enrichment analysis to identify hallmark gene sets whose expression was associated with GI scores (Online Supplementary Table S11). Our top three gene sets – interferon_alpha_response (false discovery rate <0.001), interferon_gamma_response (false discovery rate <0.001) and IL6_JAK_STAT3_signaling (false discovery rate =0.001) (Online Supplementary Figure S8A-C) – displayed inverse correlations, indicating that these immune-related pathways are upregulated in PTCL-EBV and coincide with lower GI scores.

We previously reported that the expression of PD-L1 is higher in PTCL-EBV than in ENKTL.⁴ Given that both interferon_gamma_response (IFNγ) and STAT3 are able to induce PD-L1 expression at both gene and protein levels in cancers, including ENKTL,^20,21 we correlated the gene expression of IFNγ and the IL6_JAK_STAT3 pathway with PDL1 (CD274) to understand mechanisms driving PD-L1 upregulation in our disease groups. We observed a significant correlation between the gene expression of IFNγ (R=0.55, P<0.001) and IL6_JAK_STAT3 genes (R=0.79, P<0.001) with CD274 (Figure 6B, C). Similarly, we also assessed the association between NFκB activity and CD274 since NFκB can transcriptionally upregulate CD274 expression.^22,23 A significant correlation between the median expression of NFκB transcriptional target genes and CD274 was also observed (R=0.69, P<0.001) (Figure 6D). Overall, it is possible that the upregulation of PD-L1 in PTCL-EBV may be related to activation of IFNγ, IL6_JAK_STAT3 and NFκB.

Since ENKTL and PTCL-EBV are both associated with EBV, we compared EBER expression, tumor content (via Onco -scan), and EBV miRNA (via qPCR) in both diseases. There was no significant difference in EBER positivity (Online Supplementary Figure S9A) or tumor content (Online Supplementary Figure S9B) between the two diseases. Based on the gene expression of EBNA1, EBNA2, LMP1 and LMP2A by RT-PCR, the majority (9/13, 69%) of PTCL-EBV showed a type 2 EBV latency pattern, while four cases (31%) demonstrated a type 3 latency pattern (Online Supplementary Table S12). Interestingly, PTCL-EBV (n=9) displayed a widespread lower EBV miRNA expression compared to ENKTL (n=15) and clustered separately from it (Online Supplementary Figure S10). The expression of 32 of 42 (76%) EBV miRNA was significantly lower in PTCL-EBV (adjusted P<0.05, t-test) (Online Supplementary Table S13).

To better understand the potential transcriptional impact of this differential miRNA regulation, we correlated the expression of each differentially expressed EBV miRNA and its predicted targets. Given that miRNA negatively regulate target mRNA,²⁴ all target genes that were negatively correlated (adjusted P<0.05) to their corresponding EBV miRNA were analyzed (n=172) (Online Supplementary Table S14). Strikingly, the pathways most enriched (adjusted P<0.05) within this set of genes represented either immunity or interferon signaling (Online Supplementary Table S15). After calculating a gene expression index (median expression) for these target genes, we observed significantly higher expression in PTCL-EBV (P=0.03, Mann-Whitney U) (Online Supplementary Figure S11), which is consistent with lower EBV miRNA expression in this group. Overall, these results suggest that downregulation of EBV miRNA is unlikely to be related to a difference in tumor or EBER content and could, in part, contribute to the distinctive pattern of immune gene transcription in PTCL-EBV.

Based on the evidence that GI, gene expression, as well as EBV-miRNA patterns can delineate a distinct profile for PTCL-EBV, we further investigated whether this disease harbors mutations in known driver genes of PTCL-NOS, ENKTL and solid cancers using a 35-gene T/NK lymphoid panel and 484-gene NovoPM^TM 2.0 assay (total 500 genes with 19 common genes covered in both panels).

The most commonly mutated gene was TET2 (9/14, 64%) followed by PIK3CD (3/9, 33%), STAT3 (3/16, 19%), DDX3X (2/10, 20%) and PTPRD (2/11, 18%) (Online Supplementary Figure S12). The variant allele frequency ranges for TET2 in the NovoPM^TM 2.0 panel and T/NK lymphoid panel were 22.2%-40.7% and 22%-76%, respectively. Given the tumor purity of our samples and the high frequency of TET2 mutations and their associated variant allele frequencies, it is unlikely that these mutations are attributable to clonal hematopoiesis of indeterminate potential (CHIP).²⁵ However, in the absence of blood samples from these patients, we cannot entirely rule out CHIP as a possible source of TET2 mutation in some of our PTCL-EBV cases. Eight out of nine cases positive for TET2 mutation were positive for CD8, indicating that they were not PTCL with T-follicular helper phenotype. Interestingly, TP53 mutations, commonly present in ENKTL, were not detected in our PTCLEBV cases. The median number of mutations detected was 2.5 per sample (range, 1 to 11).

The tumor mutational burden score ranges from 0 to 7.86 mutations/MB (median 4.285 mutations/MB) (Online Supplementary Figure S13A). A tumor mutational burden of less than 5 is regarded as a low mutational burden in some studies.^26,27 Eleven samples tested had a microsatellite instability score below the threshold score of 0.4 (median 0.1882) and are regarded as being microsatellite stable (Online Supplementary Figure S13B).

Given that PTCL-EBV is characterized by a cytotoxic phenotype,^2,5 we attempted to determine whether PTCL-EBV shows similarities with cytotoxic PTCL-NOS. We performed multivariate survival analysis and compared the GI and HRD scores, between PTCL-EBV, PTCL-NOS cytotoxic (n=15) and non-cytotoxic (n=11) cases. Interestingly, we observed that PTCL-EBV was significantly more aggressive than cytotoxic PTCL-NOS (HR=0.22, 95% CI: 0.08-0.58, P=0.002) but not non-cytotoxic PTCL-NOS (Online Supplementary Table S16).

PTCL-EBV exhibited significantly lower GI scores compared to both cytotoxic (P=0.041) and non-cytotoxic PTCLNOS (P<0.001) (Online Supplementary Figure S14). HRD scores were lower in PTCL-EBV than in non-cytotoxic PTCL-NOS (P=0.021) but not cytotoxic PTCL-NOS. No significant difference was detected in survival, ploidy, and GI and HRD scores between cytotoxic and non-cytotoxic PTCL-NOS and there were no genes significantly differently expressed between the two. Overall, our results show a significant difference in overall survival and GI score between PTCL-EBV and cytotoxic PTCL-NOS, suggesting biological differences between these two tumors.