Page 128 - Haematologica - Vol. 105 n. 6 - June 2020

F. Drieux et al.
cal clustering performed on 153 such cases (30 AITL, 33 TFH-PTCL, 16 NKTCL, 13 ATLL, 6 HSTL, 21 ALK-positive ALCL, and 34 ALK-negative ALCL) identified two main branches separating cytotoxic and non-cytotoxic entities (Figure 1).
The cytotoxic branch is divided into two clusters, one very homogeneous cluster (C5) comprising the 16 NKTCL, and a second cluster (C6) composed of 31 cyto- toxic ALCL (21 ALK-positive and 10 ALK-negative). The other branch is divided into four clusters (C1-4). The C1 cluster contained the six HSTL. The 63 PTCL with a TFH phenotype (AITL and TFH-PTCL) distributed along the two clusters C2 and C3. The C2 cluster comprised a major group of AITL/TFH-PTCL characterized by a TFH signature (C2, n=50). In addition to 12 ATLL, 13 TFH-PTCL and one AITL in the C3 cluster showed an enrichment in TFH and Th2 markers. Interestingly, RHOA mutations were identi- fiedin26of50(52%)and2of13(15%)oftheC2andC3 clusters, respectively (P=0.027). By immunohistochem- istry, these TFH-PTCL in the C3 cluster showed expression of GATA3 (in more than 50% of tumor cells) in 9 of 12 (75%) contributive cases (Figure 2). The C4 cluster con- tained 24 ALK-negative non-cytotoxic ALCL with Th2 signature, with 8 of 16 contributive cases showing DUSP22 rearrangement by FISH.
PTCL-NOS distribute among distinct clusters using unsupervised clustering
When applied to all 230 PTCL samples (including 77 PTCL-NOS), unsupervised clustering showed that the majority of PTCL-NOS (n=48 of 77, 62.3%) clustered within four of the six previous clusters as they showed gene signatures in common with molecular TFH-PTCL (C2, n=6), TFH/Th2 PTCL (C3, n=19), NKTCL (C5, n=5), and cytotoxic ALCL (C6, n=18) (Figure 3). Despite a variable expression of CD30 by immunohistochemistry, 18 PTCL- NOS distributed within the cluster of cytotoxic ALCL based on the expression of cytotoxic markers plus TBX21 and IFN, consistent with a possible Th1 origin. Accordingly, 8 of 12 of these cases tested for TBX21 by immunohistochemistry were positive. This molecular subgroup is further referred to as “cytotoxic/Th1 PTCL” according to its signature. In addition, 19 other PTCL- NOS cases, all with negative HTLV-1 serologies, clustered with ATLL, based on an enrichment in Th2 molecules GATA3 and CCR4, and are referred to as molecular “Th2 PTCL”. Finally, 29 PTCL-NOS did not cluster within any of the defined branches, and segregated with 19 other cases (4 AITL, 12 TFH-PTCL, 1 ATLL).
Support vector machine classifier accurately classifies specified PTCL and identifies subgroups within PTCL-NOS
We next built a support vector machine (SVM) model to assign each case to a class based on the RT-MLPA data (Figure 4A). One hundred and eighty-four PTCL corre- sponding to the molecular groups defined according to the latter clustering (Figure 3) were used for the construc- tion of the model and to define the molecular classes: 45 TFH-PTCL/AITL, 21 NKTCL, 42 Th2 PTCL-NOS/ATLL, 50 cytotoxic-ALCL, 20 non-cytotoxic ALCL, and the six HSTL. The 46 PTCL which did not cluster within these six defined branches were not considered to develop the classifier. The SVM algorithm accurately assigned 140 of 153 specified PTCL to the correct categories: 16 of 16
Table 2. Clinical, pathological and molecular features of the two sub- groups of TFH-peripheral T-cell lymphomas (PTCL).
Clinical data
Age median (range)
IPI>=3
PIT>=2
Extranodal site>=2 Stage>=3
PS>=2
LDH>1N
B signs Hypergammaglobulinemia Coombs
Anemia
Cutaneous rash
BM
Complete response
OS median (range)
PFS median (range)
Pathological data
Clear cells B blasts
EBV positivity
EBV extent >5 large blast-cells/high power field
Molecular data
TET2 mutation DNMT3A mutation
RHOA mutation IDH2 mutation
TFH signature n=50
64.4 (54-74.6)
71.8% (28/39) 71% (27/38) 50% (22/44) 100% (44/44) 40% (16/40) 64% (25/39) 70.7% (29/41) 37.5% (12/32) 46% (12/26) 61.5% (24/39) 23.8% (10/42) 48.8% (21/43) 50% (20/40) 22 (5.5-77) 10 (3-39)
65.2% (30/46) 90% (44/49)
70.2% (33/47)
29.5% (13/44)
60.6%% (20/33) 25% (8/32)
52% (26/50)
20% (10/50)
TFH and TH2 P signatures n=13
67.4 (56.7-74.7) 0.79
75% (9/12) 1 63.6% (7/11) 0.72 58.3% (7/12) 0.75 91.7% (11/12) 0.21
8.3% (1/12) 0.076 50% (6/12) 0.5 45.5% (5/11) 0.16 12.5% (1/8) 0.24
0% (0/2) 0.49 41.6% (5/12) 0.32 33% (4/12) 0.48 33.3% (4/12) 0.51 41.7% (5/12) 0.75 30.5 (6-50.5) 0.91 12 (5.5-38) 0.42
36% (4/11) 0.1 66% (8/12) 0.07
58.3% (7/12) 0.5
0% (0/12) 0.049
14.3% (1/7) 0.039 0% (0/6) 0.31
15.4% (2/13) 0.027 0% (0/13) 0.1
IPI: International Prognostic Index; EBV: Epstein-Barr virus; PIT: Prognostic Index for PTCL-not otherwise specified; PS: Performance Status; LDH: lactate dehydrogenase; BM: bone marrow; OS: overall survival; PFS: progression-free survival.
NKTCL, 13 of 13 ATLL, 6 of 6 HSTL, 31 of 31 cytotoxic ALCL, 24 of 24 non-cytotoxic ALK-negative ALCL cases, and 50 of 63 AITL/TFH-PTCL. Interestingly, 11 TFH-PTCL without RHOA mutation were classified as molecular Th2 PTCL, one as molecular cytotoxic/Th1, and one AITL was distant from the barycenter of the TFH/AITL class and not classified. Of the 77 remaining PTCL-NOS, 70 (91%) were classified by the SVM as TFH/AITL (C2, n=17), cytotoxic/Th1 PTCL (C6, n=19), ALK-negative ALCL (C6, n=5), NKTCL (C5, n=5), while 24 cases were molecularly designated Th2 PTCL (C3). Finally, seven cases, which were distant to the barycenter of their pre- dicted SVM class (2 Th2, 3 TFH/AITL, 2 cytotoxic/Th1), could not be attributed a molecular class by the SVM. Figure 4B illustrates the subgroups of PTCL-NOS as evi- denced in the principal component analysis (PCA). A cor- relation of the SVM class with the histopathological data of the 77 PTCL-NOS is presented in Online Supplementary Figure S7.
1586
haematologica | 2020; 105(6)