IHC algorithm reliably predicts GATA3 and TBX21 subtypes of PTCL-NOS

Peripheral T cell lymphomas (PTCL), are a heterogeneous group of non-Hodgkin diseases with broad morphological and immunophenotypic characteristics, and a poor prognosis.^1,2 Despite the recently updated World Health Organisation (WHO) classification, many of the subsets remain undefined and are grouped into a PTCL-not otherwise specified (PTCL-NOS) subtype.¹ Better characterisation of this subgroup would guide more targeted treatment and could improve patient outcomes. Therefore, researchers aim to identify the clinical and pathologic features that could offer prognostic value to the different entities within PTCL-NOS.

Recently, gene expression profiling (GEP) has identified two novel subtypes with different clinical outcomes, likely derived from two distinct subtypes of T-cells.^3,4,5 The PTCL-GATA3 subtype, representing 33% of PTCL-NOS, is characterised by overexpression of GATA3 and downstream genes, and is associated with a worse prognosis.³ The PTCL-TBX21 subtype, representing 49% of PTCL-NOS cases, has a better prognosis and is characterised by overexpression of TBX21 and its target genes.³ Moreover, these subtypes were found to be associated with enrichment of distinct oncogenic pathways, PI3K-mTOR activation in PTCL-GATA3 and NF-κβ in PTCL-TBX21, and are therefore vulnerable to distinct targeted therapies.^3,6,7

Although the stratification of PTCL-NOS patients has clinical utility, routine use of GEP in clinical practice is not considered feasible due to high costs and a lack of accessibility. Catalina Amador from the University of Nebraska Medical Center, US, and colleagues investigated whether using standard immunohistochemistry (IHC) on formalin-fixed, paraffin-embedded (FFPE) tissue could replicate the gene expression diagnostic signatures for routine use in clinical practice.⁸ The study results were recently published in Blood.

Methods⁸

In total, 173 cases of PTCL-NOS from multiple institutions were included in the study (49 cases in the training cohort previously evaluated by the GEP and 124 cases in the validation cohort without GEP data). All cases had data on at least two of the T-follicular helper (TFH) markers such as PD1, CXL13, ICOS, CD10, and BCL6. Cases expressing CD4 and at least two TFH markers were excluded from the study.

In order to develop the IHC algorithm, staining was performed on tissue microarrays (TMAs) using the following markers:

• GATA3 and CCR4 - for PTCL-GATA3 subtype
• TBX21 and CXCR3 - for PTCL-TBX21 subtype

Pathologists blinded to the GEP results assessed immunostaining and set the threshold of positivity. Based on multiple covariates of a percentage of positive tumor cells, an algorithm was used to assign cases to GEP-defined subtypes.

Additionally, 57 cases were also assessed for morphological features and positivity for CD4, CD8, CD30, cytotoxic markers staining and Epstein-Barr encoding region (EBER) in situ hybridisation.

Results

• Patients in the training cohort were older than in the validation cohort. Other clinical characteristics were similar (Table 1)
• GATA3, CCR4, TBX21, and CXCR3 showed a significant positive correlation between the percentage of positive cells and the corresponding mRNA expression levels
• In the training cohort, 59% of cases were positive for GATA3, 60% for CCR4, 47% for TBX21 and 57% for CXCR3, while 6% were negative for all immunostains
• There was no correlation between GATA3 and TBX21 expression
• GATA3 and CCR4 showed bi-modal distribution of staining, while TBX21 and CXCR3 had a positively skewed distribution
• Individual staining was associated with a corresponding molecular subtype, but the expression of proteins associated with GATA3 and TBX21 phenotypes was not mutually exclusive
• Double expression of GATA3 and/or CCR4, plus TBX21 and /or CXR3 was observed in 32% of cases and was more common (p=0.03) in the PTCL-TBX21 subtype (38%) compared to PTCL-GATA3 subtype (7%)
• The optimal cut-off threshold of immunostaining positivity was 50% tumor cells for GATA3 and CCR4, and 20% for TBX21 and CXCR3

Table 1. Patient characteristics by cohort IPI, International Prognostic Index

 

Training cohort
(n=49)

Validation cohort
(n=124)

p-value

IHC classification

    PTLC-GATA3

    PTLC-TBX21

    Unclassified

n=49

    15 (31%)

    31 (63%)

    3 (6%)

n=124

    46 (37%)

    69 (56%)

    9 (7%)

0.66

 

Gender

    Female

    Male

n=42

    17 (40%)

    25 (60%)

n=100

    35 (35%)

    65 (65%)

0.54

Age (years)

    ≤ 60

    > 60

n=39

    13 (33%)

    26 (67%) 

n=98

    52 (53%)

    46 (47%)

0.04

Stage

    I/II

    III/IV

n=17

    2 (12%)

    15 (88%)

n=73

    20 (27%)

    53 (73%)

0.22

Extra-nodal sites

    ≤ 1

    > 1

n=6

    4 (67%)

    2 (33%)

n=74

    54 (73%)

    20 (27%)

0.66

IPI score

    Low (0–2)

    High (3–5)

n=13

    5 (38%)

    8 (62%)

n=63

    36 (57%)

    27 (43%)

0.24

Treatment

    CHOP/CHOP-like

    Other

     None

n=7

    4 (57%)

    2 (29%)

    1 (14%)

n=63

    55 (87%)

    7 (11%)

    1 (2%)

0.07

 

• Including a second marker for each subtype improved the accuracy of classification by 8% in PTCL-GATA3 cases and 19% in PTCL-TBX21 cases when compared to GATA3/TBX21 alone
• The IHC algorithm had successfully classified 85% of the GEP-defined cases compared to the molecular subclassification (5% of the cases were unclassified)
• Similarly to the GEP classification, the IHC algorithm also predicted differences in overall survival (OS) between the subtypes, with PTLC-GATA3 having a worse prognosis in the training and validation cohorts (p=0.03 and p=0.01, respectively)
• Of the nine cases previously unclassified by GEP, one remained unclassified by the IHC algorithm, six were classed as PTCL-TBX21, and two as PTCL-GATA3
• The algorithm showed a high interobserver reproducibility
• PTCL-GATA3 IHC classification, high IPI, older age, and more than one extranodal site were associated with significantly increased risk of death (Table 2)
• The majority of PTCL-GATA3 cases had a high abundance of tumor cells with low inflammatory background and presented as a monotonous pattern of sheets of medium-size cells, with abundant cytoplasm or large clusters/sheets of large cells
• Most PTCL-TBX21 cases had a polymorphous appearance, with inflammatory cell infiltrates or a mixture of atypical small tumor cells and epithelioid histiocytes organized in a Lennert lymphoma pattern
• Monomorphic patterns seen in PTCL-GATA3 and Lennert pattern in PTCL-TBX21 were associated with reduced OS
• CD4+/CD8- immunophenotype was most frequent in PTCL-GATA3 subtype (79% of cases)
• The expression of cytotoxic markers and CD8+ was mainly a feature of PTCL-TBX21 tumors, with 54% and 32% of cases compared to 11% and 5% in PTCL-GATA3 respectively (p<0.001, p=0.04). However, the presence of a cytotoxic phenotype did not appear to have an impact on OS
• There were no differences in frequency of EBER positivity and CD30 expression between the subtypes

Table 2. A univariate and multivariate models of OS

 

Univariate analysis

Multivariate analysis

 

n

HR (95% CI)

p-value

n

HR (95% CI)

p-value

PTCL-GATA3 vs. PTCL-TBX21 by IHC

128

2.39 (1.55–3.7)

<0.0001

67

2.75 (1.51–5.01)

0.0009

High IPI vs. low IPI

74

2.13 (1.21–3.74)

0.0089

67

1.8 (1.0–3.24)

0.05

Age > 60 vs. ≤ 60 years

124

1.95 (1.27–3.01)

0.0025

 

 

 

Extra-nodal sites > 1 vs. ≤ 1

77

1.85 (1.04–3.29)

0.035

 

 

 

Conclusion

Using four commercially available antibodies, the IHC algorithm reliably predicted GATA3 and TBX21 subtypes of PTCL-NOS. Such subclassification of patients could be incorporated into patient management and guide decisions on the treatment regimen according to the underlying biology. Based on current knowledge, PI3K inhibitors may show efficacy in patients with a PTLC-GATA3 subtype due to the predominance of PI3K-mTOR activation pathways, while immunomodulators and NF-κβ inhibitors may be recommended for patients with PTCL-TBX1 due to an enrichment of NF-κβ activation pathways in this subtype.