Metabolic response prediction to ABVD from diagnostic biopsy analysis

Classical Hodgkin lymphoma (cHL) is generally treated with a combination of doxorubicin, vinblastine, vincristine, and dacarbazine (ABVD) or with bleomycin, etoposide, doxorubicin, cyclophosphamide, vincristine, procarbazine, and prednisolone (BEACOPP). Although both regimens result in similar cure rates, ABVD is associated with a higher rate of relapse and BEACOPP is associated with more severe toxicities.¹ In order to improve treatment efficacy, metabolic response has been shown to be predictive of response, enabling treatment adaptation to reduce toxic side effects or treatment intensification, as required.¹ Following initial treatment with ABVD or BEACOPP, interim positron emission tomography (iPET) is used to determine treatment response. Using genetic profiling to assess pre-treatment biopsies, a group based in Italy aimed to study the biological basis of iPET metabolic response following ABVD treatment and to identify a genetic signature capable of detecting ‘chemorefractoriness’ at diagnosis.¹

Stefano Luminari from the Hematology Unit, AUSL-IRCCS, Reggio Emilia, IT and colleagues, recently published the findings from their gene expression profiling study in Clinical Cancer Research.¹ Untreated patients with stage I–IV cHL, who underwent iPET after two courses of ABVD were recruited to the study consecutively. The study used a cohort of patients to generate a gene expression-based model (n= 121) and a different cohort to validate it (n= 117). Of these, iPET images were available for 120 and 111 patients for the training and validation cohorts, respectively. Gene expression levels of 770 genes from 24 different immune cell types were assessed. In each cohort, 23 patients with unfavorable iPET results (iPET+) were identified.

Key findings

Multivariate analysis of clinical features found a significant association only between the lymphocyte:monocyte ratio (LMR) and iPET+ in the training cohort.

After quality checks, gene expression profiles of 106 training samples were analyzed, and of these 21 were iPET+ and 84 were iPET- (one patient had no iPET images):

• Differential analysis of these identified 241 deregulated genes
• Gene ontology analysis of deregulated genes showed an enrichment of immune response, inflammation, and cell migration associated genes
• Further stringent filtering resulted in a 13-gene signature associated with iPET+ (Table 1) and included genes involved in immune modulation, cell movement, wound healing, as well as blood vessel organization and formation (Table 1)
• Further analysis of correlating genes and multivariate logistic regression analysis refined the final iPET predictive model to include five genes (ITGA5, SAA1, CXCL2, SPP1, and TREM1) plus LMR (iPET predictive score; Table 1)
• When applied to the training cohort, the model was able to accurately segregate iPET+ from iPET- (ROC analysis: AUC= 0.88; 95% CI, 0.80–96)
• Using the iPET predictive model, samples were separated into quartiles(Q1–4), with 76.2% of iPET+ patients being allocated to Q4, which suggests that the model has an accurate discriminatory capacity
• Assessing the model and how it compared to the Deauville five-point scale, all patients who would be scored as DS5 (iPET+) were positive according to the gene expression model
• Further analysis of clinical features at diagnosis revealed a correlation between iPET predictive score and SUVmax at diagnosis

In the validation cohort, 82/117 samples from patients with cHL passed quality control and went on to gene expression profile analysis. Of the patients included, 14 were iPET+ and the remaining 68 were iPET-.

• ROC analysis in the validation group achieved an AUC of 0.68 (95%CI, 0.52–0.84)
• There was no significant difference in gene expression between the training and validation of iPET+ groups
• iPET predictive score was consistently higher in iPET+ than iPET- in the validation group (p= 0.03)

Combining the training and validation cohorts, the team went on to assess any association of iPET predictive score with treatment failure (TF). TF was defined as the change of therapy after iPET+, lack of metabolic response after final PET, or progressive disease (whichever came first)

• Only patients who had at least a three-year follow-up and an iPET predictive score (n= 115) were included
• iPET predictive score was significantly higher in patients with TF

Table 1. Thirteen genes associated with iPET+ phenotype of cHL, alongside correlating clinical variable, LMR

LMR: lymphocyte:monocyte ratio, * genes with a significant p value that were included in the final model, † genes excluded from multivariate analysis due to correlation analysis

Gene name

Fold change

p value

False discovery rate

Multivariate p value

Multivariate p value including LMR

VEGFA

2.02

1.566 X 10^-7

0.0008

0.504

0.649

PLAU

2.07

4.85 X 10^-6

0.008

0.972

0.5

THBS1

2.35

1.004 X 10^-4

0.055

0.899

0.952

ITGA5

2.67

1.294 X 10^-4

0.062

0.034*

0.023*

SAA1

5.8

1.701 X 10^-4

0.063

0.139

0.072*

FN1†

3.92

1.553 X 10^-4

0.063

-

-

LRP1

2.22

2.053 X 10^-4

0.063

0.202

0.1

CXCL2

2.90

2.347 X 10^-4

0.065

0.018*

0.008*

CCL18

3.45

4.512 X 10^-4

0.082

0.793

0.665

SPP1

3.35

5.039 X 10^-4

0.082

0.094*

0.06*

CD9†

2.34

4.894 X 10^-4

0.082

-

-

CXCL3

2.98

6.138 X 10^-4

0.090

0.656

0.618

TREM1

5.29

7.439 X 10^-4

0.099

0.160

0.079*

LMR

-

-

-

-

0.055*

Conclusion

In their conclusion, the team highlighted how their iPET predictive score is the first study to identify markers to predict chemorefractoriness at diagnosis. During the study, they demonstrated that many clinical and laboratory parameters were not able to predict chemorefractoriness at diagnosis, with the exception of LMR. The authors went on to discuss the genes in their 13 gene signature model, which is largely composed of genes involved in stromal interactions, in keeping with the view that the microenvironment plays a dominant role in cHL. The genes included encode molecules involved in several oncogenic pathways, such a metastasis, angiogenesis, and cell movement. With gene ontology analysis, the genes fall into two nodes, one being mostly pro-inflammatory cytokines, and the other being molecules involved in cell movement and matrix organization. Luminari et al., went on to compare their work with another study by Dr Scott and colleagues² in which a 23-gene signature was identified that could predict overall survival in cHL patients. Comparing this 23-gene signature with the thirteen genes identified here, there was very little overlap that the authors claim to be due to different starting panels, and different research questions.

In terms of limitations, the main concern highlighted by the authors is around reproducibility, and they also mention the limited patient numbers from only two centres, resulting in a low number of iPET+ groups that could prevent generalization of the results. They advocate the further validation of this iPET predictive tool (five genes plus LMR) in larger populations.