Baseline gene expression in patients with chronic myeloid leukemia (CML) who received tyrosine kinase inhibitor (TKI) therapy in the phase 3differentiated those who achieved a good response from those with a poor response at 5 years in an exploratory analysis.
The investigators developed gene-expression models based on RNA sequencing of whole blood samples collected prior to treatment with nilotinib or imatinib in study participants who completed at least 5 years of therapy, including both good responders – those who achieved a major molecular response (MMR), defined as BCR-ABL1IS (a gene sequence found in an abnormal chromosome 22) less than 0.01% by 12 months and sustained deep molecular response (DMR) by 5 years, and poor responders – those without MMR by 12 months or with BCR-ABL1IS greater than 10% at 3 months.
A model based on the comparison of gene signatures from 47 patients who achieved a molecular response of 4.5 (MR4.5) on the International Scale (BCR-ABL1S less than 0.00032%), compared with 23 patients with a poor response, best predicted 5-year responder status (area under the receiver operating characteristic curve, 0.76),, reported during the Society of Hematologic Oncology virtual meeting.
“For this kind of work, that’s really quite good,” said Dr. Radich of the clinical research division at Fred Hutchinson Cancer Research Center, Seattle.
Notably, the differences in patient responses observed by 12 months in ENESTnd persisted for up to 10 years, he said.
The findings have potential implications for drug development and facilitation of DMR in patients on TKI therapy – a prerequisite for attempting treatment-free remission, he said.
Dr. Radich and colleagues assessed 24 clinical factors – such as Sokal risk score, TKI therapy type, age, and sex – according to responder status, and applied penalized regression to the clinical variables, to expression of 13,575 genes, and to a combination of the clinical variables and gene expression.
Clinical variables didn’t predict response in the trial, and including the clinical variables in the gene-expression model in the exploratory analysis did not improve it’s performance (AUC, 0.75). However, both the MR4.5 plus clinical variables model and the MR4.5-only model outperformed the clinical variables–only model (AUC, 0.59), he noted, adding: “So gene expression seems to be highly correlated with response.”
Of note, 458 genes were differentially expressed; those found in responders were most often associated with immune response, whereas those in poor responders were more likely to be associated with drug catabolism, WNT signaling, and cell cycle.
This suggests that good responders, compared with poor responders, have an activated immune system that is better able to engage after TKI therapy is administered to “cull through the heard, so to speak,” Dr. Radich said.
The findings were validated in an independent dataset of 19 good responders and 25 poor responders (AUC, 0.67 for the MR4.5 vs. poor-responder model).
A comparison of the expression of immune cell marker genes in good responders and poor responders further showed that T cells – particularly CD8 T cells, B cells, natural killer cells, and aggregate cytotoxic lymphocytes were expressed at significantly higher levels in good responders.
This was true in both the ENESTnd cohort and the validation dataset, he said.
Theis a randomized, open-label study comparing nilotinib and imatinib in adults with newly diagnosed Philadelphia chromosome–positive chronic-phase CML. A 5-year published in 2016 showed that 54% and 52% of patients in nilotinib 300- and 400-mg twice-daily arms, respectively, achieved MR4.5, compared with 31% of those in an imatinib 400-mg once-daily arm. In the current exploratory analysis, the gene expression model differentiated between good and poor responders regardless of the TKI used, Dr. Radich said.
The findings are of note because achieving sustained deep molecular response is necessary before CML patients can attempt treatment-free remission and because biomarkers for predicting DMR have been lacking, he explained.
“These findings could really be used, potentially, for a couple of things: One is to predict response, and that could drive patient goals, expectations, and maybe drug choice,” he said.
The findings could also be used to inform clinical trials to investigate how to best treat poor responders to improve their response, he added.
“I think there’s a lot of work to be done and a lot things to chew over, and we’re hoping that we’ll have more to talk to you about in the future,” he said.
The study was sponsored by Novartis. Dr. Radich is a paid consultant for Genentech, Cepheid, Bristol-Myers Squibb, Takeda, and Novartis.
SOURCE: Radich JP et al. SOHO 2020,