This article is more than 5 years old. Information may no longer be current.
Gene expression patterns identify high-risk chronic lymphocytic leukemia
Click Here to Manage Email Alerts
A 290-gene expression signature and IGHV mutation status stratified patients with chronic lymphocytic leukemia to identify those with high-risk disease who might benefit from prompt initiation of therapy, according to a study published in Frontiers in Oncology.
Although CLL treatment is typically delayed until disease progression, it is uncertain whether patients would benefit from treatment immediately following diagnosis, when they have a smaller tumor mass and are in better physical condition.
Researchers have discovered new drivers of CLL, and RNA sequencing offers an opportunity to identify additional biomarkers that might predict disease progression and treatment benefit in order to better risk-stratify patients, according to the researchers.
“In fact, previous efforts to improve CLL risk stratification based on RNA sequencing data have demonstrated impressive results, but the clinical application is difficult due to the expense of extensive technical and bioinformatics efforts,” Adrián Mosquera Orgueira, MD, hematology resident of cancer genomics research and bioinformatics at Health Research Institute of Santiago de Compostela, and colleagues wrote. “Therefore, there is a need for smaller transcriptomics patterns correlated with disease evolution for medical use.”
Orgueira and colleagues used RNA sequencing data from the International Cancer Genome Consortium CLL cohort to identify gene expression trends that are associated with clinical disease evolution and time to first treatment. The analysis included a test cohort of 196 patients (median age at diagnosis, 63 years; 60.7% men) and a validation cohort of 79 patients (median age at diagnosis, 62 years; 69.62% men).
The researchers created a Cox regression model with gene expression, donor sex and CLL stage at diagnosis as independent variables. They found 2,198 genes significantly correlated with time to first treatment.
After adjusting for IGHV status, researchers isolated 19 gene sets that clustered samples into two groups that appeared significantly linked to time to first treatment (Bonferroni-adjusted P < .01).
Cluster 2 — which researchers identified as the most prognostic cluster — contained 290 transcripts and showed a highly significant association with time to first treatment when controlling for IGHV mutation status (P = 6.4 x 10-7). This association persisted in the validation cohort (P = 3.05 x 10–3).
Patients in cluster 2 — which included 36.7% of patients in the test cohort and 34.1% in the validation cohort — demonstrated a more favorable prognosis than those in cluster 1. Cluster 2 comprised 51.5% of IGHV-mutated patients and 6.4% of IGHV-unmutated patients from the test cohort, and 55.5% of IGHV-mutated patients and 5.8% of IGHV-unmutated patients from the validation cohort.
Using this 290-gene expression signature and IGHV mutation status, researchers stratified patients into four groups — defined by a high- or low-risk transcriptomic profile with or without mutated IGHV — with markedly different time to first treatment.
Patients with mutated IGHV and a low-risk transcriptomic profile only needed treatment in approximately 25% of cases during disease evolution. Patients with mutated IGHV and a high-risk transcriptomic profile, as well as those with unmutated IGHV and a low-transcriptomic profile, demonstrated an intermediate risk for evolution. However, patients with unmutated IGHV and an adverse transcriptomic profile had the highest likelihood of needing treatment in the first year following diagnosis.
Researchers then sought to develop an artificial intelligence algorithm that would predict a patient with CLL’s need for therapy during the first 5 years following diagnosis. To do so, they created models with all genes associated with time to first treatment with a false discovery rate of 5%, which they tested in 222 patients who had a follow-up of at least 5 years or who had been treated in the first 5 years following diagnosis.
A model with a 2.5% learning rate showed 90% precision and 69.2% recall in determining which patients would need treatment in 5 years. This model demonstrated 88.57% precision in identifying patients who didn’t need treatment in 5 years with 96.88% recall.
“Here we present patterns of gene expression that can improve CLL patient risk stratification with a relatively small set of the transcriptome,” the researchers wrote. “These results may pave the way for the design of new treatment strategies involving early CLL treatment in high-risk patients before disease progression.” – by Jennifer Byrne
Disclosures: The researchers report no relevant disclosures.
Perspective
Back to Top
Brian Hill, MD, PhD
Time to treatment in CLL can be predicted by a number of readily available clinical and molecular features of the disease, including stage (by Rai or Binet systems), mutational status of IGHV, and chromosome copy number alterations as detected by standard fluorescent in situ hybridization panels. Orgueira and colleagues measured the gene expression level of 290 genes using RNA sequencing from baseline peripheral blood samples to classify large groups of patients with CLL into two clusters (C1 and C2).
In two separate cohorts, the authors found a dramatic difference in time to treatment (P = 10-4 to 10-6) between the two clusters. Interestingly, even within the high-risk cluster (C1), unmutated IGHV status was still associated with higher-risk disease than mutated IGHV. There were very few patients in the C2 group who had unmutated IGHV.
Also, of note, after approximately 10 years, about 75% of patients in C2 still had not needed treatment, and this level appeared to plateau with longer follow-up, suggesting that this patient population may contain a group of patients who will never require therapy.
The current study highlights the clinical heterogeneity of CLL, even within patients with high-risk (unmutated) IGHV. However, the clinical feasibility of performing RNA sequencing to risk-stratify patients with CLL at the time of diagnosis is unlikely to be practical outside of observational clinical trial protocols.
Click Here to Manage Email Alerts