Severe Chemotherapy Side Effects in NSCLC Predicted

Severe Chemotherapy Side Effects in NSCLC Predicted

Whole genome screening (WGS) and gene network modules were used to predict non-small cell lung (NSCLC) patients’ risk of developing myelosuppression in a study by researchers at Linköping University, Sweden. Blood samples from 96 patients treated for NSCLC with gemcitabine/carboplatin were analyzed to find several thousand SNVs/INDELS associated with the three main types of myelosuppression –neutropenia, leukopenia, and thrombocytopenia. Several more analytical steps condensed those into a toxicity module. That set of SNVs/INDELS was trained and tested, respectively, on two groups of the original samples to create a prediction model comprising 62 SNVs/INDELs.

This study was published in npj Systems Biology and Applications and is available online.

“It’s extremely interesting that the genes involved are associated with cell division, in particular in bone marrow. We managed not only to predict side effects for the patients, but also show that the model is biologically relevant”, says Henrik Gréen, co-leader of the study, and professor at the Department of Biomedical and Clinical Sciences, Linköping University.

Although PD-1 inhibitors or targeted therapies are the current first-line treatment for NSCLC, classic chemotherapy consisting of gemcitabine and carboplatin is still the backbone of systemic cancer treatment and is likely to remain so for a while. About 1 million patients receive chemotherapy each year in the US alone.  Many of them experience grade 3-4 myelosuppression consisting of neutropenia, leukopenia, and/or thrombocytopenia. These side effects can delay treatment, halt it altogether, or even lead to death. As a result, being able to predict which patients are at highest risk of such effects, so that dosing can be adjusted, would help oncologists better personalize cancer chemotherapy.

The Linköping team first used PLINK, to find 4594, 5019, and 5066 autosomal SNVs/INDELs with p ≤ 1 × 10−3 associated with neutropenia, leukopenia, and thrombocytopenia, respectively. Based on those, they next devised a toxicity module consisting of 215 unique overlapping genes inferred from MCODE-generated gene network modules of 350, 345, and 313 genes, each associated with one of the types of myelosuppression. This network was particularly rich in genes already associated with exposure to these drugs in earlier studies. These included differentially expressed genes in rat bone marrow, human bone marrow, and human cell lines exposed to carboplatin and gemcitabine (p < 0.05).

The final step in creating the model was to use 80% of the patients as a training data set, and apply random LASSO to reduce the number of SNVs/INDELs in the toxicity module down to 62 that accurately predict myelosuppression toxicity in both the training and the test (remaining 20%) data with high (CTCAE 3–4) and low (CTCAE 0–1). This model can be used to classify patients into one of two groups – those with high or low probability of experiencing severe side effects.

The authors say their prediction model must be tested in further studies before it can be used in the clinic. But they also note that increasingly advanced methods of genetic analysis are being introduced into the Swedish medical care system, which makes it more likely that this type of personalized prescribing tool will be used by clinicians in the future.

“We want to work towards establishing a standard within translational bioinformatics, and show that the same type of method can be applied in several medical situations. The patient material here may appear to be small, but we have even so demonstrated that this approach can be used to predict the severity of side effects for patients,” said Mika Gustafsson, the study’s co-leader and senior lecturer in the Department of Physics, Chemistry and Biology at Linköping University.