Tracking Neoantigens by Personalized Circulating Tumor DNA Sequencing during Checkpoint Blockade Immunotherapy in Non-Small Cell Lung Cancer

Abstract

The evolutionary dynamics of tumor-associated neoantigens carry information about drug sensitivity and resistance to the immune checkpoint blockade (ICB). However, the spectrum of somatic mutations is highly heterogeneous among patients, making it difficult to track neoantigens by circulating tumor DNA (ctDNA) sequencing using "one size fits all" commercial gene panels. Thus, individually customized panels (ICPs) are needed to track neoantigen evolution comprehensively during ICB treatment. Dominant neoantigens are predicted from whole exome sequencing data for treatment-naïve tumor tissues. Panels targeting predicted neoantigens are used for personalized ctDNA sequencing. Analyzing ten patients with non-small cell lung cancer, ICPs are effective for tracking most predicted dominant neoantigens (80-100%) in serial peripheral blood samples, and to detect substantially more genes (18-30) than the capacity of current commercial gene panels. A more than 50% decrease in ctDNA concentration after eight weeks of ICB administration is associated with favorable progression-free survival. Furthermore, at the individual level, the magnitude of the early ctDNA response is correlated with the subsequent change in tumor burden. The application of ICP-based ctDNA sequencing is expected to improve the understanding of ICB-driven tumor evolution and to provide personalized management strategies that optimize the clinical benefits of immunotherapies.

Keywords: ctDNA sequencing; immune checkpoint blockade; neoantigens; non‐small cell lung cancer; personalized medicine.

Figure 1

Somatic mutation profiles of pre‐treatment tissues and matched blood samples for ten patients with non‐small cell lung cancer (NSCLC). a) Left, heatmap showing inter‐individual overlap in the mutation spectrum for cases in the NSCLC dataset of TCGA. The percentage of overlap was calculated as the ratio between the numbers of intersected and union genes. Color bar, percentage of overlap. For clear visualization, 50 patients randomly selected from 1059 cases are shown. Right, histogram showing the distribution of percentages of overlap for all patients with NSCLC in TCGA. Blue line, fitting curve. b) Mutant genes detected in more than one patient in the cohort are illustrated. Total mutation burden (per MB) and clinical information are annotated in the upper two panels. Gene names are labeled on the right. The substitution spectrum and composition of clonal/subclonal mutations for all detected mutations are shown in the lower two panels. c) The heatmap shows an overview of individually customized panels (ICPs) and follow‐up ctDNA sequencing for each patient. The left column represents all genes in the ICP; the middle column summarizes ctDNA sequencing results for baseline samples; the right column indicates genes for which ctDNA could be detected in the following assays at least once. Blue square, genes detected over the duration of the treatment; red square, genes never detected in any following ctDNA sequencings; yellow square, ctDNA detected; grey square, ctDNA not detected; green square, sequencing not performed. d) Number of overlapping genes between the whole exome sequencing (WES)‐based mutation spectrum and panel gene lists. Upper panel, mutation spectrum obtained from the 10 patients; lower panel, mutation spectrum obtained from the NSCLC dataset of TCGA. Red box, number of genes detected by ICP‐based sequencing during the course of treatment; blue box, virtual validation of commercial panels, number of genes shared between the commercial panel gene list and patient mutation spectra. The boxplot shows the median value with ranges. p‐values are based on Kruskal–Wallis tests; for comparisons between the red box and each labeled box: *p < 0.05, **p < 0.01, ***p < 0.001. The red vertical line and red shadow indicate the median and range for the red box.

Figure 2

Tumor burden with respect to ctDNA changes during the course of follow‐up. a) Sequential neoantigen ctDNA‐assessment and tumor burden during treatment. The dynamic changes in the ctDNA allele frequencies are presented as lines of different colors. Tumor burden was quantified as the sum of products of perpendicular diameters (SPDs), and illustrated as black lines for each patient. SPDs were subjected to linear scaling. b) Correlation between the magnitude of the ctDNA decline and clinical outcomes after immune checkpoint blockade (ICB) administration. Grey bar plot in upper panel, fold‐change in mean ctDNA minor allele frequency (MAF) from the baseline to 12 weeks after first cycle of administration. Lower panel, heatmap showing the trend in radiological imaging‐based SPD‐measured tumor burden from the baseline evaluation to the last radiological follow‐up. Each lane represented the tumor burden for one patients serially. Each color square within the lane represented the tumor burden in each surveillance scan. The baseline tissue tumor mutational burden (TMB) and best of response (BOR) for each patient are annotated above the heatmap. The color gradient indicates the change in tumor burden. Patients are ordered according to the decline in ctDNA at 12 weeks. c) Time to tumor burden versus ctDNA decline among patients with radiological confirmation of >50% SPD decline. Statistics are based on two‐tailed paired Student's t‐tests.

Figure 3

Radiological and serological follow‐ups for P2 with unmeasurable lesions. Upper panels, surveillance CT, MRI scans (with contrast), or ECT showing the change in tumor burden over time in the mediastinal lymph node (LN), cranial bone, and lumbar vertebrae. Red arrowheads indicate the lesion site. Lower panel, clinical course. Red lines, SPD; green line with shadow, mean ctDNA MAF with the range for all detected mutations in ctDNA.