Predictive model for BNT162b2 vaccine response in cancer patients based on blood cytokines and growth factors

Abstract

Background: Patients with cancer, especially hematological cancer, are at increased risk for breakthrough COVID-19 infection. So far, a predictive biomarker that can assess compromised vaccine-induced anti-SARS-CoV-2 immunity in cancer patients has not been proposed.

Methods: We employed machine learning approaches to identify a biomarker signature based on blood cytokines, chemokines, and immune- and non-immune-related growth factors linked to vaccine immunogenicity in 199 cancer patients receiving the BNT162b2 vaccine.

Results: C-reactive protein (general marker of inflammation), interleukin (IL)-15 (a pro-inflammatory cytokine), IL-18 (interferon-gamma inducing factor), and placental growth factor (an angiogenic cytokine) correctly classified patients with a diminished vaccine response assessed at day 49 with >80% accuracy. Amongst these, CRP showed the highest predictive value for poor response to vaccine administration. Importantly, this unique signature of vaccine response was present at different studied timepoints both before and after vaccination and was not majorly affected by different anti-cancer treatments.

Conclusion: We propose a blood-based signature of cytokines and growth factors that can be employed in identifying cancer patients at persistent high risk of COVID-19 despite vaccination with BNT162b2. Our data also suggest that such a signature may reflect the inherent immunological constitution of some cancer patients who are refractive to immunotherapy.

Keywords: BNT162b2; COVID-19 vaccine; SARS-CoV-2; chemokines; cytokines; growth factors; haematological malignancies; solid cancers.

Figure 1

Timeline of the study. The BNT162b2 vaccine was administered on day 0 and 21. Heparin plasma samples for CCG analysis were collected on day 0 just prior to primer dose administration (D0), day 1 (D1), day 21 just prior to booster dose administration (D21), and day 28 (D28). For anti-RBD and anti-S1 serology, serum samples were collected on day 49 (D49) after the administration of the primer vaccine dose.

Figure 2

CCG alterations as a response to primer and booster dose vaccinations in cancer patients. (A) Differentially expressed CCGs after the administration of the primer and booster doses, compared to the CCG levels prior to vaccine administration. (B) Differentially expressed CCGs at day 21 and day 28, compared with baseline day 0. P-values were calculated using paired t-test. The vertical dotted line represents no change. The horizontal dotted line represents a p-value of 0.05.

Figure 3

CCGs alterations as a response to primer and booster vaccinations in cancer patients undergoing different treatment regimens. Cluster analyses of CCGs at different timepoints with a Partial Least Squares-Discriminant Analysis (PLS-DA) reveal minor differences between patients undergoing distinct types of anti-cancer therapies. Hematological patients included patients receiving rituximab or patients who received an allogeneic hematopoietic stem cell transplantation at least one year before the primer dose vaccination.

Figure 4

Prediction models for BNT162b2 immune response in cancer patients. (A) Correlation matrix depicting the correlation between CCG measurements (log₁₀ transformed) and quantitative anti-RBD IgG measurements at day 49. IgG antibody levels to SARS-CoV-2 RBD antigen were assessed with an enzyme-linked immunosorbent assay for quantitative detection of IgG antibody levels to SARS-CoV-2 RBD antigen. Only CCGs with significant correlations are shown. (B) Significantly different between good (≥ 200 IU/mL) (blue) and poor (< 200 IU/mL) (red) responders to the BNT162b2 vaccine. A good/poor responder threshold of anti-RBD IgG titer of 200 IU/mL used in this study predicts a neutralization response required for 50% protection against symptomatic SARS-CoV-2 infection (99%-100% specificity at a sensitivity of 95%). *p < 0.05, **p < 0.01, ***p < 0.001. (C) Area Under the Receiver Operating Characteristic (AUROC) values for 10 predictors of the binary IgG response as good or poor responders at day 0, day 1, day 21, and day 28. * Denotes significant p-values of at least < 0.05. (D) Random Forest Classifier predicted a model where a combination of CRP, IL-15, IL-18, and PlGF levels measured right before vaccine administration (day 0) and at day 1, day 21, and day 28 after the primer dose predicted good and poor responders with high accuracy (AUCs depicts averages of 10 individually constructed ROC curves).