Introduction

Checkpoint inhibitors (CPIs) including compounds targeting PD-1/PD-L1 axes have brought significant improvements in terms of overall survival in several types of advanced cancers1,2,3,4,5,6. A single response profile, such as pseudo-progression, is observed under CPIs7. Among these typically-related response profiles under CPIs is hyperprogressive disease (HPD) which was defined as an unanticipated and paradoxical acceleration of the tumor growth7,8. The incidence of HPD is variable according to the way it is defined and ranges between 4 and 29%7. Though such acceleration of the tumor growth kinetic was also observed with other agents (chemotherapy9, tyrosine kinase inhibitors10), the intensity and the frequency of the phenomenon appears to be higher with checkpoint inhibitors used alone7. A single response profile, such as pseudo-progression, is observed under CPIs7. Among these typically-related response prfiles under CPIs is hyperprogressive disease (HPD), which has been defined as an unanticipated and paradoxical acceleration of tumor growth7,8. The incidence of HPD is variable according to the way it is defined and ranges between 4 to 29%7. Although this acceleration of tumor growth kinetics was also observed with other agents (chemotherapy9, tyrosine kinase inhibitors10), the intensity and frequency of the phenomenon appears to be higher with checkpoint inhibitors used alone7. HPD may be associated with a worsening of the outcome11. Different physiopathological hypotheses have been tested to explain phenomena such as tumoral genomics variations12,13. Indeed, CPI has been shown to hasten tumor growth in a mouse model with a relative lack of PD-1 expression14. As HPD was observed in several malignant tumor types, a role for the host variations has been advocated13,15,16. Indeed, allelic variations of HLA class I genes have been shown to impact clinical outcome under CPI17. However, dedicated germinal immunogenetics studies remains rare in the context of CPI-based treatment18. To better elucidate the potential relationship between host immunogenetics and CPI treatment outcome and particularly HPD, we correlated the outcome of patients treated with CPI and selected polymorphisms described in four key genes: PD-1 (Programmed Cell Death 1 gene, 2q37.3), PD-L1 (Programmed Death Ligand 1 gene, 9p24.1), IDO1 (Indoleamine 2,3-Dioxygenase 1 gene, 8p11.21) and VEGFR2 (Vascular Endothelial Growth Factor Receptor 2 gene, 4q12).

Results

Patient characteristics and outcome

Patient baseline characteristics are given in Table 1. All patients were treated for an advanced malignancy. Non-small cell lung cancer (NSCLC) (n = 48) was the largest subgroup followed mainly by head and neck squamous cell carcinoma (HNSCC) (n = 16), renal cell carcinoma (RCC) (n = 14) and melanoma (n = 13). Importantly, all patients were treated by CPI monotherapy alone (anti-PD-1 or anti-PD-L1), with a majority of anti-PD1 (87%). Median age was 68 (range: 32–85), 65 were males (66%) and 70 were smokers (83%). Sixty-six patients had received previous irradiation (69%). The SNP genotype, gene information and genotype frequency are shown in Table 2.

Table 1 Patient characteristics.
Full size table
Table 2 Summary of genotyping results by MassARRAY (AGENA) of 98 patients.
Full size table

Median follow-up was 13.3 months (95% confidence interval [CI]; 10.6 months to 15.4 months). Median irPFS was 16.8 months (95% confidence interval [CI]; 10.2 months to NA) and median OS was not reached. Twelve-month OS and 12-month PFS were 80% (95% confidence interval [CI], 72% to 90%) and 47% (95% confidence interval [CI]; 5% to 60%), respectively.

Fifteen patients experienced grade 3–4 IrAEs (15.5%), 67 grade 1–2 IrAEs (68.25%) and 16 patients had no IrAE (16.25%). Overall response was complete for 8 patients (8%), partial for 43 patients (44%), stable disease for 28 patients (28.5%) and progressive disease for 19 patients (19.5%). TGKR could be calculated for 80 patients (15 patients had CPI as first line for advanced disease; pre-baseline scanner was not available for 3 patients). HPD was observed in 11 patients (14%). HPD was correlated with shorter OS (Fig. 1) compared with non-HPD patients (P = 0.003).

Figure 1
figure 1

Association between HPD and OS: Kaplan Meier estimates of OS of patients treated with anti PD1/anti PDL1 according to ir-RECIST criteria: clinical benefit (complete response, partial response, stable disease), PD non HPD (progressive disease) and HPD.

Full size image

HPD predictive factors

In univariate analysis (Table 3), HPD was significantly associated with age ≥70 years (25% versus 6%; P = 0.025), immune-related toxicity grade ≥3 (38.5% versus 9.5%; P = 0.016), VEGFR2 rs1870377 A/T or A/A (26% versus 4%; P = 0.005), PD-L1 rs2282055 G/T or G/G (23% versus 2.5%; P = 0.024) and PD-L1 rs2227981 G/G (4.5% versus 23.5%; P = 0.024). HPD was not significantly correlated with lactate dehydrogenase (LDH) blood levels at baseline (p = 0.055). Similarly, the neutrophil-to-lymphocyte ratio (NLR) was not linked to HPD (p = 0.936). Also, tumor burden was not associated with HPD (p = 0.732). Multivariate analysis revealed an independen t association between HPD and age ≥ 70 years (OR = 14.42; 95% confidence interval [CI];  2 to 100; P = 0.006), rs1870377 T/A or A/A, and VEGFR2 (OR = 15.36; 95% confidence interval [CI]; 1.92 to 119; P = 0.007) and rs2282055 T/G or G/G, PDL1 (OR = 17.73; 95% confidence interval [CI]; 11.55 to 227; P = 0.01).

Table 3 Univariate and multivariate analyses for hyperprogressive disease.
Full size table

A risk score was calculated by logistic regression and integrated the 3 independent variables (age, rs2282055, rs1870377) for predicting HPD. The risk for HPD was optimally estimated (OR = 18.34; 95% confidence interval [CI]; 3.38 to 99.58; P <0.001) (Table 4).

Table 4 Classification of patients based on risk group and risk evaluation of each group.
Full size table

Discussion

We observed HPD in 14% of treated patients by CPI, a figure in the range of figures reported in independent series7. We identified older age as a predictive variable for HPD in accord with previously reported series11. However, this point is controversial and observations have been reported in recent studies by Kim et al.19 and Ferrara et al.9 showingno association between HPD and age. These discrepancies may be due to the different evaluation methods used to evaluate HPD as well as to the retrospective nature of these studies. In agreement with others19, we noted that patients with HPD had higher baseline LDH levels but which did not reach statistical significance in our hands. Our negative finding contrasts with that of Kim and coworkers19 reporting that patients with HPD had baseline NLR values higher than those of patients without HPD. This discrepancy can be explained by the retrospective nature of both studies and also by the relatively small number of patients. Clearly, prospective studies based on a larger set of patients would be more likely to provide firmer conclusions regard of this possible association between baseline NLR and the risk to developing HPD under CPI. To the best of our knowledge, the present study is the first cohort that explores the link between host gene polymorphisms and HPD under CPI. Our data highlight two germinal variations with rs2282055 (PD-L1) and rs1870377 (VEGFR2) having a significant and independent influence on the occurrence of HPD.

The group of patients with rs2282055 (PD-L1) G allele, either homozygous or heterozygous, was found to be significantly associated with a higher risk of developing HPD in comparison with T/T genotype, the locus being located on chromosome 9p24.1. When expressed on tumor cells, this gene down-regulates the activation of T effector cells through a key mechanism responsible for immune response evasion20. However, the real impact of tumor PD-L1 expression on treatment outcome under CPI remains controversial21. The regulation of tumoral and non-tumoral PD-L1 expression is a complex phenomenon and is influenced by multiple molecular pathways22,23,24. rs2282055 (PD-L1) is associated with 10 other SNP all inserted in different introns of the PD-L1 gene25. It has been shown that introns may have a direct or indirect influence on mRNA expression: GTEX portal (https://gtexportal.org/home/) indicates that rs2282055 is associated with down-regulated expression of PD-L1 (CD274 gene) in brain tissue while it is overexpressed in the pancreas, suggesting that rs2282055 may impact PD-L1 expression differently in different tissues. rs2282055 (PD-L1) was recently evaluated for its association with survival of patients not treated by CPI26. In this latter study, the impact of rs2282055 (PD-L1) polymorphism on survival was found to be non-significant, thus suggesting a non-prognostic role of this polymorphism. Since PD-L1 expression was not available in our cohort, we could not examine potential links between this rs and the level of expression of PD-L1 protein In conclusion, it can be suggested that rs2282055 (PD-L1) may interfere with CPI-HPD development, while the underlying mechanism remains to be elucidated.

VEGFR2 is a gene encoding for vascular endothelial growth factor receptor 2 expressed on both endothelial cells and various immune cells27,28. VEGFR2 is a key regulator of tumor angiogenesis and tumor microenvironment by mainly promoting a high level of Tregs and by reducing the ability of T effector cells to penetrate the tumor cell bed29. Of note, rs1870377 (KDR, VEGFR2, NM_002253.3:c.1416A>T) induces a missense substitution Q472H in the fifth (out of seven) extracellular Ig-like motifs that has been shown to increase VEGF-A binding and activity inducing increased microvessel density in tumor tissue of patients with non-small cell lung cancer30. In our series, carriers of rs1870377 (VEGFR2) with any A genotype were more prone to develop HPD. Thus, VEGFR2 substitution Q472H may play a potential role in increased tumor size due to increased angiogenesis and microvessel development in these patients. It is thus conceivable that the impact of VEGFR2 on tumor and its microenvironment may differ according to the allelic inheritance of the host with an influence on HPD development under CPI.

Collectively, one can formulate a working hypothesis with HPD occurring in a subset of patients harboring unfavorable alleles which modulate the expression of different genes inducing tumor progression under CPI. It was interesting to identify key immunology-linked genes like PD-L1 and VEGFR2 gene variants using this approach. The present reported results remain challenging in clinical practice with particular attention given to the fact that most allelic variations are present at relatively low frequencies. However, this study contains a number of limitations which do not allow drawing definitive conclusion: the sample size is relatively small (11 HPD cases) and patients received two different classes of PD-1 and PD-L1 CPI. TGKR was not assessable for first-line treated patients. The study covered different histological types and some patients had been more or less heavily pretreated. According to the meta-analysis by Kim and coworkers31, the histological type of the tumor is not predictive value for the occurrence of HPD. However, it has been reported that renal cell carcinoma (RCC) patients may be at a lesser risk of HPD11,32. Of note, our cohort was also enriched with long-responding patients as all patients alive and treated with CPI in the department were asked their consent to dedicated blood sampling for the study. This explains the high response rate reported in our series (52%). Above all, the study remains original leading to identification of potential host-linked biomarkers for HPD prediction. Interestingly, it was possible to establish a powerful (OR = 18.34; 95% confidence interval [CI]; 3.38 to 99.58; P <0.001) predictive score combining host characteristics such as age and germinal gene polymorphisms. Evaluating the risk of HPD by testing host immunogenetics must remain probabilistic in nature and may differ according to ethnic population, thus limiting extrapolation of the present study outside the Caucasian population. Efforts to expand other candidate genes and their polymorphisms are currently ongoing in larger prospective cohorts. Particular attention should be paid to allelic variations of HLA class I genes.

Finally, our results support the notion of a genetic susceptibility potentially impacting the development of HPD in a Caucasian population. In a broader perspective, it is hoped that the present data can stimulate further studies integrating both somatic and germinal variability aimed at satisfying the still unmet need for faithful predictive biomarkers to ensure enhanced management of cancer therapy by CPI.

Patients and Methods

Study design and patients

This is a retrospective study covering the period April to August 2018. All data were retrieved from the clinical database of the Centre Antoine Lacassagne (Nice, France). Tumor responses were evaluated after monotherapy according to RECIST 1.1 criteria (complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD)). Objective response was evaluated as previously published33,34,35. Immune-related adverse events (irAEs) were evaluated according to National Cancer Institute Common Terminology Criteria for Adverse Events (NCI-CTCAE V5). Pre-baseline, baseline, and initial imaging results were recorded and were to calculate the TGKR (ratio of the slope of tumor growth before treatment and the slope of tumor growth during treatment), as previously reported8. The sum of the largest diameter of target lesions at baseline indicated the tumor burden at baseline. HPD was defined as a TGKR ≥2. Written informed consent was systemically obtained before collecting a study-dedicated blood sample. Patient characteristics, at baseline, also included age, gender, histology, smoker status, lactate dehydrogenase (LDH), neutrophil-to-lymphocyte (NLR) and tumor burden.

SNP selection and genotyping

Seventeen SNPs of PD-1 (rs10204525; rs11568821; rs22727981), PD-L1 (rs2282055; rs2297136; rs2297137; rs4143815; rs10815225; rs822339), IDO1 (rs3739319; rs3808606; rs373931; rs9657182; rs34820341) and VEGFR2 (rs2305948; rs1870377; rs2071559) were selected according to their functional and/or clinical relevance. Genomic DNA was extracted from a blood sample using the commercially-available Maxwell® 16 LEV Blood DNA Kit (#AS1290, Promega). The assay to screen the 17 SNPs was created by using Assay Design Suite v2.0 (AGENA Bioscience online software) with the “Genotyping Design”option. We had created the assay to screen the 17 SNPs. Data were verified and compatible with DNA controls polymorphism for 15 SNPs; the remaining 2 SNPs had been eliminated (PD-L1 rs822339 and IDO1 rs34820341) because incompatible with DNA control polymorphism (https://www.coriell.org/1/NIGMS/Collections/CEPH-Resources). For 15 SNPs minor allele frequency was ≥5% in Caucasians according to SNPpedia (http://www.snppedia.com) and the Ensemble database (http://www.Ensembl.org). All tested SNPs were in Hardy-Weinberg equilibrium (Table 2).

Statistical considerations

The link between the 15 SNPs and clinico-radiological parameters and CPI response according to ir-RECIST35 criteria and irAEs was examined. Statistical comparisons were performed using χ2 test or Fisher’s exact test for categorical data and Student’s test or Wilcoxon’s test for continuous variables. Immune-related progression-free survival (irPFS) and Overall Survival (OS) were respectively calculated from the baseline CT scan to progression (according to ir-RECIST criteria) or death and presented graphically using the Kaplan-Meier method. All variables significant at the 5% level in both univariate and multivariate logistic regression models were included. Co-linearity between all variables of the initial multivariate model was evaluated. The choice of the final model was made by performing a backward stepwise selection model. A fitted score for each participant by logistic regression was used to define two risk groups of patients (low or high risk of HPD). The optimal number of risk groups for predictive models was obtained using the Younden method36. Statistical analyses were performed using R version 3.5.0 on Windows®.

Ethical approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee (French National Commission for Informatics and Liberties N°17010).

Informed consent

All patients provided written informed consent before enrollment.