This study is part of a retrospective real-world evidence (RWE) multi-institutional study including a total of 119 pretreated mNSCLC patients who had received therapy with nivolumab or atezolizumab, according to the standard international guidelines at the Medical Oncology Unit of The Hospital of Reggio Calabria (OM-RC) and at the Radiotherapy Unit of the University Hospital of Siena (RT-SI), Italy, between September 2015 and April 2020 (16). The current study was focused on 28 of the abovementioned patients who consented to get their blood drawn and anonymously stored for scientific purposes. All procedures were undertaken in compliance with the ethical statements of the Declaration of Helsinki (1964, amended most recently in 2008) of the World Medical Association and respect of their privacy. All patients received PD-1/PD-L1 blockade in a real-world setting as recommended by the international guidelines and regulative agencies following the standard procedures of administration for each drug. All patients had received at least a previous line of platinum-based doublet ± bevacizumab prior to PD-1/PD-L1 blockade. All of them presented with Eastern Cooperative Oncology Group (ECOG) performance status ≤1 and had complete physical examination reports; histological samples; blood cell count; biochemical, immunobiological, and radiological screening; and imaging at the baseline. All patients received the immunological treatment with nivolumab (intravenous infusion of 3 mg/kg every 2 weeks) (16 patients) or atezolizumab (intravenous infusion of 1,200 mg every 3 weeks) (12 patients) until disease progression or occurrence of severe adverse events. Clinical history, physical examination, blood tests, and record of adverse events were evaluated prior to each drug infusion. A computed tomography scan (CT scan) was performed every 3 months or in any case of suspected progressive disease (PD) and evaluated according to the immune Response Evaluation Criteria in Solid Tumors (iRECIST 1.1) (19). All patients were monitored for blood cell count, and biochemistry prior to each treatment course and for their adrenal hormone profile, adrenocorticotropic hormone (ACTH), thyroid-stimulating hormone (TSH), thyroid hormones, anti-thyroid AAbs, ENA antibodies, ANAs, ASMAs, and c/p- ANCAs each month from the beginning of the treatment as reported in previous studies (9, 12).

Blood was collected into tubes containing ethylenediamine tetra-acetic acid (EDTA) at two time points, baseline and after three treatment cycles. Peripheral blood mononuclear cells (PBMCs) were isolated from blood samples by Ficoll-Hypaque density centrifugation and immediately frozen and stored in nitrogen liquid. Samples were then stained with two different approaches: panel 1 tube, which included markers cluster of differentiation (CD)3 (V450), CD4 (FITC), CD8 [labeled with peridinin chlorophyll protein complex (PerCP)-Cy5.5], CD16 (conjugated with allophycocyanin (APC)-H7], CD19 [conjugated with phycoerythrin (PE)-Cy7], CD56 (conjugated with PE), HLA-Dr (conjugated with PE), and CD45 (conjugated with PE), and panel 2, which included CD3 (conjugated with V450), CD4 [conjugated with fluorescein iso-thiocyanate (FITC)], CD8 (conjugated with PerCP-Cy5.5), CD16 (conjugated with APC-H7), CD19 (conjugated with PE-Cy7), CD56 (conjugated with PE), CCR7 (conjugated with PE-Cy7), FoxP3 (conjugated with FITC), and PD-1 (conjugated with PE), all from BD Biosciences. Samples were acquired on a fluorescence-activated cell sorting (FACS) Canto II flow cytometer (BD Biosciences). This study was not designed to investigate specific myeloid derivative PBMC subsets; however, the presence of monocytes was formally detected in the blood samples by means of forward scatter (FSC), side scatter (SSC), and CD45.

The data were analyzed with FlowJo (panel 2) or FlowCT (panel 1) (20) (R-based package) for an in-depth study of the cell subpopulation. Specifically, a semi-automated clustering and predictive modeling tool were used for synchronously analyzing 30 FCS files. After pre-processing and quality control of previously FCS files, data were normalized and clustered using self-organized maps (FlowSOM). Cell clusters were represented with uniform manifold approximation and projection (UMAP) dimensionality reduction technique. Cell clusters were annotated based on heatmap clustering, using median values for each FCS. This methodology works better on homogeneous data; thus, we limited this approach to 15 patients from the OM-RC and then manually analyzed the remaining patients from the RT-SI center.

Paired Student’s t-test was used for the analysis of each cluster in the different treatment conditions.

Samples were processed by the Microbiology Unit Laboratory at the University of Siena and by the laboratory of flow cytometry at the Grand Metropolitan Hospital, Reggio Calabria, Italy. HLA genotyping of the loci A, B, C, and DRB1 was centralized and carried out in the Tissue Typing Unit at the Grand Metropolitan Hospital in Reggio Calabria as previously described by our group (21).

Hazard ratios (HRs) and their 95% confidence intervals were estimated through the Cox regression proportional model; in the multivariate approach, a forward stepwise procedure was used, and the enter and remove limits were set to 0.05 and 0.10, respectively. The association of irAE frequency with biological parameters with clinical outcomes in the two patient cohorts was assessed by the chi-square test. Statistics were performed by the SPSS software 23.0 (International Business Machines Corp., New York, NY, USA) and R statistical software.

We retrospectively conducted a retrospective bioinformatics analysis on 28 mNSCLC patients (20 men and eight women, four with squamous and 24 with non-squamous histology) who volunteered to provide blood samples for non-profit non-interventional immunobiological analysis. These patients belonged to a larger series of individuals with mNSCLC who received immune-checkpoint blockade at OM-RC and RT-SI according to the standard guidelines for their specific disease. All of them were fit for treatment (ECOG performance status ≤1) and presented no druggable tumor driver mutation/rearrangement ( Table 1 ). All 28 patients presented received at least one chemotherapy line with platinum-based doublet ± bevacizumab prior to PD-1 blockade. In this specific series, 16 patients had received salvage therapy with nivolumab or pembrolizumab and 12 with atezolizumab between November 2015 and April 2020 at OM-RC and RT-SI units. During the treatment, after four or five treatment cycles, we recorded one or more irAEs in 14 patients. These adverse events were moderate (G1–G2) in nine patients (skin rash, poly-arthralgia, and thyroiditis) and severe (G3–G4) in further five cases [autoimmune pneumonitis (four cases), myasthenia-like syndrome (one case), hypophysitis (one case), and uveitis (one case)]. The present retrospective study was not empowered to evaluate differences in PFS or OS among patients receiving anti-PD-1 or anti-PD-L1-blocking mAbs. All the patients underwent class I HLA ABC and DRB-1 genotyping and were also monitored for serum levels of inflammatory markers [lactate dehydrogenase (LDH), C-reactive protein (CRP), and erythrocyte sedimentation rate (ESR)], AAbs (ENA, ANAs, and ANCAs) and inflammatory cell counts and neutrophil-to-lymphocyte ratio (NLR) at baseline and after three treatment cycles.

FACS analysis was performed on the PBMCs collected at baseline and after three treatment cycles. We evaluated potential treatment-related changes in different immune populations identified through a multidimensional single-cell bioinformatics approach on 15 patients from the OM-RC center (training dataset). Overall, 435,318 cells were analyzed ( Figure 1A ; interestingly, more cells were recovered after three treatment courses), and, after an automatic clustering approach, we unbiasedly identified 19 sub-clusters of mononuclear cells based on markers profile ( Figure 1B ), namely, B cells, CD16p⁺ monocytes (non-classical monocytes), CD3⁺DN (double negative), CD4⁺, CD4⁺HLADR⁺, CD8⁺, CD8⁺HLADr⁺, Debris, Eosinophils, intermediate (int) monocytes, monocytes (classical), NK, NK_CD16⁻, NK_CD8⁺, NK_CD8⁺HLADR⁺, NKT, NKT_CD4⁺, NKT_CD4⁺CD8⁺, and NKT_CD8⁺ ( Supplementary Figure 1 represents a heatmap reporting the expression of each marker in each identified cell population, and most populations have been identified thanks to their unique combination of marker expression, e.g., neutrophils were FSC/SSC high, CD45mid, CD16dim, and negative for all other markers).

After three treatment courses, we observed a significant decline in CD4⁺ T cells, B cells, and neutrophils [absolute mean difference (MD) = −4.46%, p-value = 0.0035; MD = −2.5%, p-value = 0.027; and MD = −13.03%, p-value = 0.04, respectively) and a decreasing trend for monocytes (MD = −3.88%, p-value = 0.1). Additionally, we recorded an increase in NKT_CD8⁺ cells (MD = 1.31%, p-value = 0.046) ( Figure 2A and Supplementary Figure 2 ). Likely due to the small sample size, only CD4⁺ T-cell decrease was validated in the dataset of patients from the RT-SI center ( Supplementary Figure 3 ). As shown in Figure 1B , all the cell populations were additionally clustered separately from neutrophils. Interestingly, a significant decrease in NLR (calculated by FACS analysis) was observed after three treatment cycles with a p-value = 0.015 ( Figure 2A ), perfectly overlapping the changes in the peripheral blood cell count analysis ( Figure 2B ). Again, this aspect was not validated in patients from the RT-SI center ( Supplementary Figure 3 ).

Additionally, we found a significant difference in the baseline values of NLR between patients from the two centers (mean 4.87 vs. 2.72 for OM-RC vs. RT-SI, p = 0.019) ( Figure 2B ). Interestingly, this difference disappeared after three treatment courses of anti-PD1/PD-L1 mAbs, thus ruling out the hypothesis of a laboratory-related batch effector or other possible confounding factors (sex, age, previous radiation therapy, and histology). Another plausible hypothesis may be formulated on the fact that OM-RC and RT-SI centers adopted different frontline treatments prior to immunotherapy. In fact, all of the patients enrolled in RT-SI had received frontline chemotherapy with fractioned cisplatin and oral metronomic etoposide ± bevacizumab (mPE/mPEBev), a very active regimen with powerful immunomodulating activity (22–26). On the contrary, the majority of patients enrolled at OM-RC had received more conventional frontline chemotherapy doublets (either carboplatin and taxol or carboplatin and pemetrexed). Therefore, all of the patients in the validation group derived from RT-SI had received a frontline systemic treatment with the potential ability to impair the myeloid and inflammatory components due to the long-term sequestering activity of vascular endothelial growth factor (VEGF) exerted by bevacizumab together with the immunobiological effects of metronomic chemotherapy as reported in previous studies (22, 23, 27). Considering the prognostic value of blood cell subsets in patients with NSCLC, we attempted to build a risk model to deeply evaluate their impact on NSCLC patient outcomes. Thus we conducted a statistical analysis based on fold-change modifications after three treatment cycles. Our multidimensional analysis of PBMCs showed no significant post-treatment changes in CD8⁺ T cell abundance but significant changes in CD4⁺ and B cells (p < 0.05) according to Student’s t-test ( Supplementary Figure 3 ). Interestingly, we found a significant correlation between pretreatment atypical monocytes (negative correlation), CD3⁺CD4⁻CD8⁻ (positive correlation), and CD8⁺HLADR⁺ (positive correlation) NKT cells (positive correlation) and survival ( Supplementary Figure 4 and Supplementary Table 1 ). In line with what was previously observed, we found that patients with significant post-treatment increases of CD4⁺, CD8⁺, and monocytes showed a significant improvement in survival according to the log-rank (Mantel–Cox) test (p < 0.05). Similarly, an increase in NKT also showed a clear-cut trend to improve the outcome ( Figure 2C and Supplementary Figure 5 ) of these patients. None of the other cell subset treatment-related changes was correlated with patient survival ( Supplementary Figure 5 and Supplementary Table 1 ).

For a small group of 12 patients who received treatment at RT-SI, we had sufficient material to evaluate possible changes in peripheral central memory (T_CM; CD3⁺CD8⁺CD45RO⁺CCR7⁺) and effector memory (T_EM, CD3⁺CD8⁺CD45RO⁺CCR7⁻) T cells, T_regs (CD4⁺/CD25⁺/Foxp3⁺ T cells), and PD1⁺ CTLs. In this very small cohort of patients, we recorded no change in T_CM and T_EM subsets, a significant decline in T_regs, and a paradoxical increase in CD8⁺PD1⁺ T lymphocytes after three treatment courses of nivolumab ( Figure 3A ).

In the present set of mNSCLC patients, we recorded a significant treatment-related appearance and/or increase of serum ANA (baseline vs. the third cycle = 28% vs. 64%, p < 0.0005), while no statistically changes were observed in terms of inflammatory markers (CRP, ESR, or LDH) (p > 0.5). This finding was in line with what was reported in previous studies, sustaining the good quality of the results (9). Overall, in our small set of 28 patients, there was no change in white blood cell counts with the exception of eosinophil cell counts, which were significantly increased in the post-treatment setting [140 ± 130 vs. 240 ± 180 cells/µl; p = 0.0022]. Also, these results were once again consistent with what was shown in previous studies (24).

Our analysis revealed a much lower frequency of CD8⁺ and NKT cells in patients who were positive for ANA screening ( Figure 3B ) compared with those who remained negative for the expression of these AAbs along the immunological treatment. On the contrary, we were unable to find any other immunological difference between the two cohorts of patients ( Supplementary Figure 6 ).

We also evaluated the occurrence of possible changes in specific immune populations in patients who had or did not have irAEs in the course of the treatment. Interestingly, we found an increase in neutrophils, CD16⁺ monocytes, and NKTCD4⁺CD8⁺ cells only in those patients who presented irAE signs (p < 0.1) ( Figure 3C and Supplementary Figure 7 ).

Finally, we were unable to find any correlation neither between survival and specific HLA haplotypes nor with HLA and specific immune cell changes ( Supplementary Figures 8 and 9 ) due to the small available dataset that was unable to fulfill the statistical analysis requirements.