Antibody–Fc/FcR Interaction on Macrophages as a Mechanism for Hyperprogressive Disease in Non–small Cell Lung Cancer Subsequent to PD-1/PD-L1 Blockade

This article is featured in Highlights of This Issue, p. 897

The advent of immune checkpoint inhibitors (ICI) has radically changed the paradigm of care for patients with non–small cell lung cancer (NSCLC). Several agents are now approved in the treatment of NSCLC on the basis of their superiority over chemotherapy (1, 2). Immunotherapy with antibodies targeting either the programmed death receptor 1 (PD-1) or its ligand (PD-L1) may provide long-term benefits in approximately 20% of patients (2). However, in a subset of patients, ICI paradoxically accelerates tumor growth, a phenomenon known as hyperprogression (HP; refs. 3–6). Studies have estimated that the prevalence of HP in patients with different cancer histotypes treated with ICI may range between 9% and 29% (3–6). No significant histopathologic and molecular features capable of predicting a priori HP have been identified, with the partial exception of rare MDM2 amplification and EGFR mutations (5). These observations have ignited an international debate regarding whether HP is a true phenomenon or only representative of a subset of patients with a particularly worse prognosis.

In the tumor microenvironment, the effector functions of innate immune cells may be blunted by the PD-1 receptor, as observed for T lymphocytes, suggesting that these innate cells are another potential target for ICI (7–9). Accordingly, it has been demonstrated that anti–PD-1 antibody can exert antitumor activity in immunodeficient mice via natural killer (NK) cells (8). Conversely, it has been demonstrated that, in PD-1^−/− NK cells or in NK cells pretreated with anti–PD-1 antibody, the production of lytic molecules such as perforins and granzymes is decreased (10). Moreover, tumor-infiltrating dendritic cells (9) and monocytes (11) are reported to release the immunosuppressive cytokine IL10 upon anti–PD-1 treatment. Therefore, it is possible to hypothesize that in some circumstances, PD-1 blockade might exacerbate immunosuppression upon interaction with innate immune cells.

The aim of this study was to investigate, at the clinical and pathologic level, the phenomenon of HP in patients with NSCLC and to evaluate the role of innate immunity during ICI treatment. To eliminate the interference of T lymphocytes, we exploited cell lines and patient-derived xenografts (PDX) transplanted in immunodeficient mice.

Medical records, radiologic findings, and available tumor specimens were collected from patients with NSCLC treated with ICI at the Thoracic Unit of the Istituto Nazionale dei Tumori, Milan, Italy, from July 2013 to December 2017.The study complied with the Declaration of Helsinki and was done in accordance with good clinical practice guidelines. All samples were obtained according to the Internal Review and the Ethics Boards of the Istituto Nazionale Tumori of Milan and all patients provided informed consent. All experimental protocols were approved by the ethics boards of the Istituto Nazionale Tumori of Milan (Int 22/15). Radiologic evaluation (CT scan with or without brain MRI) was performed at treatment initiation and every 8 weeks thereafter. Considering that criteria to define patients with HP described by previous authors (3–6) are applicable only in advanced lines, our multidisciplinary team (oncologists, pneumologists, radiologists, and thoracic surgeons) created institutional clinical and radiologic criteria designed to identify patients with HP also in first-line treatment. Patients with HP or those patients defined as P were classified according to predefined criteria as follows: (i) time-to-treatment failure < 2 months (time-to-treatment failure is defined as the time from the start of treatment with ICI to ICI discontinuation for any reason, including progression, patient preference, toxicity, or death); (ii) increase of ≥50% in the sum of target lesions major diameters between baseline and first radiologic evaluation; (iii) appearance of at least two new lesions in an organ already involved between baseline and first radiologic evaluation; (iv) spread of the disease to a new organ between baseline and first radiologic evaluation; (v) clinical deterioration with decrease in Eastern Cooperative Oncology Group (ECOG) performance status ≥2 during the first 2 months of treatment.

Patients who fulfilled at least three of the clinical/radiologic criteria were defined as exhibiting HP, whereas patients with RECIST 1.1 progressive disease as best response without fulfilling at least three criteria were defined as P patients. All R patients and SD patients were classified according to their RECIST 1.1 best response. Only patients who underwent at least two cycles of ICI treatment were included in this analysis.

IHC was carried out on formalin-fixed paraffin-embedded (FFPE) human or PDX tissue sections as described in Supplementary Materials and Methods. All the slides were analyzed under a Zeiss Axioscope-A1 equipped with fluorescence module and microphotographs were collected using a Zeiss Axiocam 503 Color with the Zen 2.0 Software (Zeiss). All markers were scored according to the percentage of immunoreactive cells out of the total cellularity.

All xenograft experiments were undertaken using 8- to 9-week-old female athymic nude or SCID mice (Charles River Laboratories). Human NSCLC cell line H460 tumor-bearing athymic nude mice were treated intraperitoneally or peritumorally with either 200 μg of mAb anti-mouse PD-1 (clone RMP1-14, BioXCell) or saline, and with either peritumoral anti–PD-1 F(ab)₂ or isotype control. Experiments were carried out in groups of four SCID mice, bearing a PDX sample or a cell suspension (10⁵ cells for H460 and PC9 xenograft experiments) in each flank. Mice were treated twice weekly with an intraperitoneal injection of 10 mg/kg nivolumab (Opdivo; Bristol-Myers Squibb) or nivolumab F(ab)₂ fragments. Mice were maintained in the Animal Facility of the Fondazione IRCCS Istituto Nazionale dei Tumori (Milan, Italy). Animal experiments were authorized by the Institutional Animal Welfare Body and the Italian Ministry of Health, and performed in accordance with national law (D.lgs 26/2014) and Guidelines for the Welfare of Animals in Experimental Neoplasia (12). At the end of each experiment, tumors were harvested for subsequent analyses.

Distribution of continuous and categorical biomarkers was summarized by the median as a measure of central tendency and absolute frequencies, respectively. The Cochran–Mantel–Haenszel χ² test was used to detect statistical association (i.e., P < 0.05) in univariate analysis. The median and interquartile range (IQR) follow-up was estimated using the reverse Kaplan–Meier method.

From July 2013 to December 2017, 187 patients with advanced NSCLC received treatment with ICI at the Thoracic Unit of the Medical Oncology Department at the Istituto Nazionale dei Tumori (Milan, Italy), and 152 patients were evaluable for response. We identified four categories: responders (R, 32 cases, 21%), patients with stable disease (SD, 42 cases, 27.7%), progressors (P, 39 cases, 25.7%), and patients with HP (39 cases, 25.7%). Patients' characteristics are described in Supplementary Table S1. In this population, after a median follow-up of 32.7 (IQR 15.1–39.6) months, 108 of 152 patients (71%) died. Median [95% confidence interval (CI)] overall survival (OS) in the overall population was 11.9 (95% CI, 8.8–15.5) months.

If we restrict the analysis to patients with HP, median OS significantly decreased to 4.4 (95% CI, 3.4–5.4) months as compared with 17.7 (95% CI, 13.4–24.1) in non-HP patients. Median OS was 8.7 (95% CI, 5.3–13.4), 17.7 (95% CI, 12.7–25.5), and not reached in P, SD, and R patients, respectively.

Supplementary Table S2 shows the differences between P and HP according to our criteria described in the Materials and Methods section. Of 187 patients, 64 were diagnosed in other centers and could not be included in the present histopathologic and molecular analysis. Of the remaining 123, 35 patients (11 with HP and 24 without HP) were evaluable for response and had tissue samples suitable for a wide IHC characterization and gene expression analysis. Patients' characteristics of the extensively analyzed 35 samples resemble the clinical characteristics of the whole treated population

IHC analysis was performed to assess the presence and distribution of tumor-infiltrating immune elements. The immunophenotype of 11 patients with HP was compared with that of 24 patients without HP (6 P, 11 SD, and 7 R). No significant differences were observed among all the clinical classes of patient with respect to the subsets of tumor-infiltrating T lymphocytes (TIL), evaluated by the density of CD3⁺, CD4⁺, and CD8⁺ lymphocytes and FOXP3⁺ regulatory T cells (Tregs). In addition, no differences were detected between classes of patient in the numbers of CD138⁺ plasma cells (PC), CD123⁺ plasmacytoid dendritic cells (pDC), peritumoral and stromal myeloperoxidase (MPO)⁺ myeloid cells, CD163⁺ macrophages, CD33⁺, PD-1, and PD-L1⁺ immune cells. However, MPO⁺ myeloid cells within the tumor were directly correlated (P = 0.0497) and PD-L1 expression in tumor cells was inversely correlated (P = 0.0457) with HP. Furthermore, a statistical trend was shown for the M2 macrophage/myeloid-derived suppressor cells marker Arginase-A I (ArgI) on peritumoral immune cells (P = 0.0666; Supplementary Table S3).

Gene expression profile (GEP) analysis of pretreatment tumors did not show any relevant features except for underexpression of pathways related to proliferative activity and cell metabolism in patients experiencing HP after ICI (Supplementary Fig. S1A and S1B). The analyses of selected genes representative of immune subsets by qRT-PCR revealed overexpression of the CD274 gene, encoding for PD-L1, as the only significant marker in R patients (Supplementary Fig. S1C). FISH of MDM2 and MDM4 genes, carried out on in a cohort of 30 FFPE NSCLC tissues derived from 11 patients with HP and 17 patients without HP, revealed the presence of 3 amplified tumors (2 MDM2, 1 MDM4) in patients with HP and 6 amplified tumors (4 MDM2, 1 MDM4 and 1 MDM2 and MDM4) in patients without HP (Supplementary Fig. S2).

Notably, we noticed that, in some cases, CD163⁺ tumor-infiltrating macrophages showed epithelioid morphology (alveolar macrophage-like) with the tendency to form dense clusters within neoplastic foci (Fig. 1A). In these cases, the same cells were found to coexpress CD33 and PD-L1 (Figs. 1B and 2). Such a peculiar morphology, aggregation, and immunophenotype (CD163⁺CD33⁺PD-L1⁺) of macrophages, which we define “complete immunophenotype,” was observed in all patients with HP and found to be statistically significant versus patients without HP (P < 0.0001; Supplementary Tables S4 and S5). This complete immunophenotype was also observed in 1 P patient, 2 patients with SD, and 1 R patient (Supplementary Table S5). All other cases experiencing treatment response with stable or slowly progressive disease either lacked the presence of epithelioid macrophages or showed loose clustering, or lacked some of the above markers (mainly CD33⁻ and/or PD-L1; Fig. 1C and D).

Histopathologic analyses showed the presence of clustered CD163⁺CD33⁺PD-L1⁺ epithelioid macrophages as a distinctive trait in tumors with HP. Therefore, we sought to test whether macrophages are involved in the detrimental effects associated with anti–PD-1 therapy in preclinical models. Athymic nude mice implanted with human H460 NSCLC cell line were treated either intraperitoneally (Fig. 3A) or peritumorally (Fig. 3B) with anti–PD-1 antibody (clone RMP1-14) or saline. Anti–PD-1 treatment increased tumor growth compared with the control group, regardless of route and schedule of treatment (Fig. 3A and B). Anti–PD-1 treatment was also associated with a significant increase in CD45⁺ leukocyte infiltration at the host–tumor interface, evaluated by IHC (Fig. 3C). Such an increase was mainly due to increasing numbers of intratumoral macrophages (F4/80⁺ cells) and Arginase-I⁺–expressing cells, whereas the density of B lymphocytes (CD45R/B220⁺), granulocytes (Gr-1⁺), and NK (NKp46⁺) cells was comparable with the control group (Fig. 3C). Of note, Arginase-I was also consistently expressed by the complete immunophenotype intratumor macrophages characterizing patients with HP (Fig. 1E). Tumor-associated macrophages (TAM) can express PD-1 and the blocking of this receptor restores antitumor functions (7). Thus, the detrimental boost in tumor growth may not be ascribed to such receptor blockade, but rather to the Fc domain of the antibody which is reported to modulate anti–PD-1 antibody functional activity (13). Accordingly, the same experiments were performed using anti–PD-1 F(ab)₂ fragments. The lack of the Fc portion abrogated the increase in tumor growth observed with the whole antibody (Fig. 3D).

To exclude a direct involvement of PD-1 expression in immune cells, we treated SCID mice with anti-human PD-1 (nivolumab), which does not cross-react with the murine counterpart (Supplementary Fig. S3A). Because a link between HP and EGFR mutational status has been proposed by Kato and colleagues (5), we compared two NSCLC PDXs, with and without EGFR mutation: PDX302 (P53^C135Y, EGFR^L858R, KRAS^WT, APC^WT) versus PDX305 (P53^WT, EGFR^WT, KRAS^G12C, APC^R1114L). On PDXs, before treatment, FACS analysis with human anti–PD-1 antibody showed expression of PD-1 receptor on a subset of tumor cells (around 1% PD-1⁺ cells, Supplementary Fig. S3B). In addition, IHC analysis showed that F4/80⁺ epithelioid/monocytoid elements, aggregated in clusters resembling those identified in patients with HP, were appreciable only in PDX302 (Supplementary Fig. S3C).

SCID mice carrying subcutaneous PDX302 (n = 8), but not PDX305 (n = 8), showed a significant increase in tumor growth rate compared with controls following twice weekly treatment with nivolumab (Fig. 4A). FACS analysis in lungs of PDX302 bearers showed increased cancer cell dissemination in nivolumab-treated mice but not in controls (3.88±1.99% vs. 0.87±0.33%, P = 0.0286, respectively; Fig. 4B), whereas no differences were detected in PDX305 bearers (data not shown).

FACS analysis was also performed on primary tumors for characterization of different myeloid subsets (CD11b, Ly6G, Ly6C, and F4/80) and NK cells (CD49b). In PDX302-bearing mice, an increase in CD11b⁺F4/80^high macrophages was observed in the nivolumab-treated group versus controls (Supplementary Fig. S3D), whereas no significant changes occurred in other immune subpopulations. Accordingly, IHC analysis performed on the same tumors revealed accumulation of macrophages and ArgI⁺-expressing cells (Fig. 4C; Supplementary Fig. S3E). Notably, in PDX302, the F4/80⁺ epithelioid/monocytoid clusters were enriched in nivolumab-treated tumors (Fig. 4C).

Overall, these data indicate that in the EGFR-mutated PDX302-bearing mice, nivolumab triggers a detrimental effect characterized by increased tumor growth, lung dissemination, and the accrual of macrophages, most likely M2.

To further confirm the detrimental effect induced by nivolumab in other preclinical models, PDX111 (P53^C242X, KRAS^G12V, CDKN2A^E69X, CTNNB1^T41I), PDX220 (wild-type for all tested genes), H460 (KRAS^Q61H, STK11^Q37*), and PC9 (EGFR^L858R, EGFR^E746_A750del, CDKN2A^G67V) were xenografted in SCID mice. All tested models showed low levels of PD-1 expression on tumor cells (ranging from 0.6% to 4%, Supplementary Fig. S3B). As with PDX302, an increase in tumor growth was observed in H460- and PC9-bearing mice after nivolumab treatment. PDX111 tumors showed a variable response, whereas tumors in PDX220- and PDX305-bearing mice showed no response to nivolumab (Fig. 4D).

To reinforce the role of the Fc domain of the antibody in boosting tumor growth, PDX302-bearing SCID mice were treated with nivolumab-F(ab)₂ fragments. No HP-like growth was observed (Fig. 4E), whereas in the same experiment, mice treated with the entire antibody showed HP-like growth, dissemination to lung (Fig. 4F), and regional (iliac) lymph node metastases (Fig. 3G). In the group pretreated with clodronate, which reduces F4/80⁺ macrophages, impaired nivolumab-induced tumor growth was observed (Fig. 4E). The F4/80⁺ macrophages also stained for CD206 that marks M2-like subsets and aggregated in fibrotic-like areas in nivolumab-, but not in nivolumab F(ab)₂-treated tumors. In the latter, less fibrotic areas than in control mice were observed (Fig. 4H).

These results suggest that HP is sustained by nivolumab interaction with M2-like macrophages, most likely via Fc–Fcγ receptor binding (Fig. 5).

Although ICI have changed the paradigm of care for patients with NSCLC, an in-depth examination of the Kaplan–Meier curves from the CheckMate-026 (14), CheckMate-057 (15), CheckMate-227 (16), and KEYNOTE-042 (17) trials showed an excess of disease progression and death in the immunotherapy treatment arms compared with chemotherapy in the first 3 months of treatment. This was also underscored by the European Medicines Agency in response to the appraisal for the second-line use of nivolumab in nonsquamous histologies (http://www.ema.europa.eu/docs/en_GB/document_library/EPAR_-_Product_Information/human/003985/WC500189765.pdf). Furthermore, the benefit of ICI in trials conducted in never smokers and in patients with EGFR mutation–positive or anaplastic lymphoma kinase (ALK) mutation–positive NSCLC seems unclear (1, 2, 5, 14, 15). The four most important articles on the HP topic showed prevalence rates ranging from 9% to 29% throughout tumor types, including NSCLC (3–6). In all these articles, the ratio of the tumor growth rates before and during ICI treatment was used to identify HP, although with slightly different cutoffs. None of these studies defined pathologic features able to predict HP, although MDM2/4 amplification and EGFR alterations were proposed to be associated (5). However, we did not find any significant difference in the frequency of MDM2/4 amplification between patients with and without HP, and the role of EGFR cannot be discussed due to the low number of EGFR-mutated patients included in our case series (Supplementary Table S1)

Our definition of HP is different from that used in the abovementioned studies, where radiologic imaging before, at the start and after ICI is needed to identify HP. However, in clinical practice all these radiologic evaluations are often unavailable, and as a consequence, the criteria used in literature are unable to classify patients with HP considering that ICI are starting to be widely used as first-line therapy. Furthermore, both RECIST 1.1 and Immune-related RECIST (irRECIST) criteria, used in the reported analyses, considered only changes in tumor size and did not take into account nontarget lesions, such as lymphangitis and pathologic lesions under 10 mm. In addition, functional and clinical aspects, such as deterioration in performance status, were not considered. Therefore, we decided to include both clinical and radiologic criteria to identify patients with HP in our series. These proposed criteria might overestimate the real fraction of patients experiencing HP; for these reasons, a clinical trial is ongoing within our Institute to properly validate the criteria and thereby obtain the true rate of HP as well as the distinctive immunophenotype.

Nivolumab-treated cell lines and PDX-bearing SCID mice mirrored the clinical observation of HP following treatment with ICI. Interestingly, patients and mice classified with HP share a similar tumor immunophenotype. Indeed, the population of F4/80⁺CD206⁺Arginase-A1⁺ cells emerging from PDXs with HP matches macrophage features of the human counterpart (Figs. 1A, 4C, and 4H; and Supplementary Fig. S3C). M2-like macrophages were preferentially associated with fibrotic foci in PDXs that experience HP-like tumor growth after PD-1 blockade. The recruitment of these myeloid cells may promote a peculiar cancer-associated innate response that may affect tumor growth. Indeed, Knipper and colleagues. described a cross-talk between myeloid cells and fibroblasts promoting skin fibrosis that could provide proliferative and prosurvival signals in cancer cells (18). Prominent mitotic figures can be consistently identified in nivolumab-treated tumor foci embedded in a fibrotic stroma. In human samples, the accumulation of these cells is apparently unrelated to the extent and distribution of tumor-infiltrating T-cell populations.

The role of the innate immune system in mediating the effects of ICI is now clearly emerging. Cells of myeloid origin present in the tumor microenvironment decrease the effects of ICI via PD-L1 expression (19), by “stealing” anti–PD-1 antibody from the membrane of T lymphocytes that return to anergy (20) or by secreting immunosuppressive molecules (21). Our study provides novel evidence of negative immunoregulatory role exerted by PD-L1⁺ macrophages enriched at tumor site under treatment with ICI. Pretreatment lesions from all patients classified as HP showed tumor infiltration by clustered epithelioid macrophages characterized by a CD163⁺CD33⁺PD-L1⁺ profile. Interestingly, CD163⁺ PD-L1⁺ macrophages represent a common immune landscape for different tumors. PD-L1⁺ macrophages have been recently described to accumulate in tight clusters at the tumor invasive margin in NSCLC (22). Macrophages expressing both PD-L1 and the “M2” marker CD163 have been described in MSI colorectal cancer (23), triple-negative breast cancer (24), gastric, and cervical cancer (25, 26). In some of these studies, the presence of PD-L1⁺ macrophages has been associated with poor prognosis (24, 26) and/or with immunosuppressive function through IL10 production (27). In addition, the concurrent expression of CD163, CD33, and PD-L1 has been recently described in alveolar macrophages from patients with acute respiratory distress syndrome, a nononcologic condition present in 10% of subjects admitted to intensive care units (28). In this regard, we observed an increased infiltration of M2 macrophages after anti–PD-1 administration, providing evidence for their involvement in determining HP. TAMs can also express PD-1 that, if neutralized, can restore M1-like properties (7). This likely excludes blockade of PD-1 signaling in our models that remain rather oriented to M2. Therefore, we examined the possibility of FcR engagement as a modulator of anti–PD-1 activities (13). Upon testing the F(ab)₂ moiety in comparison with whole Ab, we have shown that nivolumab without the Fc domain no longer induces HP-like disease in our models.

The anti-mouse PD-1 clone RMP1-14, utilized for the treatment of tumor-bearing athymic mice, is a rat immunoglobulin IgG2a reported to interact with the mouse inhibitory receptor, FcγRIIb (13). FcγRIIb has been shown to be involved in dampening the immune response, and impairments in FcγRIIb function are associated with an exacerbation of inflammatory processes (29). The anti-human PD-1 antibody nivolumab is an IgG4 isotype with reduced binding affinity to activating FcγRs, thereby avoiding antibody-dependent cell-mediated cytotoxicity on PD-1⁺ immune cells (30). However, nivolumab maintains the ability to bind to the inhibitory FcγRIIb receptor (13). Therefore, we suggest a possible role of FcγRIIb in the detrimental effect associated with anti–PD-1 therapy. However, since both human IgG4 and rat IgG2a can bind at lower affinity to other FcRs, the involvement of these receptors cannot be excluded. Furthermore, studies elucidating the involvement of FcRs in the development of HP are required. Very recently, a report described PD-1 expression in one NSCLC case with HP and in one mouse NSCLC cell line. The latter treated with anti-murine PD-1 Ab underwent accelerated tumor growth (31). This interesting and logical explanation of HP cannot be totally supported by the low expression/prevalence of PD-1 on tumors and by F(ab)₂ experiments in mice. We found clusters of PD-1–expressing cells in 2 of 11 patients with HP (data not shown) and in all xenografts and PDXs, although at low levels (0.6%–4%) and with no correlation with HP-like progression.

In conclusion, the ongoing debate regarding whether HP is a true phenomenon or only representative of patients with a particularly worse prognosis is confirmed in our preclinical models, where ICI are able to boost tumor growth in a manner akin to the clinical observations of HP in patients with NSCLC. Our results suggest that FcR triggering of clustered epithelioid macrophages with a specific immunophenotype by ICI delivers a signaling cascade that promotes functional reprogramming of these cells toward a more aggressive protumorigenic behavior. This eventually induces HP in a subset of patients with distinctive immune and genetic profiles (depicted in Fig. 5). Our analyses, for the first time, suggest a role of innate immunity in this process. A further prospective validation of the HP immunophenotype and its relationship with specific genotypes, as well as the new proposed clinical criteria to classify HP, is ongoing.

D. Signorelli is a consultant/advisory board member for AstraZeneca. A. Anichini reports receiving commercial research grants from Bristol-Myers Squibb. M. Garassino reports receiving commercial research grants from MSD, speakers bureau honoraria from Bristol-Myers Squibb, MSD, Roche, and AstraZeneca, and is a consultant/advisory board member for Bristol-Myers Squibb, Roche, MSD, and AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Lo Russo, M. Moro, V. Cancila, S. Marsoni, A. Anichini, L. Rivoltini, A. Balsari, G. Sozzi, M.C. Garassino

Development of methodology: G. Lo Russo, M. Moro, G. Centonze, P. Gasparini, M. Milione, L. Porcu, C. Storti, A. Anichini, M.C. Garassino

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Lo Russo, M. Moro, M. Sommariva, M. Ganzinelli, C. Proto, G. Pruneri, D. Signorelli, S. Sangaletti, L. Sfondrini, E. Tassi, G. Sozzi, M.C. Garassino

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Lo Russo, M. Sommariva, V. Cancila, M. Boeri, S. Ferro, M. Ganzinelli, V. Huber, M. Milione, L. Porcu, C. Proto, G. Pruneri, D. Signorelli, L. Sfondrini, E. Tassi, A. Bardelli, V. Torri, C. Tripodo, M.P. Colombo, A. Anichini, G. Sozzi, M.C. Garassino

Writing, review, and/or revision of the manuscript: G. Lo Russo, M. Moro, M. Sommariva, M. Ganzinelli, C. Proto, G. Pruneri, D. Signorelli, E. Tassi, A. Bardelli, S. Marsoni, V. Torri, C. Tripodo, M.P. Colombo, A. Anichini, A. Balsari, G. Sozzi, M.C. Garassino

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Lo Russo, M. Boeri, C. Proto, D. Signorelli, M.C. Garassino

Study supervision: G. Lo Russo, S. Marsoni, L. Rivoltini, G. Sozzi, M.C. Garassino

The authors wish to thank the following individuals: Dr. M. Figini for the production of nivolumab-F(ab)₂, Dr. L. De Cecco for GEP profiles, Dr. M. Dugo for bioinformatic analyses, Mrs. A. Cova and Dr. A. Berzi for technical assistance, Dr. A. Martinetti for biological sample collection, Dr. E. Tagliabue for data discussion and interpretation, Dr. G. Morello for her precious support in performing double-marker immunolocalization analyses, and Dr. G. Apolone for continuous support in the planning and execution of the study and for scientific writing. The authors would also like to acknowledge the editorial assistance provided by Chris Cammack, a professional medical writer at Ashfield Healthcare Communications, an Ashfield Company, part of UDG Healthcare plc. This work was supported by the Italian Ministry of Health (5 × 1000 Funds – 2014, to M.C. Garassino; grant no. GR-2013-02355637, to S. Sangaletti), AIRC (Associazione Italiana per la Ricerca sul Cancro; grant no.15190, to A. Balsari; grant no. 18812, to G. Sozzi; grant no. 17431, to A. Anichini; grant no. 10137, to M.P. Colombo), and EU project Horizon2020-686089 - PRECIOUS (to L. Rivoltini). E. Tassi was supported by a fellowship from Fondazione Beretta-Berlucchi.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.