Anti-GM-CSF antibodies have been described in various and not necessarily overlapping clinical situations. We queried whether the characteristics of the anti-GM-CSF antibodies (in terms of titers and neutralizing activity) varied between patients with autoimmune PAP and those with autoantibody-related severe infections. We found that (i) patients with autoimmune PAP tended to have higher titers of anti GM-CSF autoantibodies, (ii) titers and global neutralizing activity of plasma correlated, and (iii) there was no significant difference between groups in terms of neutralizing activity of anti-GM-CSF autoantibodies. The disease induced by anti-cytokine antibodies is the result of a complex combination of several factors including the characteristics of the antibody (e.g., epitope, titer, Ig subclasses, avidity), level of patient exposure to certain irritants and pathogens and potentially individual genetic susceptibility. Characteristics of the antibody have been shown to contribute to disease activity and phenotype in other autoantibody-mediated diseases [19, 20].
Anti-GM-CSF antibodies have been extensively studied in the context of autoimmune PAP. A genetic predisposition to the appearance of anti-GM-CSF antibodies has long been suspected; genetic studies have provided conflicting results depending on ethnicity or techniques used so far. [21, 22]. Serum anti-GM-CSF autoantibody levels, unlike levels in bronchoalveolar fluid, do not correlate with disease severity in patients with aPAP [23, 24]. It seems that a “critical threshold” (5 µg ml− 1) exists above which anti-GM-CSF antibodies are likely to disrupt GM-CSF-stimulated functions in alveolar macrophages and blood leukocytes [25, 26] and are strongly associated with an increased risk of autoimmune PAP . Monoclonal anti-GM-CSF antibodies from patients with autoimmune PAP have been isolated using cloned memory B cells. The authors identified several clones, mainly of the IgG1 subtype, targeting multiple non-overlapping epitopes and utilizing multiple immunoglobulin variable region V genes [28, 29]. Further, the molecular and crystal structures of anti-GM-CSF antibodies have been solved [30–32]. The targeted epitopes can be linked to differences in affinity . The antibody binding to GM-CSF is further refined by somatic mutations . It appears that it takes a combination of three or more non cross-competing antibodies to effectively neutralize GM-CSF activity in vitro and in vivo, while a single anti-GM-CSF neutralizing monoclonal antibody may not be harmful . Piccoli et al., showed that injection of a cocktail of different antibodies could lead in vivo to the formation of high molecular weight complexes sequestering GM-CSF and promoting the rapid degradation of GM-CSF-containing immune complexes in a Fc-dependent manner . Taken together, these observations show that (i) anti-GM-CSF autoantibodies in autoimmune PAP are composed of polyclonal IgG, (ii) suggest that their formation is driven by GM-CSF itself, (iii) that the memory B cell maturation process is helped by T cells, implying that somatic mutations determine antibody affinity, and (iv) are reassuring that treatments based on single anti-GM-CSF monoclonal antibodies that are currently in development for inflammatory diseases may not necessarily cause PAP or infectious disease by themselves [33, 34].
Our study is the first to address and to compare anti-GM-CSF antibody characteristics in diseases other than aPAP. Previous reports have noticed a predominance of IgG1 subclass antibodies in patients with Cryptococcus or Nocardia infections, as is true for aPAP [11, 12]. GM-CSF secretion is induced during infectious or inflammatory processes; infectious disease preceeding aPAP diagnosis may have suggested a pathogen-induced autoimmunity mechanism in some cases [12, 35]. However, the demonstration of high-titer anti-GM-CSF autoantibodies 10 years before CNS nocardiosis and 15 years before autoimmune PAP in one patient renders unlikely this hypothesis. What, if anything, distinguishes anti-GM-CSF autoantibody patients with Cryptococcus or Nocardia infection from each other and from patients with PAP alone remains obscure. Studying samples from these three types of patients, we showed that titers do not differ dramatically between groups, except for a slightly higher level in PAP patients. The in vitro neutralizing effects of anti-GM-CSF autoantibodies were not statistically different by phenotype. Plasma from PAP patients tended to be more effective at blocking GM-CSF signal (effective at 1.6% dilution) than plasma from patients with cryptococcosis (3.9%, over 2 times more). The use of whole plasmas cannot exclude possible non-antibody inhibitory effects. However, anchoring the plasma dilution to the antibody titer, we did not find any difference in IC50 determined as the concentration of anti-GM-CSF autoantibody responsible for 50% inhibition of GM-induced pSTAT5. We did not find it either regarding the EC50 for pSTAT5 experiments. For each set of experiments, PBMCs from the same healthy donor were used in order keep a constant genetic background underlying the cellular response.
These in vitro effects may not perfectly reflect the neutralizing activities of the antibodies in vivo . Further studies on the kinetics and structures of the antibodies in these three groups of patients are warranted. Alternative hypotheses could include tissue-restricted specificities. Anti-GM-CSF antibodies may act differently on pulmonary macrophages than circulating monocytes. Therefore, the lack of difference between the antibodies from these three groups, studied solely on circulating monocytes, may not accurately reflect what happens in alveolar spaces. Indeed, human circulating monocytes and monocyte-derived macrophages differ in various cellular functions and do not accurately recapitulate the biology of human lung macrophages [36, 37]. Macrophages from the central nervous system are also distinct from circulating monocytes and monocyte-derived macrophages. Finally, it may be that the antibodies are indeed all essentially the same and some other aspect of underlying host experience or genetics is the determinant of the clinical manifestation. Cigarette smoking and dust exposure have been suggested as causal or aggravating factors for aPAP, by impairing lung capillaries permeability for example and favoring the leaking out of antibodies into the alveolar space [38, 39]. Comprehensive genome sequencing in all affected are also underway to resolve the possibility of a genetic susceptibility.
Almost every case of anti-GM-CSF–related disseminated infection in our cohort involved the CNS. The search for anti-GM-CSF antibodies is now recognized as part of the workup in
“non-immunocompromised” patients with brain infection with Cryptococcus or Nocardia [40, 41].
Recent studies suggest that GM-CSF, elicited by IL-23-driven T cells, favors CNS invasion by phagocytes and secondarily myeloid-mediated tissue immunopathology [42–44]. Anti-GM-CSF antibodies are detectable in the cerebrospinal fluid of patients with Cryptococcus meningitis, albeit at lower levels than in the circulation . To what extent these autoantibodies can disrupt local immunity against fungal or bacterial pathogens and whether they predispose to CNS infection specifically needs further exploration.
Longitudinal samples may be particularly informative. In our patients who developed different manifestations of anti-GM-CSF autoantibodies, the infections usually preceded PAP. In only one patient was PAP diagnosed concomitantly with cryptococcal meningitis. Globally, PAP seems to occur later in clinical evolution. As it does not appear to be a matter of circulating antibody levels or global neutralizing effect, this temporality could be due to cumulative effects over time or changes in exposure.
Our study was retrospective, gathering samples from different centers; some clinical data may have been overlooked. The precise date of onset of the initial manifestation may have been missed. Additionally, the delay between diagnosis and date of sampling was variable between patients, with some sampling made at intervals from onset and after several lines of treatment had already been attempted. These realities make the value of our rare longitudinal samples greater. Regardless of these shortcomings, which might serve to artifactually amplify differences, we found very few.