|Home | About | Journals | Submit | Contact Us | Français|
The histopathology of type 1 diabetes is defined by a decreased β-cell mass in association with insulitis, a characteristic lymphocytic infiltration limited to the islets of Langerhans and prominent in early stage disease in children. A cytotoxic T-cell mediated destruction of insulin-producing β-cells is thought to be initiated by an unknown (auto)antigen, leading to the destruction >75% of β-cell mass at clinical diagnosis. Although considered to be pathognomonic for recent onset disease, insulitis has only been described in approximately 150 cases over the past century. This review describes the quest for this elusive lesion and gives its incidence in various patient subpopulations stratified for age of onset and duration of the disease. It discusses recent new insights into the regenerative capacity of the β-cell mass in the pre-clinical stages of the disease and relates these findings to the inflammatory processes within the islet tissue.
More than 100 years ago a characteristic inflammatory infiltration limited to the islets of Langerhans was described in a diabetic child that died in ketoacidosis.1 The lesion was later called insulitis2 and is now considered pathognomonic for type 1 diabetes in children with recent onset of disease.3 The infiltrate consists predominantly of T-cells, in which CD8+ lymphocytes dominate, but may also contain CD4+ lymphocytes, B-lymphocytes and macrophages.4 The cellular response is accompanied by a humoral response that includes autoantibodies against a wide array of b-cell antigens.5,6 However, the precipitating (auto)antigen against which the inflammatory response is directed has not been identified, nor has it been established whether the humoral response that is considered to be part of our current diagnostic criteria is a cause or a consequence of the disease. Although animal models for the disease exist, like the spontaneously diabetic NOD mouse, they are found to differ from the human disease in many key aspects7 and it is an open question whether data derived from such models will be applicable to patients. In fact, even after a century of research we know very little about the etiology and histopathology of the human disease. As will be shown in this review, one reason for this relative lack of knowledge is the very limited amount of patient data that is available for study. The pancreas is a difficult organ to biopsy and most of the material is therefore post-mortem. The islets are scattered in a matrix of exocrine tissue and thus form only 1–2% of the parenchymal tissue, with the inflammatory lesions often being rare and only affecting a small part of the islets that, in addition, are not homogeneously distributed throughout the gland, but are often located within a few lobes, while islets in other lobes remain intact. The lesions are mainly found in islets in which β-cells are still present and the lesions will largely disappear together with the β-cells against which the reaction appe ars to be directed. In addition, the few cases that were brought to autopsy often died in ketoacidosis, they may thus represent a more fulminant version of the disease that is not necessarily characteristic for the disease process in the rest of the population. Lastly, and perhaps most importantly, the histopathological lesions that we observe in cases with recent onset disease will only show the final stages of a process that has been going on for a long period of time, and until recently, we had no material available of earlier stages of the disease. Identifying patients with pre-clinical disease and studying the immunological processes occurring at this stage, may prove to be indispensable for a breakthrough in our quest for the etiology of the disease. Initial results, described in this review, will show that such observations may give rise to important new insights into the early events in the pre-diabetic islets. They indicate that human β-cells show a remarkable potential for regeneration at the pre-clinical stage and that the key question in addressing type 1 diabetes may not be why β-cells disappear, but to quote one of the early students of human insulitis, Philip LeCompte, why the β-cells are inhibited from regenerating.8
Inflammatory infiltrates in the islets of Langerhans were first described in 1902 by the German pathologist Schmidt,1 who found foci of small-cell infiltration in the periphery of islets of Langerhans from a 10 year old diabetic child with an unknown duration of disease. This islet-specific inflammation, later termed “insulitis” by the Swiss pathologist von Meyenburg,2 was long considered to be a rare event. Cecil9 described leucocytic infiltration associated with islets in 9 out of 90 patients with diabetes, but often under conditions in which a more generalized pancreatitis was present; he observed islet-specific inflammation in only a single case, involving a young adult patient with recent onset disease. In 1928 Stansfield and Warren10 were the first to draw attention to the association between insulitis and the age of the patient; they described insulitis in a six year old girl who died in a diabetic coma two months after onset of the disease, and in an 11 year old girl who died in a diabetic coma within four weeks after the initiation of symptoms. In their view, the striking lymphocytic infiltration in the islets of both cases suggested a causal relationship between the inflammation and the diabetic condition in these two young patients with recent onset fulminant disease. On the other hand, it was clear from their studies in larger groups of children that insulitis was not always observed: Warren11 observed insulitis in only one out of ten patients who developed diabetes at a young age and coming to autopsy six weeks to 29 years after onset. His focus on young patients was not so much motivated by a hypothesis that the disease in children could constitute a separate disease entity, but “since it is in children that we find the best examples of pure, uncomplicated diabetes mellitus, and since their organs are free from the various degenerative changes so often encountered in adults.” These early and inconclusive observations were revisited in 1958 by LeCompte8 who collected four cases with insulitis, all involving acute onset disease and short duration in children. He proposed four possible explanations for the presence of the cellular infiltrate: a direct invasion of the islets by an infectious agent, a manifestation of functional overstimulation or strain, a reaction to damage by some unknown nonbacterial agent and lastly an antigen-antibody reaction. Fifty years later one could still make the same list, as none of these possibilities has been excluded. LeCompte still considered insulitis to be a rare lesion, although he predicted that it was underdiagnosed as it was easily missed in microscopy using the conventional haematoxylin, aldehyde-thionin trichrome or Gomori's chromium haematoxylin-phloxine stains current at that time. The Brussels based pathologist Willy Gepts, assisted by the generous contribution of tissue samples from young recent onset cases from many different hospitals in the USA, Canada and Belgium, including the four cases from LeCompte, was the first to collect a sufficiently large collection of cases and controls to assess the true incidence and significance of insulitis in young patients. In a 1965 landmark study,12 he reported the presence of the lesion in 15/22 (68%) recent onset cases below the age of 40 and noted that it was not present in patients with a disease duration of more than a year. He also noted that β-cell mass appeared to be reduced to approximately 10% of that in non-diabetic controls, albeit with a considerable variation in the degree of β-cell loss between patients. Reanalyzing his data several years later, using newly developed immunohistochemical techniques,13 he noted that the inflammation was preferentially found in islets that still contained insulin immunoreactivity (Fig. 1A). In addition to insulitic islets, he found that the pancreas of acute-onset cases, and especially the pancreas of chronic type 1 diabetics, contained islets from which the β-cells (and the inflammatory infiltrate) had disappeared, these so called ‘pseudo-atrophic’ islets consisted mainly of α-cells (Fig. 1B) and showed a light degree of fibrosis of the islet interstitium. Interestingly, in the acute cases he also observed a variable fraction of islets that appeared to be normal in all respects, including a normal complement of β-cells and the absence of infiltrating cells (Fig. 1C). He pointed out that the remaining β-cells appeared to be hypertrophic and degranulated and that abnormally large islets were present, but that no evidence of β-cell replication could be found. He also observed that some patients presented signs of β-cell regeneration that appeared to be limited to one or more lobes of the pancreas (Fig. 1D). His description of other islet cell changes, including β-cells with an empty (hydropic) cytoplasm and β-cells with cytoplasmic inclusions (“körnchen”), still await further investigation. Gepts was prescient with his observation that the presence of lymphocytic infiltrates might point to an autoimmune etiology of the disease: referring to recent experiments in animals in which anti-insulin serum induced a form of insulitis, to the detection of autoantibodies directed against islet tissue in the serum of a young diabetic and to the simultaneous presence of autoimmune thyroiditis and juvenile diabetes in one of his cases, he stated that “these observations pointing to a possible immunologic derangement are of the greatest interest in view of the high frequency with which we have found inflammatory infiltrates in our recent onset juvenile diabetics.” In his 1978 follow up study he found additional arguments for an autoimmune process, as he observed that inflammation seemed to preferentially target β-cell containing islets, while islets without β-cells were devoid of lymphocytic infiltration, leading him to the conclusion that “insulitis represents an immune reaction of the delayed type, specifically directed against β-cells.”
Initially, Gepts's observations that a high fraction of recent-onset type 1 diabetic children presented with insulitis, met with resistance. Doniach and Morgan14 studied a group of 13 young patients with a disease duration of less than one year and were unable to find insulitis in any of these cases, concluding that lymphocytic insulitis was “less common than suggested previously.” However, other authors soon supported the findings. Junker et al. found insulitis in 6 out of 11 juvenile diabetics aged less than 30 years and dying within two months after the onset of symptoms (ref). Foulis et al.16 using a 25-year computerized survey of deaths in the UK to identify 119 young patients who died in ketoacidosis before the age of 20, in combination with immunohistochemistry to identify islets and infiltrating leucocytes, confirmed that insulitis was present in 47 out of 60 (78%) of young patients with recent onset disease (<1 year). They also confirmed the observation by Gepts that β-cell containing islets appeared to be preferentially affected, by quantifying the fraction of islets with insulitis in insulin-containing and insulin-deficient islets. They found that although the lesion was present in 23% of insulin-containing islets, it was only found in 1% of insulin-deficient islets, thus supporting the concept that insulitis represents an immunologically mediated destruction of insulin secreting β-cells. However, they also pointed out that a certain heterogeneity seemed to exist in their patient population, as they also observed young-adult patients with a short duration of the disease that showed no evidence of insulitis and in which all islets contained insulin. Together, the two landmark studies by Gepts and Foulis account for almost half of all cases with insulitis that have been published over the last century. As the number of new cases with type 1 diabetes and insulitis has been averaging approximately 1–2 cases/year worldwide, it is not surprising that large-scale follow-up studies have been lacking over the past 25 years and that the total number of (pre) diabetic patients in which insulitis was observed is limited to approximately 150 cases (Table 1).
The nature of the lymphocytic infiltrate invading the islets of Langerhans was first investigated by Bottazzo et al. in the pancreas of a 12 year old girl that died in ketoacidosis one month after the onset of diabetes. Insulitis was present in 24% of the islets and was limited to islets that were still insulin-containing. Immunophenotyping of the infiltrate showed that most cells corresponded to T cells, with T cytotoxic/suppressor cells being most abundant, although T helper cells and NK cells were also present. Macrophages or monocytes were not observed. These results were confirmed Hänninen et al (ref. 28). who observed that CD8+ T lymphocytes (T cytotoxic/suppressor) were the main infiltrating cell type, although they also observed significant numbers of macrophages in the islets of a eight year old girl that died within two weeks of onset of symptoms. Both Itoh et al.,29 who investigated pancreas biopsies from 18 newly diagnosed patients between 17 and 49 years, and Lernmark et al (ref. 32). who studied two cases of recent onset in children, supported the finding that in addition to lymphocytes, macrophages were a prominent feature of the infiltrate; although Somoza et al.(ref. 31) were unable to find such cells in insulitic lesions present in a 19 year old woman who died 10 days after onset of the disease. These somewhat conflicting findings were apparently resolved by a recent study on the composition of the infiltrate in insulitis, using 29 recent-onset patient samples from the original Foulis study.4 The composition of the cellular infiltrate was stratified according to the percent β-cells present in 279 individual islets collected across all 29 patients. It was surmised that the percent β-cells would be a surrogate marker for the stages of advent of the insulitic lesion, with 50–69% insulin-positive area taken as starting point and 0% insulin-positive area as end-stage. It was found that at all stages, CD8+ T cells were predominant, increasing in number with decreasing insulin-positive area, but disappearing when insulin-positivity was completely lost. CD20+ B-cells were found to be the second most prominent cell type, following the dynamics of CD8+ cells, while macrophages were present at relatively constant levels becoming the most prominent infiltrating cell type in insulin-deficient islets. Whether stratifying islets according to the percent insulin-positive area is a valid approach to analyze the temporal progression of insulitis, remains open for discussion. The approach does not allow for heterogeneity in the patient group, nor does it allow for attempts at regeneration of affected islets. As was recently shown,40,43 islets with insulitis contain replicating β-cells, indicating that β-cells retain a substantial capacity for growth that appears to be activated under conditions of inflammation.40,44 It cannot be excluded that such newly formed cells attract recurrent autoimmune attack and that islets with a low insulin-positive area represent islets with regeneration rather than islets in the last stages of β-cell destruction.
Phenotyping the infiltrating cells in insulitic lesions is only a first step in a process to identify the antigen against which the infiltrate is directed. The key step will be to analyse their specificity. No direct analysis of insulitic T-cell specificity has been reported to date, although CD4+ T-cells isolated from pancreas-draining lymph nodes from two chronic type 1 DM patients, in which insulitis was not observed, were shown to recognize parts of the insulin molecule.35 Although the predominantly CD3+CD8+ phenotype of the infiltrating cells is compatible with a cytotoxic T-cell mediated β-cell destruction, it has not yet been proven that the cells that we observe in the insulitic lesion are the cells that are actually responsible for the destruction of the β-cell component. It is equally well possible that a large part of such infiltrates are the consequence of β-cell destruction rather than its cause, at least at this relatively late time point in the progression of the disease. Samples from autoantibody-positive pre-diabetic individuals are now slowly becoming available,40,45,46 and are of the highest importance for studying T-cell receptor specificities. Clearly, identifying the nature of the antigen and establishing whether it is a β-cell autoantigen35 or of viral origin24,39 will be crucial in devising a therapeutic strategy.
What we observe at clinical presentation is, to quote Gepts “the final stage of a process that may have been going on for a long period of time.”12 If the presence of circulating autoantibodies against islet cell antigens is considered as a surrogate marker for β-cell destruction, than the process may take years before the clinical threshold is reached.6 Determining the incidence and time course of insulitis prior to diagnosis and correlating it to the presence (and persistence) of circulating immune markers, will be crucial for our understanding of the disease and for the development of immune intervention strategies. It will be important to correlate such information to the regenerative capacity of the β-cell mass. Not all individuals who are autoantibody-positive progress to overt disease, and the process may well involve episodes of fulminant destruction followed by episodes of repair and regeneration.
Several studies have tried to identify insulitic lesions in non-diabetic patients with circulating markers of autoimmune disease, and a total of 66 of such patients have come to autopsy or donated their pancreas for transplantation.40,45,46 In the most extensive study,40 insulitis was found in 2 out of 62 autoantibody-positive non-diabetic organ donors (Fig. 1E). The lesion was present in 3–9% of the islets in two adult donors (46 and 59 years) who were positive for multiple autoantibodies in combination with a susceptible HLA-DQ genotype, but was not found in any of the other 60 patients, most of which showed only positivity for a single autoimmune marker. Interestingly, the presence of insulitis was not accompanied by the presence of significant numbers of pseudo-atrophic islets and the relative β-cell mass was unaffected. In one patient a high level of β-cell replication was found, that was limited to the inflamed islets (1f). It is tempting to speculate that in the two autoantibody-positive donors that did not become diabetic, the insulitis-induced β-cell loss was fully compensated by the formation of new β-cells. The apparent absence of (sufficient) regeneration in auto-antibody positive individuals who progress to overt disease is a key element that warrants further study. β-cell replication appears to be absent or low in recent onset patients,37,43 raising the question whether under some conditions inflammatory infiltrates are beneficial to replication and repair, while under other conditions they are inhibitory. Studying patients that successfully compensate (or avoid) inflammation-induced β-cell loss and comparing them to patients who progress to overt disease may help to clarify this issue.
It is sobering to realize that all our knowledge about the histopathology of a major disease is based on a very limited set of cases, often studied before the advent of modern analytical techniques. Since 1902 a total of approximately 150 cases with insulitis have been described in the literature (Table 1), out of which approximately 61 are acute cases (<1 month) in young individuals (<15 years). Few of these cases have been studied in depth: early cases lacked immunohistochemistry for islet hormones and were based on classical histological stains that only allow a tentative characterization of islet cell types. Very few cases have been analyzed with molecular techniques. Detailed clinical history, immunological status and estimation of functional β-cell mass are only rarely available and perhaps most importantly, most recent onset cases involve patients who died in a diabetic coma, thus selecting in favor of patients with a fulminant version of the disease and not necessarily providing an undistorted image of the natural history of the disease. Fortunately several recent initiatives have been taken to remedy this deficit, establishing biobanks that collect and provide high quality samples, detailed clinical data, unrestricted access and most-importantly samples from potentially pre-diabetic individuals identified on the basis of autoantibody positivity. In anticipation of a more detailed and refined analysis of insulitis, the currently available data were pooled and analyzed below.
When all cases of type 1 diabetes for which minimal clinical and histopathological data are available in the literature are combined, and when only those cases are included that can be considered to be derived from population-based studies,12–15,29,37,47–49 a total of 213 cases can be included in a meta-analysis (Table 2). The combined data sets show that insulitis occurs in 73% of young (≤14 years) type 1 diabetic patients with a short (≤1 month) duration of the disease, in 60% of young patients with a disease duration between one month and one year, and only in 4% of young patients with a duration of disease longer than one year. One of the largest studies16 could not be fully included in this analysis as it did not provide clinical details about the insulitis-negative patients, but the overall outcome of this study with 39 out of 45 patients (87%) below 15 years of age and with a disease duration of <1 year showing insulitis, is in line with the current analysis.
As noted in earlier meta-analyses,32,50 insulitis is not a common lesion in older patients with acute onset of the disease. In our cumulated data set, only 29% of cases with onset between 15 and 40 years and a disease duration <1 month showed insulitis, which is significantly less than the 73% in the corresponding group of young patients. A similar observation can be made for the two groups with disease duration between one month and one year (Table 2). Several possible explanations can be brought forward to explain these differences: a first possibility is that the disease process is less fulminant in older individuals and that fewer islets are affected, leading to an underestimation of the percentage of affected individuals. Relatively small numbers of islets are examined for each patient (typically several hundred) and the presence of insulitis can be more easily missed when only small numbers of islets are affected. However, as can be calculated from previously reported data in references 16 and 49, there is a similar average fraction of islets with insulitis (versus the total number of islets, including both insulin-containing en insulin-deficient islets) in young patients (8.7%) with acute disease (<15 yrs and ≤1 month duration), as there is in the corresponding older age group (9.6%), making it unlikely that the difference in the percent affected individuals in the two groups is due to the fact that insulitic islets are present, but that they are just being missed due to their low numbers. A second possibility is that the differences are the result of a sampling bias, caused by a more focal distribution of inflammation and/or larger organ size in older patients. As sampling is often limited to a single tissue block from only one gland region, the lesion may be missed. This risk may even more pronounced in adults, as their pancreas is larger. Focal distribution of inflammatory lesions has been described by a number of authors,12,16,37,42 and appears to be one of the characteristics of the disease. The reasons for the focal distribution are unknown, but appear to be related to the lobular anatomy. A third possible explanation is that the etiology of the disease is more heterogeneous in the older age group and that part of the patients in this group have a different form of the disease that is not characterized by insulitis.50 Clearly there is a need for additional studies into the histopathology of acute non-adolescent type 1 diabetic patients. Such studies should be population-based, and should preferably not be limited to patients who died in keto-acidosis, as this in itself constitutes a bias towards the more fulminant types of the disease. Such studies should also use a clear, and preferably commonly accepted, definition of insulitis, a systematic sampling of the whole organ, correlation to clinical and immunological markers and a unique patient identification to allow unequivocal identification of the patient material when this is used in multiple studies.
Insulitis is usually limited to β-cell containing islets, and disappears together with the β-cell component, resulting in pseudo-atrophic islets in which both inflammatory infiltrates and residual β-cells are rare.12,13,16 However, not all islets appear to be affected by the disease and the effects are less pronounced in older individuals. In young patients with recent onset disease (<1 month) an average of 34% of the islets contain β-cells (Table 3), but only 33.6 ± 3.2% of these islets also show insulitis.16,49 In older patients with recent onset disease an average of 63% of the islets contain β-cells, but with only 18.3 ± 6.1% of the insulin-containing islets also showing insulitis. These observations indicate that, at least in the first month after diagnosis, an important residual β-cell mass is present that is especially large in individuals that develop the disease at a more advanced age. β-cell containing islets with a normal architecture and free from infiltrating cells are relatively frequent in recent onset patients.12,13,16 If one assumes that such islets have retained (or regained) their normal cellular composition, one can estimate that young patients have approximately 20% of their original β-cell mass in the first month after diagnosis. Such a morphological finding is supported by the analysis of the functional β-cell mass in C-peptide positive patients levels with recent onset disease, which averaged 25% of that in matched controls,51 indicating a window of therapeutic opportunity to stabilize or counteract the disease in the first weeks after diagnosis.
In patients with chronic disease (>1 year), insulitis is rare and is found in only <5% of patients. Interestingly, most patients with chronic disease still retain a significant fraction of β-cell containing islets, averaging 13% of the total islet population (Table 3). Insufficient data is available on the incidence of insulitis in such islets or on the islet architecture and cellular composition. Only two studies are available in which the β-cell mass was determined by morphometry using a systematic sampling of the whole organ: the combined data from these studies shows that in the six cases that were studied (with a disease duration of 1–34 years) the β-cell mass ranged from virtually zero to approximately 25% of that of the lowest nondiabetic control.52,53 A study on a group of 14 Japanese patients with disease duration between 13 and 46 years, also showed marked differences between patients, with the highest levels of β-cell mass in four patients with ICA positivity.30 Finally, a qualitative study on 26 patients with a disease duration of 2–54 years, using extensive and systematic sampling of the whole pancreas, showed residual β-cells in 13 out of 26 cases.54 The β-cells were found to be focally distributed and were present in a single lobe in five patients. Insulitis was not found in any of these cases. The authors inferred from their data that most of the patients whose pancreata contain β-cells after a diabetes duration of 10–20 years, would retain these cells for the rest of their lives. These observations are supported by a recent histopathological study on nine cases with a duration of diabetes of 52–84 years, in which residual β-cells were found in all pancreases, often as scattered single extrainsular cells or as significant numbers of small clusters in specific lobes.55 Both β-cell apotosis and β-cell replication were observed in some of these patients, as were signs of low level immune infiltration. A lobular distribution of residual β-cells was recently also described by Gianani et al.42 who studied 20 patients with chronic disease and found that the six patients with residual β-cells showed two different patterns of β-cell localization, either the pancreas contained a mixture of insulin-deficient and insulin-containing islets (pattern A), the latter being present only in some regions of the pancreas, next to regions with only insulin-deficient islets, or all islets contained at least some β-cells (pattern B). All of these studies suggest that variable amounts of β-cells remain present in many type 1 diabetic patients for many years after diagnosis. The often lobular distribution of the remaining cells is reminiscent of the lobular distribution of insulitis13,16 and the lobular distribution of regeneration13 that is observed in recent onset cases. Whether the remaining cells persist through survival or are continuously replenished through neogenesis and or mitosis is not known. Tentative evidence of a continuous neoformation and apoptosis cycle in chronic type 1 patients55,56 might indicate the latter possibility. Additional studies are needed in patients with chronic type 1 diabetes to determine the extent of β-cell survival. These studies should take the potentially lobular distribution of such residual cells into account. They should investigate the relationship between residual β-cells, age at onset, persistence of autoantibodies and severity of complications as a basis for the development of new therapeutic interventions.
Currently, no common definition of insulitis exists. Most early pathologists did not use pre-defined criteria to identify insulitis. More recent studies used quantitative criteria and control groups to set threshold levels of two,29 three37 or five42 infiltrating cells per islet (section) to define insulitis. Studies trying to identify insulitis in autoantibody-positive but non-diabetic organ donors used more stringent criteria with thresholds of 15 infiltrating cells per islet40 or statistically compared infiltration levels to that in a control group.45 Any operational definition of classical insulitis should take the following elements into account: the number of infiltrating cells in or closely apposed to the islet versus the number of infiltrating cells in the surrounding tissue an in matched controls, the typical clustering of islet infiltrating cells into closely apposed aggregates and the predominantly lymphocytic phenotype of the infiltrating cells with CD3+CD8+ positivity. Recent data indicating that the level of infiltrating cells in the pancreatic parenchyma varies according to the clinical conditions surrounding the patients death, with the numbers of infiltrating cells increasing with the duration of life support,44 emphasizes the need for clear quantitative criteria and especially emphasizes the need for properly matched controls.
The single most important impediment for our studies into the etiology of human type 1 diabetes is the limited availability of material to study. Most cases listed in Table 1 are only available under the form of chemically fixed and paraffin embedded tissue blocks and very few cases have been available were tissue was suitable for molecular analysis. Biobanks collecting and storing high quality tissue samples, blood samples and clinical data from autoantibody-positive pre-diabetics, type 1 diabetic patients and from matched controls will greatly facilitate our research effort. Such Biobanks, now being formed both in the US (nPOD) and in Europe (Brussels, EuroPod), will give an important new impulse to type 1 diabetes research and will hopefully lead to conclusive results as to its true relevance with regard to the etiology of the disease and the identification of the antigen(s) against which the infiltrate is directed.
These studies were supported by grants from the Juvenile Diabetes Research Foundation (26-2008-896), the Fonds voor Wetenschappelijk Onderzoek Vlaanderen (G019211) and the Vrije Universiteit Brussel. We thank Nicole Buelens for expert technical assistance.