Pathogenesis of haemophagocytic lymphohistiocytosis


Maurizio Aricò, Clinica Paediatrica, IRCCS Policlinico S.Matteo, 27100 Pavia, Italy. E-mail:

Haemophagocytic lymphohistiocytosis (HLH) is a rare, fatal disease of early infancy. The familial form of this disease was originally described by Farquhar and Claireaux (1952) and overlooked for several years. Janka (1983) detailed the main features of familial HLH in a comprehensive clinical review. Autosomal recessive inheritance has been demonstrated for familial HLH (Gencik et al, 1984). As only a limited proportion of cases have documented familial recurrence, despite similar presenting features and outcome, the term HLH is commonly used to indicate the disease in both familial or apparently sporadic cases. Its incidence has been estimated to be 0·12 per 100 000 children in a retrospective Swedish study (Henter et al, 1991a). A similar incidence was later found in a British study (cited in Henter et al, 1998). In a recent Italian study, the estimate of the incidence of HLH in the whole country over the last 25 years was 0·006 per 100 000 children per year, but in the Southern part of Italy the disease was three times more frequent than in the Northern and Central areas. Owing to the improvement in detecting HLH cases, the observed frequency of the disease has increased over time, reaching the value of 0·118 per 100 000 children per year in the last period (1990–99) (unpublished observations).

The main features of HLH include persistent fever and hepatosplenomegaly. Central nervous system involvement varies (ranging from irritability to overt meningoencephalopathy), and lymphadenopathy and skin rash may also be observed (Janka, 1983; Aricòet al, 1996, Haddad et al, 1997; Henter et al, 1998). Evidence of anaemia and thrombocytopenia may suggest the existence of acute leukaemia which can easily be ruled out by bone marrow examination. The above described clinical picture, the absence of malignant cells together with the possible association with active haemophagocytosis (which may be already evident at presentation or at repeated examinations during the disease course, in the bone marrow or in the spleen, lymph nodes or cerebrospinal fluid cells; Favara, 1989; Ost et al, 1998) are highly suggestive of HLH. Biochemical alterations largely overlap with those characteristic of the macrophage hyperactivation syndrome (Ravelli et al, 1996) and include hypertriglyceridaemia, hypofibrinogenaemia, hyponatraemia, elevated ferritin and lactase dehydrogenase (LDH) levels. Cerebrospinal fluid (CSF) pleocytosis is common. Plasma levels of soluble interleukin 2 receptor (CD25) are also increased (Komp et al, 1989; Henter et al, 1991b). Clinical and laboratory markers of associated infection were reported in 50 of the 122 (41%) children reported to the HLH registry. Of these, 25 had a positive family history of HLH and their features were not significantly different from the other 25 non-familial ones, or from the 72 children (34 familial) without evidence of associated infection at presentation. The spectrum of reported infectious agents mirrored that of common pathogens, including cytomegalovirus (CMV), Epstein-Barr virus (EBV), adenovirus, parvovirus, hepatitis B virus, herpes and Coxsackie (Aricòet al, 1996; Henter et al, 1993).

Histological examination of involved organs typically shows infiltration by histiocytes and lymphocytes, with haemophagocytosis (Akima & Sumi, 1984; Favara, 1989; Ost et al, 1998); spleen biopsy may also provide evidence of haemophagocytosis, but because of possible complications it should be applied with caution; liver biopsy is often less informative, showing a hepatitis-like lymphocytic infiltration. Thus, repeated bone marrow aspirations appear to be the best approach for the documentation of haemophagocytosis. Differently from the more common Langerhans' cell histiocytosis, bone and adjacent soft tissues are not involved in HLH (Favara et al, 1997; Aricò & Egeler, 1998).

The spontaneous course of the disease is invariably fatal, as illustrated by the median survival of 6 months and no more than 10% probability of survival beyond 3 years reported in the historical series collected by the International HLH Registry (Aricòet al, 1996). Initial therapeutic approaches showed that etoposide, with or without steroids and intrathecal methotrexate, can induce disease remission (Ambruso et al, 1980; Fischer et al, 1985). On this basis, the HLH study group of the Histiocyte Society started an international prospective therapeutic trial named HLH-94 (Henter et al, 1997). The preliminary results of this study showed that combined immunochemotherapy with dexamethasone, etoposide and cyclosporine was effective in achieving control of the disease manifestations within a few weeks, and also in prolonging the survival (J. I. Henter personal communication). The increased incidence documented in Italy, together with current availability of such an effective treatment, probably resulted in the marked reduction of cases of HLH who died in an early stage of the disease owing to uncontrolled progression. This may provide an opportunity for a patient in good clinical condition to proceed to bone marrow transplantation (BMT). After an initial report by Fischer et al (1986), BMT from matched related donors was successfully applied (Todo et al, 1990; Blanche et al, 1991). As most patients lack a familial donor, alternative sources for haemopoietic stem cells are necessary, including bone marrow (Baker et al, 1997; Jabado et al, 1997; Durken et al, 1999) or cord blood (Schwinger et al, 1998; Tanaka et al, 1998; Imashuku et al, 1999a) derived from matched unrelated donors. At present, BMT remains the treatment of choice for HLH and the only potentially curative one, as suggested also by evidence of restored natural killer (NK) cell activity in patients in long-term remission.

Hlh: a difficult diagnosis

The diagnosis of HLH is often difficult and, in the past, it frequently could not be established before the patient's death. The main problem has been the lack of a specific marker for the disease, in which the clinical picture is non-specific and mostly suggestive of disseminated infection or haematological malignancy. When details about more than one familial case are available, as for the 15 families described in the Registry study, the age at the disease onset was similar in the first and the second case (median, 2·8 versus 2·4 months; range, 0–54 months versus 2–64 months respectively), while the time required to diagnosis was longer in the first case, as documented by the older age at diagnosis (median, 8·9 months versus 4·9 months; range, 2–64 versus 0–42 respectively) (Aricòet al, 1996).

Therefore, the Histiocyte Society developed a set of diagnostic criteria to help clinicians identify, on purely clinical grounds, children with this obscure but very severe disorder (Henter et al, 1991c). Yet such criteria, although useful to identify patients with suspected HLH, are not specific enough; they are characteristic of a macrophage activation syndrome which may also be observed in different pathological conditions, including infection by viruses or other agents. Remarkably, this picture may develop not only in apparently healthy children, without any family history, but also in children with constitutional or acquired immune deficiency (Aricòet al, 1999a) or in patients with juvenile rheumatoid arthritis (McPeake et al, 1993; Ravelli et al, 1996; Stephan & Galambrun, 2000). Another potential pitfall is represented by cases of congenital lysinuric protein intolerance in which a similar clinical picture may occur (Duval et al, 1999). Most of these cases have been diagnosed as virus (or infection)-associated HLH (VAHS or IAHS) (Risdall et al, 1979, 1984; Janka et al, 1998). Visceral leishmaniasis has been reported in association with a clinical picture of HLH and should be carefully investigated even outside endemic areas (Gagnaire et al, 2000).

Differential diagnosis between these various conditions has obvious prognostic and therapeutic implications. Reports of the sporadic, infection-associated cases led to a belief that familial cases, apparently not associated with infection(s), could be those with a more serious outcome. The sporadic cases associated with viral infection were thought to be characterized by a better prognosis. As a consequence, specific HLH-directed treatment was considered unnecessary in this second group. We have learned, however, that this may not be always the case for several reasons. First, scarcity of large families, especially in western countries, may frequently hamper the ability to recognize the familial nature of the disease; second, viral infections have often been documented in familial cases (Henter et al, 1993; Aricòet al, 1996). In Japan, the propensity to discriminate between sporadic EBV-associated HLH and documented familial cases may have contributed to the tendency to spare the non-familial cases from specific therapy, which appears to have resulted in an excess mortality (Imashuku et al, 1999b). It is possible that similar behaviour, although unreported, has also been applied in other geographical areas.

Furthermore, the clinical picture of HLH in children may vary considerably. Thus, some patients develop the disease soon after birth and present with persistent spiking fever, massive hepatosplenomegaly, severe cytopenia (anaemia, progressive thrombocytopenia and, later, also neutropenia) and nervous alterations, which range from irritability to seizures or even a rapidly progressive encephalopathy (Henter & Elinder, 1992). Evidence of hypertriglyceridaemia and hypofibrinogenaemia may be of diagnostic help. As haematological malignancy is the first suspected diagnosis in most of these cases, bone marrow aspiration is pivotal in HLH. Once leukaemic infiltration has been ruled out, erythro- or haemophagocytosis should be looked for; nevertheless, this picture may be quite difficult to document and in about half of cases it is evident only at follow-up examination (Aricòet al, 1996). Thus, although it represents the eponym of the disease, the diagnostic value of haemophagocytosis cannot be overemphasized.

The clinical course of HLH may be very aggressive and sometimes initial treatment may be necessary to prevent early fatalities, even though the diagnostic work-up has not been completed (Aricòet al, 1996). In some cases, fever and/or splenomegaly may precede the onset of some of the biochemical alterations, so that the diagnostic criteria may not be fulfilled in the initial stage (Henter & Elinder, 1992). Fever and hepatosplenomegaly may eventually spontaneously subside, to recur weeks or months later. In such cases, the age at the diagnosis of HLH may be largely delayed, with a misleading effect for the clinician (Allen et al, 2001). These children may ultimately be endangered by entering an accelerated phase. Awareness of the disease and precise diagnosis are necessary to prescribe combined therapy with dexamethasone and etoposide, which represent the current standard initial therapy for HLH. This treatment has been shown to be effective in achieving disease control within the first 8 weeks in the majority of patients and in prolonging survival, despite possible reactivation, allowing more patients to proceed to bone marrow transplantation (BMT) (Henter et al, 1997; unpublished observations). It may be hypothesized that the extensive use of dexamethasone may contribute to prevent central nervous system dissemination, clearly documented by necropsies of patients who died with progressive or fulminant disease (Akima & Sumi, 1984). The possibility of regression of central nervous system (CNS) deterioration in patients achieving transient or even long-term disease control after BMT remains under evaluation.

Genetics of hlh

Genetic markers of HLH have been unsuccessfully investigated for many years. Chromosomal studies have usually been negative and occasional reports of chromosomal aberrations (Gilgenkrantz et al, 1984; Kletzel et al, 1986; Kaneko et al, 1995) have not been confirmed in larger studies (Aricòet al, 1996). Hasle et al (1996) reported one case of HLH associated with constitutional inversion of chromosome 9, but additional studies showed that the reported breakpoints were not linked to the disease gene (Aricòet al, 1999b), as also reported by Ohadi et al (1999). Linkage analysis using homozygosity mapping in four inbred HLH families of Pakistani descent isolated a putative disease gene to an approximately 7·8-cm region between markers D9S1867 and D9S1790 at 9q21.3–22 (Ohadi et al, 1999). Furthermore, linkage analysis of a group of 17 families with HLH revealed mapping of a locus linked to HLH in the proximal region of the long arm of chromosome 10 in the 10q21–22 region in 10 families but not in the remaining seven, providing evidence for genetic heterogeneity in HLH (Dufourcq-Lagelouse et al, 1999). While no further cases of HLH linked to the 9q21.3–22 locus have been reported, recently Stepp et al (1999) identified nine different mutations, three nonsense and six missense, in the two coding exons of the perforin 1 gene (PRF1) in a group of eight unrelated patients, providing the first evidence for a disease entity related to PRF1. Four novel mutations (Goransdotter et al, 2001), together with a further six (Clementi et al, 2001), have recently been demonstrated in patients with HLH of different geographical origins. Thus, at present PRF1 mutations may be considered the most frequent genetic defect underlying HLH. Although all six patients of Turkish origin with PRF1 mutations shared the same mutation, patients from Southern Italy showed a range of different PRF1 mutations, suggesting that HLH may have a common founder effect among Turkish but not Italian patients (Graham et al, 2000; Clementi et al, 2001; unpublished observations).

Pathogenesis: what is guilty?

In their original report, Farquhar & Claireaux (1952) suggested that a red cell membrane defect might induce haemophagocytosis and, consequently, cytopenia. This concept was not supported by experimental evidence and the attention of investigators focused on the frequent association of HLH with viral infections. Yet, despite repeated studies, no virus could be consistently associated with HLH, while the spectrum of reported infections simply mirrors that of common pathogens (Henter et al, 1991b; Aricòet al, 1996). Thus, it has been suggested that viruses may play a role as non-specific triggers of HLH. But why does the clinical picture of the infection degenerate into HLH? Children with HLH appear to be unable to control some infections; thus, they try to overcome them by an uncontrolled inflammatory response with sustained hyperactivation of T cells and macrophages. Hyperproduction of cytokines such as tumour necrosis factor alpha (TNF-α), interleukin 1 (IL-1), IL-6 and interferon gamma (IFN-γ) account for part of the clinical manifestation, as they may lead to severe tissue damage (Henter et al, 1991b). Hyperactivated lymphocytes and macrophages (Hansmann et al, 1989; Burgio et al, 1990) disseminate around the body and, together with haemophagocytosis, account for the eponym of HLH. Whether widespread dissemination of these cells is as a result of proliferation or reduced cell death has not yet been clarified. In one study that tested the in vitro exposure of lymphocytes from eight patients to etoposide or agonistic anti-Fas monoclonal antibodies, inherent resistance to apoptosis induction was ruled out (Fadeel et al, 1999). Whether an immune defect in HLH primarily resided in the T cell or the macrophages remained questionable for many years (Aricò & Burgio, 1989; Hansmann et al, 1989; Stark et al, 1987; Bujan et al, 1993). Association with infections suggested that HLH was secondary to a selective immune deficiency (Ladisch et al, 1978). From 1984 onwards, evidence of impaired cellular cytotoxicity in HLH has been repeatedly confirmed (Perez et al, 1984; Aricòet al, 1988; Kataoka et al, 1990). Demonstration of complete absence of natural killer (NK) cell activity by the peripheral blood lymphocytes of the patients has pointed to the inability to control certain viral infections and has been used by some clinicians as a further diagnostic tool.

Nk cell function

NK cells have long been regarded as mysterious cells (Trinchieri, 1989). Although their ability to lyse target cells has been known for many years, the mechanisms involved in their action have only recently been clarified. One main question concerns the capability of NK cells of distinguishing tumour- or virally infected cells from normal host cells. The main fail-safe mechanism against self destruction by NK cells is the expression of the major histocompatibility complex (MHC) class I molecules (Moretta et al, 1996). NK cells recognize MHC molecules via receptors that deliver inhibitory signals, thus blocking NK-cell function. In humans, these inhibitory receptors display different specificities for HLA class I molecules. Therefore, while killer inhibitory receptors (KIRs) recognize allelic specificities, Ig-like transcript 2/leucocyte Ig-like receptor 1 (ILT2/LIR1) has a broad HLA class I recognition and CD94-NKG2A binds the-non-classic HLA-E molecules, the expression of which is dependent on the expression of classic HLA class I molecules. Cells that lack the MHC class I, an event that frequently occurs in cancer and virus infections, are then lysed by NK cells (Moretta et al, 2000a). A repertoire of different subsets of NK cells, expressing HLA class I receptors with different specificities, allows the body to sense the presence or absence of different MHC class I alleles on the cell surface (Moretta et al, 1996, 2000b). Thus, the correct function of NK cells may be threatened at different levels, with functional and clinical implications that are not completely understood.

In addition to exerting a cytolytic activity, upon activation NK cells release large amount of cytokines including IFN-γ, granulocyte-macrophage colony-stimulating factor (GM-CSF), TNF and chemokines which exert a profound effect in immune response, inflammation and haemopoiesis, as well as in cell migration. Thus, for example, production of IFN-γ leads to macrophage activation and may induce the polarized response of helper cells towards Th1. It is thus evident that NK-cell activation may greatly influence the subsequent type of T-cell response and pattern of cytokines produced. Indeed, a recent report on NK cell-deficient mice showed that NK cells are major producers of IFN-γ in response to microbial products (Kim et al, 2000).

Termination of immune response

Termination of the immune response may occur following two main pathways. The first mechanism is spontaneous termination of the response by ‘ignoring’ the antigen. In normal subjects, infection triggers a normal immune response, including lymphocyte proliferation and perforin-mediated killing of infected cells. Once the infected cells are eliminated, the stimulus for immune response is abolished and further proliferation is not needed.

The alternative way to stop the immune response is that of an ‘active termination’. In this case, specific signals inhibit the response mechanisms. Such modulation depends on a different activation of T cells. The behaviour of the surface receptor B7 may be considered a good model to explain this modulation. If the receptor binds CD28, the resulting signal is one of proliferation and differentiation; in contrast, if the B7 receptor binds the CTLA4 molecule, the resulting message is to stop the process and thus induce anergy (Dahl et al, 2000). This finding is confirmed by evidence that CTLA-4 knock-out mice die of a lymphoprolipherative disease. But what decides which binding must happen? The most probable answer is that the affinity of the receptors plays the crucial role: the CD28 receptor has a high affinity compared with the CTLA-4 receptor. This may affect the ratio of binding and, thus, the ultimate response to the stimulus (Howland et al, 2000).

Perforin defect: a good model to explain hlh

Perforin is a protein that is expressed in lymphocytes but also in macrophages and other bone marrow precursors (Li et al, 1994; Berthou et al, 1995; Gasque et al, 1998). Its main role in the cytolytic process is to form pores in the membrane of target cells (Fig 1). For this purpose, perforin is stored in cytoplasmic granules that may be released as required. Following cell triggering, perforin is inserted into the plasma membrane and undergoes polymerization to form pores that may lead to osmotic lysis of the target cells. Once the target cell membrane has been ‘perforated’ by the creation of pores, granzymes and other granule components are allowed to enter the target cell. The start signal for programmed cell death (apoptosis) is given by granzymes (Berthou et al, 1995), which may enter the cell through the perforin-dependent pores. Granzyme B was recently demonstrated entering cells in a perforin-independent manner, using the cation-independent mannose-6-phosphate/insulin-like growth factor receptor (Motyka et al, 2000). Furthermore, in a mouse model deficient for dipeptidyl peptidase I, cytotoxic lymphocytes contained normal amounts of granzymes A and B but these molecules retained their prodipeptide domains and were inactive (Pham & Ley, 1999). However, in an experimental model, deficiency of granzymes A and/or B was not critical for most anti-tumour effect functions of NK cells (Davis et al, 2001). Together, the protective role of perforin appears more relevant in NK cells.

Figure 1.

Proposed model of target cell death by cytotoxic T lymphocytes (CTLs) and natural killer (NK) cells through the granule exocytosis pathway. After conjugate formation, CTL or NK cells release lytic granules containing cytotoxic proteins, such as perforin and granzymes, into the intercellular space. Here, the released perforin undergoes calcium-dependent polymerization and forms cylindrical pores in the target cell membrane. This allows the granzymes to access the cytoplasm of the target via these pores. With the integrity of the membrane destroyed and the protease activity of granzymes, the target cell rapidly dies.

Evidence obtained in an animal model showed that perforin-deficient mice cannot lyse target cells (Smyth et al, 1999), and have an impaired defence against cancer and intracellular pathogens (Stenger & Modlin, 1998; Smyth et al, 1999); knock-out mice models indicated that both perforin and interferon-γ independently contribute to anti-tumour effector functions that control the initiation, growth and spread of tumours in mice (Street et al, 2001). NK 1·1+CD3 cell knock-out mice were shown to have deficient rejection of tumour cells and suppression of tumour metastasis and outgrowth (Kim et al, 2000).

Furthermore, in a prospective study of 3625 Japanese adults observed for 11 years, subjects with low NK activity were found to be at a higher risk of developing cancer, providing strong evidence for natural immunological host defence mechanisms against cancer (Imai et al, 2000).

Moreover, antigen-presenting cell (APC) function is thought to be terminated by a mechanism that implies the involvement of perforin. If perforin is defective, APCs may not be effectively eliminated (Parajuli et al, 1999) and this may result in persistent stimulation of T cells. This might represent an additional mechanism through which perforin may affect the immune response.

In patients with HLH, as well as in other related conditions such as Chediak–Higashi and Griscelli syndromes (Klein et al, 1994; Pastural et al, 1997), the inability to release perforin-containing granules may reflect the impaired effector function of perforin.

During viral infections, the balance between the virus and the host may be variable and result in different scenarios. If the cytotoxic response is prompt and adequate, infected cells are rapidly killed and the infection is terminated owing to virus exhaustion. If the killing mechanism is less effective, viral infection may be protracted owing to the exhaustion of specific cytolytic activity. This further results in virus dissemination. Thus, modulating the cytotoxic responses following a viral infection is crucial in determining the outcome of the infection and, ultimately, the patient's clinical course.

What happens in children with PRF1 (or equivalent) mutations when they are challenged by a virus? Although T cell-mediated recognition of pathogen-derived peptides leads to T-cell activation and clonal expansion, the resulting cells fail to kill the infected cells and, thus, remove the source of antigen stimulation. Alternatively, this persistent antigen-driven T-cell activation results in the production of large quantities of cytokines, including IFN-γ and GM-CSF, both known to represent two major macrophage activators. The sustained macrophage activation and homing to the sites of T-cell activation results in tissue infiltration and the production of high levels of TNF-α, IL-1 and IL-6, which play a major role in causing tissue damage and the resulting clinical symptoms (Fig 2). It is noteworthy that the so-called ‘accelerated phase’ characteristic of Chediak–Higashi and Griscelli syndromes share the same pathological and clinical pattern. However, a different pathogenic mechanism occurs in male children with X-linked lymphoproliferative disease (XLP), a condition that may be barely distinguishable from HLH in some situations (Sumegi et al, 2000; Aricòet al, 2001). Patients with XLP are unable to control Epstein-Barr virus (EBV) infections as a consequence of a major dysfunction of the 2B4 receptor, which exerts an inhibitory instead of an activating function (Parolini et al, 2000).

Figure 2.

Schematic representation of the cytolytic effector cell defect in HLH.

In conclusion, the available data suggest that a perforin defect results in HLH, a rapidly fatal human disease, owing to reduced ability of these patients to control viral infections. Uncontrolled dissemination of the virus, together with a parallel excessive inflammatory reaction, results in extensive and disseminated tissue damage that ultimately leads to liver, lung and brain deterioration. Mutations in the perforin gene leading to perforin defects account for the clinical picture in a proportion of patients with HLH (Stepp et al, 1999; Goransdotter et al, 2001; unpublished observations).

To date, the available data do not allow the definition of a genotype–phenotype correlation, which can only be explored when a larger number of mutations are documented. The role of other molecules involved in cell-mediated cytolytic activity, such as granzymes, should be carefully explored in the remaining patients, the majority of whom are still waiting for the demonstration of an alternative underlying pathogenic defect. It appears probable that the Pakistani patients in whom linkage to chromosome 9 was found may represent an independent subset of HLH. Recent evidence of families in which HLH may have a very late onset, despite an evident NK-cell defect, strongly suggests additional genetic heterogeneity (Allen et al, 2001). The hypothesis of using impaired NK activity as a surrogate marker of the disease, particularly in the evaluation of familial donors, is being explored (Sullivan et al, 1998).

The need for a better understanding of the genetic defects in families with HLH is compelling in order to provide clues to confirm the diagnosis, and to refine treatment and genetic counselling, including prenatal diagnosis. Current knowledge of the pathogenesis of HLH may explain the curative potential of BMT. Furthermore, evidence of long-term remission from HLH after partial engraftment of BMT (Landman-Parker et al, 1993), as well as an 18-month remission in two patients following acute BMT rejection (unpublished observations), suggest that immune suppression and chimaerism may play different roles in restoring immune function and controlling the disease. At present, BMT remains the only curative treatment but a better knowledge of the pathogenic mechanisms and the underlying genetic defects could open the door to alternative therapeutic approaches, including possible gene therapy of HLH.


This work was supported in part by the following grants: Telethon Italy, Grant C30 (C.D.) and E755 (M.A.); Ricerca Corrente 80291, IRCCS Policlinico San Matteo, Pavia, Italy (M.A.); and from the ‘Associazione Antonio Pinzino’ (Petralia, Palermo, Italy). The authors are grateful to Dr Michaela Allen for her help in manuscript preparation.