Gastrointestinal Stromal Tumours: An Update

Purpose. To study the evolution of concepts concerning gastrointestinal stromal tumours (GISTs) over 30 years. Discussion. GISTs have been, for more than 30 years, the subject of considerable controversy regarding their line of differentiation as well as the prediction of their behaviour. Furthermore, once they spread within the peritoneal cavity, they are extremely hard to control. The recent findings of c-Kit mutations and the immunohistochemical detection of the product of this gene, KIT or CD117, in the mainly non-myogenic subset of this family of tumours, has led to a reappraisal of this group of lesions, which, with some exceptions, is now thought to be derived from the interstitial cells of Cajal, and this has facilitated a clearer definition of their pathological spectrum. In this article, we review chronologically the evolution of the concept of GIST with the gradual application of electron microscopy, immunohistochemistry, DNA ploidy analysis. We discuss the impact of these techniques on the pathological assessment and clinical management of GISTs.


Introduction
N eoplasm s arising from the strom al (or m ural) components of the gut can be broadly divided into two categories. Som e tum ours are not unique to the gastrointestinal w all and app ear sim ilar to their counterparts in other locations. This category includes schw annom as, usual leiomyom as (m ostly located in the oesophagus and rectal wall) and leiom yosarcom as, show ing characteristic m orphological, im m unohistochemical and ultrastructural features of sm ooth m uscle differentiation, as w ell as som e uncomm on neoplasm s such as lipomatous and vascular tum ours. T he other category is com posed of spindle and epithelioid neoplasm s histologically resem bling sm ooth m uscle tum ours but either lacking or presenting only lim ited imm unohistochem ical and ultrastructural features of m yogenic, neural or neuronal differentiation, for w hich prediction of behaviour has proved to be problem atic.
Strom al tumours of the gastrointestinal tract (G IST ) occur over a wide age range but affect predom inantly m iddle-aged and elderly individuals, w ith a slight fem ale predom inance. Bleeding is the m ost com m on initial sym ptom , and up to 20% of patients present with anem ia. Pain represents another com mon com plaint. D espite their large size, only a sm all proportion of tum ours are palp able. 1 T he biological behaviour of G ISTs is dif® cult to determ ine accurately from data availab le in the literature, given the different m orphological and diagnostic criteria used, and the tendency for late m etastases, in som e cases, with spread som etimes occurring after 20 to 30 years. 2 T he overall 5 and 10-year survivals of m alignant G IST s, have been estim ated at between 25 and 50%, 3,4 although in one series, only 10% of patients remained free of disease after a m edian follow -up of 68 m onths. 3 M ost such patients succum b to disseminated intraabdom inal disease (with m etastasis to the om entum , m esentery, peritoneum and liver), although distant m etastases (m ainly to the lungs and bone) occasionally occur. 5,6 T reatm ent of GIST is essentially surgical and the type of operation has been show n to represent one of the m ost im portant determ inants of survival. 3 Surgical excision of intraabdom inal m etastases has also been shown to slightly im prove survival. 6 Unfortunately, GISTs tend to respond poorly to chem otherapy, being even less chem osensitive than leiom yosarcom as at other sites. 7,8 D espite their relative rarity when com pared to epithelial tum ours in this anatom ical location, these tum ours, usually grouped under the non-comm ittal term of gastrointestinal strom al tum our (G IST), have been one of the m ost controversial subjects in the recent pathology literature. U ntil this year, the large num ber of studies generated by the progressive use of electron m icroscopy, im m unohistochem istry, ow cytom etry and proliferation m arkers have m ostly provided contradictory or inconclusive results regarding the`histogenesis' of these tumours and determ ination of their prognosis.
The recent identi® cation of m utations of the c-K it gene in GIST s, 9 and the im munohistological detection of its product, KIT or C D 117, 9± 11 suggesting differentiation towards the phenotype of the interstitial cell of C ajal, 9,10 has served as the basis for a new de® nition of GIST s, and has provided an elegant explanation for the previously controversial results relating to the phenotype of these tum ours, as w ell as a new springboard for future investigations.
In this article, using a chronological approach, we review the evolution of the`concept' of GIST s, from Stout' s early reports to the m ost recent studies, m entioned above, and discuss the impact of these developm ents on our understanding of GIST s, based on the literature and our personal experience.

H istogenesis/differentiation
G IST s were originally thought to arise from m ural sm ooth m uscle of the gastrointestinal tract, based on their histological resem blance to leiom yom as and leiom yosarcom as at other sites, as w ell as their usually intimate asso ciation w ith the wall of the gut. 12± 14 Epithelioid tum ours, ® rst identi® ed by M artin et al. 15 and subsequently popularised by Stout,16 w ere also considered to be sm ooth m uscle neoplasm s on the basis of the transition between spindly and epithelioid areas in some lesions. Subsequently, som e authors noticed morphological differences between these`leiom yom as' and`leiom yosarcom as' of the gastrointestinal tract and their counterparts in other sites: m ost gastrointestinal lesions appeared m ore cellular and the tum our cells had m ore elongated nuclei and less brightly eosinophilic cytoplasm . However, m ost authors attributed this to a relative lack of differentiation, rather than to the possib ility of alternative lines of differentiation. In addition, it rapidly becam e clear that GIST s, unlike their apparent counterparts in other sites, could m etastasise despite the absence of usual histological features of m alignancy (in particular, in the absence of signi® cant m itotic activity) and that their behaviour was m uch harder to predict. 12,14,17 Ultrastructural studies T he 1970s and the early 1980s saw the debate concerning these lesions focusing on the ultrastructural features of gastrointestinal sarcom as, of which the presum ed sm ooth m uscle differentiation had started to be questioned. In 1969, in their ultrastructural study of three ª gastric cellular leiom yom asº , W elsh and M eyer noticed that ultra-structural features of smooth m uscle (i.e. cytoplasm ic ® lam ents with dense bodies, extracellular basem ent m embrane, pinocytic vesicles) were identi® ed in only occasional cells and were often incom plete. 18 T hese results w ere con® rm ed by m ost subsequent studies, which failed to identify m yo® lam ents with focal densities in m ost tum ours; in fact, sm ooth m uscle differentiation in gastrointestinal strom al tum ours was most often supported only by relatively non speci® c features (such as the presence of pinocytic vesicles or focal basal lamina). 19± 23 In addition, several authors started to identify Schwannian or neuroaxonal characteristics, in som e cases m icroscopically indistinguishable from other G IST s, 21,22,24 and in 1984, a distinctive subset of tum ours, showing features of autonom ic neural differentiation was ® rst described. 25 Immunoh istochemical studies T he introduction of im munohistochemistry, in the 1980s, strengthened the debate relating to the differentiation of G IST s and initiated a profusion of publications. 22,26± 34 N um erous series of GIST have been decorated with diverse antibodies with inconsistent results. Although a large proportion (betw een 30 and 80%) of GISTs have been show n to express m uscle m arkers, 29,33,34 the m ost speci® c of these, desm in, has usually stained only a m inority of tum ours. 30,33 Variable proportions of tum ours with a neural phenotype have been identi® ed (from none to approxim ately 40%) 29,30,33 and, interestingly, divergent differentiation (coexpression of m uscular and neural m arkers) was identi® ed in up to 20% of cases by N ewm an et al. 29 U p to 41% of tum ours have been characterised by a`null' or`uncom m itted' phenotype, being stained by vim entin only. 27 M ore recently, C D 34, initially identi® ed as a m yeloid cell progenitor antigen, but also expressed in endothelial cells, in some m esenchym al cells as well as in a variety of soft tissue neoplasm s, has been show n to stain up to 80% of GIST s, with or without m arkers of other speci® c differentiation. 35± 37 Several reasons m ay explain the striking lack of consistency of these results, which has rendered their interpretation particularly dif® cult. T echnical issues, relating to the nature of ® xative, duration of ® xation, nature and dilution of antibodies are certainly partly responsible for som e of these discrepancies. V ariab le thresholds for positivity m ay have been used and the presence of norm al neural elem ents or m uscle bundles may have caused som e problems in interpretation. For exam ple, in Mazur' s study, ® rst reporting S100 protein im m unopositivity in G ISTs, seven of eight S100-positive tum ours contained only scattered elements which m ay be m ore in keeping with entrapped structures than with true nerve sheath differentiation. 22 Variable criteria for sm ooth m uscle and neural differentiation have also been used; for exam ple, while som e authors have assessed neural differentiation in G IST s using only antibodies for S100 protein, others have employed a com bination of m ore or less speci® c m arkers including N SE , PG P9.5 or Leu7.
Several authors have tried to correlate the imm unophenotype of GIST s with tum our location, histological appearance and, m ore im portantly, with prognosis. N o signi® cant differences have been dem onstrated in term s of im m unophenotype between epithelioid and spindle cell lesions. In fact, no reliable correlation between the histological features of GIST and their im m unophenotype has been achieved. 27, 29 Regarding the relationship with tum our site, the observation that, in contrast to m ost gastric and intestinal tumours, oesophageal and rectal tumours frequently fail to stain for CD 34 and tend to express desm in, has represented an interesting ® nding, indicating that these tum ours likely represent`true' smooth m uscle tumours that should be differentiated from G ISTs, which rarely occur in these locations. Som e differences in im m unophenotype betw een gastric and intestinal tum ours, the latter tending to m ore com m only express a neural phenotype, have also been suggested 29,33,38 but have not been further investigated.
The possib le relationship between the imm unophenotype and prognosis has represented one of the m ore controversial issues in GISTs. Results of a few studies have suggested the possibility of differences in prognosis betw een im m unophenotypic subsets: it has been proposed that tumours with a neural 4,29 or sm ooth m uscle 29 phenotype tend to have a better prognosis, while those w ith a null phenotype seem m ore often to behave in a m alignant fash ion. 34 How ever, these results have not been con® rm ed and this ® eld of investigation has gradually been abandoned, on the basis of the consistently inconclusive and variable results. In consequence, m any pathologists have stopped phenotyping GISTs in their routine practice. W e believe that in the absence of de® nitive results, the line of differentiation should continue to be part of the inform ation provided to the clinician in any case of GIST and to be included in future studies, until de® nitely proven irrelevant (or otherwise). In larger studies, the phenotype of G ISTs m ight not only prove to carry som e prognostic signi® cance but could also possibly show som e relation w ith treatment response, as, for example, it is w ell known that leiomyosarcom as at other sites do not usually respond w ell to chem otherapy.

Gastrointestinal autonom ic nerve tum ours (G AN T)
Am ong the spectrum of G ISTs, speci® c attention has been given in recent years to those showing autonom ic neural differentiation. T hese tumours w ere described in 1984 by H errera et al. 25 as plexosarcomas' , and subsequently designated gastrointestinal autonom ic nerve tumour (GAN T ). 39 Although they tend to be composed of syncytial sheets of cells characterised by distinctive ® brillary eosinophilic cytoplasm , com m only associated with stromal lym phocytes and extracellular nodules of eosinophilic m aterial known as`skeinoid ® bers' , 40 their histological spectrum is wide and it is largely accepted that accurate diagnosis is based on ultrastructural criteria, i.e. long cytoplasm ic processes with rudim entary cell junctions and synapse-like structures containing dense-core granules. 41 Imm unohistochem ically, these lesions often show positivity for N SE, usually in a peculiar`zoning pattern' and, interestingly, they are generally negative for C D 34 42 (personal observations). Although these tum ours have been the subject of an increasing number of publications, 25,39,41± 48 the relative frequency of G AN T and its relationship with other G IST s still requires clari® cation. At the present time, there is no convincing evidence that these tumours differ signi® cantly from other G ISTs in term s of either their clinical presentation or behaviour. N evertheless, further investigations will be required to determ ine if criteria for m alignancy can be established and w hether they differ from those app lied in other G ISTs.

Prediction of behaviour
T he dif® culty in classifying G IST s into benign and m alignant categories has been recognised since the description of`sm ooth m uscle tum ours' of the gut, by Golden and Stout in 1941, who noted that tum ours showing usual histological criteria of m alignancy did not consistently behave aggressively, while occasional well differentiated low -grade lesions gave rise to m etastases. 12 In a series of 87 G ISTs, Kempson and Ranchod identi® ed the m itotic count as the m ost useful indicator of m alignancy. H owever, while the presence of 5 or m ore m itoses per 10 H PF was closely correlated with aggressive behaviour, 40% of`leiom yosarcom as' had fewer mitotic ® gures. 14 In order to re® ne the separation of benign and m alignant GIST s, a profusion of studies subsequently analysed the correlation between m alignancy and various clinical and pathological param eters, often w ith variable results. 1,4,29,32,34,49± 52 The m itotic count has been m ost widely accepted as the best prognostic indicator 1,5,14 and it has been show n that, am ong clinically malignant tumours, a high m itotic count w as associated with a shorter disease free interval and shortened overall survival. 5 V arious cutoff levels, separating G IST s into benign and m alignant categories, som etimes including à borderline category' , or into low and high-grade subsets have been proposed. 5,51,53,54 However, because of the overlap in term s of m itotic activity between clinically benign and m alignant GIST s, and in view of the rare occurrence of metastasis in histologically bland, m itotically inactive tum ours, none of these has proven entirely reliable in the m anagement of individual patients.
The extent of disease at diagnosis (or stage) certainly also represents a strong indicator of outcom e. W hile the presence of m etastases at presentation, not surprisingly, is associated w ith a very poor prognosis, in® ltration of adjacent structures, such as the liver, pancreas or diaph ragm , is also usually regarded as indicative of m alignancy. In Shiu' s study, all tum ours which invaded adjacent organs led to the patient' s death. 50 T um our size has also been shown to be strongly correlated w ith the occurrence of m etastases. In Appelm an' s series of 127 cases, only one tumour sm aller than 6 cm m etastasised. 17 T hese results have been con® rm ed by m ost studies and, again, various cutoff levels have been proposed, usually being set at around 5± 6 cm. Unfortunately, as for the m itotic count, som e exceptions have been encountered, and, tumours as sm all as 2 cm have been reported to m etastasise. 5 In relation to tum our size, it is interesting to note that tum ours found incidentally during operation perform ed for another unrelated condition usually carry an excellent prognosis; in the study of Cooper et al. none of 19 incidentally discovered tum ours resulted in the patient' s death. 51 In order to re® ne this discrim ination between m alignant and benign lesions, num erous other clinical, m acroscopic and histological param eters have been assessed, m ost of which have been show n, at least in som e univariate studies, to have som e correlation with survival or the development of m etastases. Cellularity has been considered useful by several authors, 4,12,14,38,51,55 but this is extremely subjective and dif® cult to quantitate, and thus is subject to a signi® cant interobserver variability. M oreover, its interpretation is com plicated by the variability between areas of the sam e tum our. Although the presence of unequivocal tumour cell necrosis is usually regarded as highly suspicious for m alignancy, this has been reported in rare clinically benign cases. 55 U lceration of the overlying m ucosa has also been considered as a worrisom e feature by som e authors. 38,55,56 T he presence of atypical m itoses has been show n in som e studies to be strongly asso ciated w ith m alignancy 52 but the utility of this feature is lim ited by the fact that abnorm al m itotic ® gures are rarely encountered in GISTs to the point that, in our experience, a diagnosis of GIST is im probable in the presence of conspicuous abnorm al m itoses.
Potential differences in term s of behaviour between epithelioid and spindle cell lesions has been another controversial topic; while m ost studies have not shown any correlation between cell type and prognosis, a few recent studies have suggested that epithelioid lesions tend to behave m ore often in a m alignant fashio n. 29,38,56 In the study of N ewm an et al. all m alignant gastric tumours contained at least som e foci com posed of epithelioid cells, justifying their m ore cautious criteria for m alignancy in epithelioid G ISTs. 29 Aside from oesophageal and colorectal tum ours, which usually display fully developed features of sm ooth m uscle differentiation (and therefore can be excluded, at least conceptually, from the G IST spectrum ), signi® cant differences in terms of outcom e according to tum our site have only been observed in one study, in which the ten year survival reached 74% for gastric lesions, w hile it was only 17% for sm all bow el tumours. 32 In fact, during the last three years, a few studies have analysed prognostic factors in selected populations of tum ours from speci® c sites, such as the duodenum but, again, have not been able provide de® nitive criteria for m alignancy. 38,55,56 Interestingly, som e studies have shown a better prognosis in rare pediatric cases and in young adults, with long survival despite m etastatic disease. 57 In fact, m ost of these patients appear to be affected by C arney' s triad. In this syndrom e, of which the genetic basis is still unclear, patients tend to develop gastric epithelioid`leiom yosarcom as' , functioning extraadrenal paragangliom as and pulm onary chondromatous ham artom as (which are som etimes clinically and radiologically m isinterpreted as m etastases from the gastrointestinal tum ours). These tum ours usually appear at a relatively young age and prolonged survival (m ore than 20 years) is com m only observed in the presence of m etastases, even without surgical treatment.

Ploidy and proliferation markers
Because of this im perfect separation betw een benign and malignant GIST s using conventional pathological criteria, as well as the subjectivity and/or interobserver variability in the evaluation of som e of these param eters, ancillary techniques such as the evaluation of ploidy (by D N A¯ow cytom etry or computerised im age analysis) and proliferation m arkers were introduced with enthusiasm in the early 1990s. M ost studies have suggested that D N A ploidy, determined by¯ow cytometry, was signi® cantly correlated with histological grading and that aneuploidy was asso ciated with decreased survival. 4,51,53,58 However, m ost of these authors com pared tum our ploidy with m alignancy de® ned either clinically but with a lim ited follow -up or de® ned only by histological criteria. A subsequent study, validated by 6 years m edian follow-up, dem onstrated that ploidy lost its prognostic value in a m ultivariate m odel, which included m itotic count and the presence or absence of m etastases at diagnosis. 52 Moreover, m ost of these studies included aneuploid cases that did not show clinical evidence of m alignancy and, m ore im portantly, a few patients with diploid tum ours (even of sm all size), developed dissem inated disease. 4,52 Because of this overlap, ploidy does not app ear to be more discrim inatory when applied in Im m unoreactivity for proliferating cell nuclear antigen (PC N A), and interphase nucleolar organiser regions (AgNO R) counts have also encountered variable success. W hile som e authors failed to show any correlation between PCN A index and m itotic rate, tum our size or behaviour, 54 m ost other studies indicated a statistically signi® cant relationship with survival in univariate analysis. 4,59± 61 However, once again, the PCN A index, in m ultivariate analysis, did not appear to provide any im provement over the m itotic count in the prediction of m etastatic spread 62 and it was shown (by Yu et al.) to be less reliable than histological grading using the scheme proposed in Table 1. 61 Therefore, even if these`modern m ethods' can be som ehow correlated with survival or w ith the probability of developing m etastases, from a statistical point of view , no signi® cant advantage has been dem onstrated over a careful m itotic count or histological grading. T he clinical value of these techniques in individual uses is also lim ited, as is the case for traditional pathological criteria, by a degree of overlap between benign and malignant tumours. Although these proliferative indices may be helpful in som e`borderline' cases, and have been introduced in som e classi® cation system s, 60 their use is not warranted in the routine evaluation of G ISTs at the present time.

Practical recommendations
It appears clear from the previous discussion that, at this point, publish ed data availab le concerning phenotypic classi® cation and prognosis are extremely controversial and somewhat confusing. M any reasons m ay have contributed to the heterogeneity of previous results. Most series are relatively sm all and, in statistical terms, do not include enough cases covering the different param eters assesse d, as well as the different locations and im m unophenotypes. F ollow up is probably too short in the context of the biology of these neoplasm s, given that late m etastases, som etimes occurring after m ore than 20 or 30 years, are not uncomm on. T he de® nitions of m alignancy, and in fact, the m ethodology have varied widely between studies. M ost groups have strati® ed tum ours according to histological criteria into categories (benign, uncertain potential, m alignant) that they have then com pared with clinical behavio ur. In fact, a potentially m ore rational approach, consisting of classifying tum ours according to their behaviour, after adequate follow-up, and then analysing their clinicopathological characteristics, has not been applied in any large series. D e® nitions of GIST have also varied widely betw een authors. Som e authors have included typical cases of schwannom as, as well as conventional leiomyom as, thus introducing selection bias, while others have chosen to exclude all cases positive for m uscle m arkers, or to exclude those which were negative for C D 34.
Because of the unreliability of available criteria (when applied individually) in distinguishing tum ours likely to behave in a benign or m alignant fash ion, multifactorial app roaches have been attempted. A wide variety of prognostic schem es, including different param eters and varyin g according to tum our location, have been proposed. Actually, alm ost every single author has proposed his or her own classi® cation schem e and, at the present time, none has proven superior to the others. For practical purposes, we personally use the criteria set out in T able 1. This table, established on the basis of personal experience and of relatively simple use, has proven useful in our daily practice and has app eared m ore reliable than any m arker of proliferation. 61 How ever, other schem es, such as those presented by Suster in his review article, 63 have proven useful and, of course, they should also be considered.
It seems clear that larger series, with reproducible diagnostic and prognostic criteria, prolonged followup, including cases from various sites and covering a large num ber of param eters need to be collected. U ntil then, w e believe that one should consistently use one schem e (such as that in Table 1), and that the inevitable uncertainty which presently exists concerning the behaviour of som e GIST s, for example in the case of histologically benign but large tum ours, should be expressed in the pathology report.

Recent developm ents
C -K it is a proto-oncogene encoding a transm embrane tyrosine-kinase receptor, K IT or CD 117. 64 T he interaction of this receptor with its ligand, the stem cell factor (SC F), has been shown to play an im portant role in the development of m elanocytes, germ cells, mast cells and the interstitial cells of C ajal (IC Cs). 65,66 T he latter, which are located between the m uscular layers of the gastric and intestinal wall in association with the m yenteric plexus, are known to regulate the autonom ous contraction of the gastrointestinal tract. 67,68 These cells are characterised im m unohistochem ically by dual imm unopositivity for C D 34 and CD 117. Recently, the hypothesis that G ISTs m ight differentiate towards an IC C phenotype, which had already been raised by M ikhael et al. on the basis of their im m unopositivity for CD 34,36 has been supported by three im munohistochemical, ultrastructural and m olecular studies. 9± 11 These groups have demonstrated that im munoreactivity for CD 117 was seen in 85± 100% of G ISTs while m ost other sarcom as or other neoplasm s w ere negative. In addition, the sequencing of c-K it com plem entary D N A revealed m utations in 5 of 6 cases from one of these series and our own personal experience with larger case num bers (Rubin et al., unpublished data) is similar.
These studies included spindle cell and epithelioid neoplasm s, con® rm ing that they represent m orphological variations of the sam e entity. C D 117 positive tumours were also positive for CD 34 in 72% of cases, while the (poorly speci® c) neural m arker PG P9.5 was positive in about 70% of cases and sm ooth muscle actin w as positive in 15± 30% of cases. 10,31 Interestingly, all tumours showing imm unohistochem ical evidence of sm ooth m uscle differentiation, i.e. desm in positivity and/or diffu se actin positivity, failed to stain for C D117, further justifying the separation of GIST s and conventional sm ooth m uscle tumours in this anatom ical location. T hese ® ndings have led to a unifying concept, regarding GIST s as a m orphologically (and imm unophenotypically) heterogeneous group of tum ours differentiating towards an IC C phenotype. T he signi® cance of CD 34 negativity in a subset of these tum ours, m ore often m alignant in Sarlom o-Rikala' s study, 11 is unclear, but a parallel has been m ade by Kindblom et al. with the variable imm unophenotypes seen in the different subtypes of ICC . 10 G AN Ts, which, in our experience, are usually im m unoreactive for CD 117 but negative for C D 34, could correspond to this subset of C D 34 negative tum ours and represent a distinctive group within the spectrum of G IST s.
As discussed above, the interpretation and comparison of m ost previous studies has been im paired by the lack of consistency in the de® nition of GIST s and therefore in the criteria for inclusion. T he recognition of GIST s as a cohesive but phenotypically heterogeneous group of tumours, de® ned by the expression of c-K it, and their m ore objective separation from other m esenchym al neoplasm s (m ostly true sm ooth muscle tumours) will certainly allow m ore reproducibility between studies and could help to re® ne our criteria for m alignancy and/or prognostic factors. T he expression of c-K it also appears as a useful adjunct in the diagnosis of those G IST s arising in less usual sites, such as the m esentery, om entum or retroperitoneum , as well as in their separation from other intraabdom inal neoplasm s, such as desm oid ® brom atosis for spindle cell lesions, and m elanom a or m etastatic carcinom as, which m ay be confused with epithelioid G ISTs.
The sim ilarity between the ultrastructural features of ICC and G IST, in Kindblom ' s study, not only provided further evidence for an ICC phenotype but also represents a rational explanation for the controversial results of early electron m icroscope studies. U ltrastructural characteristics shared by IC C and G ISTs included incom plete features of m yoid differentiation, such as networks of intermediate ® lam ents, including occasional bundles of actin-type ® lam ents or incom plete external lam ina, as well as neurogenic features such as long interdigitating processes, occasional gap and desm osom e-like junctions or synapse-like contacts. 10 T his coexistence of m yoid and neurogenic features probably explains the`hesitations' betw een sm ooth-muscle and nerve sheath or neural differentiation in previous studies. U ltrastructural features of G AN T (long cytoplasm ic processes and synapse-like junctions with densecore granules) w ere also part of this spectrum . 10,41 c-K it m utations had already been demonstrated in hum an m alignant m astocytosis 69,70 and transfection of the m utant c-K it com plem entary D N A into m urine lym phoid cells has induced their autonom ous growth. 9,71 T he possibility that c-K it m ight play som e central role in tum our developm ent or progression has recently received additional support, with the identi® cation of a germ line mutation in a fam ily with m ultiple GIST s, in the sam e domain where m utations had been found in sporad ic cases. Future investigations in such patients m ight therefore provide valuable inform ation concerning the biology of GISTs. Losses in the long arm of chrom osome 14, identi® ed by one group using com parative genomic hybridization, 72 as w ell as the few reported karyo typic changes, including structural m onosom ies of chrom osomes 14 and 22, 73,74 m ight also serve as the basis for further investigations. Study of cases arising in the setting of the C arney syndrom e could also possibly generate genetic inform ation that might help in better understanding sporadic cases.

C onclusions
T he tw o fundam ental issues concerning GIST s, i.e. their phenotype and prediction of behavio ur, had already been stressed by Stout and collaborators in their early description of strom al neoplasm s of the gut in 1941. M ore than ® fty years later, the probable differentiation of GIST s towards an interstitial cell of C ajal phenotype could well represent, at last, an elegant answer to the ® rst question. Although the im plications of c-K it m utations in the pathogenesis of GIST s need to be further investigated, these m olecular ® ndings possibly represent the ® rst step towards understanding the biology of these enigm atic tumours. Furthermore, the availability of reliable m olecular and imm unohistochem ical signatures of GIST has already helped to clarify the extent of this spectrum , allowing m ore accurate distinction of G IST from other m esenchym al neoplasm s of the gut, such as`true' leiomyom as, leiom yosarcom as or nerve sheath tum ours.
However, at the present time, the reliable distinction of benign from m alignant GIST s rem ains a challenge. N one of the m ultiple prognostic factors identi® ed had proved to be reliable in the evaluation of individual cases and m ultifactorial approaches have thus far failed to im prove signi® cantly upon our m orphological classi® cation of GISTs. Hopefully, the recent ® ndings relating to c-K it m utations m ight im prove our understanding of the enigmatic biology of G IST s in the near future and should im prove the coherence of future clinical and pathological studies aim ing at re® ning our prognostic criteria and classi® cation schemes.