A combined histologic and molecular approach identifies three groups of gastric cancer with different prognosis PDF

Preview

Summary

Virchows Arch (2009) 455:197–211 DOI 10.1007/s00428-009-0813-z ORIGINAL ARTICLE A combined histologic and molecular approach identifies three groups of gastric cancer with different prognosis Enrico Solcia & Catherine Klersy & Luca Mastracci & Paola Alberizzi & Maria Elena Candusso &...

Description

Virchows Arch (2009) 455:197–211 DOI 10.1007/s00428-009-0813-z

ORIGINAL ARTICLE

A combined histologic and molecular approach identifies three groups of gastric cancer with different prognosis Enrico Solcia & Catherine Klersy & Luca Mastracci & Paola Alberizzi & Maria Elena Candusso & Marta Diegoli & Francesca Tava & Roberta Riboni & Rachele Manca & Ombretta Luinetti

Received: 27 May 2009 / Revised: 6 July 2009 / Accepted: 15 July 2009 / Published online: 12 August 2009 # Springer-Verlag 2009

Abstract The limited prognostic value of currently used histologic classifications of gastric cancer and their failure to account for the complexity of the disease as revealed by more recent investigations prompted a combined reinvestigation of histologic, molecular, and clinicopathologic patterns in 294 extensively sampled, invasive gastric cancers representing all main histotypes and stages of the disease and followed for a median of 150 months. Among histologic parameters tested, only cellular atypia, angio-lympho- or neuroinvasion, Ki67 proliferation index, expansile/infiltrative E. Solcia : P. Alberizzi : M. E. Candusso : F. Tava : R. Riboni : R. Manca : O. Luinetti Anatomic Pathology Service, IRCCS Policlinico S. Matteo Foundation, Pavia, Italy E. Solcia : M. Diegoli Department of Pathology and Genetics, University of Pavia, Pavia, Italy C. Klersy Statistics Unit, Scientific Direction, IRCCS Policlinico S. Matteo Foundation, Pavia, Italy L. Mastracci Department of Anatomic Pathology, University of Genova, Genova, Italy M. Diegoli Inherited Cardiovascular Diseases Center, IRCCS Policlinico S. Matteo Foundation, Pavia, Italy E. Solcia (*) Istituto Anatomia Patologica, via Forlanini 16, 27100 Pavia, Italy e-mail: [email protected]

type growth, and T8 cell-rich high lymphoid intra-/peritumor response (HLR) proved to be stage-independent predictors of patient survival. Among molecular tests, p53 gene exon 7 (loop 3) and 8 (loop-sheet-helix motif and S-10 band), but not p53 protein overexpression, TP53 LOH or 18qLOH, were found to worsen prognosis. Microsatellite DNA instability was a favorable prognostic factor when coupled with HLR. Patient survival analysis of the main histotypes and their subtypes confirmed the favorable prognosis of HLR, welldifferentiated tubular, muconodular, and low grade diffuse desmoplastic cancers, and highlighted the worse prognosis of anaplastic and infiltrative-lymphoinvasive mucinous cancers compared to ordinary cohesive and diffuse cancers. Distinct roles of individual morphologic and molecular factors in tumor progression of the different histotypes have been recognized. The combination of survival-predictive histotypes and individual histologic or molecular parameters allowed us to develop a classification of all gastric cancers into three grades of increasing malignancy which proved to be of high prognostic value. Keywords Gastric cancer . Prognostic factors . Lymphoid cell response . P53 gene mutations . Histotype grading

Introduction Currently used histologic classifications are of limited value in the prognostic evaluation of gastric cancer [1–3]. Separation of gland-forming ‘intestinal’ from scattered-cell ‘diffuse’ tumors according to Lauren [4] is known to predict the natural history of gastric cancer, including its association with intestinal metaplasia and blood-borne liver metastases or with direct serosal invasion and peritoneal carcinosis, respectively [5]. Different molecular changes

198

have also been detected in the two tumor types: p53 gene mutation, p53 protein accumulation, microsatellite DNA instability, 18qLOH, or c-erbB-2 expression are preferentially associated with the intestinal type and cadherin E mutation or loss of membrane expression with the diffuse type [6–12]. However, despite early suggestions that diffuse tumors had a poorer prognosis [13], Lauren’s classification failed to predict patient survival independently from tumor stage [1, 3]. Attempts to improve prediction power by considering the type of local tumor growth (expansile vs. infiltrative) [14], by introducing different degrees of glandular differentiation (well vs. moderate vs. poor) [15, 16], by combining glandular differentiation with mucin production [17–19], or by separating solid and mixed from glandular and diffuse cancers [2, 3] had limited success and failed to impact significantly on clinical practice. However, a few histologic features have consistently been found to predict tumor behavior such as vascular invasion [2, 3, 20] and intratumor lymphoid cell infiltration [21–24]. In addition, some molecular findings have also been reported to be prognostically informative, i.e., high level instability of microsatellite DNA (MSI-H) [10, 11], Epstein–Barr virus (EBV) infection [25], p53 gene mutation [8, 26], 18qLOH [27], SMAD4 or 7 gene expression [28, 29], c-erbB2 expression [30], E cadherin gene mutation [7], and gene loss or gain by comparative genomic hybridization [31]. Unfortunately, so far, none of these molecular findings have proven to be contributive and cost effective enough to enter routine diagnostic practice. Nevertheless, some of the histologic, phenotypic, or molecular patterns were helpful in identifying tumor subsets predictive of patient survival. Those showing better survival included very well-differentiated tubular cancers with intestinal phenotype [32, 33], the muconodularexpansile subtype of mucinous cancer [33], and the low grade, tumor cell-embedding variant of diffuse desmoplastic cancer [34], in addition to MSI-H [10, 11] and lymphocyte-rich cancers [21–24]. On the contrary, various kinds of anaplastic cancers with poor outcome have been identified, with or without neuroendocrine features, from small to intermediate or large cell [35–37], diffuse to solid in structure, sometimes with signs of hepatoid, choriocarcinomatous, or squamoid differentiation [3, 34, 38, 39]. Thus, in order to develop a more informative and practical classification system, a systematic reinvestigation of potentially predictive histological and molecular parameters seems to be required on a sufficiently large series of invasive gastric cancers covering all the main stages and histotypes and with long-term follow-up (fundamental in identifying long survivors with low grade tumors). The aim of this study was to: (a) further check the prognostic power of individual parameters; comparatively and simultaneously in the same patient series; (b) search for possible histotype restriction of

Virchows Arch (2009) 455:197–211

their distribution and diagnostic or prognostic power; (c) test the prognostic value of resulting tumor types and subtypes; (d) develop a simple grading system based on predictive histologic factors or histotypes, into which contributive molecular findings could be easily incorporated.

Materials and methods Tumor selection Invasive (T1b to T4) gastric cancers were selected from cases undergoing surgery with curative intent at the San Matteo General Hospital in Pavia from 1984 to 2000 or at San Martino General Hospital in Genova in 1998–1999. The study considered only cases with extensive sampling of the tumor (two to 12 blocks, depending on size) and surrounding non-tumor tissue and carefully assessed perioperative TNM stage [40]. Special attention was paid to tumors whose invasion was limited to the deep submucosa (penetrating T1b, 44 cases) or muscularis propria (T2a, 55 cases) or whose histologic patterns suggested low or high malignancy or showed a structure differing from that of ordinary glandular and diffuse cancers. In all, 294 cases were collected. Corresponding patients by the end of January 2009 had either died of the disease or other causes or were still alive and had been followed for at least 7 years, with a median follow-up of 150 months (25th–75th=90–186). Histologic and histochemical stains Paraffin sections were stained with hematoxylin–eosin, Alcian blue–PAS, Giemsa, or immunoperoxidase using specific antibodies for MUC1, MUC2, MUC5AC, and MUC6 mucins [33, 34]; cytokeratins 7, 19, 20, and AE1/AE3 (Dako, Denmark); CD8 antigen (C8/144B clone, Dako); Ki67 protein (MIB1 clone, Dako); p53 protein (D07 clone, Dako); and h-MLH1 (G168.15 clone, Pharmingen, San Diego, CA, USA) or hMSH2 (Fe11 clone, Oncogene, Cambridge, MA, USA) mismatch repair proteins. In situ hybridization for the EBER-1 gene [3] or for EBER-1 and EBER-2 genes (probe Y5200, Dako) was used to detect Epstein–Barr virus (EBV) infection. For molecular investigation, tumor tissue was carefully dissected under microscopic control to enrich the tumor cell component to more than 60%. A PCR-based denaturing gradient gel electrophoresis (DGGE) method, followed by DNA sequencing with a 3100 Applied Biosystems equipment, was used to detect p53 gene mutations at exons 5 to 8 [8]. For microsatellite instability analysis, Bat 25 and Bat 26 loci were first investigated; tumors showing concordant stability or instability at both loci were classified as either stable (MSS) or highly unstable (MSI-H), respectively. Those showing discordant findings were further tested at Bat 40, D5S346, and D2S123 loci. Only tumors with

Virchows Arch (2009) 455:197–211

instability involving at least two of the five loci were classified as MSI-H; the others were classified as low instability (MSI-L) and included in the MSI-negative tumor group together with MSS cases [10, 11, 41]. In most MSI-H cases, loss of MLH1 nuclear stain was observed immunohistochemically while only two cases lacked MSH2 reactivity. Loss of heterozygosity (LOH) at chromosome 18 was examined as previously described [12] in paired tumor and non-tumor DNAs using 12 highly polymorphic microsatellite markers (D18S56, D18S67, D18S34, D18S460, D18D450, D18S474, D18S484, DCC-vNTR, D18S69, D18S55, D18S58, and D18S70) spanning most of the long arm of the chromosome, from band 18q11.2 up to q23 near the q-ter. For TP53 LOH at 17p13.1, a CCA117-732 probe was used in accordance with Vogelstein et al. [42]. Histologic evaluation Glandular differentiation was scored 0 (predominantly diffuse or solid structure), 1 (mixed diffuse/solid and glandular structure, at least 40% each of the two components), 2 (>60% glandular structure), or 3 (well-differentiated, purely glandular structure). Assessment of tumor cell anaplasia was mainly, but not exclusively, based on nuclear features [43]. Score 1 tumors were characterized by monomorphous, small- to medium-sized nuclei of regular shape, fairly uniform chromatin distribution, and small to inapparent nucleoli inside cells with fairly abundant cytoplasm. They often presented signs of functional differentiation (including mucin secretion), with or without cell polarization and well-developed intercellular junctions or glandular structure. In contrast, score 3 tumors showed larger, polymorphic nuclei, ranging from vesicular with prominent nucleoli to dense, hyperchromatic with relatively small nucleoli, inside poorly cohesive, nonpolarized cells with scarce (small cells) to abundant (large cells) cytoplasm, usually mucin free, albeit sometimes with signs of abortive endocrine (poorly differentiated neuroendocrine cancers), hepatoid, chorionepitheliomatous, or squamoid differentiation, often with multifocal necrosis [3, 34, 36, 37, 39]. Tumors showing moderate to considerable nuclear pleomorphism in cells of various size, shape, and aggregation status, with intermediate patterns between those of score 1 and 3, were classified as score 2. Mitotic figures were counted in ten high power fields (HPF, ×400) while the percent of Ki67 positive nuclei was assessed at ×250 magnification by counting at least 1,000 cells in areas showing the highest proliferation [34]. Routine hematoxylin–eosin preparations were adequate in most cases to assess lymphatic and blood vessel invasion. When tumor cells were found in vessel-like spaces of doubtful interpretation, the CD31 endothelial cell marker (M823 antibody, Dako) was applied. Lymphoinvasion and neuroinvasion were graded 0 (absent), 1 (sporadic, i.e., no

199

more than one invaded vessel or nerve in a tumor section), or 2 (multifocal); only score 2 invasion proved to be unquestionably predictive of survival in this study. A single finding of unequivocal blood vessel invasion was sufficient to classify the tumor as angioinvasive, while the presence of either score 2 lymphoinvasion and/or angioinvasion indicated vascular invasion. Both endoneural and closely adherent perineural invasion were considered [34]. Only those cases which showed a well-demarcated regular boundary between a “pushing” tumor growth and surrounding non-tumor tissues were classified as expansile; all the remaining cases were considered as infiltrative. Tumor types Of the 294 tumors, careful investigation of hematoxylin–eosin, Alcian blue–PAS, and Giemsa-stained sections identified 74 cases with diffusely infiltrating, poorly cohesive cells dispersed in the stroma as single elements or in small aggregates with little or no gland formation (diffuse cancers) [34], 42 cases with prominent extracellular mucin (mucinous cancers) [33], and 21 cases formed predominantly or exclusively by anaplastic, highly proliferative (≥20 mitoses/10 high power fields) small to large cells with scarce stroma and frequent necrosis, regardless of their diffuse or cohesive structure and usually poor expression of endocrine or exocrine markers (anaplastic cancers) [3, 34, 36]. Among the remaining 157 tumors, 56 cases showed increased intra-peritumor lymphoid cell infiltration (lymphocyte-rich or LR cancers) [21, 24], of which 47 “high lymphoid response” (HLR) cancers showed either (a) lymphoepithelioid (LEP; 17 cases, 12 of which are EBV positive) pattern with intimate admixture of abundant lymphoid cells and epithelial tumor cells, often dissociated from each other by the infiltrate, or (b) ≥300 CD8+ cells/10 HPF inside the tumor growth, coupled with a prominent band of lymphoid cells, rich in CD8+ cells, surrounding the expansile edge of the tumor or its nodules. The remaining 110 non-HLR tumors formed the cohesive cancers group, with more or less prominent glandular (tubular, papillary, or cystopapillary) differentiation or even with solid histology. In addition, three low grade subtypes were identified according to previously published criteria, i.e., muconodular cancer, characterized by prominent mucin lakes with expansile growth and moderately numerous tumor cells freely floating in them [33], separated from the remaining mucinous cancers showing an infiltrative pattern, low grade diffuse desmoplastic cancer with prominent desmoplasia entrapping tumor cells with a limited or no invasive pattern, to be separated from ordinary diffuse cancer [34], and very well-differentiated tubular cancer with monostratified, low grade epithelium and a limited or no invasive pattern [32, 33], somewhat reminiscent of tubular breast cancer [44], to be separated from ordinary cohesive cancers.

200

Statistical analysis Data distribution was expressed as counts and percents for categorical variables and was compared by means of the Fisher exact test. The Spearman R correlation coefficient was used to measure the association of variables on a continuous scale. The follow-up extended from the date of surgery to the date of death from gastric cancer or to the last available assessment. Patients dying from other causes were censored at the date of their death. Median follow-up and its interquartile range (25th–75th) was computed by means of the inverse Kaplan–Meier method. Cox regression was used to assess the prognostic role of cytologic atypia and of a series of other potential risk factors. Uni- and multivariable models were fitted. Clinically relevant variables were included in the multivariable model in addition to stage, provided they were not collinear. Hazard ratios (HR) and their 95% confidence intervals (95% CI) were computed. The proportional hazard assumption was tested based on Schoenfeld residuals. Death rates per 100 persons per year (95% CI) were also computed as summary measures. To assess model performance, we also calculated the Royston explained variation, Harrell’s C statistic for discrimination, and the Le Cessie–van Houwellingen shrinkage coefficient (and noise in model=1—shrinkage) for calibration. A twosided p value...