To characterize cytogenetic alterations found in Barrett's adenocarcinoma (BA) and, more importantly, its premalignant stages, we studied chromosomal imbalances in various lesions in the histologically proposed metaplasia-dysplasia-carcinoma sequence using comparative genomic hybridization (CGH). Using 30 esophageal adenocarcinoma resection specimens, we were able to study 30 areas of Barrett's adenocarcinoma and 8 lymph node metastases (LN). In addition, we investigated 25 premalignant lesions adjacent to BA derived from a subset of 14 resection specimens including 11 areas of high grade dysplasia (HGD), 8 areas of low grade dysplasia (LGD), and 6 areas of intestinal metaplasia (IM), which were laser-microdissected and studied with CGH. To validate the CGH findings, fluorescence in situ hybridization analysis on 13 BA with probes specific for HER-2/neu and 20q13.2 were performed. The chromosomal alterations most often identified in BA were: gains on 8q (80%), 20q (60%), 2p, 7p and 10q (47% each), 6p (37%), 15q (33%) and 17q (30%). Losses were observed predominantly on the Y-chromosome (76%), 4q (50%), 5q and 9p (43% each), 18q (40%), 7q (33%) and 14q (30%). High-level amplifications were observed on 8q23-qter, 8p12-pter, 7p11-p14, 7q21-31, 17q11-q23. Recurrent chromosomal changes were also identified in metaplastic (gains on 8q, 6p, 10q, losses on 13q, Y, 9p) and dysplastic epithelium (gains on 8q, 20q, 2p, 10q, 15q, losses on Y, 5q, 9p, 13q, 18q). Novel amplified chromosomal regions on chromosomes 2p and 10q were detected in both Barrett's adenocarcinoma and premalignant lesions. An increase of the average number of detected chromosomal imbalances from IM (7.0 +/- 1.7), to LGD (10.8 +/- 2.2), HGD (13.4 +/- 1.1), BA (13.3 +/- 1.4), and LN (22 +/- 1.2) was seen. Although the detection of common chromosomal alterations in premalignant lesions and adjacent carcinomas suggest a process of clonal expansion, the occurrence of several chromosomal changes in an apparently random order relative to one another is striking evidence that clonal evolution is more complex than would be predicted by linear models. This is probably a reflection of the existence of many divergent neoplastic subpopulations and highlights one of the main problems associated with surveillance of Barrett's patients, namely sampling error.