Genome-wide Allelotyping of Lung Cancer Identifies New Regions of Allelic Loss, Differences between Small Cell Lung Cancer and Non-Small Cell Lung Cancer, and Loci Clustering¹ Free

Lung cancer is one of the leading cause of cancer-related deaths in industrialized countries. Its major etiological factor is tobacco smoke, a complex mixture containing many carcinogens that play a role in the genetic and epigenetic changes occurring in the respiratory epithelium during the multistage process of lung carcinogenesis. It is currently believed that 10 or more such events may have occurred before lung cancers become clinically evident (1). These genetic changes include activation of proto-oncogenes such as myc, Kras, EGFR, Her-2/neu, and BCL-2 as well as the inactivation of TSGs³like p53, Rb, FHIT, and p16(2, 3). Genetic changes affecting TSGs usually involve two events: (a) the loss of large chromosomal DNA regions of one parental DNA; and (b) a smaller mutational event(e.g., point mutation, smaller deletion, or promoter hypermethylation) affecting the second allele. The loss of chromosomal DNA can be detected by several cytogenetic techniques or by screening the tumors for LOH by allelotyping (reviewed in Refs. 2and 3). Several areas of frequent allelic loss [1p, 3p(including FHIT), 4p, 4q, 5q, 6q, 8p, 9p21(p16^INK4), 10q, 11p, 13q14 (RB1),17p13 (p53), 18q, 19p, 22q, and Xq (2, 3)]have been found previously in lung cancers. In a few cases, the underlying TSGs have been identified, whereas the search for the other putative TSGs is under way in several laboratories.

Whereas several regions (such as 3p) have received detailed analysis with many markers, to our knowledge, there has not been a comprehensive, genome-wide analysis of allelic loss with a high density of markers. In particular, there are little or no data on analysis of individual lung cancers with a battery of allele loss tests. To further refine the previous mapping of regions of allele loss and to find other potential LOH sites, we carried out a genome-wide allelotyping in 36 lung cancer cell lines (14 SCLCs and 22 NSCLCs) using 399 fluorescent microsatellite markers and an ABI 377 sequencer/genotyper. We have previously reported preliminary allelotyping of multiple chromosomal regions using 85 polymorphic microsatellite markers (4),as well as studies of selected regions such as chromosomes 3p, 4p, 4q,and 8p for analysis with high densities of markers (5, 6, 7). As a prelude to performing the present genome-wide analysis on lung cancer cell lines, we tested a subpanel of 12 NSCLC lines and their archival specimen for morphology, aneuploidy, immunohistochemical expression of HER2/neu and p53 proteins, and LOH at 13 chromosomal regions using 37 of these 85 microsatellite markers (8). We found an excellent concordance (>95%) in these parameters between the lung tumor cell lines and their corresponding uncultured tumor tissues. Thus, although tumor cell lines may have acquired additional abnormalities in culture, in the large majority of cases, they faithfully represent the uncultured tumors. The 399 markers used in the current study are widely used, commercially available, and evenly spaced across the genome. They are different from those used previously and, together with our prior tests, give us LOH data on well over 500 different microsatellite markers. In this report, several regions with LOH have been identified in lung cancer cell lines, extending previous results, and four new HDs have been detected. In addition, clustering analysis uncovered previously unknown interactions between sites of allelic loss. The lung tumor cell lines we have genotyped in this report have been deposited by us in the American Type Culture Collection and have been widely distributed to many investigators. They provide an extremely useful renewable resource for verification of results, for positional cloning and characterization of new TSGs and oncogenes, and for expression and functional studies in lung cancer. The data generated here will provide new markers that can be used for early lung cancer detection, monitoring of chemoprevention trials, and positional cloning of new TSGs.

The 36 lung cancer cell lines used in this study (14 SCLCs and 22 NSCLCs) and their normal matched BL cell line controls were established by us at the National Cancer Institute (NCI-H series) and at the Hamon Center for Therapeutic Oncology Research, University of Texas Southwestern Medical Center (HCC series; Table 1). They were grown in cell culture using previously published conditions(RPMI 1640 supplemented with 5% or 10% fetal bovine serum), and DNA was prepared by standard methods (4). These cell lines have been deposited for distribution in the American Type Culture Collection.⁴

The 400 microsatellite markers used in this study were obtained from the ABI Prism linkage mapping set v.2(Perkin-Elmer)⁵and were originally selected from the Généthon human linkage map (9). They consist of fluorescent primer pairs end-labeled with fluorochromes FAM, HEX, or NED that amplify dinucleotide-repeat fragments of 70–400 bp. Chromosomal localization of each marker was estimated by combining data from the Généthon genetic map and from the Genome Database.⁶We optimized the vendor’s conditions to perform multiplex PCR by varying the following parameters: (a) the combination of microsatellite marker primers from the various ABI Prism sets that can be coamplified and loaded together (only markers with the same fluorochrome and different size products were combined); (b)the relative concentration of marker primers within each combination;and (c) the MgCl₂ concentration. We were unable to use one of the markers (D1S2890) from the ABI Prism set,and it was discarded. The list of multiplex sets and the specific optimization conditions we used are available on request.

Optimized multiplex PCR reactions were performed in 10-μl total volumes with three to four ABI markers (0.1–0.65 μl for each individual undiluted primer pair or 0.5–3.25 pmol for each primer),1× PCR buffer, 50 ng of cell line DNA, 1.5–4.5 mmMgCl₂ (depending on the multiplex set of primers), 0.25 mm each deoxynucleotide triphosphate, and 0.4 unit of AmpliTaq Gold DNA polymerase (Perkin-Elmer). The following PCR run conditions were used: (a) 95°C for 10 min (1 cycle); (b) 94°C for 30 s, 55°C for 30 s, and 72°C for 30 s (10 cycles); (c) 89°C for 30 s,55°C for 30 s, and 72°C for 30 s (20 cycles); and(d) 72°C for 20 min (1 cycle).

The loading mix was prepared with 1.5 μl of deionized formamide, 0.3μl of dye (50 mg/ml dextran blue and 25 mm EDTA), and 0.25 μl of ABI internal size standard. Each lane contained this loading mix plus 0.3 μl each of FAM-, HEX-, and NED-labeled PCR product (the FAM-labeled product was diluted 1:2 before loading). The gels were prepared with 5% Long Ranger gel solution (Perkin-Elmer) and run at 3000 V for 2 h and 15 min. The fluorescent bands were laser scanned, and the data were stored electronically by the ABI 377 sequencer/genotyper. Representative output results are shown in Fig. 1. Complete or nearly complete loss of one allele was scored as LOH. The loss of both alleles was scored as HD. Retention of both alleles was scored as HET. A unique peak in both the tumor cell line and the matched constitutional BL cell line was scored as NI. A MA was scored when one of the tumor cell allele peaks was different from the BL peaks. Specifically, when the tumor cell had two peaks, the MA was scored as MA/HET; when the tumor cell had one peak, it was scored as either MA/LOH (when the BL cell had two peaks) or MA/NI (when the BL cell had one peak). MA/NI was interpreted as LOH because it is usually unlikely that the two BL alleles (represented by a single peak) would be shifted to the exact same position in the tumor cell.

The data were analyzed using ABI Prism Genescan software. In addition,a series of Microsoft Visual Basic programs were written specifically for various computations or repetitive tasks in Microsoft Excel,including color-coded formatting of LOH patterns, LOH characteristics(LOH frequencies, sizes of LOH, breakpoint frequencies, and so forth),and cluster analysis (see below). The program codes are in a user-ready format and are available on request.

For the purpose of this analysis, each allelotyping datapoint was assigned a number and a color code as follows: (a) +1(green), HET and MA/HET (i.e., clear examples of retention of heterozygosity); (b) +0.5 (light green), NI or a group of consecutive NIs that is bordered on either side (for a given tumor cell line) by two regions of heterozygosity (+1; such NIs are considered likely to have both copies retained); (c) 0 (gray), NI or a group of consecutive NIs that is bordered by a region of heterozygosity (+1) on one side and by a region of loss (−1) on the other (i.e.,no way of interpolating for allele loss or retention); (d)−0.5 (light blue), NI or a group of consecutive NIs that is bordered by two regions of loss (−1; such NIs are considered likely to have undergone allele loss); and (e) −1(blue), LOH, MA/LOH, MA/NI, and HD (i.e., clear examples of allele loss).

This number coding allows NI data to be assigned a “partially informative” value in a conservative manner based on their context of loss, i.e., NIs that are surrounded by LOHs for a given tumor are statistically more likely to be LOHs than HETs, and,conversely, NIs that are in a region of HETs are more likely to be HETs. Next, a concordance value is assigned for any two data points by multiplying their corresponding numerical values as defined above. Data points with the same sign thus have a positive concordance value that may vary between 0 (no concordance) and 1 (full concordance), and data points with different signs have a negative concordance value that ranges from −1 (fully discordant) to 0 (no discordance).

The concordance value of two different markers for a group of tumor samples (such as SCLC or NSCLC) is defined as the average concordance value of all pairs of corresponding data points for the two markers. In other words, two markers are concordant for a group of tumor samples if the allelotyping data for these two markers are similar for each of the samples, and, conversely, they are discordant if the allelotyping data are opposite for each tumor, with different degrees of concordance(from −1 to 1; from totally discordant to totally concordant).

The clustering algorithm is based in part on a strategy described by Eisen et al. (10) and was designed as follows. In the first round, all pairs of 399 markers (there are 79,800 such pairs) are scanned for the pair that has the highest concordance value. One of the markers from this pair is discarded, and the other is replaced by an array containing these two markers and whose LOH data are the average of the two sets of data for each marker (in their number format). The second round now has 398 elements that are either markers or arrays of markers associated with LOH data. The same procedure can be repeated on these elements, resulting in the replacement of the pair with the higher concordance by an array consisting of two or more markers made from the joining of the two elements (the average of LOH data are weighted by the number of markers in each element). This process is repeated until only one element remains that consists of an array of 399 markers listed in an order that clusters together markers with similar LOH patterns. This clustering is arranged so that markers with high LOH frequencies are preferentially positioned at the beginning of the clustered array (this is done in each round by joining the marker with the lower LOH frequency to the right of the marker with the higher LOH frequency) and chromosomal distances between adjacent markers in the cluster are minimized (this is done in each round by selecting the pair with the highest concordance, and, if there is more than one, selecting the pair with the shortest chromosomal distance). Finally, the whole clustered array is divided into nonoverlapping subgroups so that markers within each group have at least 75% concordance (i.e., have a concordance value of at least 0.75) between each other (Fig. 4).

For association between LOH frequencies and lung cancer types (SCLC versus NSCLC; Fig. 2), a one-sided binomial distribution test was performed (comparing the lower proportion against the higher one). Comparison of the averages of FAL, heterozygosity rates, numbers of breakpoints, numbers of MAs,numbers of allelic loss hot spots, average sizes of LOH (number of contiguous markers involved in LOH), and numbers of whole chromosome losses were done by using Student’s two-sided t test. Correlation between the presence of MAs and the frequency of allelic loss was assessed using both Pearson correlation coefficient and covariance.

In this report, 399 fluorescent polymorphic microsatellite markers(from ABI Prism set v.2) were used that span the entire human genome with an average intermarker distance of ∼10 cM. To increase the efficiency of the allelotyping protocol described in the ABI Prism manuals, we first carried out optimization of multiplex PCR conditions(see “Materials and Methods”). The 399 markers were combined into 110 panels containing three to four markers each that can amplify PCR fragments of different sizes. Because the markers within each panel are labeled with a common fluorescent dye (FAM, HEX, or NED), the PCR products generated from the 110 panels can be further combined and loaded into 36 lanes so that each lane contains products of different sizes and fluorochromes. One gel (36 lanes) is therefore sufficient to analyze each DNA sample for all 399 markers.

Thirty-six lung cancer cell line DNAs and their normal matched BL cell line DNA controls were allelotyped with all 399 markers (Fig. 2). Fig. 1 shows representative output data from the ABI genotyper/sequencer. Overall, the average marker heterozygosity rate for the BL DNAs was 79%, with no significant difference between SCLC and NSCLC cases(Table 1). This heterozygosity rate is an exact match with the average ABI/Généthon heterozygosity rates (79%) for these markers. Thus it appeared that the BL lines from these lung cancer patients were reasonable controls. The percentage of LOH per informative cases for both SCLC and NSCLC ranged from 0% (marker D7S519) to 92% (marker D17S938). The average percentage of LOH was 36%, with no significant differences between SCLC and NSCLC (Table 1). Previously, we and others have made the working assumption that LOH rates < 30%were probably due to random genetic changes (4). To examine this question, we plotted the number of markers showing various amounts of LOH (Fig. 3). It appears that a trough in the range 30–35% separates two regions,one (Fig. 3, left, with a peak at 20–25%) of which we assume contains mostly random losses and another (Fig. 3, right) that represents losses presumably associated with the cancer-specific phenotype. In addition, 42 markers show very frequent LOH (≥60%) for both SCLC and NSCLC and are located in regions 3p, 4q, 8p, 9q, 13q, 17p, Xp, and Xq (Fig. 2, markers in red). The 17p region loss most likely represents LOH for p53, whereas the 13q region loss predominantly represents LOH for Rb.

The FAL is a measure of the extent of allele loss in a given tumor sample and is defined as the number of LOH events in a sample divided by the total informative (heterozygous) markers in the corresponding normal DNA (11). The FAL in lung cancers ranged from 1%(HCC78) to 63% (H2882). The average FAL was 34% for SCLCs and 36%for NSCLCs, a difference that is not significant (Table 1).

At least one example of MA was found for 34% (137 of 399) of the markers, most of which occurred in only one tumor DNA sample (Table 1). No significant differences in the number of MAs between SCLC and NSCLC were observed. Two markers had four tumors with MAs (D16S3136 and D20S178) but were not located in regions of frequent allelic loss. Each lung cancer cell line had an average of five MAs in the 399 markers (Table 1). No significant correlation was found between the presence of MA and the frequency of allelic loss for individual markers; the average frequency of allelic loss for markers showing MAs were, in fact, slightly lower than the overall average (33%versus 36%).

The sizes of regional allelic loss in individual tumors were determined. The size of each of these region losses was calculated as the number of contiguous markers showing LOH, including intervening NI markers between flanking regions of LOH. The average size of contiguous LOH was calculated for each lung tumor cell line (excluding clear cases of whole chromosomal losses; Table 1). The average number of contiguous markers involved in LOH in SCLC was 5.9 ± 1.3 markers,whereas in NSCLC, the average size was 5.5 ± 1.6 markers (not a significant difference; Table 1). Assuming an average intermarker distance of 10 cM, this means that the average length of regions of allele loss in lung cancer was 50–60 cM. The number of whole chromosome losses in lung cancer ranged from 0 (H1963 and HCC78)to 13 (HCC827; average, 3–4 chromosomes lost/tumor; Table 1). HCC827 lost 13 chromosomes and had only six regions of subchromosomal allelic loss. Thus, in this cell line, chromosome nondisjunction appeared to be the predominant mechanism of allele loss. In contrast, HCC78 had no loss of whole chromosomes and only two regions of LOH. Different mechanisms of allelic loss can therefore occur for different tumors.

To determine the location of regions with very frequent breakpoints in the tumors, we calculated the percentages of breakpoints for each pair of contiguous markers (SCLC and NSCLC combined). Here, breakpoints are defined as the junction between a marker showing LOH and an adjacent marker retaining heterozygosity (HET) in a given tumor sample DNA. If a LOH and a HET are separated by NI markers, the intervening junctions are considered part of the same breakpoint. The highest percentages of breakpoints (≥20%) were found in 1q12, 1q23, 3p11–13, 3p21.3–22,5p11–13, 8p12–q11, 9p13–21.1, 10p11–q11, 16p11, and 17p11 (Fig. 2, double bars between markers). Several of these were located near a hot spot of allelic loss. The average percentage of breakpoints for all pairs of consecutive markers was 4.8% and ranged from 0–47%with a SD of 6.2%. Hence, contiguous markers showing ≥20%breakpoints are significantly different (P = 0.01) from the average markers. For individual tumors, the number of breakpoints was 18 ± 6 in SCLC and 20 ± 9 in NSCLC (not a significant difference; Table 1).

To define more clearly the regions that undergo allelic loss, areas of common allele loss (also called minimal areas of loss or hot spots)were determined by identifying the smallest common regions of frequent LOH in SCLC and in NSCLC (Fig. 2). For individual samples, these regions were scored as: (a) HET when at least one HET (or MA/HET) occurred in the region for this sample (the rest,if any, were NI); (b) NI when only NIs were observed in the region or when both HET and LOH were present; and (c) LOH when at least one LOH (or HD, MA/LOH, or MA/NI) occurred in the region for this sample (and the rest were NI). Frequencies of allele loss for these common areas of loss are calculated as usual as(LOH)/(HET + LOH). Regions of minimal loss are listed in Table 2 along with potential candidate genes.

Several such regions of allelic loss were found, the most prominent of which (>60% loss) were 1p36, 3p (multiple sites), 4p14–q13,4q21–28, 4q34–ter, 5p13–q13, 5q32–ter, 8p21–23, 9p21–22(p16 and p15), 9q22–32, 10p15, 10q22–23, 13q11,13q12–14 (RB1), 13q34, 17p12–13 (p53), 19p13,Xp–q21, and Xq22, consistent with previous results. Several loci were novel in lung cancer (Table 2). Thirty-two regions of loss were confined to one marker or located between two markers and, as such, are useful candidates for further allelotyping with higher densities of markers (Fig. 2). Note that SCLCs and NSCLCs have very distinct patterns of allele loss, as observed previously (4). In fact, one-sixth of all markers (65 of 399 markers) showed a statistically significant difference in LOH frequencies between SCLC and NSCLC (Fig. 2). On average, each SCLC has undergone loss of 17 ± 4 loci (hot spots), whereas each NSCLC has lost 22 ± 8 loci, results that were significantly different(Table 1).

Because of the importance of HDs in the positional cloning of TSGs, we searched for clear examples where both alleles were lost. Nine HDs were detected (Table 3), two of which appeared to be contiguous in the same cell line (H2052). They are located in chromosome regions 2p23.3, 3p14.2, 8q24.3, 9p21,18q11.2, and Xq22.1. Four of these occurred in LOH hot spot regions where putative TSGs are probably located: (a) 3p14(FHIT) and 9p21 (p16^INK4A), which have been described previously; and (b) 18q11.2 and Xq22,which are novel sites. Interestingly, in this series of markers, HDs were found only in NSCLCs. Three of these (H1648, HCC95, and H2052) had two HDs each (Table 3). However, HDs do occur in SCLCs as well, as has been shown previously (3).

The percentage of HDs that were detected with all 399 markers and in all 36 tumor samples was 0.06% [9/(399*36)]. This represents an estimate of the likelihood of finding a HD. It also means that one of four tumors can be expected to harbor a HD with our assay. However,because of the relatively small size of most reported HDs (typically less than 1 Mb) and the current intermarker distance of 10 cM (10 Mb),a larger number of markers would likely uncover proportionally more HDs.

This data set represents one of the first opportunities to see whether alterations (allele loss) in one chromosomal region are linked to changes in another region. We approached this by cluster analysis. The clustering algorithm we designed is based on a strategy similar to that described by Eisen et al. (10) and was implemented in Visual Basic for Excel in a user-ready format (available on request; see “Materials and Methods”).

The clustering analysis reordered all 399 markers in such a way that markers with similar LOH patterns tend to be grouped (clustered)together. It also identified discrete, nonoverlapping regions within the clustered array where LOH patterns are more than 75% concordant to one another as defined in “Materials and Methods” (vertical bars in Fig. 4, A and B). In most cases, these regions consist of markers that map to the same chromosome because their physical linkage tends to correlate with similar patterns of allelic loss. Obviously, these provide little new information. In contrast,biologically interesting results are obtained from those regions that contain clustered markers located on different chromosomes. A preliminary clustering analysis of all lung cancer samples did not reveal any concordance between markers from different chromosomes,presumably because of the different patterns of allelic loss between SCLC and NSCLC. Similarly, very little concordances were found when NSCLCs were analyzed as a group. However, analysis of SCLC and the adenocarcinoma subgroup of NSCLC (15 of 22 NSCLCs) revealed new correlations.

Fig. 4, A and B, shows the clustering analysis of all SCLCs and all adenocarcinomas, respectively. In each case, several regions of high concordance were revealed (vertical bars),and the magnifications in Fig. 4, C and D, show the most significant ones (consisting of markers from different chromosomes with high LOH frequencies). In SCLCs, concordances were observed between markers in the 3p14 locus (FHIT) and the 13q14 locus (RB), markers in the 13q14 locus (RB)and the 17p13 locus (p53), and 4q12 and 5q. In adenocarcinomas, concordances between 1q and 3p22, 3p14(FHIT) and 20p12–13, and 3q23 and 15q12 were noted. Overall, these data suggest that genetic collaboration occurs between the loss of corresponding TSGs during tumor development and that these concordant losses have a significant degree of histological specificity.

The present study describes a high-resolution, genome-wide allelotyping of 36 lung cancer cell lines using a large number(n = 399) of fluorescent microsatellite markers with a coverage of ∼10 cM on average. Previous allelotyping analyses of lung cancer by our group and others were restricted to particular chromosomal regions or arms or used relatively low densities of markers (4, 5, 6, 12). Very few reports have presented allelotyping data on multiple sites in the same tumors. Our results thus represent, to date, the highest resolution of lung cancer allelotyping with genome-wide coverage and allow for more comprehensive studies, including clustering analysis for the determination of loci interactions. To increase throughput, we have optimized the conditions for PCR multiplexing to allow analysis of each DNA sample with only one 36-lane ABI sequencer gel. In addition, several Visual Basic software programs were designed to facilitate data processing.

Several sites of frequent allelic loss have been detected in this study. The location of these sites was based on the determination of the minimal regions of loss that are defined by the occurrence of breakpoints (transitions between markers showing LOH and those retaining heterozygosity) surrounding regions of LOHs. These sites are likely to represent the major TSG regions that are affected in lung cancer and are therefore good indicators of critical regions of loss that warrant further investigation.

It was shown previously that chromosome arms 3p (several sites), 13q(Rb), and 17p (p53) have the most frequent allelic loss in both SCLC (>90%) and NSCLC (>70%; Refs.4 and 13, 14, 15). LOHs were also found in >60%of cases in 4q, 5q, 10q, and 22q for SCLC and on 6q, 8p, 9p, and 19p for NSCLC, and additional sites with moderate frequencies were also detected throughout the genome (4, 5, 6, 16). Our results agree with and extend these findings. In particular, of 38 chromosomal arms previously analyzed by us, 25 from SCLC or NSCLC had an average frequency of LOH that agrees with the present findings to within 10%,although the markers used were different (4, 5, 6, 7).

Overall, we found more than 50 different sites of frequent allelic loss(hot spots) in SCLC and NSCLC; about half of these sites have not been described previously in lung cancer. The most prominent of these novel sites (>60% LOH) occurred on 1q23, 9q22.33–32 (PTCHlocus), 10p15.3, and 13q34 in SCLC and on 13q11 and Xq22.1 in NSCLC. To our knowledge, except for PTCH on 9q, none of these novel sites with high-frequency LOH have been shown to harbor candidate TSGs. Future investigations with an even higher resolution of microsatellite markers will therefore be crucial to narrow down the sites of frequent allelic loss. Such studies have been performed previously by us in 3p (54 markers), 4p and 4q (16 markers), and 8p (26 markers), and we identified several regions of loss in addition to those reported here (5, 6, 7).

On average, 17 regions of high LOH frequency (hot spots) were found in each SCLC, and 22 regions of high LOH frequency were found in each NSCLC, giving an indication of the possible number of TSGs mutated in each lung cancer. Of course, some of the allelic losses may be caused by genomic instability or the presence of fragile sites(16). In fact, some breakpoints were identified that were recurrent in a significant fraction (>20%) of tumors. They are located on 1q, 3p, 5p11–13, 8p12–q11, 9p13–q21, 10p11–q11, 16p11,and 17p11. Most are located near a hot spot of frequent allelic loss,and some of these losses may therefore be the consequence of recurrent alterations at common, potentially fragile sites. Whether or not losses associated with these fragile sites participate in the tumor growth selection process is a question that will need to be addressed through other experimental approaches such as functional assays. Whereas we use the term breakpoint to describe these LOH-HET junctions, we stress that these could occur by physical chromosomal deletion or by other processes such as mitotic recombination (17, 18).

The statistically significant but relatively small difference in the number of hot spots in SCLC and NSCLC (17 versus 22) is possibly a consequence of the greater number of NSCLC samples that were analyzed because this tends to refine the minimal areas of loss and hence increase their number. This would be supported by the otherwise overall similarities in the characteristics of SCLC and NSCLC (FAL,average sizes of LOH, number of breakpoints, and microsatellite abnormalities) shown in Table 1. However, these two histological types do show significantly different LOH patterns for specific chromosomal regions.

Interestingly, one cell line (HCC78) appears to have a very stable genome with a low percentage of allele loss (FAL of 1%). In this cell line, only two markers (D3S1300/FHIT and D6S292)demonstrated a LOH, and both occur in a chromosomal hot spot for allelic loss [3p14.2 (FHIT locus) and 6q22.3–23],suggesting that only a few TSG mutations or losses may have been sufficient (together with some oncogene activations) for the transformation of this lung cancer. This tumor, although unusual, thus represents an important case for future detailed study.

Nine HDs were found in six regions, and all may be of potential use in identifying the resident TSGs. Of those six regions, four are localized to hot spot areas of allele loss: (a) 3p14; (b)9p21; (c) 18q11; and (d) Xq22. Two of these, 3p14 and 9p21, harbor known TSGs, namely FHIT and p16,respectively. Several HDs have been reported previously in 18q21–22 in colorectal and pancreatic cancers where DCC, SMAD2, and SMAD4 are located(19, 20, 21), but none to date were found at 18q11, and no known TSG appears to be present at this locus. Because of its high frequency of allelic loss (59%) and its specificity to NSCLC (0% in SCLC), this region appears particularly interesting and, along with another HD in 8q24, is currently under investigation. The deletion we found in Xq22 is not, strictly speaking, a HD because it was detected in a cell line from a male patient, but the absence of a wild-type allelic copy suggests that the loss of this site may also be involved in the tumorigenic process. No such Xq22 deletions have been reported previously.

MAs have been reported in 35% of SCLCs and 22% of NSCLCs using a relatively small number of markers and are associated with advanced tumor stage (2). Our results show that nearly all tumors have at least one MA, and half of them have more than five MAs. Five of the 36 tumors (14%) had more than 10 MAs and might be good candidates for screening of gene mutation in the DNA mismatch repair pathway. However, none of the MAs had the appearance of “laddering”characteristic of the replication error phenotype seen in colon cancers with a mutation in one of the mismatch repair enzymes(22).

Cluster analysis was performed using a Visual Basic program that searches for correlations between the loss or retention of different markers in lung cancer. The algorithm we designed is similar to one that has been used for cluster analysis of microarray data(10) and is well adapted to allelotyping studies. A number of correlated groups of markers that appeared to be lost or retained in a coordinate fashion were thus identified. Concordance groups might represent different TSGs whose synergistic loss would provide a growth advantage during cancer development. Because of the unavoidable presence of NI data, the groups of concordant markers listed by our clustering program may constitute only a subset of what would be detected if all markers were completely informative.

In SCLC, correlations were found between markers on 3p14–25 and 13q12–14 and between 13q14 and 17p13 that suggest synergistic interactions between the loss of putative TSGs in these loci(e.g., FHIT and RB, and RBand p53, respectively). Alternatively, these correlations may be the consequence of sequential molecular genetic changes that occur in the development of lung cancer(23, 24, 25, 26). A more novel finding is the concordance found between allelic loss on 4q12 and 5q, with both regions showing a high rate of LOH. Whereas no candidate TSGs have been described in the first region, several have been reported in 5q, including MCC, APC,DOC2, and IRF1 (27, 28, 29, 30). Further investigation of the 4q12 region may thus reveal a gene collaborating with one of the putative TSGs on 5q.

An initial clustering analysis in NSCLCs did not reveal significant concordances, presumably because of the histopathological heterogeneity of these NSCLCs. However, in the subset of 15 adenocarcinomas, two concordance groups were observed that had frequent allelic loss, one associating 1q and 3p22, and the other associating 3p14 and 20p12–13. Two novel hot spots of allelic loss were uncovered in 1q, and one novel hot spot of allelic loss was uncovered in 20p12,and neither currently harbors candidate TSGs deposited in GenBank. These regions are thus interesting candidates as collaborating TSGs in adenocarcinomas.

Future experiments will include a genome-wide mRNA expression analysis using cDNA microarrays and will relate those results with the present findings. This will allow us to find several candidate target genes that may be affected by the various chromosomal losses described here and will further refine the genomic study of lung cancer-related changes. Together with the present study, it will provide us with a comprehensive genomic characterization of several tumor cell lines that will be very useful for future functional analyses and gene discovery experiments.