Loss of Heterozygosity Mapping at Chromosome Arm 16q in 712 Breast Tumors Reveals Factors that Influence Delineation of Candidate Regions¹

Molecular genetic analysis and cytogenetic techniques indicate chromosome 16 as one of the most frequently altered chromosomes in breast cancer. LOH³is reported to range from 36% to 67% (1, 2, 3, 4, 5, 6, 7, 8, 9). The long arm of chromosome 16 is also a target for LOH in other tumor types such as prostate, lung, and hepatocellular cancer and rhabdomyosarcoma(10, 11, 12, 13, 14), suggesting that 16q harbors a multitumor suppressor gene. However, there are also tumor types that show high LOH frequencies on other chromosome arms, but not on 16q, e.g.,gastrointestinal tumors, indicating that 16q LOH is most likely a selected event.

Cytogenetic studies have implicated loss of 16q as an early event in breast carcinogenesis because it is found in tumors with few or no other cytogenetic abnormalities (15, 16, 17). LOH studies on DCIS, the preinvasive stage of ductal breast carcinoma, have also indicated 16q LOH as an early event in breast carcinogenesis. LOH on 16q was found in in situ components in 29–55% of the cases tested (18, 19, 20).

Data on 16q LOH are further corroborated by several CGH studies on invasive and in situ breast tumors (21, 22, 23, 24, 25). CGH shows that the long arm of chromosome 16 is involved in physical deletion. Percentages are lower than those obtained with LOH studies with a mean of 25%. This can be attributed to the fact that LOH is also detected when mitotic recombination has occurred, a phenomenon that does not result in loss of copy number and consequently is not detected by CGH. The occurrence of LOH due to mitotic recombination strongly suggests that haploinsufficiency is unlikely to be the genetic mechanism of the TSG at 16q.

Many studies have attempted to identify the SROs that are the target of LOH at chromosome 16q in breast cancer. At least two or three nonoverlapping regions are reported, of which 16q24.3 and 16q22.1 are most frequent. We identified the gene encoding E-cadherin, CDH1, at 16q22.1 as a target gene, but only in the histological subgroup of lobular carcinomas (26, 27). Ductal carcinomas, which comprise a much more frequent histological subtype, also show LOH of 16q22.1 but show no CDH1mutations. Therefore, at least two and maybe more TSGs that are targeted by LOH on 16q remain to be identified.

However, there is no consensus on the exact boundaries of these SROs. Published data on LOH are often confusing and contradictory. LOH data are difficult to interpret. This difficulty is exacerbated when data are pooled from different studies for a number of reasons:(a) no clear definitions of LOH are used; (b)different polymorphic markers are used; (c) SROs can be misguidedly based on a small number of tumors (which may well represent nonselected genetic events); (d) tumor series are heterogeneous; (e) tumors themselves can be heterogeneous;and (f) more than one TSG locus may be present at a chromosome arm. Any or all of these reasons may account for the lack of successful TSG identification using LOH mapping.

In this report, we describe a study of LOH at chromosome arm 16q in three sets of breast tumors originating from three different centers (a total of 712 cases). By testing a large series with a dense set of polymorphic markers that were carefully mapped, we intended to define SROs with a high degree of probability. By comparing three different tumor series, we have investigated which factors may influence detection of LOH and thus delineation of SROs. One data set was analyzed with different criteria for tumor selection and LOH detection than the other two, which enables the assessment of criteria that influence LOH data. Furthermore, by using a large data set, there is sufficient statistical power to identify possible correlations of specific LOH patterns and clinical markers.

We have found evidence for three SROs at 16q. Mucinous tumors have no LOH at 16q. Estrogen receptor-positive tumors are more prevalent in the group with LOH of 16q, but not in tumors with LOH at 16q24.3 only. We show that complex LOH patterns are more frequent when the threshold window for retention of heterozygosity is increased.

Three series of breast cancer patients were tested for AI on the long arm of chromosome 16. Histopathological classification was carried out according to the WHO criteria (28). Patients were graded histopathologically according to the modified Bloom and Richardson method (29). Patient material was obtained on approval of local medical ethics committees. DNA from tumor tissue and peripheral blood cells of the same patient was isolated as described previously(9).

Series 1 consists of 189 patients operated on between 1986 and 1993 in three Dutch hospitals, the LUMC and two peripheral centers. Tumor tissue was snap-frozen within a few hours after resection. For DNA isolation, a tissue block was selected only if it was shown to contain at least 50% tumor cells on examination of a H&E-stained section by a pathologist.

Series 2 originates from the Imperial Cancer Research Fund Breast Group at Guy’s Hospital (London, United Kingdom) and consists of 400 patients. Tumor tissue was freshly frozen and estimated to have at least 50% tumor cells.

Series 3 consists of 123 patients operated on between 1987 and 1997 at the Flinders Medical Center (Adelaide, Australia). Of these tumor tissue samples, 87 tumors were collected as fresh specimens within a few hours of surgical resection, confirmed as malignant tissue by pathological analysis, and snap-frozen in liquid nitrogen until subsequent DNA isolation. The remaining 36 tumor tissue samples were obtained from archival paraffin-embedded tumor blocks. A subset of 33 tumors was microdissected from tissue sections mounted on glass slides to yield at least 80% tumor cells. For some cases, no peripheral blood was available, and pathologically identified paraffin-embedded nonmalignant lymph node tissue was used instead.

The markers that were used in this study are listed in Fig. 1. The figure shows for which tumor series they were applied, their type,and their cytogenetic location. Details of all markers can be found in the Genome Database.⁴The marker order was deduced from data in Genome Database by mapping on a chromosome 16 somatic hybrid map (30) and by information on the genomic sequence.

Four different methods were used for AI analysis:

(a) Southern blotting was used to test RFLP and variable number of tandem repeat markers only on a subset of series 1 as described previously (method 1; Ref. 1).

(b) Microsatellite markers were amplified on normal/tumor DNA panels using PCR with ³²P-labeled nucleotides as described previously (method 2; Ref. 31). Ambiguous results were quantified using a PhosphorImager type 445 SI (Molecular Dynamics, Sunnyvale, CA). The AIF is the quotient of the peak height ratios from normal and tumor DNA. The threshold for AI is defined as 40% reduction of one allele, in agreement with an AIF of ≥1.7 or≤0.59. This threshold is in concordance with our selection of tumor tissue blocks containing at least 50% tumor cells with a 10% error range. The threshold for retention has previously been empirically determined to range from 0.76 to 1.3 (32). A so-called“gray area” with AIFs of 0.58–0.75 and 1.31–1.69 is left, for which no definite decision is made. Gray area values are depicted in Fig. 1 as gray boxes. Tumors with only gray area values are discarded completely from the analysis. When adjacent markers show clear-cut LOH or retention, the gray area values are ignored, and tumors are categorized according to their interpretable markers.

(c) The third method for AI analysis is similar to that described above, omitting the radioactive-labeled dCTP. PCR reactions of polymorphic microsatellite markers were performed with one of the PCR primers fluorescence-labeled with either FAM, TET, or HEX and subsequent analysis of PCR products on an ABI 377 automatic sequencer (PE Biosystems). Peak height values and peak sizes are analyzed with the GeneScan package. The same thresholds for AI, retention, and gray area used for the radioactive analysis are used here.

(d) Finally, an alternative fluorescence analysis was used with fluorescein- or hexachlorofluorescein-labeled primers as described previously (method 4; Ref. 33). The threshold range of AIF for allele retention was defined as 0.61–1.69, the range for allelic loss was defined as ≤0.5 or ≥2.0, and the range for the gray area was defined as 0.51–0.6 or 1.7–1.99.

Methods 1, 2, and 3 were performed at the LUMC on tumor series 1 and 2. Method 4 was applied to series 3 and performed at Flinders University(Adelaide, Australia).

Comparison of AI data for validation of the different detection methods and the different tumor series was done with theχ ² test. Correlation between allelic loss and histopathological markers was also tested with theχ ² test.

The three different sets of tumors were tested for LOH at chromosome arm 16q. LOH analysis for series 1 and 2 was performed at one center(LUMC), whereas series 3 was tested for LOH at the other center(Flinders University). These three series originated from different populations and consist of paired tumor and normal DNA from 189, 400,and 123 patients, respectively, and were tested with 37, 10, and 16 polymorphic markers as summarized in Table 1. The results of LOH analysis for the three series of invasive breast tumors are represented in Table 2. This table shows that 18 tumors from series 1 and 76 tumors from series 2 were discarded because the AIFs of all markers tested were within the gray area. This term is assigned to those results showing inconclusive AIFs.

Marker sets, LOH detection, and LOH criteria were different for each of the three series. However, these discrepancies did not result in marked differences in the overall occurrence of LOH at 16q. The effect of these discrepancies was noted only in the assignment of categories, i.e., whether there was complex LOH or LOH of the whole arm. There was a trend in series 3 toward fewer LOH events,but this difference was not significant (P = 0.12). Nor was there a significant difference between the three groups with respect to the total number of tumors with LOH of one or more markers at 16q (P = 0.14). There was,however, a marked difference in the number of tumors with loss of the whole arm and loss of only a part of 16q. This is explained by the lack of markers in the centromeric region, i.e., 16q11–21 in series 2 (only one marker) and series 3 (no markers centromeric of 16q22.1), whereas in series 1, we tested six markers in bands 16q11–21. This shows that the selection of markers influences the detection of patterns of AI. This is further illustrated by the lack of identification of a SRO at 16q23, which encompasses the fragile site Fra16D and has been implied as a LOH target by other studies(34).

The most marked difference between series 1 and 2 versusseries 3 is the number of tumors with complex LOH, alternating loss and retention of markers, which is much higher in the latter series(P = 0.004). Examples of tumors with complex LOH are shown in Fig. 2. This is most probably an effect of using different criteria for LOH and reflects that a marker is scored as showing retention of heterozygosity despite an AI. A marker in series 3 is considered to be retained when there is between 0 and 39% reduction of one of the alleles. In contrast, a marker in series 1 and 2 is judged as retained when the reduction of one of the alleles is between 0 and 24%.

Among the tumors with loss of one region, either interstitial or telomeric, there was more consensus in the three series, with a minor significant difference between the three groups for telomeric LOH(P = 0.02).

The methods used for analysis of polymorphic markers in this study are different. A major difference is the use of radioactively versus fluorescently labeled PCR products. To compare these systems, we have analyzed 16 tumor/normal DNA pairs with five polymorphic microsatellites, two tetranucleotides, and three dinucleotides using both methods. Radioactively labeled PCR products are analyzed with a PhosphorImager, and fluorescently labeled PCR products are analyzed on an ABI 377 automatic sequencer. AIFs are determined as described in “Materials and Methods.” An overview of the comparison is shown in Table 3. Sixty-nine PCRs give interpretable results for both radioactive and fluorescence labeling methods. Twenty-three PCRs are not informative; i.e., they show homozygosity in the constitutional DNA. Of the 46 remaining results, the AIF values diverge predominantly for high AI and for weak PCR reactions. In 24% of these informative cases,there is a discrepancy between the two methods for the assignment of AI. In 11% of these cases, one method gave AI, and the results using the other method were in the gray area. In 13% of these cases, one method indicated retention, whereas the results using the other method were in the gray area. AIFs in the gray area were found with both methods. There were no cases where one method showed AI and the other indicated retention, and there were no cases where the methods showed a discrepancy for informativity of a marker.

Mapping of the SRO involved in LOH is instrumental for determining the location of a putative TSG targeted by LOH. Fig. 1 shows the LOH results for tumors with small regions involved, i.e.,interstitial and telomeric LOH. For reasons elaborated in the“Discussion,” we have omitted tumors with complex LOH for assigning SROs. When comparing the three different data sets, at least three consistent regions emerge: two at the telomere in band 16q24.3 and one at 16q22.1. The region at 16q22.1 is designated A, is demarcated by markers D16S398 and D16S301, and is based on interstitial LOH events in three tumors from series 1 and one tumor from series 3. Region B at 16q24.2–3 is demarcated by markers D16S498 and D16S3407 and is based on four tumors from series 3. Region C at 16q24.3 extends from D16S3407 to the telomere and is based on one tumor from series 1, one tumor from series 2, and three tumors from series 3.

LOH limited to 16q24.3 can be found in 30 tumors and involves both regions B and C in 24 cases. 16q24.3 LOH is more frequent than LOH at other regions, e.g., 16q22.1, which is the sole target in only seven cases.

Histological subtype was known for 526 of the 618 tumors with interpretable results for LOH on chromosome arm 16q. Most tumors in these series are of the invasive ductal histological type(n = 466). The groups of lobular(n = 46), mucinous (n = 9), papillary (n = 2), and medullary(n = 2) tumors are too small for a significant comparison to stratify into the different LOH categories assigned in Table 2. When comparing LOH anywhere on 16q with no 16q LOH, there is no significant difference between ductal and lobular carcinoma. However mucinous breast cancer does not show 16q LOH in any of the eight samples tested, providing a significance of P = 0.003.

Differentiation grade was known for 424 cases: 70, 181, and 173 cases were differentiation grade I, II, and III, respectively. Distribution of differentiation grade was not significantly different in tumors with and without LOH of 16q or within the different LOH categories.

Hormonal status was available for 496 cases. As reported previously(1), there was a weak significant difference for estrogen positivity in tumors with LOH at 16q (P = 0.04). On stratification to LOH category, there was an inverse correlation for estrogen receptor-negative cases in tumors with LOH at 16q24.3 only (P = 0.008).

In this study, we have investigated LOH at chromosome arm 16q in 712 breast tumors that originated from three different centers and were tested with different polymorphic markers and different criteria. We identified 96 tumors that were not suitable for correct interpretation. This conclusion was drawn because the AIFs were in the gray area for all markers tested in these tumors. The most probable causes for this are an excess of contaminating normal cells, tumor heterogeneity,or 16q aneusomy other than LOH. In some cases, a single marker shows a dubious AIF, which is often inherent to the polymorphic marker(e.g., overlap of one allele with a shadow band of the other allele, or PCR products representing the alleles run so close on the gel that they are difficult to distinguish). For series 3, tumor tissue was microdissected when necessary to yield at least 80% tumor cells. In this series, no cases were excluded due to weak AIF,indicating the advantage of microdissection. However, the percentages of LOH categories, as assigned in Table 2, in the 33 microdissected tumors did not deviate from those in the nonmicrodissected cases in series 3, suggesting that microdissection does not necessarily give different LOH results.

A total of 52% of the remaining 618 tumors show allelic loss of one or more markers on 16q. This percentage may be higher because most of the 94 tumors with weak AIF may in fact have LOH of 16q but do not meet our stringent criteria. The majority of the tumors show loss of the whole 16q arm or of the region from 16q22 to the telomere. This is in concordance with cytogenetic data of karyotyped breast tumors(15, 17, 35), CGH (24, 36), and interphase fluorescence in situ hybridization (37).

In a previous study on LOH at chromosomal band 7q31, we showed that there is a discordant rate of LOH scoring of 12% in a double blind scoring (38). Furthermore, we have shown that artifactual LOH can be found when input of template DNA is low (39). In the current study, we have tested whether the method of LOH detection and criteria for LOH or retention influence the assignment of SROs.

Fluorescence detection and radioactive detection of AI give comparable results, although the AIFs sometimes differ. In 24% of the informative cases, we find a deviation between the two methods, but this is always a discrepancy between an AIF compliant with the gray area and an AIF for either loss or retention. In none of the cases do we find allelic loss with one method and retention with the other. There is no significant difference in the occurrence of AIFs in the gray area,indicating that fluorescence detection does not necessarily give more clear-cut results, as may have been expected from direct detection of allele intensities. This analysis suggests that different detection methods for AI cannot explain the discrepancies found in allelic loss studies.

The stringency of criteria for allelic loss, retention, or gray area may explain the difference in the percentage of LOH obtained by different research groups. Here we show that there is not a marked difference in the overall frequency of LOH when applying different criteria. However, a marked increase in the frequency of complex LOH patterns was seen when applying a stringency for LOH of >50% and retention of <40% loss of one allele intensity (method 4 as described in the “Materials and Methods”) rather than applying cutoff values of >40% and <25%, respectively (methods 1–3). Because thresholds for LOH and retention are set rather arbitrarily, this challenges the delineation of SROs as candidate tumor suppressor loci in tumors with complex LOH patterns.

16q LOH mapping reveals alternating regions of allelic loss and retention of heterozygosity in 96 tumors as well as loss of only the most distal part of 16q in 52 cases. LOH of only a small region is considered to indicate a tumor suppressor locus. As shown in previous LOH studies on chromosome arm 16q, there is more than one SRO involved in such events (1, 2, 4), strongly suggesting the presence of two or more TSGs on chromosome arm 16q. It is not clear whether all SROs that are defined by LOH maps really represent tumor suppressor loci. In our study, we consider tumors with complex LOH as not informative for SRO delineation for two reasons: (a)assigning different criteria for LOH and retention results in an increase of cases with complex LOH patterns; and (b) these tumors involve allelic loss of chromosomal regions that are unique for this tumor and do not overlap with interstitial deletion events in other tumors. Thus, these complex LOH patterns may well represent nonselected genetic events.

In this study, we have assigned three SROs, one at 16q22.1 (SRO A) and two at 16q24.3 (SRO B and C). These SROs are based on tumors with only a single region at 16q involved in LOH. Other LOH mapping studies have assigned similar regions at 16q as SRO in breast cancer (2, 4, 40, 41) The SROs defined in the current study are all based on LOH patterns observed in at least four tumors with no complex LOH as depicted in Fig. 1. Evidence for the most distal region at 16q24.3 was found in all three tumor series tested. It must be noted that the two markers that demarcate this region, i.e., D16S3407 and D16S303, were applied in all three series, suggesting that the identification of a SRO depends on the marker density in a particular region. This is also illustrated by the fact that only one marker at 16q22.1 was tested in series 2 and that not a single tumor in this series actually delineates the SRO A at 16q22.1. This region does not overlap that of the CDH1 gene, which we showed to be targeted only in lobular breast cancer (26). The four tumors that have LOH only at this region and not at SRO B and C are all of the ductal type.

The more centromeric region at 16q24.3, region B, overlaps with a locus, SEN16, which was identified by microcell-mediated transfer of chromosome 16 fragments that cause senescence in the recipient breast tumor cell lines (42, 43).

A dense transcript map has been constructed from SRO C(44), and the eight most likely candidate genes located in this region were screened for the presence of mutations in tumors with LOH restricted to 16q24: (a) SPG7(45); (b) BBC1 (46);(c) copine VII (47); (d) PISSLRE (48); (e) FAA(49); (f) MC1R; (g) GAS11; and (h) c16orf3(50). To date this analysis has not resulted in the identification of the targeted gene in the 16q24.3 region, although seven less likely candidates have not yet been screened for mutations. It may well be that other mechanisms than mutational inactivation are operational for the putative TSG at 16q24.3, e.g.,transcriptional inactivation by methylation (51).

We compared two clinical parameters, histology and differentiation grade, with LOH on 16q. Our series contained ductal, lobular, mucinous,medullary, and papillary tumors. There is no prevalence for LOH at a particular region in any of these subtypes. We could not find a difference in 16q LOH frequency when comparing the results on ductal and lobular tumors in our series. Mucinous tumors all showed retention. However, this may be due to an overall lack of LOH in this tumor type. LOH studies on mucinous breast tumors are lacking, but these tumors are often diploid, suggesting few numerical chromosomal aberrations(52).

Surprisingly, we could not find a significant difference in LOH when comparing different tumor grades. We have recently reported on LOH in DCIS, which contains less genetic alterations. Grade I DCIS shows predominantly 16q LOH, whereas grade III DCIS predominantly shows LOH at 17p (20), which may suggest two different molecular pathways. This observation is corroborated by a CGH study on DCIS(25) that showed underrepresentation of chromosome arm 16q almost exclusively in grade I DCIS. A CGH study on 40 grade I and 50 grade III invasive breast tumors (36) showed a strong prevalence of chromosome 16 copy number loss in the well-differentiated grade I group. This may suggest different mechanisms of 16q LOH, i.e., physical loss versus mitotic recombination in grade I and grade III tumors, respectively. A study by Tsuda et al. (37) showed a correlation between the mechanism of chromosome 16q loss and the histological type. However,this report had no data on LOH on chromosome 16. The question of whether grade I and grade III tumors with 16q LOH have different mechanisms for LOH will be explored in future studies.

A positive estrogen receptor was prevalent in tumors with LOH at 16q,except in cases with LOH at 16q24.3 only, which show a prevalence for estrogen receptor-negative tumors. A correlation between estrogen receptor positivity and 16q LOH has been described previously on a smaller series (1).

In conclusion, we have investigated a group of 712 primary breast tumors for LOH at 16q. Most tumors with LOH have a large region, even the whole 16q arm involved, and are therefore not informative for mapping of the TSG targeted by LOH. Tumors with complex LOH patterns should be discarded for SRO mapping because of lack of consensus within and between tumor series.

Thresholds for LOH and retention may be the cause of differences in the assignment of SROs. The location of polymorphic markers certainly determines the detection of SROs. The method used for LOH detection is most probably not the cause of interpretational differences. This study shows that even a large tumor series and dense LOH mapping may not be sufficient to decrease candidate regions for TSGs. Therefore, additional methods should be applied, e.g.,statistical modeling of LOH data (53) and high throughput screening methods like genomic and cDNA microarray technology.

We thank Drs. K. Welvaart, J. Calamé, and M. C. Gorsira for providing clinical material.