Dissecting karyotypic patterns in malignant melanomas: Temporal clustering of losses and gains in melanoma karyotypic evolution

Authors


Abstract

Malignant melanomas can be divided into two major subtypes, involving either the skin or eye melanomas. Both tumor forms exhibit highly complex karyotypes with nonrandom recurrent chromosomal imbalances. Loss of chromosome 3, the short arm of chromosome 1, and gain of 8q have been suggested to be associated with eye melanomas, whereas gain of 6p and loss of 6q have been more often seen in skin melanomas. Imbalances implicated in tumor progression include among others, −10 and +7. In spite of the abundance of cytogenetic information, with more than 300 published karyotypes, very little is known about the mode of karyotypic evolution or of the presence of possible cytogenetic pathways. In our investigation, we have used 362 melanoma karyotypes, including both the skin and eye subtypes, to identify the most frequently occurring imbalances. Tumor cases were then classified with respect to the presence or absence of these imbalances and statistically analyzed in order to assess the order of appearance of chromosomal imbalances, the presence of karyotypic pathways, as well as possible cytogenetic subtypes. We show that the melanomas develop through one mode of karyotypic evolution, common to both low and high complexity karyotypes, and we establish the temporal order by which the different imbalances occur. By applying several statistical methods, we show that at least two cytogenetic pathways of clonal evolution exist in malignant melanomas, one initiated with −3 and one with +6p, and that these pathways operate in both skin and eye melanomas. © 2003 Wiley-Liss, Inc.

Malignant melanomas are divided into 2 major subtypes, depending on their locations: skin and eye melanomas, respectively. Skin melanomas are derived from epidermal melanocytes and reliant on genetic factors such as familial incidence, skin type, race and gender and environmental factors such as ultraviolet radiation. This tumor type is one of the most rapidly increasing malignancies among Caucasians and has a high mortality rate. In sporadic forms chromosomes 1, 6, 7, 9 and 10 are the chromosomes most commonly involved in somatic mutations.1 Structural rearrangements of chromosome 1 are especially frequent and include translocations, duplications and deletions, leading to loss of 1p. The most noted chromosome 6 changes have been nonreciprocal translocations leading to the loss of 6q, also frequently occurring as a consequence of isochromosome 6p formation. Although less frequent than 1p− and 6q−, loss of 9p is an important change in cutaneous melanomas as this loss involves the CDKN2A locus, known to predispose mutation carriers for melanocyte tumors.2 The frequent changes +7 and −10 have been associated with tumor progression.1 Uveal melanoma is the most common cancer of the eye and arises in the melanocytes of the uveal tract. Cytogenetic investigations have identified several recurrent clonal abnormalities such as loss of chromosome 3, gain of 8q and deletions of 6q. The gain of 8q may occur through several mechanisms such as isochromosome formation, unbalanced translocations and gain of a complete chromosome 8. An association between gain of chromosome 8 and loss of chromosome 3 may exist since these imbalances are frequently seen together.3, 4 Given the similar cellular origin of cutaneous and uveal melanomas, attention has been focused on defining tumor-specific karyotypic profiles. In this respect, −3, +8q and 1p− have been suggested to be specific for uveal melanomas and +6p and 6q− for cutaneous melanomas. However, these conclusions have been made from relatively small series of tumors and several investigators have reported frequent loss of 6q in uveal tumors5 as well as frequent loss of 3 and gain of 8q in cutaneous tumors.6

The high degree of karyotypic complexity in malignant melanomas makes it very difficult to discern possible cytogenetic subtypes. To overcome some of the difficulties caused by the composite nature of chromosomal changes in solid tumors, we have developed and adapted several statistical methods suitable for the analysis of complex karyotypes.7, 8 In the present investigation, we applied these methods to the 307 malignant melanomas reported in the Mitelman Database of Chromosome Aberrations in Cancer9 and to 55 previously unpublished cases of uveal melanomas. We constructed a genomic imbalance map and used this map to identify the most frequent imbalances. Tumors were then classified with respect to the presence or absence of these imbalances and statistically analyzed, in order to identify and assess karyotypic profiles, the chronological order of the imbalances as well as to identify possible karyotypic pathways.

Abbreviations:

HCA, hierarchical cluster analysis; MDS, multidimensional scaling; NAPT, number of aberrations per tumor; NIPT, number of imbalances per tumor; PCA, principal component analysis; TO, time of occurrence.

MATERIAL AND METHODS

Selection of data

A total of 307 karyotypes were retrieved from the Mitelman Database of Chromosome Aberrations in Cancer.9 These karyotypes were supplemented with 55 previously unpublished cases of uveal melanomas (Appendix 1) and used to construct imbalance maps at the 368-band level. Regions that were either lost in more than 10% and or gained in more than 5% of the cases were identified and for each region the most affected segment was selected as the diagnostic imbalance. A total of 36 imbalances, excluding −Y, were identified in this way (Table I). Each karyotype was then assessed for the presence or absence of the selected imbalances. The number of imbalances per tumor (NIPT), and the total number of aberrations per tumor (NAPT), were calculated and karyotypes with at least one of the selected imbalances were retrieved, resulting in 340 cases.

Table I. Frequencies (%) of Imbalances in Malignant Melanomas
 AbbreviationMM1n = 340MM Skin1n = 92MM eye1n = 244χ2-test
  • 1

    MM, Malignant melanomas, frequencies based on all cases in the Mitelman Database of Chromosome Aberrations in Cancer classified as malignant melanomas; MM skin, frequencies based on all cases in the Mitelman Database classified as skin malignant melanomas; MM eye, frequencies based on all cases in the Mitelman Database classified as eye malignant melanomas, and the uveal melanomas from the appendix.–p-values < 0.01 are given.

+1q24-q44+1q263324
+2 111111
+3 121810
+4q27-q35+4q9910
+6p21-p25+6p282229
+7 293626
+8q10-q24+8q352538
+9q22-q34+9q10158
+12q15-q24+12q9510
+13 111210
+15 111111
+17q10-q25+17q10129
+18 1726130.0060
+19 10148
+20 1827150.0070
+21 141116
+22 121113
−1p10-p361p-292829
−3 4324500.0000
−4 2133160.0016
−5 2132160.0016
−6q10-q276q-394238
−8p10-p238p-282529
−9p10-p249p-283724
−10−103659270.0000
−11q23-q2511q-262826
−12q13-q2412q-1727140.0032
−14 2335180.0015
−15 263323
−16 2636220.0081
17p- 212619
−18 172216
−19 191719
−21 213715
−22 232323
−X 232322

Temporal analysis

Early and late imbalances were defined as in Höglund et al.7, 8 In brief, to obtain a value for lateness all tumors with a given imbalance were selected and the distributions of NIPT plotted. The modes of these distributions were used as an estimate of lateness and referred to as the time of occurrence, TO. TO is thus a function of karyotypic complexity and only takes cytogenetically detectable changes into consideration. The selected distributions were then resampled with replacement (bootstrapped) 1,000 times and the TO scored after each resampling.10, 11 The mean of the bootstrapped TO values was then used as the TO for the given imbalance. The bootstrapped 2.5, 25, 75 and 97.5 percentiles were also calculated. For the bootstrap estimates, the resampling software from Resampling Stats (Arlington, VA) was used. To evaluate if an imbalance occurred earlier or later than expected from random events, a simulation procedure was used in which the distribution of NIPT for the whole tumor population as well as the frequencies of individual imbalances were identical to the observed. The simulation was performed as described in Höglund et al.7

Principal component analysis

Principal component analysis (PCA) was performed using the Statistica software package (Statsoft, Tulsa, OH). PCA is a standard multivariate method frequently used to search for underlying structures in data sets.8, 12 In short, principal components are linear combinations of the original variables, orthogonal, and ordered with respect to their variance so that the first principal component has the largest variance. To analyze imbalances, these were used as variables and the individual tumors as the observations; this will group imbalances frequently seen in the same tumors. The factor models arrived at were evaluated at two levels: total variance accounted for and communalities. The fraction of the original variance accounted for by the specific factor model is given by the cumulative percentage of the explained variance for each component in the factor model. The communality of a given variable (imbalance) is an estimate of how well the factor model predicts the behavior of the given variable. The communality ranges from 0 to 1, where a value of 1 indicates full explanation of the variance. To analyze the tumor population, the tumors were used as variables and the imbalances as the observations; this will group tumors with similar sets of imbalances. To arrange the individual tumors according to karyotypic complexity, the NIPT for each individual tumor was used as an observation in addition to the imbalances.8 Because this parameter shows the greatest variation, the position of the tumors along the first principal component will be almost identical to karyotypic complexity. Classifying imbalances8 were identified by testing each imbalance for the presence in all members in a given cluster produced by the PCA. Communalities were not considered when evaluating the PCA of the tumors.

Hierarchical cluster analysis and multidimensional scaling

An imbalance distance matrix using Euclidean metrics was produced where the imbalances were the variables and the tumors were the observations. The TO for each imbalance was included as an observation when calculating the matrix. The resulting matrix was analyzed by hierarchical cluster analysis (HCA) and multidimensional scaling (MDS). HCA is a statistical method that groups objects into clusters and then organizes the clusters in a hierarchical tree according to similarities.13 A consequence of HCA is that one object may only be part of one cluster. Ward's method was used for cluster formation (www.stasoft.com). MDS is a statistical method that organizes objects in lower dimensional space so as to maintain as much as possible of the original distances in a higher dimensional space.14 Hence, MDS extracts the major features of the Euclidean distance matrix. For HCA and MDS the Statistica software package (Statsoft) was used.

RESULTS

A total of 36 imbalances fulfilled the inclusion criteria (Table I), the most frequent being −3, 6q−, +8q and −10, all seen in more than 30% of the cases. To detect possible differences in karyotypic profiles between skin and eye melanomas, the tumor cases were sorted with regard to site and the frequencies of the imbalances were recalculated (Table I). Eight imbalances were found to be present in significantly higher frequencies (p<0.01) in the skin melanomas, namely, +18, +20, −4, −5, −10, 12q−, −14 and −16 whereas one imbalance, −3, was seen in significantly higher frequencies in melanomas of the eye. However, after adjusting the significance level to the large number of comparisons made by a Bonferoni correction, only −3 and −10 satisfied the p<0.05 criterion and none the p<0.01. The temporal analysis (Fig. 1) revealed −3, +6p, 6q−, +8q and 8p− to be early imbalances (TO<5), followed predominately by late losses with TO=5–10.5 that were followed by even later gains TO>10.5. Thus the acquisition of late imbalances is clearly divided in two phases, one early involving losses and one late involving gains. The simulation analysis revealed that +1q, −5, −10 and +18 were seen later in the karyotypic evolution than expected from a random process in which the NIPT distribution and the frequency of the individual imbalances are kept constant.

Figure 1.

Temporal analysis of imbalances in malignant melanomas. Vertical lines in hatched boxes, the mean of the bootstrapped TO; hatched boxes, boundaries of the 25 and 75 percentiles; thin lines, boundaries of the 2.5 and 97.5 percentiles. Imbalances were first sorted according to the TO values, then imbalances with the same TO were organized with respect to the value of the 25 percentile. For Abbreviations, see Table I.

The analysis of NAPT revealed an almost monotonically decreasing distribution (Fig. 2a) with possible overrepresentation of tumors with 3–5 and 12–13 changes. The distribution of NIPT (Fig. 2b) exhibited a similar overrepresentation of cases with 3 and with 8–12 imbalances. By excluding +6p and 8p− when producing the NIPT distribution, the peak at NIPT=3 disappeared, identifying the presence of i(6p) and i(8q) as possible causes for the overrepresentation of tumors with NIPT=3–5 compared to tumors with NIPT=1–2. Since the temporal analysis revealed two separate phases for the acquisition of losses and of gains, the modes of attaining these imbalances were analyzed separately. The karyotypes were reanalyzed by scoring the late losses and the late gains individually, after which the number of losses and of gains per tumor were plotted (Fig. 2c,d). The distribution of the number of gains was monomodal, indicating that the acquisition of gains follows one mode, whereas the distribution of the number of losses was bimodal, indicating a small subpopulation of tumors that acquire losses at a higher rate.

Figure 2.

The distribution of chromosomal changes in malignant melanomas. (a) The distribution of NAPT. (b) The distribution of NIPT. (c) The distribution of the number of losses per tumor. (d) The distribution of the number of gains per tumor.

Identification of cytogenetic pathways

PCA of the imbalances exhibited a scattered pattern of imbalances with some clustering (Fig. 3a). The early imbalances were placed between two opposite clusters, one representing losses and one representing gains. Since none of the components corresponded to a time axis, the TO values for each imbalance were included as an observation and the components recalculated.8 In the resulting 3-factor representation (Fig. 3b), the imbalances are aligned along the first component according to their TO, whereas the second component separates the imbalances in two major karyotypic pathways, 1 including +6p and 6q− and possibly −16, and 1 including −3, +8q, and 8p−, and possibly 1p−. The late imbalances formed two tight clusters, one including losses and one including gains. The third factor separated −3 from +8q and showed 1p− to be closer associated with −3 than with either +8q or 8p−. When the PCA was performed on skin and eye melanomas separately, essentially the same configurations were obtained (Fig. 3c,d). However, +6p and 6q−, and +8q and 8p−, respectively, were located closer together in the PCA of the eye melanomas compared to the skin melanomas suggesting that i(6p) and i(8q) were more frequent in melanomas of the eye. These close associations were also revealed by the fact that +6p and 6q−, and +8q and 8p− showed significant correlation (p<0.01) in eye but not in skin melanomas.

Figure 3.

Identification of cytogenetic pathways in malignant melanomas by means of PCA. (a) A 2-factor representation of the PCA of the imbalances. (b) A 3-factor PCA representation of the imbalances based on the total cohort of malignant melanomas and including TO as an observation. (c) A 3-factor PCA representation of the imbalances based on skin melanomas and including TO as an observation. (d) A 3-factor PCA representation of the imbalances based on eye melanomas and including TO as an observation. Imbalances in black, early imbalances (gains and losses); imbalances in green, late losses; imbalances in red, late gains.

To further analyze the associations among the imbalances, a distance matrix based on Euclidean metrics was produced. This matrix was first analyzed by HCA, which revealed 3 clusters of imbalances, early gains and losses, late losses and very late gains (Fig, 4a). The cluster constituting the late losses also included +7. An almost identical clustering was obtained when HCA of the imbalances in skin and eye melanomas were performed separately (data not shown). The same distance matrix was then used to produce a lower dimensional representation by applying MDS. In the resulting configurations, the imbalances were organized according to their TO values along the first dimension, which accordingly represents a time axis. The 3-dimensional representation (Fig. 4b) suggested at least 3 possible pathways: one originating with −3 followed by +8q, one by −3 followed by 1p− and 8p− and one originating with 6q− or +6p followed by −16. To make the relationships among the later imbalances more explicit, the early imbalances were removed to enlarge the configuration of the remaining late imbalances. In this representation, shown in Figure 4c, the late gains formed a tight cluster surrounded by a less dense cluster of losses. Gain of chromosome 7 behaved somewhat anomalously, since this imbalance did not cluster with the other gains. The 9p− and −10 were localized on the +6p/6q− side of the representation (the right side of the Fig. 4c), suggesting these as being part of the later stages of a +6p/6q− pathway. The close association of 9p− and −10 was also seen in the HCA in which they formed a separate subcluster of late losses together with 11q− (Fig. 4a). Similarly, −X and −15 were localized on the −3 side of the configuration (the left side of the figure), suggesting these as more specific for the −3 pathway. The close association of −X and −15 was also evident in the HCA (Fig. 4a).

Figure 4.

Identification of cytogenetic pathways in malignant melanomas by means of HCA and MDS. (a) HCA of the imbalances using Euclidean distances, Ward's agglomeration principle and based on the total cohort of malignant melanomas. (b) MDS analysis of the imbalances using the same distance matrix as in a). (c) MDS of the late imbalances. Imbalances in black, early imbalances (gains and losses); in green, late losses; in red, late gains.

Since the PCA and the MDS identified −3, +6p, 6q−, +8q, and 8p− as being of particular importance in the development of malignant melanoma, the correlation matrix of these imbalances were studied in more detail. From Table II, it may be seen that +6p shows significant negative correlation with 1p−, −3 and −X. Loss of chromosome 3 showed substantial association with 8p− as well as with +8q and 1p−. This indicates a −3 pathway followed by either +8q or 8p−, or both in the form of an i(8q), and then by the somewhat later 1p−, having −X and −15 as important even-later changes (Table II, Figs. 2 and 3). The loss of 6q showed association with at least 7 other losses, of which −16 was the most early.

Table II. Correlation Matrix for +6p, +8q, −3, 6q−, and 8p−
Imbalance1Individual correlations2
+6p+8q−36q−8p−
  • 1

    For Abbreviations see Table I.

  • 2

    Significant correlations (p < 0.01) in bold.

  • 3

    The border between significant and non-significant correlations were at R = 0.143 and thus some R = 0.14 in the table are significant and others are not, depending on the third digit.

+1q0.01−0.040.040.130.12
+2−0.050.100.110.10−0.03
+30.030.03−0.220.04−0.02
+4q−0.020.050.070.140.00
+6p1.000.02−0.210.14−0.09
+70.090.01−0.080.120.04
+8q0.021.000.300.020.22
+9q−0.020.01−0.030.070.11
+12q0.110.040.03−0.06−0.09
+130.100.07−0.020.160.02
+150.070.05−0.120.020.04
+17q−0.02−0.000.030.030.06
+180.030.060.010.04−0.00
+190.010.070.010.150.06
+200.100.05−0.060.100.07
+210.060.160.120.03−0.01
+220.040.080.040.050.03
1p−−0.15−0.010.28−0.090.02
−3−0.210.301.00−0.090.23
−4−0.02−0.090.050.110.11
−50.06−0.10−0.140.06−0.02
6q−0.1430.02−0.091.000.09
8p−−0.090.220.230.091.00
9p−0.03−0.08−0.020.230.06
−10−0.020.09−0.110.170.06
11q−0.09−0.090.000.12−0.03
12q−−0.05−0.060.070.220.17
−14−0.07−0.06−0.050.19−0.06
−150.06−0.010.110.160.11
−160.060.010.070.180.13
17p−−0.04−0.06−0.030.070.10
−180.12−0.09−0.070.060.06
−190.05−0.07−0.040.070.08
−210.05−0.06−0.100.160.06
−220.06−0.07−0.090.140.10
−X−0.150.040.160.070.10

Analyses of the tumor population

The tumor cases were then subjected to PCA. The first principal component in this analysis aligned the tumors according to karyotypic complexity and the second divided the tumors into two major and extended clusters that overlapped for the highly complex tumors (Fig. 5a). By testing the cases for presence and absence of the individual imbalances, it was shown that −3 or +8q containing tumors were located at one extreme and +6p or 6q− containing tumors at the other (Fig. 5b). Thus, the malignant melanomas may tentatively be characterized as either −3/+8q containing or +6p/6q− containing tumors. To make a more detailed analysis, the three first components were used to produce a 3-dimensional representation (Fig. 5c,d). In Figure 5c,d it is obvious that the malignant melanomas may be subdivided even further. In fact, more than 10 clusters may be discerned, of which some overlap due to the angle chosen for the representation. By testing each imbalance for presence or absence in each cluster, it was shown that all clusters could be classified by having combinations of the imbalances −3, +6p, 6q− and +8q (Fig. 5d). It is evident from Figure 5d that most of the possible combinations of these imbalances are seen and that the clusters are of comparable sizes.

Figure 5.

PCA of the tumor population. (a) The two first principal components. Light blue, NIPT=1; dark blue, NIPT=2; violet, NIPT=3; red, NIPT=4; brown, NIPT=5; black, NIPT>5. (b) The identification of two tentative major cytogenetic subtypes of malignant melanomas. Blue, tumors with either −3 or +8q, or both, but not +6p or 6q−; red, tumors with +6p or 6q−, or both, but not −3 or +8q; green, tumors with −3 or +8q together with +6p or 6q−; black, tumors showing absence of −3, +6p, 6q− and +8q. (c) A 3-dimensional representation of the PCA. Color codes as in a). (d) Identification of cytogenetic subtypes of malignant melanomas. Color codes represent different combinations of the presence or absence of −3, +6p, 6q−, and +8q. Cases in black show absence of −3, +6p, 6q− and +8q and are identical to the similarly labeled cases in b).

DISCUSSION

A total of 36 imbalances, excluding −Y, were seen in the malignant melanomas. The most frequent gains were +6p, +7 and +8q, and the most frequent losses were 1p−, −3, 6q− 8p−, 9p− and −10. The comparison between uveal and cutaneous tumors with respect to the frequency of the imbalances revealed 8 imbalances that were more frequent in cutanous melanomas, whereas 1 imbalance, −3, was more frequent in uveal tumors. However, when a correction was made for the large number of comparisons made, only two passed the significance criteria: −3, and −10. Even though this may indicate that −3 is more specific for uveal melanomas and −10 for skin melanomas, both were very frequent in both subtypes. Thus, no truly subtype-specific imbalance could be discerned and hence the two tumor types seem to have highly similar karyotypic profiles.

The distribution of NAPT showed an almost monotonically decreasing distribution with some over-representation of cases with 3–5 and 12–13 changes. A similar over-representation of cases with 3 and 8–12 changes was seen when the NIPT was plotted. The seemingly higher frequencies of tumors with 3–5 changes were shown to be caused to a large degree by the presence of i(6p) and i(8q). Even though these changes did not make up the majority of the chromosome 6 and 8 changes, the high prevalence of structural rearrangements of these chromosomes made the frequency of isochromosomes substantial. Since the isochromosome represents one event, seen from a karyotypic point of view, but results in two imbalances, the presence of many isochromosomes will lead to subpeaks in the NIPT distribution. By analyzing the distribution of losses and gains separately, it became evident that the distribution of losses was bimodal, whereas the distribution of gains was monomodal. This suggests that a small fraction of the malignant melanomas acquire an increased rate of attaining losses, but not gains, and that this may explain the slight overrepresentation of cases with 12–13 changes. The monomodal nature of the distributions indicates that the tumor population is homogenous with respect to the mode of karyotypic evolution: tumors with complex karyotypes seem to be produced by the same underlying process as the less complex tumors. The observed mode of karyotypic evolution in malignant melanomas is thus similar to the ones observed in bladder,15 breast,11 and kidney cancer7 as well as colorectal tumors.16 This may indicate that one common process for karyotypic evolution is operating in all five tumor types.

The temporal analysis showed a distinct pattern of early and late imbalances. Even though clinical data were not available to correlate the attained temporal order of imbalances with tumor progression, we have previously shown that results obtained with identical methods show a good correlation with pathological data.7, 11, 15, 16 The analysis showed −3, +6p, 6q−, +8q, and 8p− to be early imbalances. These imbalances were followed first by losses and then by gains, and thus the karyotypic evolution was partioned into two phases. The temporal clustering of losses and gains was also evident from the PCA and MDS analyses in which the losses formed a distinct cluster and the gains formed a denser separate cluster. A similar temporal clustering was seen in both uveal and cutaneous tumors when analyzed separately, and is thus a feature of both tumor subtypes. In kidney and colorectal carcinomas both gains and losses are frequent, but in these tumor types gains and losses constitute separate karyotypic pathways.7, 16 Hence, the temporal clustering of late losses and late gains seems to be a characteristic of malignant melanomas. The shift from the acquisition of losses to the acquisition of gains was not linked to a change in the mode of karyotypic evolution. Excluding the few tumors with increased propensity for late losses discussed above, the distribution of NIPT for late losses was almost identical to the equivalent NIPT distribution for late gains; both were close to monotonic and decreasing, and conformed to the overall NIPT distribution. This indicates that gains and losses are acquired by the same mode.

Loss of chromosome 10 seems to be of particular importance in the progression of melanomas since it was seen at high frequencies in both eye and skin tumors. Loss of chromosome 10 also occurred later than expected from the simulation, and thus the selective advantage of this imbalance may be dependent on preceding changes, stressing its importance in tumor progression.17 Likewise, +1q was a frequent late event and seen later than expected from the simulation. This is in contrast to its behavior in breast carcinomas, where +1q is an early and classifying imbalance. A similar unusual behavior was the late appearance of +7. This imbalance is frequent in many carcinomas, e.g., kidney, bladder and breast cancer as well as colorectal tumors.7, 11, 15, 16 However, in those tumor types, +7 is an early event and functions as a classifying imbalance in at least two of the tumor types. A particular characteristic of melanomas was the frequent involvement of chromosome 6, seen as either +6p or 6q−, frequently caused by i(6p) in the uveal melanomas. Furthermore, both +6p and 6q− were early events. This is in contrast to many other tumor types where +6p, when present, is a late event and 6q− is a moderately late event.7 The highly characteristic chromosome 6 changes in melanomas are only paralleled by the similarly typical −9 in bladder cancer.15 Hence, the high frequency of −10, the late appearance of +1q and +7, and the early appearance of +6p and 6q−, make the overall karyotypic profile of malignant melanomas highly characteristic.

To detect possible karyotypic pathways, we analyzed imbalances that were frequently seen in the same cases, the rationale being that imbalances belonging to the same pathways would be seen more frequently in the same cases than imbalances belonging to different pathways. Thus, correlations between imbalances would reveal possible cytogenetic pathways. To extract the central features of the correlation matrix, we performed principal component analysis. We have previously shown the adequacy of this method by comparing it with correspondence analysis,11 a method similar to PCA but based on normalized weighted chi-square distances.18 In the present investigation, we also applied multidimensional scaling, MDS. The purpose of this method is to arrange objects, in the present case imbalances, in a space with a limited number of dimensions so as to reproduce the observed distances in a multidimensional space. The advantage with MDS is that any distance measure may be used. In our investigation, Euclidean distances between imbalances were calculated by which two imbalances frequently seen in the same cases will be separated by a small distance. The fact that the PCA and the MDS produced almost identical representations indicates that the results are robust and that they may correspond to biologically relevant structures in the data.

The early imbalances −3 and +6p were seen at opposite positions in the PCA and the MDS representations. Accordingly, these two imbalances showed negative correlation and may thus represent two alternative starting points of karyotypic evolution (Fig. 6). An important change secondary to −3 was 1p−, which was seen in 42% of all the tumors with −3, and showed significant association with this imbalance. Both gain of 8q and loss of 8p were associated with −3 and thus represent additional important secondary changes in the −3 pathway. Of particular importance was the association between −3 and +8q, which was stronger than between 8p− and +8q, even though i(8q) was a frequent change. Late losses in the −3 pathway were −X and −15. Both the PCA and MDS results indicated a second pathway initiated with +6p and with 6q− as an important secondary event. The interpretation of 6q− is, however, somewhat ambiguous. First, the temporal analysis indicated that +6p was earlier than 6q−. Second, the PCA of the uveal melanomas indicated a strong linkage between +6p and 6q−, indicating 6q− as being secondary to +6p in this pathway, whereas the PCA of the cutaneous melanomas indicated 6q− to be secondary to −3. Furthermore, 6q− was present in 32% of the −3 tumors and 49% of the +6p tumors. The best interpretation of these results is that 6q− is an important secondary imbalance in both pathways, with some preference for the +6p pathway. In addition to showing a negative association with −3, +6p also showed a negative association with 1p− and −X, both of which were important imbalances in the −3 pathway. Later losses in the +6p pathway were 9p−, −10 and 11q−. After the acquisition of the pathway-specific late losses, the routes converged to a common set of late losses followed by late gains common to both pathways (Fig. 6). Similar pathways were obtained when the analyses were performed on cutaneous and uveal melanomas separately, suggesting a similar cytogenetic evolution in both tumor subtypes.

Figure 6.

A summary of the cytogenetic pathways in malignant based on the temporal analysis, the PCA and the MDS of the imbalances, the individual correlations between the imbalances and the PCA of the tumor cases.

In conclusion, the accumulated cytogenetic data on malignant melanomas revealed a complex pattern of chromosomal alterations. The distribution of NIPT was found to be monotonously decreasing, which is compatible with the conclusion that the same mechanisms produce low as well as high complex karyotypes. By the analysis of the NIPT distributions of tumor cases with specific imbalances, we were able to establish a temporal order by which the imbalances appeared in the melanoma karyotypic evolution. In addition, this analysis revealed a temporal clustering of losses and of gains. To reveal potential karyotypic pathways, we performed several statistical analyses. These investigations resulted in the identification of two major cytogenetic pathways, both of which were present in uveal and in cutaneous melanomas, suggesting that these two subtypes of melanomas develop though similar genetic pathways.

APPENDIX

 

Table  . Karyotypes from 55 cases of Uveal Melanomas1
  • 1

    The karyotypes were obtained from 1989 to 2000 at the British Columbia Cancer Agency. The biopsies were cultured in RPMI 1640 medium with added L-glutamine and 15% fetal calf serum. Dissaggrated cells from the tumors were grown in petri dishes for 5 to 15 days. Metaphases were harvested from culture supernatants and G-banded using standard methods. The karyotypes are given according to ISCN (1995).19

1.46,XX,t(1;11)(p36;q22)[13]
2.43,X,−Y,del(1)(p13),−3,i(4)(p10),+8,−13,−18[11]
3.43,X,−Y,−3,−14,der(17)t(14;17)(q10;q10)[6]
4.38–46,XY,del(1)(p21),+add(1)(p13),−3,add(7)(p11),−14,−15,−16[cp3]
5.43,X,−X,?dup(1)(q11q21),−3,−4,i(8)(q10)[cp10]/44,idem,+i(8)(q10)[cp5]/88,idem,+i(8)(q10)×2[cp5]
6.45,XY,i(6)(p10),−7[31]
7.43,X,−X,del(1)(p12),−3,+i(8)(q10),−13,−18[5]
8.a 46,XX,−3,i(8)(q10),+i(8)(q10)[2]/47,idem,+i(8)(q10)[2]/45,idem,der(1)t(1;16)(q10;p10),−16[3]
b 46,XX,−3,+8[2]/46,XX,−3,+i(8)(q10)[13]/46,XX,−3,i(8)(q10),+i(8)(q10)[4]
9.47,XX,+der(8)t(6;8)(p11;p11),add(15)(p11),t(8;22)(p11;p11)[11]
10.42–46,XX,add(1)(p13),−3,+8,+i(8)(q10),−22[cp5]
11.90,XXY,−Y,−3,−3,+i(8)(q10),−10,−10,+2-3mar[cp18]
12.44,X,−Y,−3,i(8)(q10)[20]/44,idem,i(6)(p10)[4]
13.46,X,−Y,−3,add(6)(q12),i(8)(q10),+i(8)(q10)×2,−13,−15,−22,+3mar[11]
14.44,X,−Y,del(1)(?p22p36),−3,−4,+8[12]/46,X,−Y,+7[cp4]
15.47,XY,+7[3]
16.48,XX,i(8)(q10)×2,+2mar[cp2]
17.46–47,XY,+2,der(4)add(4)(p15)add(4)(q26),−5,add(6)(q13),+add(6)(q13),−9,−10,+13,−15,+20,add(21)(p11),+mar[cp3]
18.48,XX,add(1)(q11),der(6)t(1;6)(q21;q21),i(6)(p10),i(8)(q10),+i(8)(q10),der(9)t(1;9)(q12;p11),+11,−16[5]
19.44,X,−Y,−3,i(8)(q10)[15]
20.47,XY,+del(6)(q15),add(14)(p11),add(19)(q13)[20]/47,idem,add(11)(q24)[10]
21.47–48,XY,−1,−3,der(5)t(1;5)(q21;q11),+?del(6)(q11),+i(8)(q10),del(16)(q22),+22[cp15]
22.45,XX,i(1)(q10),−2,−3,del(7)(q32),add(10)(p11),der(11)t(?4;11)(?q27;q22),−15,add(17)(p11),add(21)(p11),+mar[16]
23.46,XY,t(13;14)(p10;q10),−14,add(17)(p13),add(22)(p11),+mar[2]/46,idem,add(7)(p22)[4]
24.49–51,X,−Y,+i(1)(q10),−3,i(6)(p10),i(7)(p10),i(8)(q10),+i(8)(q10)×2,+mar[4]
25.46–48,XY,−3,i(8)(q10),+i(8)(q10),+i(8)(q10),+2−3mar[cp5]
26.70,XX,−X,i(6)(p10)×2,+7,i(8)(q10)×2,add(9)(p10),+14,+14,+14,+20,+20,3−8mar,inc[cp20]
27.45,X,−Y,der(17)t(6;17)(p11;p13),−20,+mar[16]
28.45,idem,der(6;8)(p10;q10)×2[6]/46,idem,der(6;8)(p10;q10)×3[5]
29.44,X,−X,−3,der(22)t(8;22)(q11;p11)[33]/45,X,−X,−3,+i(8)(q10),der(22)t(8;22)(q11;p11)[2]/90,XX,−X,−X,−3,−3,+i(8)(q10), +i(8)(q10),der(22)t(8;22)(q11;p11)×2[2]
30.42,X,−Y,der(1)t(1;3)(p11;p11),+del(1)(q10),−3,del(3)(p10),−6,i(8)(q10),der(12)?t(6;12)(p11;p13),add(13)(p11),−15,−20, der(22)t(8;22)(q10;p10),+mar[2]
31.42,X,−Y,der(1)t(1;3)(p11;p11),+del(1)(q10),−3,del(3)(p10),−6,i(8)(q10),der(12)?t(6;12)(p11;p13),add(13)(p11),−15,−20, der(22)t(8;22)(q10;p10),+mar[2]
32.46,XX,add(3)(q21)[5]
33.46,XX,t(1;11)(q11;q11)[2]
34.64–86,XXX,der(1)t(1;21)(p10;q10),−3,−6,i(6)(p10),+7,+8,+8,+i(8)(q10),−10,+15,+20,−21,+22,+mar[cp15]
35.49,XXXc,i(6)(p10),+21,+mar[18]/50,idem,+15[2]
36.40–47,XY,der(1)t(1;6)(p11;p21.1),+der(1)t(1;6)(p11;p21.1),−4,−6,i(8)(q10),−14,−15,add(19)(p13.3),+20,+mar[cp13]
37.46,add(X)(p22),Y,−1,add(1)(q21),add(3)(q21),del(6)(q15),+add(8)(p11),der(12)t(1;12)(q23;q13),−15,add(19)(p13.3),+mar[14]
38.42–47,X,−Y,add(4)(p16),−6,der(8)?t(6;8)(q15;p12),−10,add(13)(p11),−17,+1−4mar[cp3]
39.45,X,−Y,−3,+i(8)(q10)[6]/46,idem,+20[7]
40.46,X,?add(X)(p22),−3,+mar,3−5dmin[17]
41.47,XX,+6[22]
42.88,XX,−Y,−Y,i(1)(q10),+i(1)(q10),−3,i(6)(p10)×2,der(8)t(8;?16)(p11;p11)×2,+i(8)(q10)×3,−11,−11,−16,−16,−19[22]
43.46,XY,add(6)(q13),add(8)(p21),der(9)t(?8;9)(?q34;q11),add(11)(p14),der(16)(t?8;16)(?q13;q24),add(19)(q13)[10]/47,idem,+der(11)(p14),+19,-add(19)(q13)[8]
44.43,X,−X,−1,−3[28]
45.46,XY,add(6)(q13)×2,+der(11)t(11;22)(p11;q11),add(12)(q13),der(15)t(12;15)(q15;p11),−16,−22[10]
46.44,X,−Y,−3[2]/43,X,−Y,der(1)t(1;9)(p11;q11),−3,−9[14]
47.45,X,−Y,−3,+i(8)(q10)[4]/44,idem,−13[4]
48.44,XY,i(1)(q10),−3,−4,add(6)(q12),−10,add(11)(p15),−16,+r,+mar[cp14]
49.45,XX,add(6)(q11),add(6)(?q21),i(6)(p10),−11[cp8]
50.45,X,−Y,−3,+i(8)(q10)[8]/46,XY,−13,+mar[2]
51.45,X,−Y[13]
52.45,XY,−9,der(12)t(9;12)(q11;p11)[15]
53.46,der(X)add(X)(p22)del(X)(?q24q26),Y,add(6)(q15),add(7)(p11),der(?20)t(8;20)(q13;p11)[18]
54.45,XY,add(2)(p25),i(6)(p10),der(11)t(11;18)(q25;q11.2),−18,add(22)(p11)[8]
55.41–42,X,add(X)(p21),der(5)dic(5;19)(p14;p13.3),i(6)(p10),−10,−10,+ins(11)(q21;?),−13,t(13;15)(p11;q11),der(14)?t(14;10) (p11;q11), −15,−17,der(18)?dic(18;20)(p11;p13),−19,−20,−22,−22,+2−3mar[cp12]

Ancillary