| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Tumor Biology |
Department of Clinical Genetics, University Hospital, SE-221 85 Lund [M. H., D. G., G. B. H., F. M.], and Department of Genetics, Lund University, SE-223 62 Lund [T. S], Sweden
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
400 published karyotypes, very little is known about the mode of karyotypic evolution and the possible presence of cytogenetic pathways related to tumor development. In the present investigation we used 387 ovarian carcinoma karyotypes to identify the most frequent genomic imbalances. Tumor cases were then classified with respect to the presence or absence of these imbalances and statistically analyzed to assess the order of appearance of chromosomal imbalances, as well as possible karyotypic pathways and cytogenetic subtypes. We establish the temporal order by which the different imbalances occur and show that at least two cytogenetic pathways exist, one characterized by +7, +8q, and +12, and one by 6q- and 1q-. We show that ovarian carcinomas develop through at least three phases of karyotypic evolution. At the early stages, Phase I, the karyotypic evolution seems to proceed though step-wise acquisition of changes. The transition to Phase II showed signs of an increased chromosomal instability, most probably caused by extensive telomere crisis and the onset of breakage-fusion-bridge cycles. This process was linked to the presence of imbalances characteristic for the 6q-/1q- pathway. The transition to Phase III involved triploidization and was also linked to the presence of the 6q-/1q- pathway. |
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Several attempts have been made to systematize the cytogenetics of ovarian cancer (4 , 7) . However, because of the karyotypic complexity an extensive amount of cytogenetic and/or CGH data are needed to reveal possible karyotypic pathways for tumor development or cytogenetic subtypes. In the present investigation we have identified all of the cytogenetically aberrant ovarian carcinomas, altogether 387, present in the Mitelman Database of Chromosome Aberrations in Cancer.5 To identify the most frequent imbalances we constructed a genomic imbalance map. Tumors were then classified with respect to the presence or absence of these imbalances and statistically analyzed (8 , 9) to assess the order of appearance of imbalances, possible karyotypic pathways, and cytogenetic subtypes. These statistics were then compared with clinical data and in vitro measurements of mitotic instability.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
. Each karyotype was then assessed for the presence or absence of the selected imbalances. The NIPT was then calculated, and the 316 cases with at least one imbalance were selected for additional analysis.
|
HCA and MDS.
HCA and MDS were performed on an imbalance distance matrix based on Euclidean distances produced by using imbalances as the variables and the tumors as the observations. An HCA group objects into clusters and then organizes the clusters in a hierarchical tree according to similarity (10)
. A consequence of HCA is that one object may only be part of one cluster. Ward’s method was used for cluster formation.6
MDS organizes a higher dimensional matrix in a lower dimensional space so as to maintain as much as possible of the original distances (10)
, and, hence, MDS extracts the major features of a distance matrix. HCA and MDS were performed using the Statistica software package (Statsoft, Tulsa, OH).
PCA.
To search for possible patterns of correlations between the imbalances, PCA was performed using the Statistica software package (Statsoft). PCA is a standard multivariate method used frequently to search for underlying structures in data sets (11)
. 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 seen frequently in the same tumors. The factor models arrived at were evaluated at two levels: total variance accounted for and the 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. Only components with eigenvectors >1 were considered. 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. The NIPT for each individual tumor was used as an observation in addition to the imbalances (9)
. Because this parameter shows the greatest variation, the first principal component will be almost completely dominated by this variable and the position of the tumors along the first principal component will be ordered according to karyotypic complexity. Classifying imbalances (9)
were identified by testing each imbalance for the presence in all of the members in a given cluster produced by the PCA.
Measurement of Anaphase Bridging.
The frequencies of nuclear blebs, micronuclei, and chromatin strings in interphase cells were assessed in short-term cultures from nine ovarian adenocarcinomas, using routine cytogenetic preparation methods (12)
. The frequencies of these abnormal interphase structures are strongly correlated (r = 0.98) to the frequency of bridges at anaphase and may, therefore, be used as an indirect measure of this type of CIN (13)
.
|
RESULTS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
revealed a highly heterogeneous tumor population. The polyploid (CN >57) group constituted 36% of the tumors with CNs ranging from 58 to 127. Hyperdiploid (CN = 48–57) tumors were seen in 28%, and hypodiploid (CN <45) in 22%, whereas the peridiploid tumors (CN = 45–47) was a minority, seen in 14% of the cases. The distribution of the NIPT (Fig. 1b)
showed a sharp decline in frequencies of tumors with increasing NIPT and was similar to a geometrical distribution for NIPT = 1–8. However, above NIPT = 8 the frequencies were constant or slightly increasing, suggesting a multimodal distribution. To investigate the possible multimodality additionally, the NIPT distributions of tumors containing specific imbalances were analyzed separately. Tumors with 1p-, 1q-, 7p-, 10p-, -13, -15, -16, -17, and -18, and with +1q, showed clear bi- or trimodal distributions. In Fig. 1c
, cases containing these imbalances have been pooled and the NIPT distribution plotted. Three subsets of tumors are revealed by this analysis, those with 1–7, those with 8–16, and those with >16 imbalances. We next pooled and analyzed the hyper-, hypo-, and the peridiploid tumors. These near-diploid cases were dominated by tumors with 1–7 changes, showed a shoulder at NIPT = 8–15, and had very few tumors with >15 imbalances (Fig. 1d)
. The polyploid tumors showed a clear bimodal distribution, with tumors having 1–4 changes or >16 changes as the dominating subtypes (Fig. 1e)
. The data thus suggest that ovarian carcinomas may be divided into three classes of karyotypes representing three different phases of karyotypic evolution. Phase I is seen in both poly- and nonpolyploid tumors, and is characterized by a few imbalances. Phase II is mainly seen in near-diploid tumors and show 7–16 imbalances, and Phase III is mostly seen in polyploid tumors and contains >16 imbalances.
|
; Ref. 9
). Early imbalances (TO = 1–2) were 6q-, +7, +8q, and 1q-; intermediate imbalances (TO = 3–4) were +12, -8, and +1q; and the remaining were late (TO = 5–6). Imbalances not included in this analysis, and thus mostly seen in Phase II and Phase III tumors, were considered late and with no specific TO attached. Of the gains, the early +8q, and +12 showed significant (P < 0.01) associations with FIGO stage I and the late +6p with stage IV. Similarly, +12 was associated (P < 0.01) with histopathological grade 1 and the late +2 with grade 3. The majority of the late losses, -4, -5, 7p-, 7q-, -8, 10p-, 11p-, -13, -14, -15, -16, -17, -19, and -22, were significantly (P < 0.01) associated with grade 3.
|
. The MDS analysis of the same distance matrix produced one central cluster of late losses, whereas the early losses were positioned at one side of the graph (Fig. 3b)
. The gains were scattered at some distance from the central cluster and at the opposite side of the early losses. The PCA of the imbalances produced one tight cluster of the gains +3q, +7, +8q, +12, and +20, whereas the remaining imbalances formed a more dispersed cluster with no obvious substructures (data not shown). Thus, these analyses suggest two subtypes of tumors, one dominated by gains and one by losses. As no time axis was obvious in the previous PCA, the TO values were included as an observation and the PCA repeated. In the resulting two-dimensional configuration (Fig. 3c)
, the first principal component correlates with the TO values and, thus, represents a time axis. The second component separated all of the gains, except +1q, from the losses. The PCA thus suggests the presence of two cytogenetic pathways; one hyperdiploid initiated by either +7 or +8q, followed by +12 and then by +20 and +3q as late imbalances, and one hypodiploid pathway initiated by 6q-, followed by 1q- or +1q, and then by the remaining losses (Fig. 3c)
. Of the two pathways, the latter seems to lead to more aggressive tumors, as 6q-/1q- tumors were over-represented in FIGO stages III and IV (P = 0.002) as well as in grade 2 and 3 tumors (P = 0.012). The two-factor model explained 67% of the variability, and the communalities ranged from 0.23 to 0.85. However, a separate three-dimensional PCA of the hypodiploid pathway revealed that the 6q-/1q- pathway could be resolved into one initiated by 6q- followed by +1q and -8 as intermediate changes, and another initiated by 1q- followed by 1p- and then by -8 (Fig. 3d)
. These two pathways converged to a set of common late imbalances. The dichotomy of the hypo- and diploid pathway was also evident from the MDS analysis, where 6q- and 1q- were well separated (Fig. 3b)
. The correlation matrix on which the PCA was based revealed significant associations (P < 0.01) between many of the gains, including the early imbalances +7 and +8q but no negative correlations.
|
. Classifying imbalances (9)
were identified by testing each imbalance for its presence in all of the members of a given PCA cluster. Loss of 6q and loss of 1q behaved as classifying imbalances, and identified one major cluster each, whereas the three imbalances +7, +8q, and +12 were needed to determine the remaining major cluster (Fig. 4b)
.
|
. The fact that +2 and +6p clustered with the other gains indicates that these imbalances function as very late imbalances in the +7/+8q/+12 pathway. The early losses 1q- and 6q- were, as in the MDS of the Phase I tumors, placed at some distance from each other, and, thus, the 1q- and the 6q- pathways are still separated in Phase II tumors. The frequency of the individual imbalances in Phase II tumors were calculated (Table 1)
and compared with the corresponding frequencies in Phase I cases. The majority of the imbalances showed a significant (P < 0.01) increase in frequency, as determined by X2 tests, with the exception of +7, +8q, +12, +20, 1q-, and 1p-. The fact that +7, +8q, and 1q- did not increase in frequency, whereas 6q- did, may indicate that the 6q- pathway is more prone to transition into Phase II. This was also supported by the finding that only 25% of the near-diploid +7/+8q/+12, and the 1q- cases, respectively, were Phase II tumors, whereas 41% of the near-diploid 6q- cases belonged to this class. The hyper- and the hypodiploid pathways were also evaluated by comparing the frequencies of the imbalances in the respective pathways. This was accomplished by selecting tumors with the characteristic changes +7, +8q, or +12, and 6q- or 1q-, and then calculate the frequencies of the imbalances. An extensive overlap between the pathways was then observed (Table 2)
. However, the cases were dominated by 6q-/1q- and by 6q-/1q-/+7/+8/+12 tumors; the proportion of the +7/+8q/+12 tumors without concomitant 6q- or 1q- was significantly (P = 0.036) smaller than 6q-/1q- tumors without +7, +8q, or +12.
|
|
. The under-representation of +8q may be because of the frequent -8 in polyploid tumors. The MDS of the polyploid Phase III cases revealed two major clusters of imbalances, one consisting of gains and one of losses (Fig. 5b)
. This shows that the +7/+8q/+12 and the 6q-/1q- pathways are separated to some extent also in Phase III polyploid tumors.
Terminal Deletions.
The most common type of abnormality was deletions, either caused by terminal deletions or unbalanced translocations. To study these losses in more detail, the chromosome arms 1p, 3p, 6q, 7p, and 11p were selected, and the frequency of losses calculated for each cytogenetic band. The frequencies of deletions were then plotted (Fig. 6)
relative to the frequency of the most terminal band in near-diploid Phase I, triploid Phase I, near-diploid Phase II, and triploid Phase III tumors, respectively. In such a graph a random distribution of breakpoints between the chromosome bands would result in a linear increase of chromosomal losses from centromere to telomere. An elevated deletion frequency was observed for the very terminal bands of Phase I tumors. This was particularly evident for 1p, 7p, and 11p in polyploid Phase I tumors. When the tumors progress to Phase II and III the deletion profiles become close to linear, and the frequency by which the pericentromeric bands are lost conforms to a greater extent to the frequencies seen for the terminal bands, suggesting a high frequency of whole-arm losses. The graph for 3p displayed a pattern distinctly different from the other arms, with a high frequency of 3p14/3p21 breaks, and 3p14–21 interstitial losses in Phases II.
|
|
DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
In previous investigations we have used the total tumor population when analyzing the temporal order of imbalances (8 , 14, 15, 16) .7 However, this is only applicable when the distribution of NIPT is monomodal and, hence, is expected to follow one single mode of karyotypic evolution. To evaluate the temporal order among the imbalances in ovarian carcinoma, we therefore limited the analysis to the near-diploid Phase I tumors. We used the modal value of the NIPT distribution of tumors with a given imbalance as a measure for the TO of that imbalance (9) . In this way we have arrived previously at chronological orders of karyotypic events that correlate with histopathological staging (14) . The temporal analysis revealed 1q-, 6q-, +7, and +8q to be early imbalances, -4, -8, +1q, +12, and +20 to be intermediate, and the remaining imbalances to be late. The determined temporal order of imbalances correlated very well with stage and grade where early imbalances were predominantly seen in low-stage and low-grade tumors, whereas late imbalances were seen in high-stage and high-grade tumors. Hence, the deduced temporal order of imbalances most likely reflects the order occurring in vivo.
To screen for possible cytogenetic pathways, we reasoned that imbalances frequently present in the same cases would belong to the same pathway. Thus, the pattern of correlations among imbalances would reveal possible cytogenetic pathways. To condense the large correlation matrix produced by the 30 imbalances, and to extract the central features of the matrix, we performed PCA. We have shown previously the adequacy of this method by comparing it with multiple correspondence analysis (15) , a method similar to PCA but based on X2 statistics. However, to obtain additional information of the structure of the imbalance data, HCA and MDS were performed on a matrix of Euclidean distances between the imbalances. These analyses revealed a striking clustering of gains and of losses, with the exception of +1q that was placed together with the losses. Thus, the ovarian carcinomas form two separate cytogenetic subgroups, one hyperdiploid and one hypodiploid. To obtain a more comprehensive model of the cytogenetic pathways we performed a PCA in which the TO for each imbalance was included as an observation. This organizes the imbalances along the first principal component according to the TO values and, thus, by including the second and third components, a three-dimensional representation of the cytogenetic pathways will be obtained (9) . The first analysis suggested two pathways, one hyperdiploid, initiated by +7, +8q, or +12, and one hypodiploid initiated by 6q- followed by 1q-. However, the separate three-dimensional analysis of the hypodiploid pathway revealed the presence of two pathways, one initiated by 6q-, followed by +1q and -8, and one initiated by 1q-, followed by 1p- and -8. This conclusion was supported by the fact that 6q- and 1q- were placed in different clusters and at some distance in the HCA and MDS analyses, respectively. As the two early imbalances in the hyperdiploid pathway, +7 and +8q, showed positive correlation they do not represent two exclusive starting points. Both of these imbalances were followed by the much more frequent +12, followed by +20 and +3q, and by +6p and +2 in Phase II tumors. The particular importance of 1q-, 6q-, +7, +8q, and +12 in the development of ovarian carcinoma was also evident by the PCA of the tumor cases in which these were shown to be classifying imbalances. The fact that the three latter imbalances were needed to determine one of the major clusters indicates that they may substitute for each other and that they, together, distinguish one subtype of ovarian carcinomas. The +7/+8q/+12 pathway was found to be less aggressive, as the Phase I tumors belonging to this group showed a significantly lower proportion of stage III/IV and grade 2/3 tumors, compared with the Phase I 6q-/1q- tumors. This is in accordance with previous studies that have demonstrated that gain of the former chromosomes, in the absence of more complex changes, is a characteristic of borderline and low-grade ovarian carcinomas (17 , 18) .
The MDS of the imbalances present in Phase II tumors showed clustering of gains and of losses and, thus, the two major pathways seem to be separated to some extent in these tumors as well. In addition, 6q- and 1q- were positioned at some distance emphasizing the separation of the two alternative hypodiploid pathways also at Phase II. The frequencies of the individual imbalances in +7/+8q/+12 Phase I and in 6q-/1q- Phase I cases showed clear and distinct differences between the pathways (Table 2)
, but these differences were less distinct at Phase II. In fact, a high degree of overlap was seen, i.e., 6q- and 1q- were frequently seen together with +7, +8q, or +12. This could indicate that some Phase II tumors develop from Phase I through a recapitulation of Phase I. Another possibility is that Phase II tumors represent an altogether different subgroup of tumors, with a higher rate of acquiring imbalances, and in which both pathways may operate simultaneously, albeit with a slightly higher affinity for either of the hypo- or the hyperdiploid routes in individual cases. This would explain both the apparent separation of gains from losses in the MDS and the less distinct characteristics of the pathways at Phase II. However, a closer inspection of the Phase II tumors revealed that +7/+8q/+12 tumors without the concomitant presence of 6q- or 1q- were underrepresented, whereas 6q-/1q- tumors without +7, +8q, or +12 were not. This implies that +7/+8q/+12 tumors do not progress to Phase II as frequently as 6q-/1q- tumors do. The common simultaneous presence of +7/+8q/+12 and 6q-/1q- then indicates that the acquisition of the latter set of imbalances is a crucial step for progression from low-grade and borderline carcinomas to high-grade ovarian cancer. Hence, Phase II tumors seem to originate either from Phase I 6q-/1q- or from Phase I +7/+8q/+12 tumors having also acquired 6q- and/or 1q-. The present analyses thereby indicate at least two distinct pathways for ovarian epithelial carcinogenesis: one initiated by chromosomal gain in low-grade/borderline lesions, which in rare cases dedifferentiate into high-grade tumors through the acquisition of chromosomal losses, including 6q- and/or 1q-, and another pathway initiated by 6q-/1q- with a higher propensity to progress to Phase II (Fig. 7)
.
|
In Phase I tumors, frequent losses of the most telomeric bands were seen. In Phases II and III, this was replaced by a random pattern of breaks along the chromosome arms, and eventually by the loss of the entire arm. An exception was 3p, where breaks at 3p14/3p21 were the dominating events. A telomeric-centromeric shift of breakpoints has been associated with CIN caused by telomere shortening, eventually leading to a telomere crisis (19) . The telomere dysfunction may, in turn, trigger the evolution of highly complex karyotypes through repeated BFB events. Measurements of terminal restriction fragment lengths have demonstrated that telomeres in the majority of ovarian carcinomas are indeed shorter than in normal ovary surface epithelium (20) and that the shortening is more pronounced at higher FIGO stages (21) . Also, the presence of BFB events has been demonstrated in both primary tumors and established cell lines from ovarian adenocarcinomas (13 , 22) . In this study we complemented the statistical analyses with measurements of anaphase bridging in a few ovarian carcinomas. Evidence of bridging events was found only in complex tumors with losses but not in those with simple karyotypes including +7, +12, or +8q. This indicates that at least two distinct mechanisms for chromosomal evolution are operating in ovarian carcinomas: on the one hand, a thus far unknown mechanism, possibly related to mitotic checkpoint failure (23) , generating a smaller number of changes in a step-wise manner in low and moderate grade tumors, and on the other, BFB events leading to the evolution of multiple chromosomal changes, mostly seen as chromosomal deletions in high-grade tumors. The progression from hyperdiploid Phase I to Phase II indicates that the latter mechanism may sometimes be superimposed on the former during progression toward a more aggressive phenotype. BFB events have been implicated in mitotic failure and polyploidization in head and neck carcinomas (24) . Thus, the absence of BFB instability in hyperdiploid tumors may possibly also explain why these rarely, compared with the hypodiploid Phase I and Phase II tumors, undergo polyploidization. The unique deletion pattern of 3p may, finally, indicate that another, third mechanism of chromosome mutation operates at these stages. The breakpoints in these tumors were concentrated to 3p14/3p21, containing the FRA3B fragile site (25) .
In conclusion, the karyotypic evolution in ovarian carcinomas is characterized by at least two different cytogenetic pathways (Fig. 7)
. One characterized by chromosomal gains, the +7/+8q/+12 pathway, corresponding to low-stage and low-grade tumors, and a second, the 6q-/1q- pathway, characterized by losses and showing moderate levels of stage and grade. At the early stages the karyotypic evolution show similarities to many other tumor types by the seemingly step-wise gain of changes, resulting in the Phase I tumors. The transition to Phase II involved the acquisition of an increased CIN, most probably caused by extensive telomere crisis and the onset of BFB cycles, a process linked to the presence of imbalances characteristic for the 6q-/1q- pathway. Hence, low-grade and borderline +7/+8q/+12 tumors could not progress unless they displayed mixed-pathway characteristics. The onset of triploidization was linked to the presence of the 6q-/1q- pathway, and possibly to the predisposition of such tumors to engage in BFB events. Thus, the key element in the progression of ovarian carcinomas is the 6q-/1q- pathway.
|
FOOTNOTES |
|---|
1 Supported by the Swedish Cancer Society, the Crafoord Foundation, and the Nilsson Family Foundation. 
2 To whom requests for reprints should be addressed, at Department of Clinical Genetics, University Hospital, SE-221 85 Lund, Sweden. Phone: 46-46-173739; Fax: 46-46-131061; E-mail: mattias.hoglund{at}klingen.lu.se
3 Internet address: http://www.nci.nih.gov/statistics/.
4 The abbreviations used are: CN, chromosome number; BFB, breakage-fusion-breakage; NIPT, number of imbalances per tumor; TO, time of occurrence; PCA, principal component analysis; HCA, hierarchical cluster analysis; MDS, multidimensional scaling; FIGO, International Federation of Gynecology and Obstetrics; CIN, chromosomal instability; CGH, comparative genome hybridization.
5 Internet address: http://cgap.nci.nih.gov/Chromosomes/Mitelman.
6 Internet address: http://www.stasoft.com.
7 M. Höglund, C. Jin, D. Gisselsson, G. B. Hansen, F. Mitelman, and F. Mertens. The karyotype evolution of head and neck squamous cell carcinoma (HNSCC) occurs through three distinct phases, submitted for publication.
8 M. Höglund, D. Gisselsson, G. B. Hansen, T. Säll, and F. Mitelman. Dissecting karyotypic patterns in neuroglial tumors: chromosomal evolution through two pathways initiated by +7 with -22 functioning as an independent imbalance, submitted for publication.
Received 11/ 8/02. Accepted 4/15/03.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  K. L. Gorringe, S. Jacobs, E. R. Thompson, A. Sridhar, W. Qiu, D. Y.H. Choong, and I. G. Campbell High-Resolution Single Nucleotide Polymorphism Array Analysis of Epithelial Ovarian Cancer Reveals Numerous Microdeletions and Amplifications Clin. Cancer Res., August 15, 2007; 13(16): 4731 - 4739. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Heidenblad, E. F. P. M. Schoenmakers, T. Jonson, L. Gorunova, J. A. Veltman, A. G. van Kessel, and M. Hoglund Genome-Wide Array-Based Comparative Genomic Hybridization Reveals Multiple Amplification Targets and Novel Homozygous Deletions in Pancreatic Carcinoma Cell Lines Cancer Res., May 1, 2004; 64(9): 3052 - 3059. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
, boundaries of the 25th and 75th 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 1
