Automated detection of residual cells after sex-mismatched stem-cell transplantation – evidence for presence of disease-marker negative residual cells

Background A new chimerism analysis based on automated interphase fluorescence in situ hybridization (FISH) evaluation was established to detect residual cells after allogene sex-mismatched bone marrow or blood stem-cell transplantation. Cells of 58 patients were characterized as disease-associated due to presence of a bcr/abl-gene-fusion or a trisomy 8 and/or a simultaneous hybridization of gonosome-specific centromeric probes. The automatic slide scanning platform Metafer with its module MetaCyte was used to analyse 3,000 cells per sample. Results Overall 454 assays of 58 patients were analyzed. 13 of 58 patients showed residual recipient cells at one stage of more than 4% and 12 of 58 showed residual recipient cells less than 4%, respectively. As to be expected, patients of the latter group were associated with a higher survival rate (48 vs. 34 month). In only two of seven patients with disease-marker positive residual cells between 0.1–1.3% a relapse was observed. Besides, disease-marker negative residual cells were found in two patients without relapse at a rate of 2.8% and 3.3%, respectively. Conclusion The definite origin and meaning of disease-marker negative residual cells is still unclear. Overall, with the presented automatic chimerism analysis of interphase FISH slides, a sensitive method for detection of disease-marker positive residual cells is on hand.


Background
Malignant hematological diseases represent 5.5% of all cancers in Germany [1]. One way to cure these fatal diseases is allogenic bone marrow or blood stem cell transplantation. In case of a male donor and female recipient (and vice versa) we talk of a sex-mismatched transplantation. In such a setting it is relatively simple to classifiy donor and acceptor cells in the bone marrow or blood cell system. The existence of 100 percent donor cells is called complete chimerism, in contrast a mixture of both donor and acceptor cells mixed chimerism. Chimerism analysis is done on these sex-mismatched transplants to monitor minimal residual disease and to plan further immunotherapy like donor lymphocyte infusion (DLI) [2]. In routine diagnostics, fluorescence in situ hybridization (FISH) is frequently applied for chimerism analysis [3][4][5][6], which demands manual time-consuming counting of cells. An experienced technician needs about 2.5 hours for approximately 3,000 cells. Therefore, an automatic FISH chimerism analysis is extremely valuable for diagnostics and correct treatment of affected patients, as it can be carried out in a fraction of time. Thus, the presented single cell based approach becomes now competitive in comparison to PCR based chimerism analysis [7].
Frequently observed disease-markers are the bcr/ablfusion-gene as present in more than 95% of chronic myeloid leukemia (CML) cases [8,9], and trisomy 8 found in 11% of acute myeloid leukemia (AML) [10]. The simultaneous detection of the gonosomal constitution and a tumor marker enables the identification of residual tumor cells. The latter was already proposed 1994 by Nagler and coworkers [11], however, it was not often carried out before [12][13][14], and not studied under routine conditions. Here we tested an automated interphase FISH analysis for the characterization of chimerism in 58 patients after allogenic stem cell transplantation with different hematological malignancies.

Determination of cut off levels
FISH-analysis of residual cells after sex-mismatched transplantation is mainly based on simultaneous labeling of the centromeres of the X-and Y-chromosomes. Because of possible false positive and false negative results e.g. due to background or hybridization problems, it was necessary to determine the cut off level. Therefore, a total of 26,633 cells from 10 healthy female and 35,783 cells from 11 healthy male were analyzed with the described automated system. The automated analysis showed in the female controls 257 cells with apparent male signal constellation (XY), and the male controls had 142 cells with apparent female signal constellation (XX). To control these automated results we investigated all questionable cells; only 38 out of 257 and 27 out of 142 could be confirmed to be real XY-positive cells or in the male case XX-positive cells. This corresponds to a false positive rate of 0.14% in female and 0.08% in male. An additional random control of 4,841 XX cells in female and of 4,535 XY cells in male showed that there was no further failure of automatic counting.
As the cut off level depends on the amount of analyzed cells, all mentioned female and male cells were listed in spreadsheets with random order and arranged in blocks of 50, 100, 200, 400, 800, 1,500, 2,000, 2,500, 3,000 and 4,000 cells. Subsequently, the mean and standard deviation was assessed for each block. The cut off level was defined as the mean plus twice the standard deviation. The respective cut of levels for each block size were fitted by a trend line enabling the calculation of cut off levels for arbitrary cell numbers up to 4000. Fig. 1 shows the determined/calculated cut off values for female and male cells including trend lines.
In order to determine the false positive rate for trisomy 8 another 15,882 cells from 5 healthy people were analyzed with centromere 8 probes. The mean false positive rate was 1.2%. In the same manner 11,453 cells of 11 healthy controls were investigated using the LSI-probe against the bcr/abl-fusion gene. The mean false positive rate for the bcr/abl-probe was 0.7%. For estimating the cut off level for XX/XY in combination with trisomy 8 (XX or XY+trisomy 8) or bcr/abl (XX or XY+bcr/abl) the 95-quantil with the following formula was used: Thus, the cut off level was as follows: • for XX+trisomy 8 and XX+bcr/abl = 0.005% • for XY+trisomy 8 and XY+bcr/abl = 0.003%.

Minimum number of cells to be analyzed in a blood sample
Two statistical aspects of the present study were further assessed. First, the minimum number of cells to be analyzed was determined in order to achieve a predefined accuracy for the estimated fraction of acceptor cells in the total blood of the patient.
Given the total blood of a transplanted patient consists of N cells of which N A are acceptor cells, the fraction of acceptor cells in the patient is P A = N A /N. A blood sample contains fewer than N cells and the fraction of acceptor cells in the sample p A can only be an estimate for the true fraction P A . But how many cells must be analyzed in a blood sample in order to achieve a predefined accuracy?
The probability of finding M A acceptor cells in a blood sample of size M drawn from a total of N blood cells of which N A are acceptor cells is given by a hypergeometric distribution. However, the sample size M is generally much smaller than the total number of cells N. Thus, the true fraction of acceptor cells P A can be assumed to be the same before and after the blood sample has been drawn from the patient. The hypergeometric distribution is then well approximated by the binomial distribution . In the present case the exact value of P A = N A /N is unknown but is to be estimated by the sample ratio p A = M A /M. The mean and standard deviation of the random variable p A are given by , respectively. For determining the minimal number of cells to be analyzed in a blood sample we use the order of magnitude estimate P A = 0.01 according to the measurements. The coefficient of variation (equivalent to the relative standard deviation given in %) is then given by the ratio of standard deviation and mean . For the relative standard deviation to be smaller than q the sample size must exceed M* = 99/q 2 . Hence, for q = {100%, 50%, 25%} the required sample sizes must be larger then M* = {99, 396, 1584}, respectively.

Error bounds for the fraction of acceptor cells due to classification errors of the automated cell recognition device
Second statistical aspect, the error bounds for the fraction of acceptor cells in the blood sample were calculated with respect to classification errors introduced by the automated cell recognition device. The measured number of acceptor cells may not reflect the real number of acceptor cells in the sample due to measurement errors of the automated device. In order to assess the error rate of the device we classified 10 and 11 samples from healthy females and males, respectively, and obtained histograms for the fraction of misclassified cells. These were similar for females and males and thus pooled in a single distribution as shown in Fig. 2. Using this error distribution we calculated the probability density function for the true number of acceptor cells in the sample as described in the following paragraph. The resulting probability density function indicates that the measurements rather overestimate the number of true acceptor states.

Probability density function for the true number of acceptor cells in the sample
In order to assess the error introduced by the measuring device we calculated the probability distribution Histogram for the misclassification error as estimated from the number of wrongly classified cells per sample Sample plots of the conditional probability distribution are shown in Fig. 3 for the empirical prior.

Quantification and characterization of residual cells in patients after sex-mismatched transplantation
Overall 454 samples of 58 patients were investigated with X-and Y-chromosome specific probes as described in detail in Tab In 12 patients cytogenetic disease-markers were detected before transplantation (bcr/abl-fusion (n = 9); trisomy 8 (n = 3)). For these cases a simultaneous hybridization of the centromeres X and Y together with the bcr/abl-or centromere 8-probe was performed. As shown above it is possible to decrease the cut off level for acceptor cells by targeting gonosomes and tumour specific genome alterations in a single hybridization. 55 samples of twelve patients were investigated. In nine of these samples resid- In these remaining groups with small numbers of patients the disease-marker gave no additional information. The amount of disease-marker-positive or negative residual cells showed no correlation with clinical outcome like relapse or death. Table 2 shows the course of 9 patients with residual cells with known disease-marker.

Discussion
Advantage of automatic scanning system "Metafer" Chimerism analysis after sex-mismatched bone marrow or peripheral blood stem-cell transplantation is an important diagnostic component to monitor transplantation and minimal residual disease and DLI [15][16][17]. The FISH technique progresses in importance but demands high personnel skills and costs. An automatic chimerism analysis system could be the solution for that dilemma and was evaluated here. As advantage of automatic analysis using Metafer turned out, that the picture and the coordinates of each cell are memorized. Via basing points it is possible to relocate each cell for further investigation. Moreover, it is possible to analyze huge amounts of cells and to detect small subpopulations of residual cells. Automatic analysis correlates linearly with manual analysis (R 2 = 0.985) [18]. This permits to compare automatic and manual chimerism analysis. Because of the small fraction of the targeted cells in the whole population one should analyze in future studies more than 1600 cells for a reasonable precision (rel. standard deviation of 25%) as described in the statistical part.

FISH vs. PCR
Comparing FISH and PCR in chimerism analysis it was shown that the results are in concordance [7]. The sensitivity of PCR is between 3-5% and allows only semiquantitative analysis [4,19] whereas FISH is more sensitive (1%) and is absolutely quantitative [20]. Because of the different sensitivities it is recommended to use just one method [21], in patients with complete chimerism the method with the highest sensitivity should be used for early detection of residual cells [17,22].

Evaluation of cut off levels and possible source of error
For cut off levels in XY-FISH-analysis it could be shown that they depend on the sum of evaluated cells per sample (e.g. 3,000 counted cells: cut off 1.2% and 0.6% respectively -see Fig. 3. Trakhtenbrot and coworkers [23] described a very alike cut off level in female cells.
In women with sons, male cells could be found 27 [24] or 38 years [25] postpartum in blood samples. The authors showed that up to 40.000 male cells could be transplanted in a normal PBSCT by female donors. In case male cells were found in patients after transplantation this would be incorrectly interpreted as residual cells. Therefore, listing of female donors with sons is recommended in anamnesis and should be considered in chimerism analysis. Unfortunately, this information was not available for the investigated patients.
In the present study in two samples residual cells were arranged in conglomerates (patient 36, in sample 20 and 26 month after transplantation). The histological origin of these cells was not investigated. Potentially these cells resembled endothelial cells that derived from injury of the endothelium. To prevent contamination with endothelial cells samples from the third aspiration of a single venous puncture is recommended for the cytogenetic analysis. The false positive rate of 1.2% for trisomy 8 we determined was in concordance with Jenkins et al. [26] and Cuneo et al. [27]. The bcr/abl-false positive rate of 0.7% was identical with Amiel et al. [28], Van den Berg et al. [29] and Mühlmann et al. [30]. However, overall a big variation can be found in the literature concering false posi-tive rate of bcr/abl which is given between 2-10% [20,[31][32][33][34]. Possible reasons for these differences could be: 1) different tissue samples (bone marrow vs. peripheral blood), 2) different cell cycle stage (G1, G2) or 3) different chromatin structure in healthy and moribund cells [35] and 4) different probes. With 95-quantil the cut off level for simultaneous hybridization of gonosomes and diseasemarkers were estimated and represent 0.005% in XX+trisomy 8/XX+bcr/abl and 0.003% in XY+trisomy 8/XY+bcr/ abl. This allows detecting one disease-marker positive residual cell in 20.000 analyzed cells which was claimed already in 1994 [36]. PCR as alternative diagnostic method does not have this high sensitivity. 33     Bcr/abl negative cell (left) and bcr/abl positive cell (right) Figure 5 Bcr/abl negative cell (left) and bcr/abl positive cell (right). The right cell shows the bcr/abl-gene-fusion (arrowhead). The LSI ES bcr/abl probe of Vysis/Abbott was applied here.

Automatic scanning applied on sex-mismatched patients
residual cells and relapse was 6 month and is identical with the data published by Uzunel et al. [39]. Other studies could not find a correlation between mixed chimerism and relapse [40][41][42]. To what extend mixed chimerism gives evidence about relapse is discussed controversially. In here presented data the occurrence of residual cells was not a marker for relapse. One reason might be the retrospective analysis of patients in this work.
In 12 patients a simultaneous hybridization of gonosomes and disease-markers was applied. 7 patients had disease-marker positive residual cells. But the study showed also that disease-marker positive and diseasemarker negative residual cells can be verified within a sample. The detection of disease-marker positive residual cells had no impact on relapse or survival. In contrast Führer et al. [14] could detect disease-marker positive residual cells before relapse. Thiele et al. [13] also arranged a simultaneous hybridization of gonosomes and disease-markers and assumed that cells carrying the disease-marker represent the source for later relapse.
In 3 samples only disease-marker negative residual cells were found. They might represent 1) healthy (benign) leucocytes, 2) precursor tumor cells which do not yet carry the disease-marker, 3) false negative disease-marker positive cells, 4) endothelial cells or 5) cells from female donors with sons. U X% = under cut off level of X%, in brakets are samples were simultaneous hybrization of gonosomes and disease-marker probes was applied, first percentage = amount of residual cells, second percentage = fraction of disease-marker positive residual cells. This parameter was activated to use the counterstain mask for integration time adjustment. As bright artifacts within the image field would usually interfere with the automatic integration time adjustment, using the counterstain mask enabled correct adjustment for image fields where such artifacts were only present outside the nuclei.

Image Processing
MedianV MedianV An image processing operation was applied to the signal channels to reduce the noise level without significantly reducing the sharpness of the image by vertical median filtering. This filtering was used to remove small "hot spots" of one pixel size in the images which appear in CCD camera images after long integrations or due to camera pixel defects.

Cell Selection
Obj. Threshold 23% An object threshold of 23% in the counterstain channel was used to segment the cell nuclei.

Conclusion
Automated chimerism analysis is a robust and sensitive method which can be used in routine diagnosis to detect residual cells effectively and economically. Simultaneous hybridization of gonosomes and disease-marker represent a sensitive method to detect disease-marker positive residual cells with a very low cut off level. The amount of residual cells correlates with survival. There are patients with residual cells < 4% without tendency of relapse. The detection of disease-marker positive residual cells up to 1.3% does not correlate with relapse. Disease-marker positive and disease-marker negative residual cells can appear at the same time in one sample. The definite origin of disease-marker negative residual cells is unclear and should be investigated in a large multicenter study.

Controls
Peripheral blood samples of 21 clinically healthy male (11) and female (10) between 6 and 67 years were studied as controls.

Patients
A total of 28 female and 30 male patients were analyzed retrospectively after sex-mismatched stem cell transplantation which were performed between 1995 and 2006 at the University Medical Centre Jena. As shown in Table 1, there were 24 acute myloid leukemia (AML), 16 chronic myeloid leukemia (CML), 5 acute lymphatic leukemia (ALL), 4 multiple myeloma (MM), 3 severe aplastic anemia (SAA), 2 myelodysplastic syndrome (MDS), and 1 patient each with Non-Hodgkin lymphoma (NHL), chronic lymphatic leukemia (CLL), lymphoma (Lym) and osteopetrosis (OP). Conditioning regimens were dose reduced in 30 patients or myeloablative in 23 patients [44][45][46] 54 patients underwent peripheral blood stem-cell transplantation (PBCST) and the remaining 4 bone mar-row transplantation (BMT). The median age of the transplanted patients was 46 years (2-89 years). 12 patients showed cytogenetic disease-marker in their malignant cells, i.e. a bcr/abl-fusion in nine and a trisomy 8 in three patients. Overall, 19 patients died, either due to relapse (n = 9), a graft-versus-host-disease (n = 3) or an infection (n = 4). In 3 patients the reason of death remained unclear.

Cytogenetics and molecular cytogenetics including FISH analysis
Standard techniques were used to cultivate leukocytes out of venous blood, prepare chromosome-preparations [43], and to perform interphase FISH analysis [44]. Commercially available probes (Abbott, Wiesbaden, Germany) for LSI-ES bcr/abl, centromere 8, X and Y were applied.
The evaluation procedure of FISH-slides was as followed: 8 slides were automatically scanned over night and the amount of residual cells was registered. Cells which did not have the characteristic signal combination for XX and XY were excluded. All detected potential residual cells were visually controlled by microscope and each valid cell was further examined wether the residual cell carried a disease-marker or not. The system allowed repositioning of all residual cells in order to visually control the group of interest. Most important parameters for the classifier used for analyzing the patient samples.  Table 3 shows the parameters used for automated scanning.