High-Throughput Association Testing on DNA Pools to Identify Genetic Variants that Confer Susceptibility to Acute Myeloid Leukemia

A large number of candidate genes and variants exist that may affect the risk of developing acute leukemia (1, 2). Genotyping the large number of variants on adequate numbers of cases and controls to provide sufficient power to detect statistically significant disease associations is a technical challenge. DNA pooling when applied to large case-control studies offers a novel approach that may be used to preselect disease-associated genetic variants for subsequent genotyping at the case-control level. DNA from many individuals is combined or “pooled,” the pool is then treated as a single sample. Allele-specific PCR (AS PCR) can then be used to differentiate between the two alleles, which when combined with real-time quantitative PCR, can be used to determine the relative frequency of each allele. Disease associations can be estimated by comparing the relative frequency of the alleles between pools constructed from the cases and controls (3).

Several exposures have been implicated in the aetiology of acute myeloid leukemia (AML), including benzene, cigarette smoking, ionizing radiation, and cytotoxic chemotherapy (4). The effects of genetic variants capable of modulating cellular response to these exposures have been examined, and variants in GSTT1 (5), CYP2D6, CYP2C19 (6), NQO1 (7), and GSTP1 (8) may affect risk of AML. However, the genetic component of AML aetiology is likely to be polygenic, and the identification of further candidate genes is essential for future studies. A critical feature of leukemic transformation is the accumulation of DNA damage. Thus, genetic variants in genes that affect the ability of the cell to repair DNA, or engage cell cycle checkpoints and apoptosis are good candidates for modulating leukemia risk. Reduced DNA repair capacity has been shown to be a polymorphic phenotypic trait, with reduced DNA repair capacity associated with increased risk for a number of malignancies (9). Evidence implicating variation in DNA repair capacity on AML risk is derived from the association of an increased risk of leukemia seen in a number of inherited DNA repair disorders, including Fanconi anaemia (10), Bloom's syndrome (11), and Ataxia telangiectasia (12). However, the alleles responsible for these disorders are rare (13–15) and account for only a small proportion of AML cases. The genetic component contributing to the remaining cases is likely to be associated with more common genetic variants. We have applied the technique of DNA pooling to a case-control study of AML, looking at the association of variants in DNA repair and cell cycle/apoptotic regulatory genes, to determine the validity of the pooling approach, and also to determine whether genetic variants could be identified which mediate the risk of AML.

As part of a large case-control study conducted in the North and South-West of England, we extracted DNA from the peripheral blood of 482 Caucasian AML cases and 838 age, sex, and General Practice matched Caucasian controls. Full details of the case-control study can be found elsewhere (16).

Sample DNA concentrations were determined using real-time quantitative PCR against a standard curve constructed from human genomic DNA (Promega, Madison, WI) (22 ng/μl–0.5 μg/μl), using primers specific to chromosome 20, a chromosome rarely affected by cytogenetic aberrations in AML (forward: 5′ CCCACCACCCAGGCTAGCT 3′, reverse: 5′ CGTCGATGTCCGTGCAGAT 3′). The amplicon region was examined to ensure that no known polymorphisms/microsatellites were present that could affect amplification efficiency between samples. Reactions were performed using machine standard cycling conditions on an Applied Biosystems 5700 (Applied Biosystems, Foster City, CA) and were set up as follows: DNA dilution 9 μl, 9 pm of each primer, 10 μl 2× Sybr green master mix (Applied Biosystems). During quantitation, if samples fell out of the range of the DNA standard curve, samples were either diluted or concentrated and re-quantified. All DNA samples were diluted to 1 ng/μl before pooling. Before pool construction, a proportion (approximately 10%) of diluted samples were re-quantified to verify the concentration, all samples checked fell within the 0.95–1.05 ng/μl concentration range.

Two pools were constructed, one combining 444 AML cases (398 de novo AML samples and 46 secondary AML samples as defined by the presence of a previous malignancy or prior diagnosis of a myelodysplastic syndrome(16), 100 ng of each sample), and one containing 828 controls (55 ng of each sample), to give a final volume for each pool of approximately 45 ml.

While the majority of the peripheral blood samples used in this analysis are composed predominantly of normal lymphocytes and granulocytes, a small proportion may have been contaminated with leukaemic blasts. Within the blast cells, there is potential for loss or gain of genetic material that could affect allele frequency estimates. However, none of the genes investigated in this study are situated in regions known to be commonly deleted or duplicated in AML. As such, it is unlikely that the presence of low levels of malignant blast cells will significantly affect the results.

For each variant, two allele specific (AS) primers were designed in the same orientation, the sequence of each being identical except for the terminal 3′ base, which was allele specific. A third common primer in the reverse orientation was also designed to be used with each of the allele-specific primers in separate reactions. AS primers were designed to have a similar melting temperature (T_M), close to but above the annealing temperature used (58°C), while the common primer was chosen to have a T_M significantly higher than the annealing temperature (a minimum of 65°C). Stoffel DNA polymerase was used for amplification of allele-specific products, as this enzyme is highly intolerant of 3′ mismatches and will therefore provide a high degree of differential amplification between perfect and mismatched AS primers. However, the processivity of Stoffel is low compared to most other DNA polymerases. Furthermore, Stoffel is intolerant of A/T-rich template regions. To compensate for these limitations, the amplicon length was kept as short as possible (between 60 and 65 bp), and primers were designed to avoid A/T-rich regions. As a consequence of these strict design guidelines, suitable primers could often only be designed in one orientation around the variant. The ability of the primers to produce a single specific amplicon was verified by using the Applied Biosystems 5700 (Applied Biosystems) disassociation curve function.

Real-time PCR was used to determine the relative amount of each allele in both the case and control pools (3). Frequencies for each allele were independently determined using an AS primer and a reverse primer, the latter common to each AS reaction (for primer sequences, see Appendix A). For each allele, four reactions were performed. Ten nanograms of pooled DNA (approximately 3.75 human diploid genome copies per case sample pooled, and two copies per control sample pooled) were amplified in a total reaction volume of 50 μl, consisting of 0.2 μm of each primer; 6 units of Stoffel gold DNA polymerase (Roche Molecular Systems, Alameda, CA); 1× Stoffel buffer (10 mm Tris-HCl, 10 mm KCl, pH 8.0); 30 mm KCl; 2 mm MgCl₂; 0.2× Sybr green (Molecular Probes, Eugene, OR); 2 μm Rox (Molecular Probes); 5% DMSO; and 2.5 % glycerol. All products were amplified on an Applied Biosystems 7700 using the following amplification conditions; 2 min at 50°C; 12 min at 95°C; followed by 45 cycles of 20 s at 58°C and 20 s at 95°C, with a final extension step of 20 min at 72°C. Samples were repeated if the SD of the C_t value between the replicates was greater than 0.2.

Variants were selected for study on the basis of an established alternative genotyping technique being available. Two variants were selected for pool validation; the GSTP1 I105V (G>A) (17) and the LTα +252 (G>A) (18). An additional nine variants were selected for further study, that is: Cyclin D1, codon 870 splice site variant (A>G) (19); BRCA1 P871L (C>T) (20); ERCC2 K751Q (A>C) (21); FAS −1377 (G>A) (22); hMLH1 −93 promoter (G>A) (23) and V219I (G>A) (24); p21 S31R (C>A) (25); and the XRCC1 R194W (C>T) and R399Q (G>A) (26) variants.

One hundred individual samples were randomly selected from each of the sample sets used to construct the case and control pools, and individually genotyped using the AS PCR protocol detailed above. When genotyping using AS primers, only samples that amplified before 38 cycles were used to assign genotypic status. To be assigned as homozygous for either allele, at least eight cycles had to separate the amplification plots of alleles 1 and 2, while for heterozygote samples, amplification plots were required to be within one cycle of each other. Samples that did not meet these criteria were repeated, and then discarded from analysis if they did not fulfil these criteria (2–7% of the total 200 samples, depending on the variant genotyped, fell into this category). Controls of known genotype were run on each plate. The validity of these genotypes was verified using the referenced RFLP method for the variants in GSTP1 (17), LTα (18), Cyclin D1 (27), p21 S31R (28), hMLH1 −93 (23, 29), ERCC2 (21), XRCC1 R399Q (21), and R194W (30). Allelic discrimination using the Applied Biosystems Taqman approach was carried out for the hMLH1 V219I, BRCA1 P871L, and the FAS −1377 variants (standard Taqman probes were used for the V219I and FAS −1377 variants, while MGB probes were used for the BRCA1 P871L variant; for primer and probe sequences, see Appendix A). RFLP and Taqman assays were quality controlled by sequencing 10 samples selected at random, as follows: 10 picomoles of each primer, 200 μm dNTPs, 1.5 mm MgCl₂, 1× reaction buffer, 0.1 unit Amplitaq gold polymerase (Applied Biosystems), in a total reaction volume of 20 μl. The following amplification conditions were used: 10 min at 95°C; followed by 30 cycles of 1 min at 56°C, 1 min at 72°C, and 1 min at 95°C, with a final extension step of 5 min at 72°C. Products were gel purified using a Qiagen (West Sussex, United Kingdom) gel extraction kit, and sequenced using the ABI BigDye dye primer ready reaction kit (Applied Biosystems). The sequencing reactions being visualized on an ABI 377 sequencer (for sequencing primers, see Appendix A). For samples discarded from analysis after genotyping using AS PCR, no obvious bias was seen in the amplification plots toward samples that appeared to be more heterozygote or homozygote like. Similarly, when comparison was made with a genotyping result from an alternate technique, there was no obvious bias toward samples that genotyped as heterozygote or homozygote for either allele.

Allele frequencies were calculated using C_t values derived from each set of replicate amplifications. The C_t value is the cycle number at which the level of amplification passes a user set threshold where all samples are in the logarithmic phase of amplification, and correlates to the amount of starting template in the reaction. A delay of one cycle in amplification is equivalent to half the amount of template, and the difference in the C_t values for allele 1 and 2 (d), can easily be converted into the frequency (p), of allele 1 in the pool using the equation, p = 1/2^d+ 1 (d = C_t of allele 1 − C_t of allele 2) (3). Amplification efficiencies for the two allele-specific primers often differed slightly, with one AS primer amplifying more efficiently than the other, resulting in non-identical C_t values for heterozygote samples. This bias toward one allele can affect the accuracy of the allele frequency estimates. To attempt to compensate for this, a known heterozygote sample was amplified and the deviation in C_t values for each primer set obtained. By subtracting this value from d, obtained using the DNA pool, this differential amplification can be adjusted to obtain more accurate allele frequencies. The allele frequency in the case pool was compared to that in the control pool and used to calculate an odds ratio (OR) for the lower frequency allele in the control pool. The 95% confidence interval (CI) was calculated by incorporating both the sampling error and measurement error derived from the four replicates.

As with all laboratory techniques, error due to measurement is present. We have been able to taken into account the magnitude of this variation in the calculated CIs. Combining the error due to measurement with that due to sampling has allowed us to maintain statistical power (31). The number of false-positive results increases as the magnitude of measurement error increases; the effect of taking this into account is to widen CIs, thus reducing the probability of false positives. These adjusted CIs are combined with ORs to assess the association between each variant and disease.

The allele frequency in the case pool was compared to that in the control pool and used to calculate an OR for the lower frequency control pool allele. The 95% CI was calculated by incorporating both sampling error and measurement error derived from the four replicates. Attributable/preventable fractions were calculated using the methods from Schlesselman (27).

All samples were individually genotyped for the glutathione S-transferase P1 (GSTP1) A>G polymorphism at codon 105 and the Lymphotoxin α (LTα) G>A polymorphism at +252 using the methods of Warzocha et al. (18) and Welfare et al. (17), respectively. To ensure accurate allele frequency estimation, the allele-specific primers should discriminate between the two alleles. To quality control these primers, 47 samples were individually genotyped for both the GSTP1 and LTα polymorphisms using the AS PCR technique, and the resulting genotypes compared to those obtained using the Warzocha and Welfare techniques. The concordance between the two techniques was excellent with 98% of samples giving identical results for each polymorphism. Allele frequencies were then estimated in the pools using AS PCR for the GSTP1 and LTα polymorphisms, and the C_t values were converted into estimates of allele frequency. Actual and pool estimated allele frequencies for the C allele of GSTP1 were identical for both the case and control pools (0.34). For the G allele of LTα, actual and pool estimated allele frequencies were 0.37 and 0.38, respectively, for the case pool, and 0.37 and 0.36, respectively, for the control pool. Thus, there was an excellent degree of concordance between estimated and observed allele frequencies.

One approach to screening pools is the use of non-validated primers. However, this approach may be associated with an increased risk of identifying false associations. Our chosen definition of primer validation was a minimum of 95% concordance between individual genotyping using AS PCR and another individual-level genotyping technique. To investigate this possibility, we designed AS PCR assays for nine polymorphisms where established alternative genotyping techniques were available, and used these two methods for each polymorphism to genotype 200 randomly selected cases and controls taken from the samples used to construct the pools. Five of the nine AS PCR assays showed at least 95% concordance with the established genotyping technique, as follows: p21 S31R (97% concordance rate), XRCC1 R194W (98%), ERCC2 K751Q (100%), FAS −1377 (99%), and marginally the hMLH1 −93 with a concordance rate of 94.8%. Concordance rates of 67–93% were observed for primer sets that failed this initial validation (XRCC1 R399Q, Cyclin D1 codon 870, BRCA1 P871L, and hMLH1 V219I). Consequently, we designed alternate primers for the hMLH1 V219I and −93 G>A variants and performed primer validation using an alternate DNA polymerase (CEAZ). CEAZ is an alternate form of Stoffel polymerase, which has a greater tolerance for AT-rich sequences, but with a similarly high ability to discriminate between perfect and mismatched AS primers. Unfortunately, the use of this alternate polymerase alone did not improve concordance rates for the previous primer pairs. However, by using CEAZ polymerase in conjunction with alternate primers, we were able to obtain concordance of greater than 98% for the hMLH1 −93 and V291I variants.

We applied the six validated AS PCR assays to the pools for the following variants: ERCC2 K751Q, FAS −1377, XRCC1 R194W, p21 S31R, hMLH1 V291I, and hMLH1 −93. Allele frequencies were estimated using these AS PCR assays on the case and control pools, and compared with those determined using alternative genotyping methods on individual samples. All the samples comprising both the pools were individually genotyped for ERCC2 K751Q, hMLH1 −93, and FAS −1377, while only a subset of cases and controls (100 of each) was genotyped for XRCC1 R194W, p21 S31R, and hMLH1 V291I. All six validated AS PCR assays applied to both pools gave allele frequency estimates that correlated well with the results of individual genotyping (Table 1). Assays using primers that failed validation resulted in inaccurate results (data not shown), supporting the conclusion that only validated assays should be applied for use on pool screening.

The allele frequencies determined using the AS PCR assays using the control DNA pool are in agreement with those previously published for the ERCC2 K751Q (21), FAS −1377 (22), XRCC1 R194W (30), p21 S31R (28), hmlh1 V291I (24), and hMLH1 −93 (23) variants (Table 1). Using the AS PCR technique, the frequency of the FAS −1377 A allele was estimated to be at a significantly higher frequency among the cases compared to the controls, a finding confirmed by individual genotyping using Applied Biosystems Taqman allelic discrimination, suggesting that the FAS −1377 A allele is associated with an increased risk of developing AML (OR 1.76, 95% CI 1.26–2.44) (32) (Table 1). For the ERCC2 K751Q XRCC1 R194W, p21 S31R, hMLH1 V291I, and hMLH1 −93 variants, no significant associations were seen with AML.

The rapid screening of polymorphisms for potential disease association remains a significant issue, hindering the full understanding of the genetic components of polygenic diseases. We have evaluated the use of AS PCR for the estimation of allele frequencies on DNA pools created from a case-control study of AML, and have shown that it represents a practical and efficient tool for the purpose of screening polymorphisms for potential disease associations, provided the appropriate validation criteria are used.

DNA pooling offers a significant reduction in the quantity of DNA used for each assay, an important consideration when samples are limited. For individual genotyping, approximately 10 ng of DNA are required. In contrast, the same amount of DNA when used in a pool of 1000 samples can contribute to 1000 allele estimates. There is also a significant saving in the cost per assay when using DNA pools. For example, the cost of determining allele frequency by individually genotyping 1000 samples is approximately 3 times the cost by pooling. Furthermore, the cost for each variant is independent of pool size, while the cost of genotyping individual samples increases with sample size. Thus, the larger the pool size, the more economical pooling becomes. Despite the loss of individual genotype information, smaller sub-pools may be used to examine haplotypes (33, 34), gene-environment (35), and gene-gene interactions. For such larger studies, the use of multiple pools will increase the number of reactions required; despite this, the economic savings will be still be substantial.

A potential problem associated with DNA pools is a loss of statistical power due to error. The sources of error come from two components, sampling and measurement. While sampling error is reduced with pool size, measurement error (the error inherent in the assay) is constant for each assay. To prevent loss of power due to measurement error, we have used the variation present in the four replicates to better define the CIs around the OR. By increasing the CIs around the point estimates, the likelihood of false-positive associations is decreased, thus maintaining power (31). We suggest that the minimum pool size used should be 50 samples (3); however, in common with several other studies, an increase in pool size is associated with a decrease in sampling error (31, 36). Techniques such as DHPLC do not appear to be affected to the same degree by sampling error (37), as such these techniques may be superior for small pool sizes. However, when using AS PCR for case-control screening, it is not so much the accuracy of the technique, but rather the ability to detect real differences in allele frequency between pools that is important; thus, in the case-control setting, both techniques are equally valid, and the availability of technology should dictate the technique used.

Using the AS PCR approach, we have generated results for case-control comparisons for AML for six variants in five DNA repair or cell cycle/apoptosis regulatory genes, together with comparisons of GSTP1 and LTα. A significant association was suggested for the FAS variant at −1377 that we have subsequently confirmed by individual genotyping, and also identified significant evidence for biological functionality (32). Non-significant results were obtained for the ERCC2 K751Q, hMLH1 −93, XRCC1 R194W, p21 S31R, and the hMLH1 V219I variants, and all these results were confirmed by individual genotyping. The lowest single nucleotide polymorphism (SNP) allele frequency that can be estimated using the AS PCR technique on DNA pools has been reported as 5% (3), and we have confirmed through individual genotyping that variants with an allele frequency as low as 5% can be accurately estimated, that is, the XRCC1 R194W and p21 S31R variants.

On the basis of our experience of using AS PCR for the estimation of allele frequencies in DNA pools, we would suggest the following approach be used. Validation of constructed pools is essential, preferably using multiple assays to remove the chance of random associations resulting in false validation. All assays designed for use on the pools (including those used for validation) cannot simply be designed and applied to the pools, but must be initially validated on a subset of samples with known genotypes, to minimize the generation of spurious results. While DNA pooling does not allow for the determination of individual genotypes, the method does offer a highly efficient screening protocol to determine the distribution of alleles between the case and control pools for any given polymorphism, as long as assay validation is rigorous. Once DNA pools have been generated, it will be possible to screen large numbers of candidate variants, to select variants that may be associated with disease for individual genotyping and further investigation.

AS Primer Sequences

LTα (G>A +252) A allele 5′ GAAGGGAACAGAGAGGAAT 3′; G allele 5′ GAAGGGAACAGAGAGGAAC 3′; Common 5′ CTGTCACACATTCTCTGTTTCTGCC 3′.

GSTP1 (A>G Codon 105) T allele 5′ TTGGTGTAGATGAGGGAGAT 3′; C allele 5′ TTGGTGTAGATGAGGGAGAC 3′; Common 5′ TGGAGGACCTCCGCTGCAA 3′.

hMLH1 (G>A promoter); G allele 5′ GTGCTCACGTTCTTCCTTC 3′; A allele 5′ GTGCTCACGTTCTTCCTTT 3′; Common 5′ ATCAATAGCTGCCGCTGAAGGGT 3′.

hMLH1 (G>A codon 219); G allele 5′ ACTAACAGCATTTCCAAAGAC 3′, A allele 5′ ACTAACAGCATTTCCAAAGAT 3′; Common 5′ AGCAAGGAGAGACAGTAGCTGA 3′.

ERCC2 (A>C codon 751); C allele 5′ GAATCAGAGGAGACGCTGC 3′; A allele 5′ GAATCAGAGGAGACGCTGA 3′; Common 5′ AGGAGTCACCAGGAACCGTTTAT 3′.

XRCC1 (G>A codon 399); G allele 5′ GTGTGAGGCCTTACCTCC 3′; A allele 5′ GTGTGAGGCCTTACCTCT 3′; Common 5′ AGGAGTGGGTGCTGGACTGTCA 3′.

BRCA1 (C>T codon 871); T allele 5′ GCGCCAGTCATTTGCTCT 3′; C allele 5′ GCGCCAGTCATTTGCTCC 3′; Common 5′ AGAGAATGTTGCACATTCCTCTTCT 3′.

XRCC1 (T>C codon 194); C allele 5′ GGGATGTCTTGTTGATCCG 3′; T allele 5′ GGGATGTCTTGTTGATCCA 3′; Common 5′ AGGATGAGAGCGCCAACTCTCT 3′.

p21 (C>A codon 31); C allele 5′ ACAGCGAGCAGCTGAGC 3′; A allele 5′ ACAGCGAGCAGCTGAGA 3′; Common 5′ ATCGCTCACGGGCCTCCTGGA 3′.

FAS (G>A −1377); A allele 5′ ATGAGGAAGACCCTGGGT 3′; G allele 5′ ATGAGGAAGACCCTGGGC 3′; Common 5′ CTATTAGATGCTCAGAGTGTGTGCA 3′.

Cyclin D1 (A>G codon 870); G allele 5′ AGTGATCAAGTGTGACCCG 3′; A allele 5′AGTGATCAAGTGTGACCCA 3′; Common 5′ CTGTAAGCCCCGGCAAGGCTGC 3′.

hMLH1 −93 re-designed primer G allele 5′ ATGGCGTAAGCATCAGCTG 3′; A allele 5′ ATGGCGTAAGCATCAGCTA 3′; used in combination with previously designed reverse primer.

hMLH1 codon 219 re-designed primer A allele 5′ GTGGACAATATTCGCTCCA 3′; G allele 5′ GTGGACAATATTCGCTCCG 3′; Common 5′ TATAGGTTATCGACATACCGACT 3′.

Sequencing Primers

p21 codon 31 sequencing forward 5′ GCCTTCCTTGTATCTCTG 3′; reverse 5′ TGCCTCCTCCCAACTCATCC 3′. XRCC1 194 sequencing forward 5′ AGGTAAGCTGTACCTGTCACTC 3′; reverse 5′ ATTGTTGCCAAAACCCAC 3′. BRCA1 P871L sequencing forward 5′ AACCACAGTCGGGAAACAAG 3′, reverse 5′ ACAGGAAAGCGGCAGTGAT 3′.

hMLH1 219 sequencing forward 5′ CTCAGCCATGAGACAATAAATC 3′ and FAS sequencing forward 5′ CTGTCACTGCACTTACCACC 3′ used in combination with the AS common primer.

For all other assays, the primers used for RFLP were used for sequencing analysis.

Taqman Allelic Discrimination Assays

hMLH1 219 assay, 10 ng of DNA, 900 pm of each primer (forward primer 5′ GGACACTACCCAATGCCTCAA 3′, reverse primer 5′ TAATGTGATGGAATGATAAACCAAGATA 3′), and 100 nm of each probe (G probe Fam labelled 5′ ACAATATTCGCTCCGTCTTTGGAAATGCTG 3′, A probe Vic labelled 5′ ACAATATTCGCTCCATCTTTGGAAATGCTG 3′).

FAS −1377 assay, 10 ng of DNA, 900 pm of each primer (forward primer 5′ GGACACTACCCAATGCCTCAA 3′, reverse primer 5′ TAATGTGATGGAATGATAAACCAAGATA 3′), and 100 nm of each probe (A probe Fam labelled 5′ AAGACCCTGGGTGTGCCAGCCT 3′, G probe Vic labelled 5′ CCCTGGGCGTGCCAGCCT 3′).

BRCA1 P871L assay, 10 ng of DNA, 900 pm each primer (forward primer 5′ GGTTTCAAAGCGCCAGTCAT 3′; reverse primer 5′ CACATTCCTCTTCTGCATTTCCT 3′, and 100 nm of each probe (C allele MGB probe 5′ TGCTCCGTTTTCAAA 3′ Vic labelled, T allele MGB probe 5′ TTGCTCTGTTTTCAAAT 3′ Fam labelled).

All assays were carried out using the Applied Biosystems universal master mix; assays were carried out using an annealing temperature of 60°C, using an Applied Biosystems 9700 and machine standard cycling conditions. Analysis was carried out using the SDS sequence detection software.

We thank the Leukaemia Research Fund for supporting this work. We also thank Jan Parker and Denise Robinson, and all the consultants and staff involved in the Leukaemia Research Fund's Adult Acute Leukaemia study.