Minimal residual disease analysis for the prediction of relapse in children with standard-risk acute lymphoblastic leukaemia

Authors

  • Nicholas J. Goulden,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    2. Department of Paediatric Haematology, Royal Hospital for Sick Children, Bristol
    Search for more papers by this author
  • Christopher J. C. Knechtli,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    2. Department of Paediatric Haematology, Royal Hospital for Sick Children, Bristol
    Search for more papers by this author
  • Russell J. Garland,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    Search for more papers by this author
  • Kenneth Langlands,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    Search for more papers by this author
  • Jeremy P. Hancock,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    Search for more papers by this author
  • Michael N. Potter,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    2. Department of Paediatric Haematology, Royal Hospital for Sick Children, Bristol
    Search for more papers by this author
  • Colin G. Steward,

    1. Department of Pathology and Microbiology, School of Medical Sciences, University of Bristol,
    2. Department of Paediatric Haematology, Royal Hospital for Sick Children, Bristol
    Search for more papers by this author
  • Anthony Oakhill

    1. Department of Paediatric Haematology, Royal Hospital for Sick Children, Bristol
    Search for more papers by this author

Dr Colin G. Steward Oncology Day Beds, Royal Hospital for Sick Children, St Michael’s Hill, Bristol BS2 8BJ.

Abstract

We report a largely retrospective analysis of minimal residual disease (MRD) in a cohort of 66 children suffering from acute lymphoblastic leukaemia (ALL). All patients lacked high-risk features at diagnosis, i.e. the presenting white cell count was <50 × 109/l, age 1–16 years and translocations t(9;22) and t(4;11) were not present. All were treated according to either the MRC protocols UKALL X or XI. PCR of IgH, TCRδ and TCRγ gene rearrangements and allele-specific oligoprobing were employed for the detection of MRD. Sensitivity was at least 10−4 in 78/82 (93%) probes examined.

A total of 33 patients relapsed (seven on therapy and 26 off) and 33 remain in continuing complete remission (CCR) (median follow-up 69 months from diagnosis). Of those who remain in CCR, MRD was present in the bone marrow in 32%, 10% and 0% at 1, 3 and 5 months into therapy respectively. This is in marked contrast to the presence of MRD at these times in 82%, 60% and 41% of patients who relapsed (P<0.001, P<0.005 and P<0.005). These results provide further evidence of a strong correlation between clearance of MRD early in therapy and clinical outcome in childhood ALL.

There are two major approaches to identifying patients likely to relapse following conventional chemotherapy for acute lymphoblastic leukaemia (ALL). The first relies on identification of poor-risk features at presentation ( Pui, 1995), the second on disease clearance in the early weeks of therapy as assessed by light microscopy ( Steinherz et al, 1996 ; Lilleyman et al, 1997 ). However, both fail to identify the majority of those destined to relapse. This has led to a widespread interest in the tracking of minimal residual disease (leukaemia present below the 5% threshold of light microscopy), principally by methods based on PCR amplification of immunoglobulin or T-cell receptor gene rearrangements or RT-PCR of specific translocations. Such methods allow disease clearance to be followed to levels up to 5 logs lower than that possible with the light microscope ( Ouspenskaia et al, 1995 ), thereby allowing more sophisticated analysis of an individual patient's response to treatment.

The most widely applicable techniques are those which utilize PCR of immunoglobulin heavy chain (IgH) or T-cell receptor (TCR) subunit gene rearrangements ( Knechtli et al, 1995 ; Roberts et al, 1996 ), since translocations have only been identified so far in approximately 40% of children with ALL. Unfortunately, early studies tended to confuse, due mostly to a combination of small patient numbers with relatively short follow-up and a range of methods with differing limits of detection. There was also excessive concern about the risk of false negative disease detection due to a propensity for some rearrangements to change during the course of disease (termed clonal progression) ( Steward et al, 1994 ; Beishuizen et al, 1993 ; Steenbergen et al, 1993 ). Consequently these techniques have not yet been widely adopted as a clinical tool.

We and others have therefore concentrated on the development of techniques which are highly sensitive, which minimize potential problems with clonal progression phenomena, and which allow the vast majority of patients to be studied. The method presented relies on PCR amplification of three loci, IgH, TCRδ and TCRγ, electroblotting of PCR products and probing with radio-labelled patient-specific probes. It has enabled us to study minimal residual disease (MRD) clearance retrospectively in 66 patients with standard-risk ALL undergoing treatment according to the Medical Research Council (MRC) protocols UKALL X and XI. The results clearly demonstrate a slower average clearance of disease in those who go on to relapse. Furthermore, they are in excellent agreement with other groups both in the U.K. and internationally, further raising hopes that more appropriate individualization of therapy will soon be possible.

PATIENTS, MATERIALS AND METHODS

Samples. A total of 206 remission bone marrow samples obtained from 66 children with ALL were analysed for the presence of MRD. Bone marrow samples were taken at the time points stipulated in the relevant UKALL protocol (1, 3 and 5 months from diagnosis and at the end of treatment). Latterly, local ethical approval was obtained to study bone marrow samples taken 2–3 months following the end of chemotherapy. Samples taken from children treated according to UKALL X were analysed retrospectively, whereas those treated with the UKALL XI protocol were studied within 5 months of finishing therapy.

Patients. The diagnosis of ALL was confirmed by morphological and immunophenotypic analysis ( Bennett et al, 1976 ; Campana et al, 1990 ). Cytogenetic analysis was attempted in all patients but failed in seven. All patients who lacked high-risk features at diagnosis were deemed to have standard-risk disease and were eligible for inclusion in the study. Standard risk was defined by a presenting white cell count < 50 × 109/l, age at diagnosis >1 year and <16 years and absence of the high-risk translocations t(9;22) and t(4;11). Age at diagnosis ranged from 1.03 to 15.96 years and total presenting white cell counts from 1.0 to 46.7 × 109/l. All children who responded to induction therapy entered morphologic remission by day 28. With reference to T-lineage ALL, all but one patient had to be rejected from the study because their peripheral blood counts at presentation were >50 × 109/l. 55 children were treated at the Bristol Royal Hospital for Sick Children, nine at the Hospital for Sick Children, Great Ormond Street, London, and one at Addenbrooke's Hospital, Cambridge. The patients studied presented between 1985 and 1994 and were treated according to MRC UKALL X (33 patients), MRC UKALL XI pilot (11 patients) or MRC UKALL XI (21 patients) protocols. These clinical details are summarized in 1Tables I(a) and I(b). Relapsed patients are over-represented in this study because the samples from the Great Ormond Street Hospital for Sick Children had been studied previously as part of a project looking at the stability of gene rearrangements between presentation and relapse ( Steward et al, 1994 ).

Table 1. Table I. (a) Patients who have relapsed; (b) patients remaining in continuing complete remission. Patient characteristics including immunophenotype, sex, total presenting white cell count, MRC treatment protocol used and age at presentation. Number of bands found by PCR of each locus, type of probe used and probe sensitivity are also given. Cont IT, continuing intrathecal methotrexate; hdMTX, high-dose intravenous methotrexate; 3rd int, third block of intensification treatment; nt = not tested; H = IgH DNJ probe; D = TCRδ probe, G = TCRγ probe.Thumbnail image of

Treatment protocols. Patients in UKALL X received induction with prednisolone, vincristine, asparaginase, daunorubicin and intrathecal methotrexate. They were then randomized to receive either no intensification (arm A), early intensification at week 5 (arm B), late intensification at week 20 (arm C) or two intensification blocks at weeks 5 and 20 (arm D). Intensification consisted of prednisolone, vincristine, daunorubicin, etoposide, cytosine arabinoside and thioguanine. Cranial radiotherapy was given as central nervous system prophylaxis. Maintenance therapy (12 weekly cycles of prednisolone, vincristine, mercaptopurine and oral and intrathecal methotrexate) was continued until week 104 ( Chessells et al, 1995 ). Patients treated with the early UKALL XI pilot protocol received remission induction as for UKALL X. In 1992 the anthracycline was dropped from induction in the subsequent UKALL XI protocol. Initially all patients on UKALL XI received two blocks of intensification therapy starting on weeks 5 and 20 as for UKALL X. Latterly, children were randomized to receive or not an additional third intensification block of treatment at week 35 (dexamethasone, vincristine, asparaginase, cyclophosphamide, cytosine arabinoside and thioguanine). With respect to central nervous system prophylaxis between weeks 9 and 14, there was a randomization between intrathecal methotrexate alone or in combination with high-dose intravenous methotrexate. Continuing therapy as for UKALL X was maintained for 100 weeks.

DNA preparation. Bone marrow mononuclear cells (BM MNCs) were obtained by density gradient centrifugation over LymphoprepTM (Nycomed, U.K.) washed in phosphate-buffered saline and stored at −70°C. DNA was extracted using QIAmp® kits according to the manufacturer's instructions (Qiagen GmbH, Hilden, Germany) or by conventional phenol–chloroform extraction and ethanol precipitation ( Sambrook et al, 1989 ). In 10 cases stored presentation mononuclear cells were not available and DNA was obtained from archival bone marrow aspirate slides as described previously ( Steward et al, 1994 ). Because of concerns over the variable quality of DNA obtained from archival slides, these were not used as a source of DNA for the analysis of remission samples.

Polymerase chain reaction. 100 μl PCR mixes contained 100 ng to 1 μg extracted DNA or 2 μl of material from scraped slides, 200 μm dATP, dCTP, dGTP, dTTP, PCR buffer (10 m m Tris HCl pH 9.0, 50 m m KCl, 1.5 m m MgCl2, 0.1% Triton X-100, 0.01% (w/v) gelatin), 0.25 μm of each primer (FR3A and JPS) and 0.2 units of SuperTaq (HT Biotechnology, Cambridge, U.K.). 3.0 m m MgCl2 was necessary for the Vγ1/9-JγI/II PCR mix. PCR conditions were identical for each set of amplimers and comprised an initial denaturation step at 94°C for 3 min, followed by 35 cycles of denaturation at 94°C for 1 min, annealing at 55°C for 1 min and extension at 72°C for 1 min, with a final extension step at 72°C for 10 min. In order to avoid the risk of contamination ( Kwok & Higuchi, 1989), all pre-PCR procedures (preparation of stock solutions, DNA isolation and setting up of PCR reactions) were performed in a separate room from the PCR reaction itself and all post-PCR procedures. PCR reactions were prepared in a laminar flow hood using aerosol-resistant, disposable, filter pipette tips. Non-DNA-containing reactions were amplified in parallel as negative controls.

Primer sequences

FR3A5′-ACACGGC(C/T)(G/C)TGTATTACTGT-3′

JPS5′-GTGACCAGGGT(C/T)CC(C/T)TGGCCCCAG-3′

LJH5′-ACCTGAGGAGACGGTGACC-3′

Vδ25′-CTTGCACCATCAGAGAGAGA-3′

Dδ3 screen5′-GTTTTTGTACAGGTCTCTGT-3′

Dδ3 intron5′-AGGGAAATGGCACTTTTGCC-3′

Vγ15′-TG(A/C)(C/T)TCTGG(A/G)GTCTATTACTGT-3′

Vγ95′-AAACAGGACATAGCTACCTACT-3′

JγI/II5′-CGATACTTACCTGTGACAAC(C/A)AG-3′

Vδ2 lead5′-GTCATGTCAGCCATTGAGTT-3′

anti-Vδ25′-TCTCTCTCTGATGGTGCAAG-3′

Characterization of clone-specific rearrangements. Presentation DNA from all 65 children with B-cell precursor (BCP) ALL was screened by IgH FR3-JH and TCR Vδ2-Dδ3 PCR. In 38 cases presentation DNA was also subjected to TCR Vγ1/9-JγI/II amplification. Presentation DNA from the single case of T-ALL was analysed by Vδ1-Jδ1 and Vγ1/9-JγI/II PCR. 30 μl of the resultant PCR product was then resolved by 8% non-denaturing polyacrylamide gel electrophoresis (PAGE). Following ethidium bromide staining, clonal bands were crush eluted and sequenced directly as described previously ( Langlands et al, 1995 ). In a minority of cases multiple bands of similar size were seen and a TA cloning kit (Invitrogen, San Diego, Calif., U.S.A.) was employed according to the manufacturer's instructions. Several colonies were picked until at least three yielded the same sequence which was then assumed to represent the sequence of the leukaemic clone of interest. The V, (D) and J segment usage of the leukaemic clone was identified and 20 base ‘allele-specific’ probes complementary to the junctional sequence were synthesized (Oswel, Southampton). In order to minimize both the number of probes required and the impact of clonal evolution, the IgH probes were designed to each unique DNJ junction.

Investigation of remission specimens. 1 μg (equivalent to 150 000 genomes) of BM MNC DNA from each remission marrow aspirate was amplified with the appropriate primer pair. PCR conditions applied were the same as those used for the screening PCR (above) except that a 3′ anti-sense primer was used in the case of FR3 (LJH instead of JPS) and Vδ2-Dδ3 (Dδ3 intron instead of Dδ3 screen) as a further manoeuvre to avoid contamination. The quality of DNA extracted from the remission specimens was tested with a ‘housekeeping’ PCR using the Vδ2 lead anti-Vδ2 primer pair noted above. A non-DNA-containing negative control and two samples each containing 1 μg of normal BM MNC DNA as well as 10-fold dilutions of leukaemia cell DNA in normal BM MNC DNA were amplified on all occasions. A 30 μl aliquot of each PCR product was resolved by 8% PAGE, visualized with ethidium bromide and transferred to a nylon membrane (HybondTM-N+, Amersham International, Bucks.) by electroblotting at 350–400 mA for 30 min (Millipore Blotter, WEP Ltd). The products were then fixed in 0.4 m NaOH for 1 min followed by 2× SSC for 30 s. Subsequently, the membrane was probed with the relevant γ32P-dATP end-labelled oligonucleotide followed by autoradiography as previously described ( Goulden et al, 1994 ; Langlands et al, 1994 ). Exposure of the autoradiograph was continued until low-level non-specific hybridization of the probe to the normal control was just visible. The presence of MRD was defined as stronger hybridization of the probe to a rearrangement of the correct size in the patient sample when compared to that seen with the normal bone marrow. A semi-quantitative estimate of the amount of leukaemia present in each remission sample and an assessment of the sensitivity of the allele-specific probe was provided by comparison of the hybridization between the sample in question and that of the logarithmic dilutions of leukaemia cell DNA.

Statistical analysis. The χ2 test with Yates' continuity correction was used to assess the statistical significance of any difference in proportions of MRD positivity between the relapse and remission groups at each time point. Fisher's exact test was substituted for the χ2 test in cases where low expected frequencies (<5) were observed. Where possible, odds ratios with 95% confidence intervals were also calculated.

RESULTS

Patients

Thirty-three patients have relapsed with a median time to relapse of 33 months (range 8–74 months). Five (UPN 25, 28, 29, 30, 31) relapsed after completion of MRD analysis, but the study was otherwise unblinded. 33 remain in continuing complete remission with a median follow-up of 69 months from diagnosis (range 44–125 months) as of 1 August 1997.

Clone-specific rearrangements

Patients were selected for inclusion in the study on the basis of the presence of at least one clonal IgH or TCR rearrangement at diagnosis. An assessment of the applicability of this multilocus screening approach can be gained from consideration of a larger unselected cohort of 102 BCP ALL patients screened in our laboratory by each of the FR3-JH, Vδ2-Dδ3 and Vγ1/9-JγI/II systems. 93 of these produced at least one clonal band, of which 35 were positive at two loci and a further 19 had rearrangements at all three loci. Application of the Vδ1-Jδ1 and Vγ1/9-JγI/II systems generated at least one clonal band in 18/21 patients with T-ALL screened.

Sensitivity of clone-specific probes

A total of 84 oligonucleotide (61 IgH, 17 Vδ2-Dδ3 and six Vγ1/9-JγI/II) probes were used in the study. In 78 (93%) cases the sensitivity of these was estimated to be equivalent to the detection of one leukaemic cell in at least 10 000 normal cells.

Stability of rearrangements between presentation and relapse

The presence of the diagnostic rearrangement or a related secondary rearrangement was not documented in bone marrow samples taken at relapse in two of the 33 cases. In one case the relapsed disease had undergone complete clonal change with respect to all rearrangements documented at diagnosis. This child (UPN 33) with an initial total white cell count of 3.4 × 109/l relapsed with a total white cell count of 206 × 109/l 6 months after the end of therapy. Neither the diagnostic or relapse rearrangements were detected at day 28 of induction treatment. In the other case (UPN 11), MRD was not present in the marrow at the time of isolated relapse in the central nervous system. Cerebrospinal fluid was not available for analysis and therefore complete clonal change cannot be ruled out.

Patterns of minimal residual disease

These results are illustrated in detail in Fig 1 and summarized in Table II. The positive and negative predictive values of MRD assessment at each time point are depicted in Fig 2.

Figure 1.

2 = 2–3 months after the end of therapy; ○, MRD negativity; ●, high-level MRD evident after PAGE only; ●, low-level MRD detected after PAGE and allele-specific oligoprobing. In patients who relapsed (a), the occurrence of overt haematological relapse is depicted by ▾ followed by the site of relapse (BM, bone marrow; TEST, testicular; CNS, central nervous system) and the time to relapse in months in parentheses. In those remaining in remission (b), the duration of follow-up is given in months on the right.

Table 2. Table II. Proportion of patients MRD positive at each time point divided by outcome. Statistical significance values, the method used to calculate them (C = χ2 test, F = Fisher's exact test), odds ratios (OR) and 95% confidence intervals for the odds rations (OR 95% CI) are given. EOT = end of therapy; EOT + 2–3/12 = 2–3 months after the end of therapy. Thumbnail image of
Figure 2.

–3 months after the end of therapy.

DISCUSSION

The results presented constitute the largest MRD study undertaken on a group of patients treated according to Medical Research Council UKALL protocols. In patients lacking established high-risk factors they demonstrate a clear relationship between the interval to achieving MRD negativity and risk of relapse. Most notably, all seven patients with persisting detectable disease at 5 months were known to have relapsed. In contrast, none of the 20 patients from the remission cohort tested at this time had evidence of MRD.

There are several potential criticisms of this study. Patients were largely studied retrospectively, those who relapsed are over-represented (33/66 patients studied), specimens were not available at all desired time points for each patient and the workers were not blinded to outcome in most cases. Furthermore, the patients studied come from all treatment arms of both protocols (UKALL X and XI) and even within the UKALL XI group some received daunorubicin on days 1 and 2 of induction whereas others did not. These concerns, however, should not be allowed to obscure the fact that very similar results have been obtained from studies of patients on a spectrum of protocols from the U.K., Australia, America and Europe. For instance, we detected MRD at the end of induction therapy (day 28) in 56% of patients in long-term CCR compared with 50% or 66% (by FR3- and FR1-PCR respectively) of a comparable cohort examined by workers in Leeds ( Evans et al, 1997 ). These results are akin to those from Adelaide [36% of patients at day 42 on Australian and New Zealand Children's Cancer Study Group IV and V ( Brisco et al, 1994 )], Amsterdam [35% of those studied from Dutch Childhood Leukaemia Study Groups V–VIII ( Steenbergen et al, 1995 )] and from Ulm 26/60 (42%) in the prospective BFM trial ( Seriu et al, 1997 ).

At the end of induction therapy we found a highly significant difference in the proportion of patients MRD positive between the relapse group and those remaining in CCR (82% v 32%, P = 0.0003). This divergence has been described previously and has led to great interest in accurate quantitation of disease at this time in an attempt to improve yet further the discrimination of high-risk patients ( Brisco et al, 1994 ; Cavéet al, 1994 ). For example, those with persistent clonal bands (indicative of MRD at or above 1 in 103) or quantified MRD in excess of 1 in 103 at this time have been shown to have a high propensity to relapse. However, we have several anxieties regarding such an approach. Firstly, it places heavy reliance on accurate quantitation of MRD in a single marrow sample and assumes that disease is evenly distributed throughout the bone marrow. This is in direct contradiction to animal studies which suggest a patchy distribution of MRD ( Martens et al, 1987 ). Secondly, accurate quantitation requires either limiting dilution experiments ( Brisco et al, 1994 ; Ouspenskaia et al, 1995 ) or co-amplification of a competitor molecule ( Cavéet al, 1994 ). Such approaches increase complexity, significantly complicating the conduct of large-scale clinical studies. Finally, it fails to take into account the impact of post-induction consolidation therapy, which may explain why in some series a proportion of the patients with persisting high-level MRD level at the end of induction therapy remained in sustained remission ( Steenbergen et al, 1995 ). Patients UPN 103, 106 and 121 would be typical examples of the latter.

This has led us to scrutinize later time points during therapy. By so doing it is clear (Fig 2) that the specificity of relapse prediction increases with time on treatment, but only at the expense of sensitivity. Therefore, at the 5-month time point, all seven patients MRD positive by our technique relapsed, but we failed to identify 59% of those who went on to relapse. If MRD detection is to be used as a method of selecting patients for highly intensive or novel therapies, or even for disintensification of therapy, how may this situation be improved?

It could be argued that a more sensitive method of detection would pick up more relapse candidates at 5 months. The current method has a median sensitivity of detection of one leukaemic cell in 104 normals (0.01%) and cannot detect MRD below one in 105 since 1 μg of DNA contains only 2 × 105 genomes. In contrast, the method employed by Ouspenskaia et al (1995 ) has a median sensitivity up to 1.5 logs greater due to a combination of enrichment procedures and nested PCR. However, this technique detects MRD even after completion of therapy in most patients, suggesting that complete eradication of leukaemia may not be necessary for long-term remission ( Roberts et al, 1997 ). It is therefore possible that the use of a more sensitive test at 5 months would improve the detection of patients destined to relapse, but only at the expense of including candidates for CCR who might be given unnecessary additional therapy as a result. This hypothesis remains to be tested.

It must also be remembered that false negative prediction is unavoidable to some degree. Although MRD is almost always evident in the marrow at the time of isolated extramedullary relapse ( Goulden et al, 1994 ; Neale et al, 1994 ; O'Reilly et al, 1995 ), we and others have observed consistently poor prediction of such relapses by tracking MRD in bone marrow before they occur. Further, our previous large study of marker stability suggested that, even probing multiple loci and employing DNJ probes for IgH rearrangements, complete clonal change would cause false negative disease detection in approximately 8% of cases ( Steward et al, 1994 ). This was borne out in the present study where only two of 33 patients (6%) who relapsed did so with rearrangements totally unrelated to those seen at diagnosis.

We had previously suggested that MRD analysis at the end of therapy would allow detection of a considerable proportion of off-treatment relapses ( Potter et al, 1993 ). However, although it is true that all of the four patients positive at this time went on to relapse, we also failed to detect MRD in 16/20 patients who relapsed off therapy. In this study the presence of MRD at 2–3 months off therapy enabled us to anticipate relapse in three patients who had been MRD negative at the end of treatment (with no false positives amongst those who remain in CCR), but two relapsed within 1 and 2 months of this sample and the clinical value of such information is debatable.

In summary, we favour this method for MRD detection because it combines wide applicability (> 90% of patients) with high sensitivity (disease detectable to at least 1 in 104 in >90% of patients studied). The combination of size and sequence specificity results in a robust test which could be used for clinical decision-making with a high degree of confidence. Furthermore, as a consequence of its inherent threshold of sensitivity, this method is able to distinguish — at either 3 or 5 months into treatment — patients who relapse from those who remain in remission with very low false positive prediction rates.

The results of this and similar studies now pose a new series of questions. Does MRD decay have independent prognostic significance over light microscopic assessment of blast clearance, a widely applicable, rapid and inexpensive test? Will these studies prove robust when workers are completely blinded to the clinical characteristics and outcome of the patient? Will a relatively simple technique, such as automated fluorescent analysis of amplified rearrangements ( Owen et al, 1997 ), withstand direct comparison with more complex, sequence-based methods? These important questions will now be addressed in a series of studies on patients treated according to UKALL protocols organized by the U.K. Collaborative Group on MRD.

Acknowledgements

We thank all colleagues involved in sample collection and patient care at the Royal Hospital for Sick Children, Bristol, in particular Dr H. Kershaw and the nursing staff of Oncology Day Beds. We also thank Professor J. M. Chessells and Dr F. Katz (The Hospital for Sick Children, Great Ormond Street, London) and Dr D. Williams (Addenbrooke's Hospital, Cambridge) for their help with providing bone marrow material and clinical information on some of the patients in the study. We are grateful to Dr L. Hunt (Institute of Child Health, Bristol) for her advice on the statistical analysis of the data, to the Leukaemia Research Fund and the Ben Drewer Trust for their financial assistance and the COGENT Trust for providing laboratory facilities. We are also indebted to P.G. for his continued support.

Ancillary