Hematopoiesis

ATAC-seq data were processed as previously described^{23 (link)} with notable exceptions. In brief, reads were trimmed using a custom script and aligned using Bowtie2. To call peaks, data were aggregated by each unique cell type, peak summits were called using MACS2, and filtered using a custom blacklist, as previously described^{23 (link)}.
To generate a non-redundant list of hematopoiesis and cancer peaks we first extended summits to 500 bp windows (+/− 250 bps). We then ranked 500 bp peaks by their summit significance value (defined by MACS2) and chose a list of non-overlapping, maximally significant, peaks. The complete data set comprised a total of 590,650 peaks. To annotate peaks with promoter/distal labels, and nearest gene, we used the Homer package, with the command “annotatePeaks.pl”. As described previously^{23 (link)}, we counted fragments for each sample across all 590,650 peaks to provide a count matrix. To obtain normalized fragment counts, which were used for all downstream processing, we first performed quantile normalization followed by GC normalization (CQN R package^{47 (link)}). Data tracks, used solely for visualization, were normalized to the number of fragments falling within all peaks for each sample. Coverage tracks were visualized using the Gviz R-package. Fragment yield (Supplementary Fig. 1e), was computed by multiplying the library diversity calculated using PICARD tools with the number of reads falling within peaks, values were then divided by the number of cells used in each assay.
For information on TF-based analyses, see Supplementary Note 1.

The NCI IBMFS cohort is an open retrospective/prospective cohort, established in January 2002, with approval from the NCI Institutional Review Board. Data reported here include individuals enrolled prior to December, 2007, with follow-up through to December, 2008. The protocol, NCI 02-C-0052 [NCT00056121] (http://www.marrowfailure.cancer.gov), was advertised by mailing to paediatric haematologists/oncologists, medical geneticists, and IBMFS family support groups. Voluntary enrollment by the family contact (usually a parent or proband; a proxy was used for deceased patients) began with a telephone interview. Discussion at a team meeting determined whether the proband met the criteria for the suspected syndrome or needed further testing. A Family History Questionnaire provided medical information about relatives. Written informed consent and medical record release forms were signed. Individual Information Questionnaires (medical history, cancer risk factors, etc.) were sent to the proband (or proxy) and first-degree relatives. Biannual follow-up was obtained on all participants. Cancer diagnoses were confirmed by medical record review. All participants were enrolled in the ‘Field Cohort.’ Those who visited the National Institutes of Health (NIH) Warren G. Magnuson Clinical Center were reassigned to the ‘Clinic Cohort.’ Families in the Clinic Cohort visited the NIH for 5 d, for thorough review of medical histories and physical examinations by haematologists and multiple subspecialists, as well as aetiologically-focused laboratory tests.
Participants were assigned to a specific syndrome according to standard criteria and confirmed by syndrome-specific tests where available (Alter, 2003 ). FA was diagnosed by abnormal chromosome breakage in peripheral blood lymphocytes, using both diepoxybutane and mitomycin C (Cervenka et al, 1981 (link); Auerbach et al, 1989 ). Skin fibroblasts were analysed when lymphocytes were normal but FA remained highly suspect (seeking evidence for haematopoietic mosaicism) (Alter et al, 2005 (link)). FA complementation group analyses were performed using retroviral correction (Chandra et al, 2005 (link)).
The clinical diagnosis of DC was made in individuals with components of the diagnostic triad (nail dystrophy, reticular pigmentation, and oral leucoplakia), or those with at least one other typical physical finding (Vulliamy et al, 2006 (link)), in association with marrow failure. We expanded the inclusion criteria to patients with marrow failure, any of the above physical parameters, and blood leucocyte subset telomere lengths below the first percentile of normal-for-age (Alter et al, 2007a (link)). We also classified as ‘DC’ probands and healthy family members who had pathogenic mutations in known DC genes, such as DKC1, TERC, TERT, and TINF2, including those with none of the typical physical findings (Savage & Alter, 2009 (link)).
The diagnosis of DBA was made in those with macrocytic pure red cell aplasia, and supported by finding increased red cell adenosine deaminase (Glader & Backer, 1988 (link)). Patients with SDS had neutropenia and exocrine pancreatic insufficiency, confirmed by detection of sub-normal levels of serum pancreatic trypsinogen and isoamylase (Ip et al, 2002 (link)).
All living affected individuals were specifically screened for all of the major IBMFS; genotyping was performed when possible (Ameziane et al, 2008 (link); Moghrabi et al, 2009 (link)). Affected individuals who had not received a transplant had bone marrow aspirations, biopsies and cytogenetic studies. Individuals who could not be classified as having a specific IBMFS were designated as ‘Others.’ Categories of ‘DC-like,’ ‘FA-like,’ and ‘SDS-like’ were used for individuals whose features initially suggested DC, FA, or SDS but who failed to meet diagnostic criteria. Severe bone marrow failure was defined as impaired haematopoiesis sufficiently severe to lead to bone marrow transplant (BMT) or death (Rosenberg et al, 2003 (link)); MDS required severe pancytopenia and dyspoietic morphology, with or without a cytogenetic clone (Alter et al, 2000 (link)).
Analyses used Microsoft Excel 11.0 (Microsoft, Redmond, WA, USA), Stata 10.1 (StataCorp, College Station, TX, USA), and MATLAB 2008b software (The MathWorks, Natick, MA, USA). The Kaplan-Meier product limit estimator was used to calculate actuarial survival probabilities by age and cumulative incidences in the absence of competing risks; subjects were censored at death (Kaplan & Meier, 1958 ). Subgroup survivals were compared using the log-rank test for equality of survivor functions. Cause-specific hazards and cumulative incidence curves accounting for competing risks were calculated as described previously (Rosenberg et al, 2003 (link)). The observed number of cancers was compared with the expected number (O/E ratio), based on general population incidence data from the Surveillance, Epidemiology, and End Results (SEER) Program, adjusting for age, sex, race, and birth cohort (Ries et al, 2008 ). Sex ratios were examined using the binomial test of comparison with a male:female ratio of 1:1. Statistical tests were 2-sided, and P-values ≤0·05 were considered significant.