Cornelia de Lange syndrome (CdLS; OMIM 122470) is a
dominantly inherited multisystem developmental disorder
characterized by growth and cognitive retardation;
abnormalities of the upper limbs; gastroesophageal dysfunction;
cardiac, ophthalmologic and genitourinary anomalies;
hirsutism; and characteristic facial features1–3. Genital
anomalies, pyloric stenosis, congenital diaphragmatic hernias,
cardiac septal defects, hearing loss and autistic and self-
injurious tendencies also frequently occur2. Prevalence is
estimated to be as high as 1 in 10,000 (ref. 4). We carried out
genome-wide linkage exclusion analysis in 12 families with
CdLS and identified four candidate regions, of which
chromosome 5p13.1 gave the highest multipoint lod score of
2.7. This information, together with the previous identification
of a child with CdLS with a de novo t(5;13)(p13.1;q12.1)
translocation, allowed delineation of a 1.1-Mb critical region
on chromosome 5 for the gene mutated in CdLS. We identified
mutations in one gene in this region, which we named NIPBL,
in four sporadic and two familial cases of CdLS. We
characterized the genomic structure of NIPBL and found that it
is widely expressed in fetal and adult tissues. The fly homolog of
NIPBL, Nipped-B, facilitates enhancer-promoter
communication and regulates Notch signaling and other
developmental pathways in Drosophila melanogaster5.
CdLS is a dominantly inherited disorder with characteristic facial
appearance, limb defects (Fig. 1) and growth and cognitive retardation.
We carried out a genome-wide linkage analysis in nine families with
CdLS with more than one affected family member. Under a model of
genetic homogeneity, we used a linkage exclusion mapping approach,
1Division of Human Genetics and Molecular Biology, The Children’s Hospital of Philadelphia and The University of Pennsylvania School of Medicine, Philadelphia,
Pennsylvania 19104, USA. 2Division of Medical Genetics, Geneva University Hospital, Geneva, Switzerland. 3University of Nevada School of Medicine, Las Vegas,
Nevada, USA. 4University Medical Centre, Hamilton, Ontario, Canada. 5Michigan State University, East Lansing, Michigan, USA. 6Department of Pediatrics and
Department of Human Genetics, University of Utah Health Sciences Center, Utah, USA. 7Department of Developmental and Cell Biology and 8Department of Anatomy
and Neurobiology, University of California, Irvine, California, USA. 9Nemours Children’s Clinic, Wilmington, Delaware, USA. 10Department of Oncology, Biology, and
Genetics, University of Genoa, Italy. 11The Division of Obstetrics and Gynecology, Drexel University School of Medicine, Philadelphia, Pennsylvania, USA.
Correspondence should be addressed to I.D.K. (email@example.com).
Published online 16 May 2004; doi:10.1038/ng1364
Cornelia de Lange syndrome is caused by mutations in
NIPBL, the human homolog of Drosophila melanogaster
Ian D Krantz1, Jennifer McCallum1, Cheryl DeScipio1, Maninder Kaur1, Lynette A Gillis1, Dinah Yaeger1,
Lori Jukofsky1, Nora Wasserman1, Armand Bottani2, Colleen A Morris3, Malgorzata J M Nowaczyk4,
Helga Toriello5, Michael J Bamshad6, John C Carey6, Eric Rappaport1, Shimako Kawauchi7,8, Arthur D Lander7,
Anne L Calof7,8, Hui-hua Li9, Marcella Devoto9,10& Laird G Jackson1,11
NATURE GENETICS VOLUME 36 | NUMBER 6 | JUNE 2004
Figure 1 Characteristic features of CdLS.
(a) Full face view of a child with CdLS showing
characteristic facial features, including arched
eyebrows with synophrys, long eyelashes,
ptosis, depressed nasal bridge with anteverted
nares, long philtrum with thin upper lip and
micrognathia. (b,c) Variability of upper limb
abnormalities in CdLS. (b) Oligodactyly defect
with absence of ulna and most digital
structures (note also hirsutism of back).
(c) Distal reduction defect with missing and
excluding all markers for which the affected individuals in one or more
families did not share both parental alleles (if both parents were unaf-
fected) or the allele transmitted by the affected parent. This analysis
identified five regions containing one or more markers with positive
lod scores in the nine families (chromosomes 2q37, 5p13, 10p13, 14q24
and 17p13; Table 1). We analyzed these five regions in the original nine
families and in three additional families with CdLS (total of 12 families)
and obtained negative lod scores for D17S938 in one family, excluding
chromosome 17. All other markers gave positive lod scores (Table 1).
We carried out fine mapping in all 12 families with additional mark-
ers at an average density of 1–1.5 cM in the defined regions on chro-
mosomes 2, 5, 10 and 14. Multipoint linkage analysis did not improve
the odds for linkage to chromosomes 2, 10 or 14 but resulted in a max-
imum lod score of 2.7 for chromosome 5p13, which was the highest
score for the entire genome analysis. We refined the critical region on
chromosome 5p13 by obligate recombination events to a region of
∼7.4 Mb spanning 5p13.1–13.3 flanked by markers D5S477 distally
and D5S1376 proximally (Supplementary Fig. 1 online) and contain-
ing 58 putative genes (Fig. 2a).
We looked for other corroborating evidence to target one or more of
the four candidate regions. We previously identified a child with clas-
sic features of CdLS and a balanced de novo t(5;13)(p13.1;q12.1)
translocation, and another child with classic features of CdLS and a de
novo chromosome 5p13.1–p14.2 deletion (the only reported case of a
constitutional deletion of 5p13.2) was recently described6. These cases
supported the association of 5p13 with CdLS. We next refined the 5p
breakpoint in the child with the translocation (samples were not avail-
able from the child with the 5p deletion, who died shortly after birth).
We carried out fluorescence in situhybridization (FISH) analysis using
clones in the minimal critical region on 5p13 of the child with the
translocation (Fig. 2b). Owing to sample limitations, we could not ini-
tially identify a clone that spanned the translocation breakpoint, but
VOLUME 36 | NUMBER 6 | JUNE 2004 NATURE GENETICS
Table 1 Results of linkage analysis for markers with highest two-point lod scores
Marker cMI IIIII VIVIIXIII XIVXV XVII XXXXIXXIV families families
Markers were selected from five positive regions after genome-wide scan. The numbers of genes are approximate. Families I–XVII were included in the whole-genome scan.
Families XX, XXI and XXIV were typed only for markers that gave positive lod scores in the whole-genome scan.
Total lod score
Figure 2 Identification of NIPBL as the gene
underlying CdLS. (a) Critical region on
chromosome 5p13, with microsatellite markers
and their distances (in Mb) from the p-terminal
arm of chromosome 5 indicated above the
diagram. Arrows mark the refined critical region
after high-resolution analysis and identification of
obligate recombination events. BAC clones used
for FISH analysis are indicated by their RP11
addresses. FISH analysis further narrowed the
critical region, as shown by arrows below the
diagram. The expanded view identifies genes in
the defined critical region (from the July 2003
build of the University of California Santa Cruz
genome browser). (b) FISH analysis using BACs
from the linkage-defined chromosome 5p13
critical region. The chromosome 5p telomeric
control probe is labeled in green and the 5q
telomeric control probe is labeled in red. In the
left panel, BAC RP11-252F20 is labeled in
green. Both signals from BAC RP11-252F20 are
on chromosome 5 p (arrows), indicating that it is
proximal to the translocation breakpoint. In the
middle panel, BAC RP11-14I21, which contains
NIPBL, is labeled in green. There is signal on
both of the chromosome 5p arms as well as on
chromosome 13q (arrows), indicating that the
probe was split on the translocated chromosome.
In the right panel, BAC RP11-317I23 is labeled
in red. One signal is present on the normal
chromosome 5p arm and the other is present on
13q, indicating that this probe is distal to the
we narrowed the critical region to an interval of 1.1 Mb containing 11
putative genes (Fig. 2a).
We carried out mutational analysis of the first three exons of all 11
genes by conformation-sensitive gel electrophoresis (CSGE)7and
identified mutations in two overlapping transcripts, BX538178(3,653-
bp mRNA) and IDN3(8,124-bp mRNA; Fig. 2a). The identification of
mutations in both transcripts (in BX538178-specific sequence, in
IDN3-specific sequence and in the overlap region) and their exact
sequence identity over a 2,259-bp region of overlap suggested that they
were part of a larger transcript, which we called NIPBL (Nipped-B
like). CSGE analysis of the complete coding sequence of NIPBL in 30
probands (including the 12 familial probands) identified mutations in
2 familial and 4 sporadic cases of CdLS (20% mutation detection rate;
Table 2 and Supplementary Fig. 2 online). In three of the four spo-
radic cases for which samples from both parents were available, the
mutations were de novo. In one sporadic case (3023-3027delTGTCT),
samples were available from only the mother, and she did not carry the
mutation. All four mutations identified in sporadic cases were
frameshift mutations (three deletions and one insertion). The two
familial mutations (a missense mutation in family II in the first codon
(2G→A, causing the amino acid substitution M1K) and a splice site
mutation in family XXI (6763+5G→T) in intron 39) were identified in
all affected siblings and were not present in any of the parents, impli-
cating germline mosaicism as a mechanism in familial recurrences
where neither parent manifests features of the disorder.
All mutations are expected to result in a truncated or, in the case of
the M1K mutation, untranslated protein. The mutations are spread
throughout the gene and were not seen in 300 normal ethnically
matched control chromosomes. We identified seven sequence poly-
morphisms (Supplementary Table 1 online).
We studied expression patterns by northern-blot and in situ analy-
ses. Northern blots of fetal and adult samples for multiple probes
detected transcripts of ∼6 kb and 1.9 kb transcripts and, in fetal sam-
ples, additional bands of ∼9.5 kb and 7.2 kb (Supplementary Fig. 3
online). The presence of multiple transcripts is suggestive of alterna-
tive splicing for this gene. Transcripts of the mouse homolog of NIPBL
were detected widely at gestation days 9.5 and 10.5 (Fig. 3), with
notable accumulations in limb bud, branchial arch and craniofacial
mesenchyme. These regions are involved in patterning of the skeleton
and soft tissues of the limbs, jaw and face (among others).
NATURE GENETICS VOLUME 36 | NUMBER 6 | JUNE 2004
Table 2 Mutations and clinical features in individuals with CdLS
Mutation Exon Effect Clinical features
2G→A (M1K)2 Altered start codonFamily II: mutation identified in all three affected siblings (who each have a different father)
but was not present in their mother or in the two fathers from whom samples were available.
All three siblings (an 8-year-old girl and 17-year-old and 3-year-old boys) have moderate growth
and cognitive delays, small hands without reduction defects, hirsutism and typical facial features.
Male child seen at 4.5 months of age with severe bilateral upper limb
reduction defects (oligodactyly, single digit), severe growth and cognitive delays, typical facial
features, hirsutism and a cleft palate.
Female adult seen at 23 years of age with severe growth and cognitive delays,
reduction defect of the right limb (oligodactyly, four digits) and small left hand with no reduction
defect, typical facial features, hirsutism, cleft palate and hearing loss.
Male child seen at 3.5 years of age with severe growth and cognitive delays, small
hands with no reduction defects, typical facial features, hirsutism, undescended testes and
bilateral sensorineural hearing loss.
Female child seen at 10 years of age with severe growth and cognitive delays
severe bilateral upper limb reduction defects (oligodactyly, single digit), hirsutism and typical
Family XXI: mutation identified in the two affected siblings but neither parent. The two siblings,
a 6-year-old boy and a 5-year-old girl, both have growth and cognitive delays, small hands without
reduction defects, typical facial features and hirsutism.
150delG3Stop codon 28 amino
1546–1547insG10 Stop codon 3 amino
2520delT10 Stop codon 5 amino
3023–3027delTGTCT 10 Stop codon 1 amino
6735+5 G→T 39Splice site disruption
Figure 3 Expression of NIPBL in the developing mouse.
(a–c) Embryonic day (E) 9.5 embryos, whole-mount in situ
hybridization. (d–i) E10.5 embryos, vibratome sections (200 µm) of
embryos processed for whole-mount in situ hybridization. (a,d,g) Sense
control. (b,e,h) Mouse homolog of NIPBL. (c,f,i) Fgf8 (positive control).
(b) The mouse homolog of NIPBL is expressed throughout the embryo,
especially in the limb buds and branchial arches (arrow, fore limb bud;
arrowhead, first branchial arch). (c) Fgf8 expression marks a portion of
the surface ectoderm of the same structures (arrow and arrowhead as in
b). (e,f) At E10.5, sections through the forelimb bud show that
expression of the mouse homolog of NIPBL is concentrated in the
mesenchyme (asterisk in e marks ventral limb bud mesenchyme; dorsal
mesenchyme is also stained), whereas Fgf8 expression marks the apical
ectodermal ridge (arrow in f). No significant differences in intensity
between fore- and hindlimb buds were observed (data not shown).
Coronal sections at the level of the head show expression of the mouse
homolog of NIPBL in the mesenchyme of both the lateral (L) and
medial (M) nasal processes (h). In contrast, Fgf8 hybridization marks
the ectoderm surrounding the developing nasal pit. Scale bars: a, 0.5
mm (for a–c); g, 0.5 mm (for d–i).
We amplified cDNA isolated from lymphoblastoid cell lines, com-
pared it with sequences in the University of California Santa Cruz and
National Center for Biotechnology Information genomic databases and
determined that NIPBL is represented by two overlapping transcripts:
BX538178 (3,653-bp mRNA) and IDN3 (8,124-bp mRNA). We con-
firmed this by northern-blot analyses using probes generated from
sequence-specific regions of these two transcripts. The genomic
sequence spans 188 kb, and the mRNA is 9,505 bp (coding region, bases
127–8,539), encoding a protein of 2,804 amino acids. The mRNA com-
prises 47 exons, with one 5′ noncoding exon. The protein sequence of
human NIPBL shares 92% identity with mouse, 88% with rat and 37%
with the fruit fly Nipped-B gene product (SIM alignment). In a BLAST
search of the National Center for Biotechnology Information database,
NIPBL also had substantial homology with the Saccharomyces cerevisiae
sister chromatid cohesion protein 2, which forms a complex with SCC4
and is required for the association of the cohesin complex with chromo-
somes8. We used the PROSITE program to search for conserved motifs
and found that NIPBL has a bipartite nuclear targeting sequence (amino
acids 1,108–1,124) and a putative HEAT repeat. HEAT repeats (origi-
nally identified in the huntingtin protein) are found in condensins,
cohesins and other complexes with chromosome-related functions9.
Nipped-B is an essential regulator of cut, Ultrabithorax and Notch
receptor signaling. Its protein product belongs to the family of chro-
mosomal adherins, and genetic evidence suggests that it has an archi-
tectural role in facilitating long-distance interactions between
enhancers and promoters5. The involvement of Nipped-B in regulat-
ing Notch signaling is of interest, as two other genes involved in Notch
signaling are implicated in human developmental disorders (muta-
tions in JAG1 result in Alagille syndrome10, and mutations in DLL3
result in spondylocostal dysostosis11).
The identification of mutations in a single allele of NIPBL in indi-
viduals with CdLS is consistent with a dominant pattern of inheri-
tance. All mutations identified so far predict a truncated protein
product and probably result in functional haploinsufficiency. That
haploinsufficiency is a mechanism in CdLS is confirmed by the child
with a large deletion of the region (encompassing NIPBL) and severe
manifestations of CdLS6, and by the child with the translocation
reported here, who also has severe manifestations.
In this report we show that mutations in NIPBLcause CdLS. Because
the paucity of familial cases and consistent cytogenetic rearrangements
did not allow for standard positional cloning approaches, we identified
this gene by combining information on candidate regions not excluded
by linkage analysis with other supporting data (cytogenetic rearrange-
ments). The expression pattern of NIPBL, and the mechanism of action
suggested by its structural homologs, provides insight into the patho-
genesis of the defects seen in the multiple systems involved in CdLS.
Individuals with CdLS. We verified that all affected individuals enrolled in the
study were diagnosed with CdLS. All affected individuals and unaffected family
members were enrolled in the study under a protocol of informed consent
approved by the Institutional Review Board at The Children’s Hospital of
Genome-wide linkage analysis. We carried out linkage studies using the ABI
linkage mapping set version 2, consisting of 400 fluorescently labeled polymor-
phic markers spaced at intervals of ∼10 cM throughout the genome. We esti-
mated marker allele frequencies used in the lod score analysis based on alleles
observed in the families’ founders. We carried out model-based two-point and
multipoint linkage analysis on data from the whole-genome scan and from the
fine mapping of chromosomes 2, 5, 10 and 14 in all families using the GENE-
HUNTER computer program version 2.0 (ref. 12). For lod score analysis, we
assumed the disease to follow an autosomal dominant mode of inheritance
with disease allele frequency of 0.00001. To account for the possibility that the
disease in families with unaffected parents was due to germline mosaicism in
one of the parents, we coded all unaffected individuals (parents and siblings)
from whom samples were available for genotyping as unknown at the disease
phenotype. Thus, we did not have to assume anything about the unknown pen-
etrance of the putative mutation underlying CdLS. We retained marker geno-
type information from unaffected siblings when such information was
available and used it to reconstruct phase for haplotyping. Marker maps used in
multipoint linkage analysis were sex-averaged genetic maps from the Center for
Medical Genetics of the Marshfield Clinic Research Foundation.
FISH. We carried out FISH analysis using standard techniques as described pre-
viously13. We used BAC clones to the critical region on chromosome 5p13.2
(from telomere to centromere: RP11-8C23, RP11-67N10, RP11-317I21
(AC026463.5), RP11-14I21, RP11-7M4, RP11-252F20, RP11-90P7, RP11-
60A21 and RP11-138C1) identified through the University of California Santa
Cruz genome browser as probes to refine the position of the chromosome 5p13
breakpoint of the t(5;13)(p13.1;q12.1) translocation. We obtained the BACs
from Children’s Hospital of Oakland Research Institute. We extracted total BAC
DNA (Perfectprep Plasmid XL, Eppendorf Scientific) and labeled it with spec-
trum orange or green dUTP by nick translation using a commercially available
kit (Vysis). We combined labeled DNA with Cot-1 DNA. We carried out
hybridization and washes using standard conditions.
Mutational analysis and CSGE. We carried out CSGE according to standard
protocols7. Oligonucleotide primer sequences and PCR conditions used for
amplification of all exons of the NIPBL are available on request. We purified
PCR products corresponding to all altered migration patterns (shifts) using
QIAquick PCR purification kit (QIAGEN Sciences) and sequenced them on an
ABI 377 sequencer.
Northern-blot analysis. We hybridized poly(A)+RNA northern blots of multi-
ple adult human tissues (Human 12-Lane Multiple Tissue Northern (MTN)
Blot BD Biosciences Clontech) and human fetal tissues (MessageMap Northern
Blot, Stratagene) with a 301-bp probe from BX538178-specific cDNA sequence
(NIPBL exon 2 and 3), a 344-bp probe from IDN3-specific cDNA sequence
(NIPBL exon 46 and 47) and a 252-bp probe from a region of overlap between
the two putative transcripts (NIPBL exon 10; all primer sequences available on
request). We used the BD SpotLight Random Primer Labeling Kit (BD
Bioscience Clontech) to label probes and SpotLight Chemiluminescent
Hybridization & Detection Kit (BD Bioscience Clontech) for hybridization and
visualization. Experiments were duplicated using Ready-to-go DNA labeling
beads (- dCTP; Amersham) with 32P-dCTP and purified on ProbeQuant G-50
microcolumns (Amersham). We blocked blots with yeast tRNA and herring
sperm DNA. We visualized the signal by exposure to autoradiograph film for
1–5 min (chemiluminescent) and 1–4 h (32P).
In situhybridization. We generated a probe for the mouse homolog of NIPBL
by PCR from an EST clone (oligonucleotide primer sequences available on
request), which yielded a 389-bp product corresponding to the last 190 bp of
exon 10 and all of exon 11 of human NIPBL (Table 2). We subcloned this frag-
ment into pCRII-TOPO (Invitrogen) to generate antisense and sense digoxi-
genin-labeled cRNA probes. We generated an Fgf8 probe (positive control)
from a 422-bp NcoI-PstI fragment of the Fgf8 cDNA (bp 59–481 of GenBank
Z48746) cloned into pBluescript. We dissected CD-1 (Charles River) mouse
embryos at days 9.5 and 10.5 of gestation and fixed and processed them for
whole-mount in situ hybridization, with detection using alkaline phos-
phatase–conjugated sheep antibodies to digoxigenin and 5-bromo-4-chloro-3-
indolyl phosphate/nitroblue tetrazolium as the chromagenic substrate14.
URLs. The University of California Santa Cruz genome browser is available
at http://genome.ucsc.edu/cgi-bin/hgGateway. The National Center for
Biotechnology Information genome database is available at http://www.ncbi.
nlm.nih.gov/. The PROSITE program is available at http://us.expasy.org/
cgi-bin/scanprosite. SIM alignment is available at http://us.expasy.org/cgi-
bin/sim.pl. The Center for Medical Genetics of the Marshfield Research
Foundation is available at http://research.marshfieldclinic.org/genetics/.
VOLUME 36 | NUMBER 6 | JUNE 2004 NATURE GENETICS
LETTERS Download full-text
Accession numbers. GenBank: Human IDN3, NM_133433; mouse IDN3
homolog BG070859 and XM_127929; rat IDN3homolog, XM_238213; NIPBL,
BK005151. GenBank protein: Saccharomyces cerevisiae sister chromatid cohe-
sion protein 2, Q04002.
Note: Supplementary information is available on the Nature Genetics website.
We thank the individuals with CdLS and their families for their support and
willingness to donate samples; the Cornelia de Lange Syndrome Foundation, their
staff and their director J. Mairano for their support; and N. Spinner, M. Jackson,
A. Kline, J. Morrissette, M. Budarf and the staff of the clinical cytogenetics
laboratory and the sequencing core at The Children’s Hospital of Philadelphia for
their comments and guidance. This work was supported by grants from the
National Institutes of Health, National Institute of Child Health and Human
Development (to I.D.K., M.D., A.D.L. and A.L.C.).
COMPETING INTERESTS STATEMENT
The authors declare that they have no competing financial interests.
Received 27 January; accepted 31 March 2004
Published online at http://www.nature.com/naturegenetics/
1. de Lange, C. Sur un type nouveau de dégénératio (typus Amstelodamensis). Arch.
Méd. Enfants 36, 713–719 (1933).
2. Jackson, L., Kline, A.D., Barr, M.A. & Koch, S. de Lange syndrome: a clinical review
of 310 individuals. Am. J. Med. Genet. 47, 940–946 (1993).
3. Ireland, M., Donnai, D. & Burn, J. Brachmann-de Lange syndrome. Delineation of the
clinical phenotype. Am. J. Med. Genet. 47, 959–964 (1993).
4. Opitz, J.M. The Brachmann-de Lange syndrome. Am. J. Med. Genet. 22, 89–102
5. Rollins, R.A., Morcillo, P. & Dorsett, D. Nipped-B, a Drosophila homologue of chro-
mosomal adherins, participates in activation by remote enhancers in the cut and
Ultrabithorax genes. Genetics 152, 577–593 (1999).
6. Hulinsky, R. et al. Prenatal diagnosis dilemma: fetus with del(5)(p13.1p14.2) diag-
nosed postnatally with Cornelia de Lange syndrome. Am. J. Hum. Genet. 73 Suppl.,
7. Ganguly, A., Rock, M.J. & Prockop, D.J. Conformation-sensitive gel electrophoresis
for rapid detection of single-base differences in double-stranded PCR products and
DNA fragments: evidence for solvent-induced bends in DNA heteroduplexes. Proc.
Natl. Acad. Sci. USA 90, 10325–10329 (1993).
8. Ciosk, R. et al. Cohesin’s binding to chromosomes depends on a separate complex
consisting of Scc2 and Scc4 proteins. Mol. Cell 5, 243–254 (2000).
9. Neuwald, A.F. & Hirano, T. HEAT repeats associated with condensins, cohesins, and
other complexes involved in chromosome-related functions. Genome Res. 10,
10. Li, L. et al. Alagille syndrome is caused by mutations in human Jagged1, which
encodes a ligand for Notch1. Nat. Genet. 16, 243–251 (1997).
11. Bulman, M.P. et al. Mutations in the human delta homologue, DLL3, cause axial
skeletal defects in spondylocostal dysostosis. Nat. Genet. 24, 438–441 (2000).
12. Kruglyak, L., Daly, M.J., Reeve-Daly, M.P. & Lander, E.S. Parametric and nonpara-
metric linkage analysis: a unified multipoint approach. Am. J. Hum. Genet. 58,
13. Krantz, I.D. et al. Deletions of 20p12 in Alagille syndrome: frequency and molecular
characterization. Am. J. Med. Genet. 70, 80–86 (1997).
14. Kawauchi, S. et al. Regulation of lens fiber cell differentiation by transcription factor
c-Maf. J. Biol. Chem. 274, 19254–19260 (1999).
NATURE GENETICS VOLUME 36 | NUMBER 6 | JUNE 2004