ICF, An Immunodeficiency Syndrome: DNA Methyltransferase 3B Involvement, Chromosome Anomalies, and Gene Dysregulation

The immunodeficiency, centromeric region instability, and facial anomalies syndrome (ICF) is the only disease known to result from a mutated DNA methyltransferase gene, namely, DNMT3B. Characteristic of this recessive disease are decreases in serum immunoglobulins despite the presence of B cells and, in the juxtacentromeric heterochromatin of chromosomes 1 and 16, chromatin decondensation, distinctive rearrangements, and satellite DNA hypomethylation. Although DNMT3B is involved in specific associations with histone deacetylases, HP1, other DNMTs, chromatin remodelling proteins, condensin, and other nuclear proteins, it is probably the partial loss of catalytic activity that is responsible for the disease. In microarray experiments and real-time RT-PCR assays, we observed significant differences in RNA levels from ICF vs. control lymphoblasts for pro- and anti-apoptotic genes (BCL2L10, CASP1, and PTPN13); nitrous oxide, carbon monoxide, NF-κB, and TNFa signalling pathway genes (PRKCH, GUCY1A3, GUCY1B3, MAPK13; HMOX1, and MAP4K4); and transcription control genes (NR2F2 and SMARCA2). This gene dysregulation could contribute to the immunodeficiency and other symptoms of ICF and might result from the limited losses of DNA methylation although ICF-related promoter hypomethylation was not observed for six of the above examined genes. We propose that hypomethylation of satellite 2at1qh and 16qh might provoke this dysregulation gene expression by trans effects from altered sequestration of transcription factors, changes in nuclear architecture, or expression of noncoding RNAs.

In this review,w ew ill briefly describe the ICF phenotype, then atureo fk nown ICF-associated mutations in DNMT3B ,a nd why it is likelyt hatI CF is actually due to loss of the enzymatic activity of DNMT3B andn ot to alterationsi ni ts specific protein -protein interactions. The relationships of DNA hypomethylation andc hromosome abnormalities in the ICF syndrome and in cancer will also be discussed. Lastly, the questiono fa bnormal gene expression in ICFl ymphoblastoid cells will be addressed in some detail with previously unreported data from an expression microarray analysis anda n inferred model of how ICF-related DNA hypomethylation leads to the disease.
DNMT3Bm utations in ICF patients(ICF type 1) ICF type 1i st he only formo fI CF whose genetic etiology is known.I ti nvolves biallelic DNMT3B mutations [10]. Unless otherwise noted, ICF will denote type 1inthis review.The ICF-linked DNMT3B mutations are often missense mutations and are usually found in the partofthe gene encoding the catalytically active C-terminal portion of the protein, namely,one of ten motifs conserved among all cytosine-C5 methyltransferases [1,2,4,5,10,15,28].
The involvement of DNAh ypomethylation in the phenotype of ICF is supported at the cytogenetic level. ICF-specific rearrangements in mitogen-treated lymphocytes from patientsa re the samei nf requency, spectrum and chromosomal specificity as those that we found in an ormalp ro-B lymphoblastoid cell line treatedw itht he DNA methylationi nhibitors5azacytidine or 5-azadeoxycytidine [29,30]. The invariant hypomethylation of certain portions of the genome in ICF cells and tissues, most notably Sat2 [1,13], is also consistent with ICF being due to DNA methylation deficiency.
The only ICF-associated missense mutation outsidet he catalyticC -terminalh alfo fD NMT3B (S282P)i sw ithin the PWWPd omaina nd was found in both DNMT3Ba lleles in two related ICF patients [28]. The analogous mutation in mouse cells resulted in thel osso fd etectablet argetingt o constitutive heterochromatin in interphase and metaphase [38]. This redistributionc ould explain the hypomethylation of Sat2 in the juxtacentromeric constitutive heterochromatino ft hese twoI CF patients [28]d espite the persistence of methyltransferasea ctivity in vitro [38].
The functional importanceofthe non-catalyticroles of DNMT3B/Dnmt3b was illustrated by studies of differentiation of rat pheochromocytoma cells (PC12) into neuronal cells. Induction of differentiation in PC12 cellsm ediated by nerve growth factor was inhibited by antisenseo rs mall interfering RNA (siRNA) for Dnmt3b [41]. This inhibition couldb e largely reversed by transfectionw ithaplasmid encoding Dnmt3b, either wild-type or mutated in the C-terminal catalytic domain so as to inactivate catalytic activity,but not by mutantsmissingthe central ATRX domain or N-terminal PWWP domain. One of the noncatalytic, neuronaldifferentiation-related targets of the Dnmt3b is the T-cadherinpromoter whose activity was repressed by Dnmt3b irrespectiveo ft his promoter's methylation status [42].
Generally,m utant DNMT3B proteins from ICF cellsare still able to engage in normal protein -protein interactions [34]. One exception is an ICF mutation that alterst he amino-terminal region of DNMT3B's catalytic domain, and, in am ouse mutant interferes with acatalysis-enhancing interaction of Dnmt3a and Dnmt3b [17]. The other exceptionsare two mutations in the Ct erminal half of DNMT3B that did not appreciablyr educe catalytic activity but did greatly reduce its interaction with Dnmt3L [43](see below). These exceptionsf urther implicatel osso fD NA methylation, and not some other biological activity of DNMT3B, as the upstream molecular defect causing the ICF syndrome.

Stimulation of DNMT3Bcatalytic activity by interaction with DNMT3L
There is as pecific DNMT3L interaction domain located in the C-terminus of Dnmt3b/DNMT3B [44]. DNMT3L is an on-catalyticp rotein neededf or establishment of DNA methylation patterns in the germline, includingm aternal methylation imprinting and normal levels of methylation of satellite DNA sequences [20]. DNMT3L stimulates the catalytic activity of Dnmt3a and Dnmt3b methyltransferases [45]. This mayi nvolve contact of DNMT3Ba nd DNMT3L leading to ac hanges in the higher-order organization of DNMT3B, just as DNTM3L reorganizes theo ligomerizationo ft he DNMT3A2 isoformo fD NMT3A [46]. Importantly, two ICFassociated DNMT3B mutations (A766P and R840Q) that result in proteins with approximately wild-type basal methylation activity exhibit decreased associationw ithD NMT3La nd as trongd ecreasei n stimulation by DNMT3L [43]. These resultsi mplicate interaction of DNMT3B with DNMT3L as critical for normalDNA methylation that is protective against the ICF phenotype.

ICF type 2: Not associated withDNMT3Bm utations
About 40% of ICF patientshavenomutations in exons of DNMT3B [7,10]. There might be mutations in the promoter or other transcription control elementso r mutations affecting splicing or polyadenylation. However, for several of these ICF patientswithout detected DNMT3B mutations, the most common isoformo f DNMT3B RNA was still observed [7], and for the one patient examined in two putative promoter regions of DNMT3B ,n om utations were found [8]. Moreover, mRNA froma nI CF patientw ithout detected mutations in DNMT3B was examined by RT-PCR with primers or amplicon sizes specific for several of the DNMT3B RNA splice variants 3B4, 3B3, and3 B1 [47].N oe videncef or pathogenic alterations in DNMT3B splicing patterns was uncovered.
ICFp atientsw ithout DNMT3B coding region mutations seem to be derived from adifferent subtype of the disease. Lymphocytes or fibroblasts from all nine patientsinthis category displayed hypomethylation in satellite a (centromeric; Figure 1) DNA while all five examined patientsw ho had mutations in DNMT3B did not havet he pericentromeric DNA hypomethylation extending into the centromere [7]. ICF patients without mutations in DNMT3B exhibit the same Sat2 hypomethylation and chromosomal anomalies at 1qh and 16qh in mitogen-stimulatedcells seen in patients with DNMT3B mutations [8,13,14,48].

Immunodeficiency
The immunodeficiencyd isplayed by ICFp atients, despitet he presence of Bc ells, results in severe recurrent infectionsoften seen in early childhood [49 -51]. ICF immunodeficiencyi sv ariable ranging from agammaglobulinemia to am ild reduction in immune response [10,52]. In one study,2 7o ut of 44 patients presented with agammaglobulinemia althought hey had Bcells in the peripheral blood [10]. All but oneof ther est hadh ypogammaglobulinemia, onew ith selective IgA, two with IgM, and one with IgG2 subclassdeficiency.Normal percentagesofTcells were observed in 36 of 38 ICF patients, the expected stimulation of T-cell proliferation in 25 of 28 tested patients, andCD4 þ cellswere decreased in only five of 38 patients.
How similarare the Dnmt3b mutant mice to humanICF patients?
The missense Dnmt3b mutant mice went to term unlike the equivalent Dnmt3b homozygousn ullm ice, which died between 13.5 and 16.5 day post coitum [17]. Most of the missense mutant mice died within 1d ay of birth althoughs ome survived through to adulthood. Those that survived displayed low birth weight andf acial anomalies (shorter nose and wider nasalb ridge) reminiscent of theI CF syndrome. Although their immunes ystem was abnormal, the identified abnormalitywas in the T-cell lineage.One or two daysa fter birth, theyh ad reduced amounts of thymocytes, apparently due to apoptosis, and high levels of fragmented nuclei in the thymus.Incontrast, the immunodeficiencyo fI CF patients only infrequently involves decreased levelsofTcells [1,10] but always is characterized by reduced levels of one or more of the types of immunoglobulins. Normal levels of B cellswere observed in these mice as is usually the case for ICF patients.
As describedabove, only ICF Type 2patients (who do not display DNMT3B mutations) exhibit hypomethylation of centromeric satellite DNA ( Figure 1). The Dnmt3b missense mutant mice displayed hypomethylation of murine minor satellite DNA,w hich is centromeric, as well as of major satellite DNA,which is juxtacentromeric [ 17]. No mention was made in this studyo fc hromosomal rearrangements.H owever, in an earlier reportf rom this group, aneuploidy, polyploidy andn umerous breaksa nd fusions in chromosomes were observedi nm urine embryonic fibroblasts with double knockout of Dnmt3b [59]. This is veryd ifferent from the chromosomal abnormalities in ICF,which do not include aneuploidyorpolyploidy, and are localized almost exclusively to the juxtacentromericheterochromatin of only afew chromosomes, those with long regions of Sat2 [1,14] (see below). Therefore,the DNAhypomethylation and chromosomal abnormalities in mice with Dnmt3b mutations or derived cell cultures is more extensive (including centromeric as well as juxtacentromericDNA)than in the DNMT3B -mutant ICF syndrome in humans, even when comparing the same DNMT3B mutations, and the nature of the immune dysfunction is different. These differences between species in ad isease that most prominently affects satellite DNA-rich heterochromatin are not surprising. Mice do not haveSat2or Sat3-like sequences in their genome, where so much of the ICF DNA hypomethylation is concentrated [60]. Moreover, murine chromosomes are acrocentric, rather than mostly metacentric, as forh uman chromosomes, including those harbouring most of the Sat2 or Sat3 DNA. At the end of this review,w e will discuss our hypothesis thatSat2 hypomethylation plays an indirect causative role in ICF.
In additiontoSat2 andSat3, satellite DNA at Yqh is hypomethylated in ICF cells [13,60]. Another class of ICF-hypomethylated heterochromatic sequences is in facultative heterochromatin, the inactive Xc hromosome of females(X i ) [60,63,64]. Hypomethylation of examined CpG islands on the Xchromosome has been found in some, but nota ll, ICF patientso nt heir X i [1,51,63]. ICF-related Yqh andX i hypomethylation is unlikely to be of biological significanceb ecause no gender-specific differences in symptomsh aveb een reported for this disease, and, for X i genes, the hypomethylation is often inconsistent from patient to patient. Analysis of methylation of imprinted genes in ICF has not revealed any gene region with consistent hypomethylationa mong examinedp atients [52]. Therefore, these gene regions tooa re unlikely to contribute to the ICF phenotype.
By HPLC analysis of DNA digests, we demonstrated that the hypomethylation of the genome in ICF involves only ar athers mall percentage of the 5methylcytosine residues, namely 7% hypomethylation in brain DNA [14]. We confirmed that the methylation abnormality of ICF is confined to as mall fraction of the genome by two-dimensional electrophoresis of restriction digests of DNA from fourI CF vs. four control LCLs [65]. Only 13 of the approximately one thousand spotsd isplayedc onsistentI CF-specific differences,a nd all but one of these was derived from NBL2 andD4Z4 tandemrepeats. These repeats are dissimilar from each other and from Sat2 or Sat3 in sequence, G þ Ccontent, andchromosomal location (acrocentric,s ubtelomeric, or juxtacentromeric) despitet heir common ICF hypomethylation [65][66][67].T hat only al imiteda mount of consistent DNA demethylation is associated with ICF,a nd mostly in DNA repeats, should be takeninto account in models of howDNMT3B mutation gives rise to the disease.
Given the involvement of DNMT3Bi nt he ICF syndromea nd ICF-linked hypomethylation of the aboved escribedt andemr epeats,i ti sc lear that this enzyme is necessaryd uring development for normal methylation of these sequences in human somatic cells. The low but appreciable levels of methylation in these repeats in ICF somatic cells, whichare higher than those of sperm, might be due to either DNMT3A and/or residualD NMT3B activity.T he restructuring of chromatin composition during spermatogenesis [73] might inhibit access of satellite DNA and other large tandem repeats to DNMT3B or facilitate access to as yet uncertain DNA demethylation machinery [74,75] so as to explain the low levels of methylation of these sequences in sperm.
Frequent increases in methylation of someD NA sequences andd ecreases in methylation of othersa re seen in aw ide varietyo fc ancers [ 61].T here is often more hypomethylationt hanh ypermethylationo f DNA during carcinogenesis, leading to anet decrease in theg enomic 5-methylcytosine content [ 76]. Although the exact methylation changes between different cancerso ft he same type are not the same, therea re cancer type-specific differences in the frequency of hypermethylation or hypomethylation of certain genomicsequences. These opposite types of DNAm ethylation changesa ppear to be mostly independent of one another, althought heym ay arise because of as imilara bnormality leading to longlasting epigenetic instability in cancers [77]. Evidence of hypermethylation of some DNA sequences in ICF has been sought but has notb een found. Therefore, unlike cancers,I CF DNAh as exhibitedo nly hypomethylation.
The decondensationo f1 qh and1 6qh in ICF lymphoblastoid cells is likely to be critical to the ICFtype rearrangements in these regions. Indirect evidence for more frequentsomatic paring of 1qh and1 6qh in ICF lymphocytes than in control lymphocytes was reported [79]. In ICF LCLs, multiradials composedof arms of both Chr1 and Chr16 had been shown to be favoured over homologous associations [ 14]. We had proposed that those multiradials represent unresolved Holliday junctions. In collaborationw ith David Gisselsson, we analyzed chromosome dynamicsa t mitosis and the frequency of genomic imbalances in ICF LCLs [80]. Consistent with our model,the results suggest that illegitimate recombination of heterochromatic sequences at interphase due to increased 1qh and 16qh associations in ICF LCLs leads to severe perturbations at mitosis. DNMT3B co-localizes throughoutm itosis with components of thec hromosomec ondensation machinery( hSNF2H,K IF4A,h CAP-C,a nd hCAP-G) in HeLa cells [35]. These proteins were associated with Sat2 and rDNA in interphase as determined by chromatin immunoprecipitation of ac ontrol LCL, althoughthe extentofassociation was not quantitated. In ICF LCLs,o ne of these proteins, hSNF2H, was tested for its association with DNMT3B andfound to still coprecipitate with the mutant DNMT3B [34]. Nonetheless, these proteins and HP1, whichishighly concentrated in G2-phase ICF lymphoblasts in an anomalous giant nuclear body containing Sat2 DNA [40], might have arole in the ICF-specific chromosomal abnormalities at 1qh and16qh.
Moreover, the cell type and cell growth conditions influence the association of rearrangements with Sat2 hypomethylation. WhileS at2-hypomethylated chorionic villus cultures havea ni ncreased frequencyo f these rearrangements, they are only seen in averylow percentage of the metaphases in cultures at low passage numbers [84]. That frequencyi ncreases dramatically at higher passages. In ICF cells, the frequencyo f chromosomalr earrangementsa t1 qh and1 6qh depends on cell growth [1,9]. In vitro mitogen stimulation of lymphocytesg reatly increasest he development of these aberrant chromosomes independent of its role in inducing cycling. Amuch higher frequency of juxtacentromeric (pericentromeric) rearrangements of Chr1 andC hr16 per metaphase is seen 72 or 96 ha fter mitogen stimulation of ICF lymphocytest hana t4 8h,a lthough the frequent abnormal decondensation of 1qha nd 16qh can be observed in metaphases at 48 h [ 12,53,85].T he rearrangements observedi nm itogen-stimulated ICF lymphocytes andi nu ntreated ICF LCLs may occur in vivo,but at averylow rate, as deduced from studies of micronucleus formation in unstimulatedb one marrow andl ymphocytes fromI CF patients [53,58,86]. The viability of ICF patientsand cell-type specificity of the disease, mostly an immunodeficiency disease, indicates that ageneralized breakdown in 1qh and 16qh chromatin stability is not manifest throughout the tissues of ICF patients.

Canceri ncidence in ICF patients
Studiesf romD NA-hypomethylated mice give evidence of ac ausal relationship between genomic hypomethylation and cancer, but only certain types of cancer [77,94]. Although ICF had notbeen associated with cancer,f ewer than 50 patients( mostly children) havebeen identified. Their usually very short average lifespan would preclude detectiono facancer predisposition that was not very high andd id not result in tumorsr atherq uickly. However, recently an ICF patient of 7y earsw as reported to haveH odgkin lymphoma [95], and previously an unrelatedI CF patient was describeda sh aving an adrenocortical adenoma [ 8]. Then ormalo rh eightened DNA damager esponse observed in ICF lymphocytes [96] might helpe xplain why tumorsh aven ot been found more frequently in ICF patients comparedtopatients with shortl ife expectancies due to several otherr are chromosomal instabilitys yndromes [97].

Overview of ICF microarrayexpression analyses
Because ICF patients can haveverylarge decreases in specific serum immunoglobulins despite the usually normal levels of Bc ells [5,10], transcriptional dysregulation in Bcells or both Band Tcellsislikely to be the predominant cause of their immunodeficiency. We showed that ICF LCLs havep lentiful surface IgM and normal IgM RNA levels [5] despite extremely low levels of serum IgM. Therefore, the immuned efect in ICF occursa tastep prior to class switching. It was suggested thatp eripheral bloodderived Bcells and BLCLs in ICF patientsmay display an altered expression patternonly as result of being less mature than their normal counterparts [98]. However, the absence of detectable IgM in serum from 12 out of 45 patients [10]indicates an intrinsic defect in Bcells in ICF patients. Moreover, the differences in the expression patterns of ICF LCLs compared to control LCLs described below argue for more than just alossof maturity of Bcells. Forexample, we found significant differences in RNA levels in ICF vs. control LCLs for some genes expressed only in mature Bc ells, others known to be expressed mainly in the T-cell lineage,and yet othersw ith no known or expected relationship to lymphogenesis.
The summarydatafor 20 of these genes of interest from the second experiment are shown in Ta bles Iand II. Nine were also found to be dysregulated in the first microarray experiment and eight were similarlyup-or down-regulated in two different probe sets in the second experiment. By quantitative real-time RT-PCR (qRT-PCR), we verified that the following RNAswere overexpressed in eight ICF LCLs compared to eight Ta ble I. Immune system-related genes with significant differences in RNAlevels in ICF vs.control lymphoblasts.  The first [5] and secondexperiments on ICF vs.control B-cell lines were done similarly on HuGeneFL and HG-U133 microarrays, respectively (Affymetrix). The ICF LCLs for Expt.#2were those in Expt.#1(only one of the twoICF Bstocks) with three additional LCLs, patients 1and 3from one study [29] and patient 5from another [6].The control LCLs for Expt.#2were from the mothers and fathers of patient ICF Band Cand AG15022 and AG14953(Coriell Institute). Apositivefold change (FC)denotes RNA overexpression in ICF and is the mean signal for ICF divided by that for the controls for that probeset. AnegativeFCindicates underexpression in ICF and is the negativeofthe mean signal for controls divided by that for ICF. P -values are fortwo-sample t -tests to evaluate the significance of ICF-associated increases or decreases in mean RNA levels relativetothe controls.For IGHG3, PRKCH, CD44, and CKLF, another probeset in the microarraygavesimilar results in Expt.
#2 (not shown); b Real-time quantitativeR T-PCR(MyIQ Cycler and iQ SYBRGreen Supermix, Bio-Rad)was done on all the ICF LCL samplesused for Expt. #2 and 6-12 control LCLs (including additional B-cell lines)t hat were prepared from random hexanucleotide or oligo(dT)-primed cDNA. Primers were designed fort he HG-U133 microarrayp rober egions and optimal annealing temperatures for PCR were determinedbygradient PCR. The data were normalized to those from GAPDH. The fold change is described as for the microarray data.The P -values are for the differences in the mean RNAl evels using log-transformed data.NA, not assayed. control LCLs:t he transcriptionf actor NR2F2 ( COUP-TFII); SMARCA2 ( BRM), encoding a SNF2 subunit of achromatin remodelling complex; PRKCH and PTPN13 ( FAP-1 ), whichr egulate apoptosis; GUCY1B3 and GUCY1A3,w hiche ncode the two subunitsofasoluble guanylate cyclase, and CD44 and CKLF,w hicha re implicated in the laters tages of Bcell development. Although the terms up-or downregulation are used, acaveat is that differences in posttranscriptional processing are sometimes responsible for changes in the steady-state levels of RNA,t he parameter monitored in these studies. Twelveg enes with significant differences from the microarray data between ICF and control LCLs were then tested by RT-PCR. Only one of these did not exhibit RT-PCR resultsc oncordant with the microarray results (data not shown). In addition, onegene, RGS1 ,w hichd isplayed ICF-specific changes in its RNA levels only in the first microarrayexperiment, did not show significant differences in RNAlevels in ICF vs. controlLCLs by qRT-PCR (data not shown). Some otherg enes also displayed significant differences in RNA levels between ICF and control LCLs only in the first experiment [5]. However, the microarrayf or the first experiment was smaller, the gene annotation was muchless thorough, andthe probe sets were different from those in the second experiment (usually 20 oligonucleotidesper set in the first experiment and11 in the second). These factorsand the absence of some of the probe sets or the apparently poorerhybridization of others can explain why some of the genes that appearedtobesignificantly dysregulated in ICF LCLs compared to controlLCLs in the first array experiment were not seen as dysregulated in the second.
In view of the extreme scarcity of ICF patients and their median age of 8yearsatdeath [10], we examined Epstein Barrv irus-transformedB -cell lines rather than lymphocytes. The activation associated with this transformation might hide some in vivo defects in Bcell activation in ICF patients and only as ubset of B cells will be transformed. Nonetheless, we found muchc onsistencyi nL CLs from eight unrelated patientsa nd new insightsi nto transcriptional regulation of the immune system.

Geneswith lymphocyte functionsthat had alteredRNA levels in ICF LCLs
The most dramatic differenceinRNA levelsinICF vs. controlLCLs was seen for IGHG3 (Table I). This was expected based upon results givenr esultsf rom the patient sera and surface immunoglobulin expression [5]. The second largest difference in RNA levelsi n diseasecompared to control LCLs in Ta ble IorIIwas observed for MAP4K4, which has been implicated in antigen-mediated T-cell activation [99] but not in Bcell function.T hisu biquitously expressedk inase participates in the JNK/TNF-a and p53s ignalling pathways andc an be controlleda tt he levelo f transcription [ 100 -102]. The maximum microarray signalfor MAP4K4 in the controlLCLs in the second microarray experiment was4 6w hile,w itht he exception of one LCL (ICF K), the ICF LCLs had muchhighersignals (371, 2893, 295, 291, 1856, 348 and 1060).M AP4K4 seems to haveab road role in fostering cell migration [ 103]. Anotherg enew hose RNA was upregulated in ICF vs. control LCLs was NT5E,w hich was found to be expressed mostly in B cells, ratherthan Tcells, and usually only after isotype switching [104].
CD27, whichplays akey role in T-cell memoryand in the stimulation of Bcells to produce immunoglobulins [105], hadlower mRNA levels in ICF LCLs relative to control LCLs (Table I). This is consistent with ICF immunedysfunction. The ability of peripheralbloodderived Bcells (CD19 þ )toexpresscell-surface CD27 after stimulation [98]makes the down-regulation of its RNA in ICF lymphoblastoid cells unlikely to be causally involved in ICF immunodeficiency. While CD40, CD44, CKLF and ITGB2( CD18) mRNAs were significantly upregulated in ICF vs. controlLCLs, their positive functions in lymphogenesis [106 -112] makethem unlikely candidates for active playersinthe immunodeficiencyo fI CF patients.S omeg ene candidates for interfering with later stages of B-cell differentiation or activation did not display significant differences in RNA levels in ICF vs. control LCLs in the microarray analysis, namely, BTK, PRDM1, PAX5, IRF4, BCL6 , XBP1, BACH2 and MAPBPIP .

Cell death or growth arrest genes that had altered RNA levels in ICFL CLs
Genesinvolved in cell deathorarrest of the cell cycle mayb ei mportant contributors to thei mmune dysfunctiono fI CF patients. CASP1 , BCL2L10, PTPN13, HMOX1 , MAPK13 and PRKCH ,w hich displayed ICF-specific differences in their RNA levels, might be involved in abnormal regulation of apoptosis or cell cycle arrest in lymphoid cells in ICF patients (Tables Iand II). However, major decreases in numbers of lymphocytes are notusually found to be associated with ICF.Low levelsofTcellsare present in only half of ICF patientsand the levels of Bcells are even less likely to be lower than normal in ICF patients and are never undetectable [ 1,9]. Nonetheless, too much cell death induced by B-cell activation just in the laters tages of B-cell development couldlead to the loss of plasma cells and decreased serum immunoglobulin without compromising total levels of Bcells. CASP1(Ta bleI)isacytokineinvolvedinavariety of inflammatoryp rocesses,i ncludingt he proteolytic maturation of thei nactive precursoro ft he inflammatory cytokines interleukin-1 (IL1)and IL18 [113]. It is pro-apoptotic in various cell typesa nd mayb e associated with IgAdeficiency andincreased apoptosis in Bc ells [114]. Therefore, theo bservedi ncreased levels of CASP1R NA in ICFv s. controlL CLsa re consistent with theh eighteneds usceptibilityo fI CF lymphoid cellst oa poptosis [80,86,115,116].O verexpression of CASP1R NA in ICFc ells coulda lso perturbN F-k Bs ignallingp athwaysi mpacting expression of othergenes (see below).Underexpression of RNAf or thea nti-apoptoticH MOX1 stress protein andoverexpressionofMAPK13and PRKCHRNA in ICFv s. controlLCLs(Ta bleII) mightalsocontribute to pro-apoptotictendenciesofICF lymphoid cells.
There was significantly more BCL2L10RNA in ICF LCLs compared to controlL CLs (Table II). This widely expressedm embero ft he BCL2 family has a polypeptide structure compatible with both pro-a nd anti-apoptotic effects depending on the cell type and conditions [117]. It appearstobeanegative regulator of cell death in humanglioma cells provoking them to exit from the cell cycle [118]. While ICF LCLs were hypersensitive to g radiationcomparedtocontrols, we demonstrated that this was mostly due to irreversible growth inhibition, secondarily to non-apoptotic cell death,and thirdly to apoptosis [96]. All three of these types of responses to irradiationwere significantly more frequentfor ICF cellsthan for controlcells despitethe functional cell cycle checkpoints in ICF cells.
Anothers ignallingp athway with ICF-specific differences in RNA for several of its membersi st he carbonmonoxide (CO)pathway (Table II). HMOX1, which displayed ICF-specific decreases in RNA levels, catabolizes heme and therebyr eleases gaseous CO, which is responsible for its anti-apoptotic effects [129]. This pathwaya lsoi nvolvesd ownstream activation of NF-k B, which,i nt urn, by promoter interactions, activates transcriptionf rom as ubset of NF-k B-dependent anti-apoptotic genes [129][130][131]. Among the diverse effects of CO signalling,itseems to be am odulator of autoimmunity [129].
Anotherl inkt ot he NF-k Bs ignalp athways among the ICF-overexpressedR NAsi nvolves the abovementioned proteink inase Cf amily member, PRKCH. This calcium-independent, serine-a nd threonine-specific enzyme is activated by diacylglyceroltophosphorylate awide range of cellular proteins and therebyi nfluence many aspects of physiology [132][133][134][135].Unlikelymost protein kinase Cisoforms, transcriptiono f PRKCH is highly tissue specific. Its expression primarily,b ut not exclusively, in epithelial tissues is probably due to an enhancer, as ilencer and trans-actingf actors [ 136].I ns kin,P RKCH is associated with terminal differentiation of keratinocytes. Genetic polymorphismsi n PRKCH are implicated in increasingthe risk of rheumatoid arthritis and cerebral infarction [135,137]. Overexpression of PRKCH mayplay arole in tumor progression through downstream ERK and ELK effectors [132]. PRKCH RNA was overexpressed 4-to 14-foldinICF vs. control LCLs as seen in both microarray experiments andqRT-PCR( Ta ble II). It is implicated as ap ro-apoptotic protein in early B-cell development [138]. PRKCH RNA was reportedt ob ei nduced upon lymphocyte activation but was present at much lower levelsi nB cellsthan in Tcells, whether resting or activated [137]. Among the processes subject to its regulation, PRKCH helps controlt he activation of NF-k Bu pon lipopolysaccharide induction of primaryr at astrocytes [133]. PRKCH activity can result in the production of nitric oxide (NO), another important signalling gas, via the inducible nitric oxide synthetase gene(iNOS) [134].
Some of the genes in Ta bles Ia nd II that exhibited ICF-related changes in RNA levels in the Bl ymphoblastoid cellsare much more closely associated with T cells than with Bc ells. This might reflect coordinate dysregulation in the B-cell and T-cell lineages. For those genes, it might be that only dysregulation in T cells is relevant to the diseaseort hat the role of these genes in Bc ells is insufficiently appreciated. Alternatively,t hese changesm ight contribute to the pathology of ICF because of inappropriate expression of T-cell specific genes in the B-cell lineage.

Dysregulation of genes that may help explain non-immune symptomsofI CF
While the significantly altered RNA levelsf or some proteins in ICF vs. control LCLs, such as the CNN actin-binding protein,m ight have no biological consequences, othersm ay be af actor in the nonimmunes ymptoms of ICF.A lthough we examined RNA only in BL CLs,a bnormal RNA levels in the lymphoid cell lineage might be found in other lineages too and altered regulation in lymphocytes can sometimes mirror more physiologically important dysregulation in av eryd ifferent tissue [142]. Overexpression of PTPN13 RNA in ICF LCLs (Table II) might pertaint on eurologicalfi ndingsi nI CF patients because high levels of PTPN13 in fetal brain and its ability to bind to the neurotrophinr eceptor implicate this protein in controlling neuronal celld eath [120,143]. One of the neurological abnormalities in ICF is seizures, which was reported in three of 45 patients [10]. SLC1A1, aneuronalprotein involved in transporting glutamatet hati sp rotectivea gainst seizures and neuronal death [ 144], was significantly underexpressed at the RNA levelinICF LCLs relative to controlLCLs (TableI I). In addition, the observed BCL2L10R NA dysregulation may be involved in ICF-related neurological disturbances because BCL2L10m RNA increases appreciably from av ery low level during in vitro differentiation of rat astrocytes [145]. Moreover, GUCY1B3 and iNOS are associated with each other in certain areas of the hippocampus in mice [ 146]a nd CASP1 overexpression has been linkedtocognitive impairment with aging [147].

Lack of detectable methylation changes in the promoter regions of examined ICF-upregulatedgenes
We recently examined methylation of five genes with qRT-PCR-confirmed ICF-associated upregulation of their RNA,n amely, GUCY1A3, PTPN13, NR2F2, SMARCA2 and CKLF (Tables Iand II). Their 5 0 gene region (for GUCY1A3)o ru pstream regions (fort he othergenes) were assayed for methylation in ICF and control LCLs as well as in several normal tissues by combined bisulfiter estrictiona nalysis( COBRA). COBRA allows quantitationo fD NA methylation levels at restriction endonuclease sites in agiven DNA sequence that is amplified by PCR [150]. Genes with up-regulationo fR NA were chosen because of the frequent association of DNAh ypomethylation in promotersw itht ranscription [151]. By COBRA, we saw no consistent ICF-specific differences in DNA methylation in the examined regions despite their ICF-related increases in the levels of the corresponding mRNAs (Figure 2and data not shown). This was similart oo ur previous finding for GUCY1B3 in ICF cells [5]. Almost allofthe immediate upstream regions were constitutively unmethylated. There were various amounts of partialm ethylationi nt heir further upstream sequences but no correlations between methylationa nd RNA levels amongi ndividual LCLs. Therefore,t he differential mRNA levelsf or these genes in ICF vs. control LCLs could notb e explained by differences in methylation in or near their promoters. Ac aveat in this analysis is that only the top gene in an ICF-activated transcription pathway might have ICF-specific promoter hypomethylation.
Hypothesis: The relationship of DNMT3B mutations to the ICF phenotype Methylation analysis of various genes [51,52,63,64] (Figure 2), and whole-genomer estriction analysis [65] have revealed consistent hypomethylation only in tandem DNA repeats (including satellite DNA)a nd X-linked sequences in X i .H owever, unconventional genes,w hoseb road biological influence has been appreciated only recently (especially microRNAa nd anti-sense RNA genes) [152], were not specifically examined. While acritical genethat is hypomethylated specifically in ICF cells mayh aveb een missed, we favour ad ifferent explanation for the connection of selective DNA hypomethylation to the ICF syndrome. We propose that the pathogenicity of DNMT3B mutations in ICF patients is due to the hypomethylation of constitutive heterochromatin.T hiss ame explanation can be applied to patientsw ith ICF type 2, who have no detected DNMT3B mutations but do exhibit the characteristic hypomethylation of juxtacentromeric satellite DNA [7].Because no genderbias has beenr eported for ICF,o ur proposal for the involvement of satellite DNA is limited to the long juxtacentromeric heterochromatin butn ot Yqh (Figure 1) nor the facultativeh eterochromatin of X i . We favour the involvement of Sat2-containing1 qh and1 6qho verS at3-containing9 qh (Figure1 ) because of the more frequent cytogenetic abnormalities in the formerr egions.
Evidence is mounting that constitutive heterochromatin is biologically important andn ot just an inert filler in the genome,a sm any previously thought. In fission yeast and drosophila, transcription of noncoding RNA is important for the establishment of constitutive heterochromatin [153 -155]; these organisms have littleo rn om ethylation of their DNA [16]. Some,b ut not all, of various tested normal or cancer samples andh alf of ICF LCLs that we analyzed for Sat2 RNA were positive by RT-PCR (which included controls for DNA contamination) andb ya ssays for RNA polymerasee ngagement [156]. However, these signals were very low andw ed id not see the large increase in Sat2 transcripts upon heat shock [156] that is found for 9qh Sat3 transcripts [157].
Besides constitutive heterochromatin yielding noncoding transcripts that might affect expression of protein-coding genes, its intranuclearl ocation may help organize chromatin throughout the nucleus so as to modulateg ene expression in trans [85]. Evidence fort hisp henomenonh as been reported in the lymphoid lineage [158,159]. The intranucleard istributionofcentromeres in lymphoid cells is distinct for the cell type and stage of differentiation and involves genesa ssociated with lymphogenesis [160]. The importance of the spatial location of chromatini n the nucleus is illustratedbythe finding that much gene expression occursi nt ranscriptionf actories that are specific for different functional groups of genes [161]. RepresentativeCOBRA analysis for DNA methylation of agene that had RNA upregulatedi nI CF vs. control LCLs. DNA samples had been modified with bisulfite and amplified by PCR with primers at the indicated positions as previously described [5]. The PCR products could be cleavedb yT aqI or BstUI only if they had been methylated at the CpG dinucleotidei nt he indicated sites in genomic DNA; the pre-TaqI site, CCGA, would be converted to aT aqI site, TCGA, upon bisulfite treatment and PCR if it was Cm5 CGA in genomic DNA. (A) Diagram of the 5 0 PTPN region showing the transcription startsite (TSS) [123] at Chr4 87,734,909 (hg18, UCSC database), the 5 0 CpG island ( 2 701 to 2 150), and PCR primers;positions are given relativetothe TSS. (B) and (C), electrophoresis gels stained with ethidium bromide and visualized for fluorescent bands from the PCR products ( 2 628 to 2 119 and 2 1250 to 2 940) digested with Bst UI or Ta q I. PBMC, peripheral blood mononuclear cells; ICF LCLs are described in the legend to Ta ble I, with the addition of another control LCL (AG14832, Coriell Institute). Sizes are giveninbpfor the expected and obtained restriction products.
Recently,o ne dramatic changei np ositioning of hypomethylated constitutive heterochromatin specifically in ICF lymphoblasts and lymphocytes has been described. It is the formation of agiant promyelocytic leukemia (PML) type nuclear bodyt hat correlated with undercondensed1 qh or 16qh, but not 9qh, in a large percentage of ICFG 2n uclei [ 39,40]. Allt hree isoforms of HP1a sw ell as SP100, SUMO-1, transcriptionf actorsC BP and DAXX,t he DNA helicase BLM, and the SWI-SNF remodelling protein ATRX co-localize in this single nuclear body. Much smaller PML-type bodies containing these proteins are observedi nG 2-phase nuclei of normal cells but the association of 1qh Sat2 DNA with these normal bodiesisless frequentthan for the giant PML nuclear body in ICF lymphocytes and LCLs. This abnormal concentrationo fs atellite DNA heterochromatin and nuclear proteins in ICF G2-stage lymphoid cellsh as been proposed to be linked to undercondensation and chromosomal abnormalities at 1qh and 16qh [ 40]. However, it might also reflect an abnormal distributionofc hromatinproteins in interphase that could influencee xpressiono fg enes elsewherei nt he genome.
There are more andmore examples of transcription controlp roteins that bind selectively to constitutive heterochromatin [162 -168]. Furthermore, there is evidence that the binding of at least some of these transcriptionf actorst ot he highlyr epetitive DNA of constitutive heterochromatin sequesterst hese proteins in ar eversible manner so as to modulate expression of sets of genes [5,168 -172]. Methylation of satellite DNA can dramatically alter binding of certain transcription controlp roteins to DNA, in general [173] or constitutive heterochromatin [174], in particular. Therefore, pathogenic dysregulation of a subset of genes in ICF might be due to altered transcriptionf actorb inding to satelliteD NA in response to its disease-related hypomethylation. This would be an ew type of DNAm ethylation control of gene expression in trans mediatedbychromatin acting as ad ynamic reservoir fors torage andp ossibly deliveryo ft ranscriptionm odulatoryp roteins, which in the case of ICF,m ight explain the life-threatening immunodeficiency.