Jarid2 Associates with the PRC2 Complex in Mouse ES Cells
To screen for novel PRC2 partners we immunopurified and identified Eed-associated proteins using clonal mouse ES transgenic lines stably-expressing FLAG epitope-tagged Eed, as diagrammed in Figure S1A
, available online. In addition to previously characterized PRC2 components—Eed, Suz12, Ezh2 and Aepb2—mass spectrometry analysis identified Jarid2 in Eed-FLAG immunoprecipitates, but not control extracts (; all identified peptides are listed in Table S1
). Anti-Jarid2 immunoblot analysis of Eed-FLAG eluates confirmed association between Jarid2 and Eed (). To address whether Jarid2 interacts with the intact PRC2 complex, we subjected the Eed-FLAG eluate to another round of immunoaffinity purification with anti-Jarid2 IgG or control IgG (Figure S1B
). Mass spectrometry analysis after this two-step purification identified all core PRC2 subunits in addition to Jarid2, indicating that Jarid2 interacts with the intact PRC2 complex (; peptides listed in Table S1
Isolation of the PRC2 Complex from ES Cells Identified Jarid2 as a Novel Component
Next, we showed that endogenous Suz12 and Ezh2 immunoprecipitated endogenous Jarid2 from mouse ES cell nuclear extracts and conversely, Jarid2 immunoprecipitated Suz12 and Ezh2 (). Furthermore, Jarid2 co-sedimented with PRC2 in two high-density peaks in the glycerol-gradient sedimentation analysis of the Eed-FLAG eluates (). Eed, Suz12 and Ezh2 co-sedimented in fractions 3-5 in the absence of Jarid2, suggesting that Jarid2 is not required for the assembly of core PRC2 complex in mouse ES cells, consistent with previous reports that the three core subunits—Eed, Ezh2, and Suz12—form a stable complex (Cao and Zhang, 2004
; Martin et al., 2006
). Nevertheless, the majority of PRC2 in ES cells appears to be bound to Jarid2 (). Sedimentation analyses of nuclear extracts from Jarid2 shRNA-expressing cells (in which Jarid2 is downregulated to 30%–40% of wild-type levels; described in detail below), revealed that the remaining Jarid2 co-sediments with PRC2 in a single peak (Figure S1C
). These data suggest that the formation and/or stability of the largest complex is sensitive to Jarid2 levels.
Numerous studies demonstrated that Jarid2 expression is under the control of ES transcriptional circuitry, including transcription factors Nanog, Oct4, Sox2, Klf4 and Tcf3 (Boyer et al., 2005
; Cole et al., 2008
; Kim et al., 2008
; Loh et al., 2006
; Zhou et al., 2007
mRNA is among the transcripts most highly enriched in undifferentiated mouse and human ES cells and human oocytes (Assou et al., 2009
; Sun et al., 2008
; Zhou et al., 2007
). Interestingly, due to conserved amino acid changes that preclude cofactor binding (Figure S2A
), Jarid2 lacks histone demethylase activity characteristic of other JmjC domain proteins (Klose et al., 2006
; Lan et al., 2008
; Shirato et al., 2009
). We reasoned that Jarid2 represents an attractive candidate regulator of the PRC2 function.
A Short Motif Conserved in Jarid Proteins Is Required for Interaction of Jarid2 and Jarid1a with Suz12
The mass spectrometry analysis of Jarid2 associated proteins showed Suz12 as the most enriched PRC2 member (), suggesting a direct interaction. In agreement, purified recombinant GST-Jarid2 recovered recombinant Suz12 in a pull-down assay (). Two N-terminal regions of Suz12 bound GST-Jarid2 (), and a single Jarid2 region corresponding to amino acids 726-913 (fragment d) was sufficient for binding to both Suz12 N-terminal fragments (). In contrast, we failed to detect interaction of Jarid2 with recombinant Ezh2, Eed, Aebp2 or RbAp48 (Figures S2B and S2C
Jarid2 and Jarid1a Directly Bind Suz12 via a Conserved Amino Acid Motif
Jarid2 is most closely related to the four Jarid1 family proteins, which also contain JmjC domain and ARID (AT-rich interaction domain) (Kortschak et al., 2000
). Jarid1 proteins are active H3K4me3 demethylases (Lan et al., 2008
), and Jarid1a/Rbp2 was shown to associate with PRC2 (Pasini et al., 2008
). Although we failed to identify Jarid1a peptides in mass spectrometry analysis of Eed-FLAG eluates (Table S1
), we detected Jarid1a-PRC2 association by coimmunoprecipitation and immunoblotting (Figure S2D
), and hypothesized that investigating the mode of Jarid1 association with PRC2 may shed light on the molecular recognition of PRC2 by Jarid proteins.
First, we tested whether Jarid1a, like Jarid2, binds Suz12. Recombinant GST-fusion protein corresponding to amino acids 250-500 of human Jarid1a (fragment b) bound the second, but not the first, N-terminal region of Suz12 (). Thus, both Jarid2 and Jarid1a interact with Suz12 amino acid region 185-370, but Jarid2 recognizes additional residues within the first N-terminal fragment.
Jarid2 and Jarid1a regions responsible for Suz12 binding do not overlap with any discernible structural domains () and display low similarity, with the exception of a highly homologous short sequence “GSGFP.” We hypothesized that this motif may play a role in Suz12 recognition. Indeed, mutations of GSGFP to GAGAA diminished binding of Jarid2 and Jarid1a fragments to full-length Suz12 (compare d and f in in ). This motif is conserved in all vertebrate Jarid2 proteins, as well as in C. elegans Jarid2 (), whereas D. melanogaster and other Drosophila species contain a non-conservative substitution within the motif (GYGFP). The GSGFP motif is also conserved in all four Jarid1 family proteins: Jarid1a/RBP2, Jarid1b/PLU-1, Jarid1c/SMCX and Jarid1d/SMCY, as well as in the single Jarid1 homolog in Drosophila, Lid (). The presence of the GSGFP motif in metazoan Jarid proteins suggests that the association with Suz12 may be a common feature of Jarid family members. However, we cannot exclude the possibility that additional molecular interactions control Jarid-PRC2 complex formation in vivo.
Mouse ES Cells Contain High Levels of Jarid2 Protein
The preferential recovery of Jarid2 in the Eed-FLAG purification indicates that Jarid2 is the major Jarid family member associated with PRC2 in ES cells. To estimate the relative molar amounts of Jarid2 and Jarid1a proteins in ES nuclear extracts we compared immunoblot signals of endogenous Jarid2 or Jarid1a to signals from a serial dilution of purified, recombinant protein fragments of known concentrations (). From this analysis we calculated that 1 mg of ES nuclear extract contains 6 pmols of Jarid2 and 0.12 pmole of Jarid1a. Although such measurements are not precise, we further estimated that a single ES cell nucleus contains about 50,000 Jarid2 and about 1000 Jarid1a molecules. Interestingly, downregulation of Jarid2 results in a modest upregulation of Jarid1a protein, but not RNA levels (Figures S2E and S2F
), perhaps via stabilization of Jarid1a through PRC2 association.
Jarid2 Occupies PRC2 Targets Genome-Wide
To determine the genome-wide occupancy of Jarid2 in mouse ES cells and to analyze the extent to which it overlaps with PRC2 and Jarid1a binding, we used chromatin immunoprecipitation coupled with massively parallel DNA sequencing (ChIPseq; (Barski et al., 2007
; Johnson et al., 2007
). Illumina Genome Analyzer was used to generate 13.5, 12.7, 18.7, 11, and 28.5 million mapped sequence reads from Jarid2, Ezh2, Suz12, Jarid1a, and control libraries, respectively. QuEST ChIP-Seq analysis software (Valouev et al., 2008
) identified 1337, 1692, 2073 and 1764 “significant regions” enriched within Jarid2, Ezh2, Suz12 and Jarid1a ChIP-seq datasets [at the false discovery rate (FDR) of less than 2.8%]. Significant regions were identified for all proteins with high stringency threshold of having at least one position with 50-fold or higher enrichment.
Comparison of Jarid2, Suz12 and Ezh2 ChIP-Seq enrichment signals revealed a nearly complete overlap of binding patterns (as illustrated by the Hoxd
gene cluster and Sox9
gene locus, Figures and S3A
). We also observed good overlaps between our binding data and those reported in the literature (Figures and S3A
, compare top three panels with the panels displaying Ezh2 data [Ku et al., 2008
] and H3K27me3 data [Mikkelsen et al., 2007
]). However, we did not detect significant binding of Jarid1a over PRC2 bound loci (Figures and S3A
). Next, we analyzed genome-wide co-occupancy of the interrogated proteins. Within Jarid2, Ezh2 and Jarid1a significant regions (defined as having a peak of 50-fold or higher enrichment) we identified relative fold enrichment of Suz12, Ezh2, Jarid2 and Jarid1a (defined as the enrichment of sequence tags relative to control across the entire region; less than 3-fold enrichment falls within experimental variability and was considered as not enriched). 99.2% and 99.8% of Jarid2 significant regions was enriched for Suz12 and Ezh2, respectively, with the majority of regions enriched over 10-fold (). Conversely, 99.7% and 99.7% of Ezh2 significant regions was enriched for Suz12 and Jarid2, respectively. The majority of Ezh2 regions were enriched for Jarid2 more than 10-fold (). Colocalization of Jarid2 with PRC2 was also supported by high correlation (0.66-0.89) of ChIP-seq binding signals among the Jarid2, Ezh2 and Suz12 datasets (). There was also good correlation of Jarid2 and Ezh2 binding with H3K27me3 ().
Jarid2 and PRC2 Occupy Same Genomic Targets in ES Cells
In sharp contrast, only 1% and 2% of Jarid2 and Ezh2 significant regions, respectively, was enriched for Jarid1a signals, and none had over 10-fold enrichment (). Overall, Jarid1a occupancy showed little correlation with PRC2 binding (0.18-0.37, Figure S4A
) or with enrichment for H3K27me3 (). Instead, Jarid1a significant regions overlapped with H3K4me3 ( and S3B
Jarid2-PRC2 Targets Are Enriched for Unique DNA Sequence Motifs
De novo search for motifs overrepresented within the Jarid2, Ezh2 and Suz12 binding peaks identified two significant motifs () that were enriched at peaks and also throughout the bound regions. The first motif is a tandem repeat of CCG and is present within 61%–72% of Jarid2, Ezh2 and Suz12 regions (5% FDR). A second, GA-rich motif was present within 56%–66% of PRC2 and Jarid2 regions (5% FDR). Both motifs were also significantly enriched within a previously reported Ezh2 dataset from Ku et al. (2008)
(60% and 57% of regions contained these motifs, respectively).
Jarid2 and PRC2 Co-occupy Promoters of Developmental Genes in Mouse and Human ES Cells
Jarid2 bound regions typically overlap with transcription start sites (TSS; 69% overlap, Figure S5A
) and exhibit a mean and median size of 3.3 kb and 2.7 kb, respectively (Figure S5D
). Functional classification of identified targets via GO term analysis (Beissbarth and Speed, 2004
) showed highly significant enrichment in genes involved in development, morphogenesis, and transcription (Figure S5B
), similarly to what was previously observed for PRC2 (Boyer et al., 2006
). A complete list of Jarid2 bound genes is provided in Table S2
To validate ChIP-seq results, we performed ChIP-qPCR analyses of selected PRC2 target genes and, as a control, a gene not bound by the PRC2 complex (Mcm6) using independently isolated DNA from mouse and human ES cells. Relative Jarid2, Ezh2 and Suz12 occupancy levels were correlated for all tested PRC2 target genes in both mouse and human ES cells, indicating that Jarid2 association with PRC2 targets is conserved between mouse and human ().
Jarid2 and PRC2 Simultaneously Bind Target Genes
To demonstrate that Jarid2 and PRC2 can simultaneously bind to the same chromatin regions, we performed sequential ChIP analysis. Chromatin from Eed-FLAG mouse ES cells was first immunoprecipitated with FLAG antibody, followed by a second ChIP step with either Jarid2, Suz12, or non-specific IgG antibody. For all interrogated PRC2 targets, we detected simultaneous Jarid2 and PRC2 binding (). In sum, biochemical association, direct binding between Jarid2 and Suz12 in vitro, and target co-occupancy across the genome strongly suggest Jarid2 is an integral PRC2 subunit in ES cells.
Jarid1a/Rbp2 Occupies Promoters of Genes Involved in RNA Metabolism and Mitochondrial Function
Although we identified 1764 genomic regions enriched for Jarid1a, we failed to detect an overlap with PRC2 binding. Consistently, functional classification of identified Jarid1a targets showed enrichment of genes involved in RNA processing and mitochondrial function, but not in development (Figure S5C
). 82% of Jarid1a significant regions overlapped with H3K4me3 (representing 15% of all H3K4me3 significant regions in mouse ES cells). These targets were not bivalently marked, however, as we failed to detect an overlap between Jarid1a and H3K27me3 (). We also noted a quantitative difference in the size of Jarid1a and Jarid2 bound regions (Figure S5D
). De novo motif search analysis revealed enrichment for a consensus recognition site of the Ets family transcription factor GABP in 55% of Jarid1a bound regions (Figure S4B
). Intriguingly, our observations parallel those made in a previous ChIP-chip study of JARID1A occupancy in human promonocytic U937 cells (Lopez-Bigas et al., 2008
). These parallels include: (i) overrepresentation of RNA metabolism and mitochondrial gene targets, (ii) enrichment for Ets family binding sites, and (iii) strong overlap with H3K4me3 in the absence of H3K27me3. Strikingly, 79% (185 out of 232) of JARID1A targets reported by Lopez-Bigas et al. (2008)
in U937 cells are found in our Jarid1a dataset, despite differences in cell type, species and antibody used in the two studies. The findings outlined above suggest that a significant subset of Jarid1a targets represents house-keeping genes active across many cell types in humans and mice.
To confirm that our results are not an artifact of antibody cross-reactivity, we performed ChIP-qPCR analyses using five different anti-Jarid1a antibodies. We failed to detect Jarid1a binding to PRC2 targets with any of the antibodies (Figure S6A
), but observed Jarid1a enrichment at all interrogated Jarid1a target genes identified by ChIP-seq with all tested antibodies (Figure S6B
). Taken altogether, our data strongly argue that Jarid2, not Jarid1a, is the major PRC2 partner at chromatin targets in ES cells. We therefore focused our subsequent analyses on the mechanisms by which Jarid2 regulates PRC2 functions.
Jarid2 Is Important for Recruitment and/or Stabilization of PRC2 at Target Genes
To address whether Jarid2 downregulation affects PRC2 target occupancy we developed stable, clonal mouse ES cell lines expressing shRNA targeting Jarid2 or, as a control, a non-targeting shRNA, in a doxycycline (Dox) inducible manner. Dox treatment of Jarid2 shRNA cells resulted in Jarid2 downregulation to 30%–40% of control levels (Figure S7A
), without affecting total Ezh2 and Suz12 protein levels (Figure S7A
), Oct4 expression (), or ES cell proliferation (data not shown). ChIP-qPCR analysis showed varied Jarid2 levels at different targets, nevertheless in all cases both Jarid2 and Ezh2 occupancy were significantly reduced upon Jarid2 knockdown ().
Jarid2 Controls PRC2 Target Occupancy in ES Cells
To exclude a possibility that diminished PRC2 binding is an artifact of shRNA off-target effects, we used the heterozygous Jarid2
gene trap mouse ES line (Davisson, 2006
) to assay Jarid2 and Ezh2 protein levels and target occupancy. As this line was developed in a different genetic background than the one used throughout this study (E14 versus LF2), we compared wt LF2 cells, wt E14 cells and Jarid2
−/+ E14 (CSA 131) cells. Total levels of Jarid2 protein were higher in LF2 as compared to E14 cells, and were further diminished in Jarid2 −/+ E14 cells (Figure S7B
). Nevertheless, Jarid2 target occupancy was comparable between LF2 and E14 wt cells, with reduced binding in Jarid2
−/+ E14 cells (), indicating that even a modest two-fold reduction of Jarid2 levels is sufficient to diminish Ezh2 association with target genes.
Jarid2 and PRC2 Association with Target Genes Is Mutually Dependent
If Jarid2 and PRC2 cooperate in target recognition, then Jarid2 chromatin binding should be PRC2-dependent. To test this prediction, we assayed occupancy of Ezh2, Suz12 and Jarid2 at selected target genes in Eed
−/− ES cells. Consistent with Eed serving as a linchpin for the PRC2 complex, Suz12 and Ezh2 binding was diminished in Eed
−/− as compared to wt cells (). Jarid2 binding was concomitantly reduced, indicating that Jarid2 and PRC2 target association is mutually dependent (). However, we observed a two-fold down-regulation of Jarid2 protein levels in Eed
−/− cells (Figure S7C
), which may in part account for the diminished association of Jarid2 with targets.
Jarid2 Knockdown Results in Derepression of PRC2 Target Genes in ES Cells
Loss of PRC2 components in ES cells results in upregulation of PRC2 target genes (Boyer et al., 2006
; Pasini et al., 2007
). To test whether Jarid2 knockdown leads to a similar effect, we assayed mRNA expression levels of selected PRC2 target genes in ES cell lines expressing Jarid2 or non-silencing shRNAs. Whereas Oct4
mRNA levels were comparable among all shRNA lines, expression of interrogated PRC2 target genes was upregulated upon Jarid2 knockdown (). Similarly, expression of PRC2 target genes was also upregulated in Jarid2
−/+ E14 cells, as compared to wt E14 and LF2 ES lines ().
Jarid2 Negatively Regulates PRC2 Enzymatic Activity
To test whether, in addition to targeting, Jarid2 can also regulate PRC2 enzymatic activity, we performed histone methyltransferase (HMT) assays with purified reconstituted recombinant PRC2 complex either in the absence or in the presence of the recombinant full-length Jarid2, and using native core HeLa histones as a substrate. Addition of sub-stoichiometric amounts of Jarid2 was sufficient to inhibit PRC2 HMT activity in a dose-dependent manner (), while addition of other recombinant chromatin proteins had no effect (Figure S8A
). Similarly, Jarid2 inhibited PRC2 HMT activity on nucleosomal substrates (). However, addition of Jarid2 to H3K4 methyltransferase complexes purified via the Ash2 core subunit had no effect on its HMT activity (Figure S8B
). Moreover, Jarid2 stimulated HMT activity of recombinant, purified H3K9 methyltransferase G9a (Figure S8C
), which was previously shown to bind Jarid2 (Shirato et al., 2009
). In sum, these results demonstrate that Jarid2 specifically inhibits PRC2 HMT activity in vitro and suggest that it plays diverse roles in regulation of histone methyltransferases in distinct cellular contexts. As HMT assays were performed in the absence of cofactors other than S-adenosyl-methionine (SAM), Jarid2-mediated inhibition of PRC2 HMT likely occurs through a nonenzymatic mechanism. Interestingly, we found that recombinant Jarid1a also inhibited PRC2 HMT activity in vitro (Figure S8D
), indicating that other Jarid family members may regulate PRC2 enzymatic function.
Jarid2 Directly Inhibits PRC2 Histone Methyltransferase Activity
Next, we studied the effects of Jarid2 downregulation on H3K27me3 levels in vivo. Although a modest reduction of Jarid2 levels achieved in our ES lines is sufficient to reduce Ezh2 target occupancy by 2-6 fold (), we observed little to no effect on H3K27me3 levels at the same gene targets (). For example, the Sox9 gene exhibited the strongest relative downregulation of Ezh2 occupancy, yet a slight upregulation of H3K27me3 levels (), suggesting that the Jarid2-containing PRC2 complex is less active. Interestingly, despite unaffected H3K27me3 levels, expression of PRC2 target genes was upregulated upon Jarid2 knockdown (), and this upregulation correlated with a modest increase in H3K4me3 levels (). Our results demonstrate that Jarid2 negatively regulates PRC2 HMT activity.
Jarid2 Depletion Results in Gastrulation Defects in Xenopus Embryos
Jarid2 promotes PRC2 recruitment to the target genes while inhibiting PRC2 enzymatic activity, suggesting that it modulates PRC2 function at developmental genes, perhaps to sensitize them for subsequent activation during differentiation (). To address whether Jarid2 is important for gene regulation during early embryogenesis, we downregulated Jarid2 levels in Xenopus laevis
embryos by injecting morpholino oligonucleotides targeting the translation start sites of both non-allelic copies of X. laevis
Jarid2 (Jarid2 MO1). Jarid2 amino acid sequence and domain composition is highly conserved between frogs and mammals (Figure S9
Immunoblot analysis of embryonic extracts showed that Jarid2 MO1 injection resulted in the downregulation of Jarid2, without affecting Suz12 protein levels (). Jarid2 MO1 injected embryos exhibited gastrulation arrest, whereas control embryos proceeded to develop normally and were assayed at the neurula stage (; phenotype penetrance in ). To ensure that the observed phenotype does not result from off-target effects, we designed two additional translation-blocking MOs, each matching one of the two non-allelic Jarid2 copies (Jarid2 MO2a and Jarid2 MO2b). Injection of either of the MOs resulted in phenocopy of Jarid2 MO1 phenotype. When co-injected together in equimolar amounts (referred to as Jarid2 MO2) at 2-4 fold lower concentration than either Jarid2 MO1, Jarid2 MO2a, or Jarid2 MO2b alone, these two MOs showed a strong synergistic effect resulting in 100% penetrant developmental arrest at the late blastula stage (; quantified in , protein knockdown verified in Figure S10A
). Injection of Suz12 MO also resulted in arrest prior to completion of gastrulation (), consistent with gastrulation defects reported in Suz12
−/− mice (Pasini et al., 2004
). In contrast, injection of MOs targeting H3K4 methyltransferase MLL1 or Jarid1a had no effects on gastrulation (not shown).
Jarid2 Is Required for Induction of Gastrulation Programs in Xenopus Embryos
Jarid2 Is Required for Activation of Gastrulation Gene Expression Program
embryos, genes necessary for orchestrating gastrulation events are induced during the mid- and late blastula stage. The late blastula stage corresponds to the critical transition period during which cells exit pluripotency and restrict their developmental potential, but have not yet differentiated to form three germ layers. In agreement with a recent report (Akkers et al., 2009
), H3K27me3 levels at selected early differentiation genes significantly increased during the transition from mid- to late blastula stage (Figure S10B and C
, compare % input recovery scales on y
axis in B and C). Importantly, Jarid2 knockdown resulted in further upregulation of H3K27me3, evident particularly at the late blastula stage (Figure 10B and C
) and consistent with repression of PRC2 HMT activity by Jarid2.
To assay the effects of Jarid2 downregulation on gene expression, we performed quantitative RT-qPCR analyses from late blastula stage embryos, either untreated or injected with Jarid2 MO1, Jarid2 MO2 or Suz12 MO. Expression of genes involved in germ layer formation was downregulated by all assayed MOs (), although generally Jarid2 MOs displayed a more prominent effect than Suz12 MO. In contrast, analysis of a cohort of non-developmental genes revealed no significant effect on expression (), indicating that MO treatment did not cause a global transcriptional failure. Furthermore, whole-mount RNA in situ hybridization analysis of gastrula embryos asymmetrically injected with Jarid2 MO1 at the two-cell stage (resulting in Jarid2 depletion in half of the embryo) revealed diminished expression of gastrulation genes Brachyury, Wnt8 and Xnot on the MO-injected side of the embryo (). In sum, our observations indicate that Jarid2 function is essential for induction of gastrulation programs at the exit from pluripotency.
Jarid2 Knockdown Results in Failure to Induce Mesoderm in Response to Activin Signaling
ectodermal explants isolated from blastula stage embryos can be induced in vitro to form mesoderm by treatment with Activin, a ligand for Nodal signaling, but the competence to induce mesoderm is lost by the end of gastrulation (Kimelman, 2006
To test whether Jarid2 depletion affects induction of mesodermal genes or, alternatively, results in the delayed or prolonged competence for induction, we isolated ectodermal explants from the early blastula stage control or Jarid2 MO-treated embryos, cultured them either in the absence of Activin or in the presence of Activin pulse at blastula, early gastrula or late gastrula stages, respectively, and analyzed gene expression by RT-qPCR (). Control explants not exposed to Activin established a default epidermal fate, as evidenced by the expression of epidermal Keratin. Activin treatment of control explants at blastula stages resulted in the induction of the mesodermal markers XHex and GATA6 and suppression of the default epidermal fate, whereas treatment at early gastrula stages lead to the induction of cardiac and muscle actin, again accompanied by the suppression of the epidermal fate (). The competence for mesoderm induction was lost by late gastrula stages.
In contrast, Jarid2 depleted embryos failed to induce mesodermal markers in response to Activin treatment at any of the assayed stages, suggesting a differentiation defect rather than altered timing of developmental competence. Remarkably, Jarid2 depleted explants also failed to realize their default developmental potential as evidenced by the absence of epidermal Keratin expression (), indicating that Jarid2 depletion results in differentiation failure.