As shown in Figure Figure2A,2A, the most dramatic effect of erythroid maturation on β-globin position is a steep drop in the number of peripheral loci between stages 2 and 4. While 28% and 36% of alleles are near the nuclear envelope in the immature cell fractions (stages 1 and 2, respectively), this peripheral fraction falls to 8% in the most mature cells. As in previous two-dimensional (2D) studies, loci within 20% of the nuclear radius from the envelope were considered to be peripheral (shell 1) (Kosak et al. 2002). The mean of the distances from the periphery of the most mature cell fraction (stage 4) was statistically significantly different from that of the more immature cells (for stages 1–3, P < 0.001). As the measurements are corrected for nuclear radius, the smaller dimensions of the mature cells cannot account for this shift in position. In addition, if the distribution of β-globin loci was random, we would expect 22% of the alleles to be located within the peripheral shell (Kosak et al. 2002). Thus, the fact that only 8% of βglobin loci are peripheral clearly demonstrates a nonrandom positioning toward the nuclear interior with the final stage of erythroid maturity. To further validate our 2D results, we also performed three-dimensional (3D) ImmunoFISH analyses of the position of the β-globin locus relative to lamin B staining on the four maturation stages. In immature cells (stages 1 and 2), 66% and 75%, respectively, of β-globin loci are in contact with the lamina and thus are perinuclear, whereas at stage 3, 47% of the loci are peripheral, and in the most mature cells (stage 4), only 27% of the loci are peripheral (Supplementary Fig. 1B). These data reinforce the shift in nuclear position of the β-globin locus with erythroid maturation.

Indirect immunostaining shows Pol II exists in freely diffusible form and in distinct foci (transcription factories) that are the sites of the majority of RNA synthesis and contain multiple transcription factors (Grande et al. 1997). These transcription factories are attached to the underlying nucleoskeleton and show very little mobility and may thus represent one facet of nuclear architecture (Kimura et al. 1999).

We performed immunostaining experiments on sorted immature and fully differentiated nucleated erythroid fetal liver cells using an antibody recognizing both hyperphosphorylated and nonphosphorylated Pol II (for review, see Dahmus 1996). In the immature c-kit-positive population (stage 1), Pol II is distributed over hundreds of foci closely packed throughout the nucleoplasm with the exclusion of PCH, nucleoli, and most of the immediate nuclear periphery (Fig. 5a–c). As cells differentiate along the erythroid lineage, Pol II foci gradually diminish in number and no longer fill out most of the nucleoplasm. In the last sorted maturation stage, Pol II staining frequently resembles a cage-like structure, centered mostly in the nuclear interior but also reaching out to the periphery (Fig. 5d–f). These results demonstrate a changing nuclear landscape with erythroid differentiation, reflecting an increasingly restricted compartment available for gene expression.

As shown in Figure Figure7A,7A, an increasing fraction of cells show globin gene expression with progressing maturation. Detectable transcription foci were largely absent in the most immature cells (<2%). Both α (21%) and β-globin (13%) transcripts began to emerge with the appearance of the erythroid cell surface marker ERY-1 (stage 2). However, almost three-quarters of nuclei scoring positive for transcription foci exhibited only a single RNA FISH signal (either α-globin or β^major-globin), suggesting that this cell population is only at the threshold of globin gene activation. By the intermediate stage (stage 3), most globin loci are actively transcribing. At this stage, only 69% of β-globin alleles appear active compared with 83% of α-globin alleles. Thus, α-globin may undergo longer or more frequent transcription bursts, as suggested by previous reports that α-globin expression slightly precedes that of β-globin (de Krom et al. 2002). This notion is supported by the observation that while 60% of nuclei express all α-globin and β-globin alleles (four RNA FISH spots), another 21% show two α-globin transcript signals but only one or no β-globin signal (data not shown). In contrast, in the most mature cell population, 89% of all α-globin and 86% of β-globin alleles exhibit primary transcript signals, and 79% of nuclei contain signals from all four loci. However, we detected a small percentage (~7%) of nuclei without active α-globin or β^major-globin transcription. These nuclei were frequently also the smallest and most compact; the absence of a visible RNA FISH signal may thus signal the final shutdown of globin transcription prior to enucleation. In support of this finding, in a human transcriptional analysis, Brown et al. (2006) observed a decrease in globin gene expression in their most mature isolated erythroid population.

Both a peripheral position or proximity to PCH has been correlated with transcriptional repression during differentiation (Brown et al. 1997; Brown et al. 1999; Francastel et al. 1999; Su et al. 2004). For example, we demonstrated that up to 60% of β-globin loci colocalized with PCH in undifferentiated murine erythroleukemia (MEL) cells in which the β-globin genes are transcriptionally inactive, but only 8% of β-globin alleles contacted PCH after induction of differentiation (Francastel et al. 2001). In contrast, the β-globin locus infrequently colocalizes with PCH in primary fetal liver cells, even in immature cells prior to transcriptional activation. Instead, the locus is often at the nuclear periphery. This discrepancy may reflect a biological difference between the nuclear organization of transformed cell lines and primary cells and/or differences revealed by the current 3D and former 2D analyses. Regardless, the bias of the β-globin locus for the nuclear periphery rather than PCH when inactive in immature fetal liver (i.e., stages 1 and 2) and ES cells might reflect functional differences between these two repressive compartments. In fact, several reports suggest that any particular silenced gene associates either with PCH or the periphery, but rarely both (Skok et al. 2001; Kosak et al. 2002; Zink et al. 2004; Williams et al. 2006). PCH consists of constitutive heterochromatin that replicates late in S phase, is marked by high levels of trimethyl-H3K9 (Dimitrova and Gilbert 1999; Wu et al. 2005), and represents a compartment for stable epigenetic gene repression (as observed in T- and B-cell differentiation) (Brown et al. 1997, 1999; Su et al. 2004). In contrast, the bulk of the chromatin at the nuclear periphery replicates in mid-S phase and is enriched in dimethyl-H3K9, consistent with facultative heterochromatin (Wu et al. 2005). Furthermore, experiments by Hutchison and Weintraub (1985) demonstrated that, unlike PCH, the nuclear periphery is enriched in DNase I-sensitive chromatin, suggesting the presence of poised or active genes. As such, the periphery may harbor gene loci that are silent but responsive to developmental cues to become activated or repressed. In ES cells, in which the globin locus is peripheral, differentiation along the hematopoietic and then erythroid lineages is one possible fate. Therefore, the locus may be competent for activation in the nuclear periphery of ES cells, while in immature fetal liver cells it is poised to activate the globin genes. This is in contrast to activated T cells where the β-globin locus is stably silenced by recruitment to PCH (Brown et al. 2001). We find that the β-globin locus associates with PCH only at a very low frequency in splenic B cells. However, most of these B cells are in a quiescent state, and recruitment of heritably silenced genes to PCH only occurs in cycling but not resting lymphoid cells (Brown et al. 1997, 1999).

One striking finding is that robust β-globin expression (commencing at stage 2) precedes the most dramatic relocation of the β-globin locus away from the nuclear periphery (stages 3 and 4). In fact, many β-globin loci that associate with strong transcription foci colocalize with the nuclear lamina. Thus, our results provide further evidence that robust transcription is possible in the immediate vicinity of the nuclear periphery in mammalian systems, and that the relocation of the β-globin locus to the nuclear interior is more likely a consequence of transcription rather than the basis for transcription.

Although the nuclear periphery has been described traditionally as a repressive compartment and repository for gene-poor chromosome regions (Gottschling et al. 1990; Gotta and Gasser 1996; Andrulis et al. 1998, 2002; Laroche et al. 1998; Croft et al. 1999; Galy et al. 2000; Boyle et al. 2001), several recent studies in yeast have demonstrated that repressive and activating domains can coexist at the periphery. A genome-wide analysis revealed that a subset of highly active genes (e.g., induced GAL genes) associate with the nuclear export machinery and are relocated to the periphery (Casolari et al. 2004). In addition, the majority of α-factor-induced genes, involving loci from multiple chromosomes, associate with components of the nuclear envelope upon activation, suggesting a global nuclear reorganization (Casolari et al. 2005). INO1, a nucleoplasmic gene locus when repressed, relocates to the periphery when activated by the unfolded protein response (Brickner and Walter 2004). Furthermore, Ishii et al. (2002) demonstrated that genes can be tethered to the periphery and still be transcriptionally active when protected by boundary activities conferred by components of the nuclear transport machinery. More recently, Schmid et al. (2006) reported that a transient association of genes with the nuclear pore basket may represent an integral part of the activation process for many genes.

In mammalian systems, the periphery is largely considered to be a transcriptionally inactive compartment, particularly in undifferentiated or progenitor cells (Imai et al. 1997; Croft et al. 1999; Boyle et al. 2001; Kosak et al. 2002; Zhou et al. 2002; Zink et al. 2004; Williams et al. 2006). Moreover, movement away from the periphery has in several instances been correlated with transcriptional activity. For example, in a Chinese hamster ovary (CHO) cell clone, a lacO array relocates away from the periphery when targeted by VP16 and induced to express a reporter gene (Tumbar and Belmont 2001). Rudolph and colleagues (Zink et al. 2004) showed that genes adjacent to CFTR independently adopt tissue-specific positions relative to the nuclear periphery that reflect their individual transcriptional state with interior location correlating with activity. The Mash1 locus is positioned at the periphery when repressed in ES cells but is relocated to the interior when it becomes transcriptionally active during neuronal differentiation (Williams et al. 2006). However, Fisher’s group (Hewitt et al. 2004) has also shown that the mammalian nuclear periphery is not universally repressive, as the IFNγ gene remains peripheral in developing T-helper cells (as well as B cells and fibroblasts) even when transcriptionally active.

It is tempting to invoke the gene gating hypothesis, first proposed by Blobel (1985), to explain the significant number of β-globin loci we observe expressing at the nuclear periphery (e.g., >60% of active loci at stage 2 are in contact with the nuclear lamina). Placement of active genes in the proximity of nuclear pore complexes (NPCs) would ensure the rapid export of mature transcripts from the nucleoplasm and thus promote efficient downstream processing. This proximity could be particularly advantageous during the earliest bursts of β-globin gene expression, when it may be imperative to quickly reach a threshold of β-globin protein synthesis to balance α-globin levels. In fact, as assayed by ImmunoFISH, peripheral loci also colocalize with foci of NPC components (data not shown). While this observation is in accord with the gene gating model, NPCs are embedded within the lamina and are present at high density in the erythroid cells. Thus, at this resolution, it is not possible to distinguish association of the β-globin locus with the lamina versus with NPCs. Furthermore, the fraction of transcriptionally active β-globin loci at the periphery decreases with erythroid maturation and higher frequency of expression. This result is more consistent with a model of stochastic initiation of globin gene activation (de Krom et al. 2002), where a fraction of loci may be triggered while still at the periphery and subsequently relocate to the interior. On the other hand, the β-globin locus may not need to be near the nuclear periphery in the more mature erythroblasts with their high levels of globin transcripts and steady flow of newly synthesized RNA to the cytoplasm.

We thank Arthur Sytkowski for the generous gift of the ERY-1 antibody, David Scalzo for discussions on and assistance with statistical analyses, and Barbara Trask, Steven Henikoff, Tomoyuki Sawado, and Steven Kosak for critical reading of the manuscript. This work was supported by American Society of Hematology Scholar and Cooley’s Anemia Foundation Awards (to M.A.B.) and NIH Grants DK 44746 and HL57620 (to M.G.).