Here we have identified proliferation and quiescence signatures of HSCs. Our experimental design utilized a combination of pair-wise comparisons and time course microarray experiments. The pair-wise analysis allowed us to find the genes different between quiescent and cycling HSCs, while the time course data allowed us to order these genes in a time-dependent manner. The power of our overall methodology is reflected in the remarkable overlaps between the gene lists presented and those extracted from published data (Akashi et al. 2003
), in particular the common P-sig and common Q-sig.
Applying a novel approach to utilizing the GO annotations, we calculated the statistical significance of the enrichment of particular GO categories in our lists. We also devised a new method for calculating the distance between gene lists based on the GO structure. This allows one to assess the functional similarity, in “GO space,” of gene lists that may not have any actual genes in common (such as our TOM groups). Furthermore, since the GO vocabulary is not specific to any species, this method allows for cross-species and cross-platform comparisons of gene lists. Re-analysis of data from previous studies may reveal a functional stem cell signature in GO space that was not evident at the gene level (Ivanova et al. 2002
; Ramalho-Santos et al. 2002
; Fortunel et al. 2003
Applying GO analysis to the TOM groupings revealed elemental subgroups within the signature lists that allowed us to construct a molecular model of the HSC activation cycle. The majority of unperturbed HSCs reside in a quiescence niche and express receptors, for example the metabolism- and ageing-associated receptor IGF1R and the receptor tyrosine kinase Tie1, that allow them to respond to multiple mitogenic signals (A). They also express high levels of transcription factors, such as c-fos and GATA-2, that enable swift activation of HSCs. This expression profile, found in the TOM 0 and 30 groups, suggests that although adult HSCs are quiescent, they are in a “state of readiness” to react to changes in their environment.
Immediately after activation is triggered (here by 5FU), HSCs enter a superquiescent “pause.” This state, found at TOM 1 and also observed by cell cycle analysis (Randall and Weissman 1997
), is mediated by antiproliferative genes such as Tob1, p21,
(B). Interestingly, p21
-null mice have defects in HSC self renewal (Cheng et al. 2000
). We observed up-regulation of TIMP3 and the serine proteinase inhibitor A-3 g, which inhibit cell migration (Qi et al. 2003
). At least six interferon-γ-induced genes were also up-regulated at this point, suggesting that HSCs are responding to proinflammatory signals. We speculate that the pause in HSC proliferation and migration allows HSCs to survive 5FU cytotoxicity while the cells simultaneously “prepare” to proliferate and repopulate the bone marrow to ensure survival of the animal.
In the early phase of proliferation starting at day 3, when increased numbers of HSCs in cell cycle are first detected (Randall and Weissman 1997
), HSCs have committed to cell division, as can be seen by the maximal expression of genes involved in DNA replication and repair (C). At day 6, the late phase of proliferation, when the greatest number of HSCs are in cycle, we see expression of genes involved with energy production, indicating an overall increase in metabolic activity in the HSCs (D). Prior work has linked HSC mobilization with proliferation (Wright et al. 2001
; Heissig et al. 2002
), and our data indicate that the opposite is also true: to proliferate, HSCs need to move out of their quiescence niche and into a proliferative zone (C and D). We see the up-regulation of α4-integrin
at day 3 followed by a dramatic decrease at day 6 post 5FU treatment. Experiments that block α4-integrin function by blocking antibodies or via knockout technology have previously shown that down-regulation induces increased mobilization and delays recovery after 5FU treatment (Craddock et al. 1997
; Scott et al. 2003
). The gene expression pattern displayed by α4-integrin predicts that down-regulation of α4-integrin is necessary for 5FU-induced proliferation. As stated above, down-regulation of α4-integrin is sufficient to alter recovery of bone marrow progenitors after 5FU treatment, supporting the link between HSC proliferation and migration in our model. Down-regulation of c-Kit has also been linked to mobilization of HSCs (Heissig et al. 2002
), and its expression is lowest at day 6 post treatment.
In order to “reset quiescence,” HSCs need to return to their niche (E). This process begins at day 10, when the number of cycling HSCs falls and HSCs express the high levels of specific antiproliferative genes such as Btg1
and several components of the JAK/STAT signal transduction pathway. Both SOCS3 (Soriano et al. 2002
) and STAT3 (Levy and Lee 2002
) have been associated with both positive and negative regulation of proliferation and differentiation of various hematopoietic cell types. Simultaneous expression of SOCS3, STAT3,
suggests a complex regulation of HSC quiescence, but earlier work examining STAT signaling in other stem cell populations gave us insight into the role of JAK/STAT signaling in HSCs. Expression of STATs has been shown to establish and maintain stem cell pluripotency in embryonic stem cells (Raz et al. 1999
). However in Drosophila
testes, JAK/STAT activation is crucial for stem cell self renewal; perturbations by both loss and increase in expression lead to dramatic changes in the stem cell compartment (Kiger et al. 2001
). Notably, activation of the JAK/STAT pathway by PKD1 induces cell cycle arrest through p21-dependent mechanisms (Bhunia et al. 2002
). This supports our hypothesis that JAK/STAT signaling is important for inducing quiescence at day 10, since we have shown that p21 is likely involved in HSC cell cycle arrest. The involvement of JAK/STAT signaling in both stem cell pluripotency and HSC quiescence suggests that these processes may be linked in HSCs.
Endoglin, also found in the TOM 10 group, is known to be expressed on both murine (Chen et al. 2002
) and human (Pierelli et al. 2001
) HSCs, and has been shown to decrease cell migration by increasing cell–cell adhesion (Liu et al. 2002
). Its expression pattern was negatively correlated with 5FU-HSC proliferation: it was lowest at day 6 after 5FU treatment, and highest at day 10. Our data suggest that HSC proliferation requires mobilization from the niche, and that restoration of quiescence is accompanied by a return to the niche. Endoglin's expression pattern makes it an ideal candidate for mediating HSC-to-niche homing and long-term association.
Our model derived from gene expression profiles correlates well with the literature on HSC cell cycle and mobilization. Although other models of HSC mobilization and 5FU treatment have previously been proposed (Heissig et al. 2002
), our data allow association of specific genes with particular stages of HSC activation and recovery. Our model predicted that CD48 might preferentially mark cycling HSCs, and our cell cycle analysis of CD48+
HSCs confirmed this prediction. Our model also postulates the presence of “quiescence” and “proliferative” zones in the bone marrow; osteoblasts may be a component of this quiescence niche (Calvi et al. 2003
; Zhang et al. 2003
In summary, we present proliferation and quiescence signatures for HSCs that show remarkable overlap with published literature. In addition, this study revealed new, uncharacterized genes whose role in HSC self renewal needs to be explored: some of these genes may play as yet undiscovered roles in the development of cancer or may aid in the manipulation of stem cells for therapeutic uses. Finally, harnessing the GO using novel bioinformatics approaches to analyze our data at a global level allowed us to propose a model of the HSC activation cycle.