Re-entry into quiescence protects hematopoietic stem cells from the killing effect of chronic exposure to type I interferons

To understand how acute and chronic IFN-1 exposure affects hematopoiesis, we analyzed the changes occurring in the BM of WT mice injected with poly I:C every second day over a time course ranging from 1 d (1 injection) to 30 d (15 injections; Fig. 1 A). We confirmed surface expression of the IFNAR1 receptor subunit on all mature and immature BM compartments (not depicted), and IFN-1 production in the sera of poly I:C–treated mice (Fig. 1 A). After an acute spike of IFN-1s produced by the first poly I:C injection, we measured low but detectable chronic production of IFN-1s from 3 to 30 d of continuous poly I:C treatment. In response to IFN-1 production, we observed a rapid and sustained IFNAR-dependent depletion of BM cells, corresponding to a decrease in mature Gr-1⁺/Mac-1⁺ granulocytes (Gr) and B220⁺ B lymphocytes (B; Fig. 1, B and C; and Fig. S1 A). This BM cell depletion was associated with a widespread and sustained IFNAR-dependent apoptosis in the BM cavity as measured by immunofluorescence staining for cleaved caspase-3 (CC3; Fig. 1 D). We also found an IFNAR-dependent increase in serum levels of proinflammatory cytokines, including TNF and the IFN-inducible chemokine MCP-1 in poly I:C–treated mice (Fig. 1 E). These systemic changes were associated with a rapid decrease in lymphoid cells in the blood and peritoneal cavity, and a late increase in Gr in the spleen and peritoneal cavity (Fig. 1 F), which together argue against a major mobilizing effect, at least during the first 2 wk of poly I:C treatment. Collectively, these results demonstrate that 30 d of continuous poly I:C treatment results in a sustained BM aplastic phenotype in the context of elevated BM apoptosis and chronic systemic inflammation, similar to what is observed in human patients with chronic IFN-1 production (Gokce et al., 2012). We therefore consider 1–3 d of poly I:C treatment to represent acute IFN-1 exposure, and 5–30 d of continuous poly I:C treatment to accurately mimic chronic IFN-1 exposure.

We then addressed how chronic IFN-1 exposure affects the number of HSCs and other early progenitor cells in the BM. Similar to a previous study using IFN-γ (Zhao et al., 2009), we observed a significant expansion in the frequency of Lin⁻/c-Kit⁺/Sca1⁺ (LSK) cells in poly I:C–treated mice, which normally consists of Flk2⁺ LSK multipotent progenitors (MPP^F+), Flk2⁻/CD48⁺ LSK MPPs (MPP^F-), and Flk2⁻/CD48⁻/CD150⁺ LSK HSCs (Fig. 2 A and Fig. S1 B). This increase mainly reflected an elevated frequency of MPP^F− in the LSK compartment of poly I:C–treated mice, and was accompanied by a substantial decrease in the frequency of CD34⁺/FcγR⁻/Lin⁻/c-Kit⁺/Sca1⁻ common myeloid progenitors (CMP) in the Lin⁻/c-Kit⁺/Sca1⁻ myeloid progenitor (MP) compartment (Fig. 2 A). These changes were entirely dependent on IFNAR signaling and not observed in the BM of poly I:C–injected Ifnar1^−/− mice (Fig. 2 B). As a consequence of this increase in MPP^F− frequency, the percentage of HSCs in the LSK compartment of poly I:C–treated mice also decreased over time (Fig. S1 B). However, when the mean cell number for each of these populations was calculated based on their frequency in the BM and the actual BM cellularity at each time point, we found a significant increase in LSK number, whereas the total number of HSCs remained essentially unchanged in poly I:C–treated mice even after 30 d of continuous treatment (Fig. 2 C).

Because IFN-1s are well-known inducers of Sca-1 expression (Dumont and Coker, 1986) resulting in increased Sca-1 levels over the poly I:C treatment period (Fig. S1 B), we directly investigated the purity of the HSC compartments using the HSC-specific marker ESAM (Ooi et al., 2009). As in untreated HSCs, we found that the vast majority of IFN-1–exposed HSCs were ESAM⁺ (Fig. 2 D), hence confirming their positive identification and lack of contamination by ESAM⁻ MPs (Fig. 2 E). In contrast, we observed an increased proportion of ESAM⁻ MPs in the phenotypic MPP^F− gate (Fig. 2 D). To gain further insights into the identity of these contaminating cells, we directly treated naive HSCs and early progenitors with IFN-α (100 ng/ml) for 12 h in vitro (Fig. 3 A). As expected, IFN-1s increased Sca-1 expression on already Sca-1-expressing HSCs, MPP^F+ and MPP^F− (Fig. 3 B), but also reactivated Sca-1 expression on initially Sca-1 negative myeloid progenitors CMPs, GMPs, and MEPs (Fig. 3 C). In line with our in vivo data, contaminating progenitor cells that have regained or augmented Sca-1 expression only increased the frequency of MPP^F−, but not the frequency of HSCs, within the LSK gate (Fig. 3, B and C). Importantly, no other surface markers defining HSC identity were affected by addition of IFN-α, and none of the IFN-1–treated progenitors acquired a Flk2⁻/CD48⁻/CD150⁺ HSC surface phenotype. These results indicate that the phenotypic distinction between MPs and MPPs becomes blurred upon IFN-1 exposure, and that the LSK expansion in poly I:C–treated mice is mainly the result of contaminating MPs that have reacquired Sca-1 expression. They also show that the Flk2⁻/CD48⁻/CD150⁺ HSC compartment remains free of contaminating progenitor cells even in conditions of IFN-1–driven Sca-1 expression. Collectively, these data indicate that IFN-1s induce a rapid phenotypic BM remodeling, particularly in the LSK compartment, but do not deplete HSCs.

Acute IFN-1 exposure rapidly induces HSC proliferation (Essers et al., 2009), but whether HSCs continue to cycle in the presence of chronic IFN-1s is unknown. To investigate how IFN-1s affect HSC cell cycle activity over time, we injected 5-bromodeoxyuridine (BrdU) 3 h before sacrificing poly I:C–treated mice and measured BrdU incorporation in HSCs (Fig. 4 A and Fig. S2 A). Consistent with prior results (Essers et al., 2009), we found that IFN-1s rapidly induced HSC proliferation by 12 h, with BrdU incorporation peaking after 1–3 d of poly I:C treatment. However, HSC BrdU incorporation rapidly returned to basal levels by 5 d of poly I:C treatment, and then largely remained steady even after 30 d of continuous poly I:C administration (Fig. 4 B). We then characterized the cell cycle distribution of HSCs during this period using intracellular Ki67/DAPI FACS staining (Fig. S2 B). In agreement with the BrdU results, we observed a significant decrease in quiescent (G₀) HSCs by 3 d of poly I:C treatment, with a corresponding increase in the proportion of HSCs in G₁ and S-G₂/M phases of the cell cycle (Fig. 4 C). However, the HSC compartment returned to a predominantly quiescent phenotype by 7 d of poly I:C treatment, though this quiescent phenotype appeared to gradually erode over prolonged IFN-1 exposure. We also observed a similar transient increase in HSC proliferation in WT mice injected with 1 × 10⁴ units of mouse recombinant IFN-α every 12 h for up to 7 d, with increased BrdU incorporation by day 1 and a return to steady-state levels by day 7 (unpublished data). This indicates that the short HSC cycling response is not restricted to poly I:C treatment or is the direct consequence of the spike in IFN-1 production after the initial poly I:C injection (Fig. 1 B). We also generated WT:Ifnar1^−/− BM chimeric mice to directly test the involvement of IFNAR signaling. We confirmed the 50:50 distribution of WT:Ifnar1^−/− cells in the blood and HSC compartment 2 mo after transplantation (unpublished data), and used BrdU incorporation to analyze the proliferation of WT and Ifnar1^−/− HSCs side-by-side in the same chimeric animal after poly I:C injection (Fig. 4 D). Notably, WT but not Ifnar1^−/− HSCs proliferated for a brief period of time, thus demonstrating that the transient IFN-mediated HSC proliferation is a direct consequence of IFNAR signaling. Collectively, these results indicate that IFN-1 exposure induces a rapid but transient, rather than sustained, proliferation in HSCs, which quickly return to a quiescent state during chronic exposure.

We then investigated how chronic IFN-1s could induce transient HSC proliferation in vivo. We first assessed the status of the AKT/PKB–FoxO3a pathway, which is directly activated by IFN-1 (Yamazaki et al., 2006; Essers et al., 2009) and is essential to regulate the switch between quiescence and proliferation in HSCs (Pietras et al., 2011), by measuring AKT phosphorylation levels in HSCs using intracellular FACS staining (Fig. S2 C). Interestingly, we found that the kinetics of AKT phosphorylation mirrored those of HSC proliferation, with a peak in phosphorylation at day 3, followed by a return to basal phospho-AKT levels by day 7 of poly I:C treatment (Fig. 5 A). Furthermore, the quiescence-enforcing transcription factor FoxO3a was predominantly sequestered in the cytoplasm and therefore inactivated at day 3 when HSCs were cycling, and relocalized in the nucleus at day 7 when HSCs had relapsed into quiescence (Fig. 5 B). We then used a custom-made SA Biosciences PCR array to assess expression of cell cycle genes and key genes associated with quiescence-enforcing programs in HSCs at days 0, 3, 7, and 30 of poly I:C treatment (Fig. 5 C). We confirmed induction of IFN-1 target genes by poly I:C treatment at all time points, while expression of IFNAR1 and other IFN signaling pathway components did not decrease (Fig. 5 C and not depicted), indicating that HSCs were capable of responding to IFN-1s throughout the time course. Notably, expression of Irgm1, a negative regulator of IFN-mediated HSC cell cycle activity (King et al., 2011), was maximal in day 7 when HSCs had reentered quiescence (Fig. 5 C). However, except for IFN-1 targets, we observed a broad transcriptional repression throughout the treatment time course, which was noticeably more accentuated in cycling day 3 HSCs. In particular, we did not find substantial induction of cyclin and cyclin-dependent kinase (CDK) genes that drive HSC proliferation other than a transient increase in Ccne1 (cyclin E1) expression in cycling day 3 HSCs (Fig. 5 D). Instead, we found significantly decreased expression of many quiescence-enforcing genes, including the cyclin-dependent kinase inhibitors (CKI) Cdkn1b (p27) and Cdkn1c (p57), the transcription factors Foxo1, Foxo3a, and Trp53, and the AKT inhibitor Pten in cycling day 3 HSCs, which subsequently returned to steady-state levels at day 7 when HSCs had relapsed into quiescence (Fig. 5 E and not depicted). Moreover, we observed a similar trend in two other quiescence-enforcing signaling mechanisms, with a transient decrease in expression of TGF-β and Notch pathway components and targets in cycling day 3 HSCs and restoration in day 7 HSCs that have reentered quiescence (Fig. 5 F). Collectively, these results demonstrate that cell cycle machinery is not directly activated by IFN-1s in vivo and, instead, that multiple quiescence-enforcing mechanisms are transiently antagonized and subsequently reactivated during chronic IFN-1 exposure.

We next tested whether IFN-1s could also exert a proproliferative effect in vitro by culturing purified HSCs in the presence of increasing concentrations of IFN-α. We found that addition of IFN-α did not enhance HSC proliferation in vitro, and directly antagonized HSC growth at a 100 ng/ml concentration, a dose previously shown to induce HSC proliferation in vitro in IFN-γ (Baldridge et al., 2010; Fig. 6 A). We confirmed this result using methylcellulose cultures of purified HSCs, where treatment with 100 ng/ml IFN-α significantly decreased cloning efficiency and colony size (Fig. 6 B). We next measured BrdU incorporation after 12 h of culture with or without 100 ng/ml IFN-α (Fig. 6 C), and found no increase in HSC proliferation upon direct IFN-1 stimulation. To test whether signals from other BM cells are required to induce HSCs to proliferate in response to IFN-1 in vivo, we co-cultured purified and CFSE-labeled WT or IFNAR-deficient HSCs with unfractionated WT BM cells in the presence or absence of IFN-α, and analyzed BrdU incorporation by flow cytometry during the culture period (Fig. 6 D and Fig. S3 A). However, even in this co-culture system, IFN-α did not enhance HSC proliferation or cell division at low concentrations and remained antiproliferative at higher concentrations (Fig. 6 D and not depicted). This inhibitory effect required IFNAR expression on HSCs, as BrdU incorporation by co-cultured Ifnar1^−/− HSCs/BM cells was unaffected by IFN-α compared with WT HSCs/BM cells (Fig. 6 D). Moreover, co-culture of WT HSCs with BM cells from 1-d poly I:C–treated mice also reduced HSC proliferation (unpublished data). To further try and recapitulate the conditions in the BM niche, we co-cultured purified Violet-dye–labeled HSCs on OP9-DL1 stromal cells in the presence or absence of IFN-α and analyzed EdU incorporation during the 12-h culture period by immunofluorescence (Fig. 6 E and Fig. S3 B). However, even in these conditions, IFN-α inhibited HSC proliferation in vitro. This antiproliferative effect appeared to be a property common to all IFNs, as HSC proliferation after 12 h of culture with 100 ng/ml IFN-γ was also inhibited (unpublished data). We then investigated the expression of cell cycle genes and key genes associated with quiescence-enforcing programs in HSCs cultured for 12 h in the presence or absence of IFN-α (Fig. 6 F). As expected, we found no evidence of activated cell cycle machinery in HSCs cultured with IFN-α, but also did not observe the relaxation of quiescence-enforcing mechanisms occurring in day 3 cycling HSCs in vivo. Instead, we found significantly decreased levels of Ccnd1 (Cyclin D1) and increased expression of Cdkn1b (p27) and Cdkn1c (p57), similar to in HSCs cultured with IFN-γ (de Bruin et al., 2013). Together, these results demonstrate that direct IFN-1 stimulation, regardless of the concentration, fails to induce HSC proliferation in vitro. This reinforces our finding that the transient proliferative effect observed in vivo likely results from a relaxation of multiple HSC quiescence-enforcing mechanisms rather than a direct activation of the cell cycle machinery.

Because loss of function has been associated with HSC proliferation induced by IFNs (Essers et al., 2009; King et al., 2011), we investigated whether the reacquisition of quiescence in chronic IFN-1–exposed HSCs confers functional protection. We used HSCs isolated from mice treated with poly I:C for either 3 (cycling) or 7 (quiescent) days to assess their short-term potential in vitro and long-term potential in vivo (Fig. 7 A). Strikingly, both colony formation in methylcellulose and expansion rates in liquid culture were severely impaired in IFN-1–exposed HSCs regardless of the duration of poly I:C treatment and HSC cell cycle status at the time of harvest (Fig. 7 B). Consistent with these in vitro results, we found similarly decreased engraftment in recipient mice transplanted with IFN-exposed day 3 (cycling) or day 7 (quiescent) HSCs (Fig. 7 C). We confirmed that defective reconstitution was not the result of changes in lineage distribution (not depicted), but was caused by reduced BM chimerism and the number of engrafted donor-derived IFN-1–exposed HSCs (Fig. 7, D and E). Moreover, we found similar functional impairments in HSCs isolated from mice treated for 30 d with poly I:C (unpublished data). Collectively, these data indicate that chronic IFN-1 treatment impairs HSC function regardless of the length of exposure and its transient effect on HSC proliferation in vivo.

We next assessed whether the loss of function observed in IFN-1–exposed HSCs could be caused by induction of apoptosis. IFN-1s have been shown to be proapoptotic in numerous cell types, including hematopoietic cells (Kiladjian et al., 2011), but their effect on HSC survival has never been directly tested. We first measured apoptosis levels in HSCs isolated from mice treated with poly I:C for either 3 or 7 d, and analyzed either directly upon harvest or after 12 h in vitro culture to force quiescent HSCs to proliferate (Fig. 8 A). When assayed directly upon isolation, we only observed increased CC3 activity in cycling day 3 HSCs, in contrast to quiescent day 7 HSCs (Fig. 8 B). However, upon culture, both day 3 and 7 HSCs exhibited a significant and IFNAR-dependent increase in CC3 activity (Fig. 8 C). Consistently, we also found elevated CC3 levels in HSCs isolated from day 7 IFN-α–injected mice after 12-h in vitro culture (Fig. 8 D). These proapoptotic effects were reversible, as a 2-wk recovery period after 7 d of poly I:C treatment in vivo (day 7 + 2 wk) largely restored HSC survival and function in methylcellulose cultures and upon transplantation (Fig. 8, E and F). To gain molecular insights, we then used our SA Biosciences PCR array and direct qRT-PCR analyses to investigate the status of the apoptotic machinery and expression of Bcl2 family members in IFN-1–exposed HSCs. Strikingly, we did not observe increased expression of any of the proapoptotic genes in freshly isolated day 3 HSCs (Fig. 8 G). Instead, we found a significant decrease in the expression of the essential prosurvival gene Mcl1 in cycling day 3 HSCs, which likely contributes to their decreased survival compared with quiescent day 7 HSCs. In sharp contrast, day 7 HSCs cultured for 12 h showed a rapid and significant increase in expression of proapoptotic Bax, Puma, and Noxa genes (Fig. 8 H). Collectively, these results indicate that IFN-1 exposure, regardless of its duration, poises HSCs for apoptosis but only results in direct cell killing in proliferating HSCs either due to a reduction in prosurvival factors in cells with transiently antagonized quiescence-enforcing mechanisms, or to activation of a proapoptotic gene program in cells that are actively forced to enter the cell cycle.

IFN-1s are known to regulate the proapoptotic machinery via activation of p53 (Takaoka, et al., 2003). To probe the role of p53 in mediating the effects of IFN-1s on HSCs, we treated HSCs isolated from control (Ctrl) or Trp53^−/− mice with IFN-α for 12 h in vitro (Fig. 9 A). Although IFN-α exposure induced CC3 activity and triggered Bax, Puma, and Noxa expression in control HSCs (Fig. 9, B and C), no CC3 induction and significantly decreased activation of the proapoptotic genes was found in IFN-α–treated Trp53^−/− HSCs (Fig. 9, B and D). Moreover, no CC3 induction was observed in BakBax-deficient HSCs upon treatment with IFN-α for 12 h in vitro (Fig. 9 B), thus validating the role of the intrinsic mitochondrial apoptotic pathway in the proapoptotic effects of IFN-1s on HSCs. Because p53 is also a critical enforcer of HSC quiescence (Pietras et al., 2011), we next investigated its importance for the in vivo proproliferative effect of IFN-1s on HSCs by measuring BrdU incorporation in poly I:C–treated control or Trp53^−/− mice (Fig. 9 E). However, IFN-1 exposure induced HSC proliferation with similar kinetics in both control and Trp53^−/− mice, and HSCs from Trp53^−/− mice showed no significant delay in their return to quiescence compared with control HSCs (Fig. 9 F). Moreover, we observed similar results upon poly I:C treatment of Foxo3a^−/− mice, which lack another important HSC quiescence-enforcer (Fig. 9 F). Collectively, these results demonstrate a key function of p53 in regulating the mitochondrial proapoptotic machinery in response to IFN-1 stimulation and indicate that no single transcription factor regulates the switch from quiescence to proliferation that briefly accompanied IFN-1 exposure in vivo, which instead relies on the concerted deactivation/reactivation of multiple quiescence-enforcing mechanisms (Fig. 9 G).

Lastly, we investigated whether quiescence reentry during chronic IFN-1 treatment could protect IFN-exposed HSCs from the killing effect of myeloablative drugs and adapted the protocol previously used to demonstrate the impaired function of proliferating HSCs after acute IFN-1 exposure (Essers et al., 2009). Instead of injecting 5-fluorouracil (5-FU) at varying time points after acute (3-d) treatment with poly I:C, we injected mice treated with poly I:C for either 3 or 7 d with a single dose of 5-FU (150 mg/kg) directly after the poly I:C treatment period to study the impact of the initial cell cycle status of IFN-exposed HSCs on their survival and regeneration potential (Fig. 10 A). We also injected mice with 5-FU in the middle of a 7+7 d poly I:C treatment course to directly assess the effect of chronic IFN-1 exposure on HSC survival and regeneration potential (Fig. 10 A). As expected, we found significantly decreased survival in mice injected with 5-FU after acute IFN-induced HSC proliferation (day 3), compared with untreated mice (day 0) or mice with chronic IFN-exposed HSCs that had reentered quiescence (day 7; Fig. 10 B). Besides some delay in kinetics, we also observed recovery of all blood parameters in day 7 mice treated with 5-FU, indicating that IFN-exposed HSCs are still able to maintain blood production (Fig. 10 C). Strikingly, forcing IFN-exposed HSCs to proliferate in the constant presence of IFN-1s (day 7 + 7) resulted in a dramatic decrease in viability with 80% of the mice dead by 15 d after 5-FU injection (Fig. 10 B), and severely impaired blood recovery by 12 d after 5-FU injection (Fig. 10 C). Analyses performed at 10 d after 5-FU injection confirmed a substantial loss of BM cellularity in day 7 + 7 mice, with reduced Gr and B cell numbers compared with non–5-FU injected control mice and recovering day 0 and 7 mice (Fig. 10 D). We also quantified HSC numbers in these conditions using stringent HSC gating and the ESAM marker to confirm the absence of contaminating Sca-1–expressing progenitors (Fig. S4). Although recovering day 0 and 7 mice both showed increased HSC numbers, the HSC pool in day 7 + 7 mice was severely depleted to levels even lower than in non–5-FU injected control mice (Fig. 10 E). Collectively, these data indicate that whereas relapsing into quiescence protects IFN-1–exposed HSCs from apoptosis, forcing them to cycle in the constant presence of IFN-1s results in their depletion and subsequent inability to maintain blood homeostasis (Fig. 10 F).

In this study, we elucidate some of the complex effects of IFN-1s on HSC function. We show that HSC proliferation in vivo is a transient event despite chronic exposure to IFN-1s, which results from a brief down-regulation of a large number of quiescence-enforcing mechanisms that only occurs in vivo in the BM niche and not in vitro upon HSC culture. Importantly, we find that IFN-exposed HSCs quickly return to a quiescent state, where they remain largely protected from the apoptotic effects of IFN-1s unless forced to reenter the cell cycle after transplantation, culture, or regeneration from myeloablative treatments, at which point they activate a p53-dependent proapoptotic gene program. Altogether, our findings resolve numerous ambiguities related to the proliferative and suppressive effects of IFN-1s on HSC function. They identify the primary effect of chronic IFN-1s on HSCs as proapoptotic rather than proproliferative, and demonstrate that reentry into quiescence acts as a safeguard mechanism ensuring the survival of the HSC pool during chronic IFN-1 exposure.

Although the ability of acute IFN-1s to induce HSC proliferation is well known, as is the capacity of HSCs to return to quiescence after a single poly I:C injection (Essers et al., 2009), the effects of chronic IFN-1s on HSC cell cycle status have not been closely investigated. The observation that WT HSCs are selectively lost in WT:Ifnar1^−/− BM chimeric mice upon poly I:C treatment supported the notion that chronic IFN-1 exposure could induce HSC proliferation and thereby their subsequent functional exhaustion (Essers et al., 2009). The idea that HSCs would proliferate continuously in the presence of IFNs is further supported by two studies directly investigating HSC proliferation 4 wk after either chronic Mycobacterium avium bacterial infection (which induces IFN-γ production) or continuous treatment with purified IFN-α, which both found elevated HSC cell cycle activity using BrdU incorporation and Ki67/Hoechst staining, respectively (Baldridge et al., 2010; Mullally et al., 2013). However, these studies only used a CD150⁺ LSK as phenotypic definition for HSCs that, as we show here, is not sufficient in the context of IFN exposure to exclude contamination by some actively cycling myeloid progenitors that have reacquired Sca-1 expression. In contrast, using the more stringent Flk2⁻/CD48⁻/CD150⁺ LSK phenotypic definition for HSCs and validation with the additional HSC marker ESAM, we find that after a short burst of acute proliferation, IFN-treated HSCs quickly reenter quiescence during chronic poly I:C treatment and then mostly remain dormant, despite continued IFN-1 exposure. Moreover, we show that the number of phenotypic HSCs remains constant through this period, suggesting that the depletion of WT HSCs in WT:Ifnar1^−/− BM chimeric mice is likely the result of other IFN-1–mediated effects such as apoptosis, rather than continuous proliferation. Thus, our data support the idea that IFN-1–induced HSC proliferation is a transient event even in the presence of chronic IFN-1s, and is not sufficient to deplete the HSC pool.

The finding that IFN-1 induces acute HSC proliferation in vivo is paradoxical given that IFN-1 has long been described as an inhibitor of cell cycle activity in multiple cell types, including hematopoietic progenitors (Verma et al., 2002; Kiladjian et al., 2011). Indeed, we show that IFNs themselves are not intrinsically proproliferative, as neither IFN-α nor IFN-γ (de Bruin et al., 2013) are able to stimulate purified HSCs to proliferate in vitro, regardless of the concentration used. In addition, we found that IFN-α treatment directly blocks the cell cycle machinery in cultured HSCs, most likely by inducing expression of the CKIs p27 and p57, and that co-culture of HSCs with unfractionated BM cells or stromal cells does not reverse this direct antiproliferative effect of IFN-α exposure. These results suggest that IFNs do not directly trigger the cell cycle machinery beyond the level of activation that already occurs in culture, and highlights the limitations of in vitro culture systems in accurately reproducing the physiological BM niche. They are also in sharp contrast with another published work showing increased BrdU incorporation in HSCs cultured with BM cells in the presence of IFN-γ for 12 h (Baldridge et al., 2010). However, these HSCs were identified after culture using the CD150⁺ LSK phenotypic definition and, therefore, are likely contaminated by some actively cycling myeloid progenitors that have reacquired Sca-1 expression upon IFN-γ treatment. Collectively, these results indicate that the classical role of IFN-1s as anti-proliferative cytokines applies to HSCs, but only in vitro and not in an in vivo context.

So, why do IFN-1s induce transient HSC proliferation in vivo? Although we observe some of the previously reported changes in cell cycle genes in acutely proliferating IFN-1–exposed HSCs (i.e., increased Ccne1 expression and decreased Pten levels; Essers et al., 2009), we also could not identify a broad gene signature reflecting direct activation of the proliferation machinery in vivo. Instead, we find decreased expression of many genes enforcing quiescence, including Foxo3a, p53, p27, p57, and components of the Notch and TGFβ pathways in acutely proliferating IFN-1–exposed HSCs in vivo. Consistently, we observe a complete recovery in the expression of quiescence-enforcing genes when HSCs reenter quiescence during chronic IFN-1 exposure. These changes likely reflect transient inactivation of quiescence-enforcing mechanisms to a threshold that allows a brief proliferation of the HSC pool. Moreover, our data suggest that this effect is based on inactivation of a broad set of mechanisms, as genetic deficiency in individual quiescence-enforcing transcription factors such as Trp53 and Foxo3a are not sufficient to perturb the cell cycle behavior of IFN-1–exposed HSCs. As we do not observe decreased expression of IFNAR or IFN-1–regulated genes, the return of the HSC pool to quiescence likely reflects the activation of mechanisms known to restrain HSC proliferation in response to IFN-1s, such as Irgm1, Irf2, CD81, and/or TGFβ signaling (Sato et al., 2009; King et al., 2011; Lin et al., 2011; Brenet et al., 2013), rather than induction of an IFN-1–refractory state. Thus, our data support a model in which IFN-1s do not actively trigger the proliferative machinery of HSCs in vivo, but instead indirectly license HSCs to proliferate by transiently reducing the expression and activity of the complex regulatory network that enforces HSC quiescence in the BM niche. Such a model is supported by recent evidence that proliferating IFN-1–exposed HSCs alter their physical localization in the BM away from quiescence-enforcing periarteriolar niche cells (Kunisaki et al., 2013). It would also explain why the proliferative effect of IFN-1s is not observed in vitro, as most quiescence-enforcing mechanisms such as FoxO3a are already rapidly inactivated, regardless of the presence of IFN-1s, upon removal of HSCs from the BM niche and subsequent culture (Yamazaki et al., 2009).

IFN-1s have classically been described as death inducers, capable of triggering apoptosis via activation of a variety of mechanisms including proapoptotic Bcl2 proteins such as Bax, and death ligands such as TRAIL and TNF (Kiladjian et al., 2011). Although HSC survival relies in large part on balanced expression of prosurvival and proapoptotic Bcl2 genes, (Mohrin et al., 2010), a role for apoptosis in the effect of IFN-1s on HSC function has not been carefully investigated. Here, we find increased apoptosis in acutely proliferating IFN-treated HSCs, which likely results from decreased expression of the essential prosurvival factor Mcl1 (Opferman et al., 2005) and tilts the balance of Bcl2 family members toward cell death. However, once IFN-treated HSCs have reentered quiescence, they do not show detectable apoptosis levels unless they are subsequently forced to proliferate due to culture, transplantation or myeloablation treatment. In this context, IFN-1–exposed HSCs that reenter the cell cycle show a striking activation of a proapoptotic gene program with induction of Bax, Noxa, and Puma, which likely tilts the balance of Bcl2 family members toward cell death. Our data therefore support a model in which IFN-1s poise HSCs for apoptosis but only trigger direct cell death in proliferating HSCs. This raises the intriguing possibility that the relapse of the HSC pool to a quiescent state after chronic IFN-1 treatment could result from the elimination of a subset of proliferating HSCs. Such a hypothesis would be consistent with models of an HSC hierarchy with dormant and active subpopulations (Wilson et al., 2008), but this would need to be directly tested. Our results are also consistent with the demonstration that IFN-1s can induce apoptosis via activation of cell cycle checkpoint mechanisms, including p53 and the IRF family of transcription factors (Takaoka, et al., 2003; Tamura et al., 2008). Indeed, Bax, Puma, and Noxa are well-known p53 target genes, and our data identify p53 as a key trigger mechanism for the induction of apoptosis in response to IFN-1 exposure. Because different splice isoforms of Bcl2 family members can exhibit unique pro- or antiapoptotic activities (Goff et al., 2013), whether changes in splicing contribute to IFN-1–mediated apoptosis would also need to be investigated. Collectively, our results provide a mechanism for the proapoptotic effects of IFN-1 on HSCs, and an explanation for how chronic IFN-1 exposure can suppress blood production without exhausting the HSC compartment by demonstrating that reentry into quiescence acts as a safeguard mechanism ensuring the survival of the HSC pool during chronic IFN-1 exposure.

Our findings are relevant to human patients and may explain the relative rarity of BM failure syndromes caused by chronic IFN-1 production in the absence of other complicating factors (Zhou et al., 2007; Ioannou et al., 2010). In these conditions, the HSC compartment may remain quiescent and thus largely protected from the proapoptotic effects of IFN-1 exposure. Notably, patients suffering from an IFN-1–driven diseases, such as SLE, who undergo autologous BM transplantation experience a postprocedure mortality rate nearly 10-fold higher than other autoimmune patients (Tyndall, 2011; Szodoray et al., 2012). It would be interesting to assess whether the increase in transplant failure in SLE patients might be linked to decreased or delayed hematopoietic reconstitution, which in turn could be driven by an IFN-1–driven proapoptotic program that is activated after mobilization and/or transplantation. Indeed, we directly illustrate the inherent fragility of quiescent IFN-exposed HSCs that are forced back into the cell cycle by showing decreased HSC engraftment after transplantation, as well as pronounced HSC depletion and poor survival upon 5-FU–mediated myeloablation in mice with chronic exposure to IFN-1s. Our results therefore support the idea that blockade of IFN-1s or their prodeath mechanisms may restore HSC potential in patients with chronic IFN-1, as a 2-wk recovery from IFN-1 exposure in mice returns HSC apoptosis and function to relatively normal levels. Interestingly, a recently published work shows that GATA3 activation via the p38 MAP kinase pathway also limits the long-term reconstitution capacity of IFN-1–exposed HSCs (Frelin et al., 2013). This suggests that IFN-1 may trigger a set of complementary mechanisms aimed at limiting the function of HSCs, which could be relevant in the context of viral infection. Collectively, these results suggest that blockade of IFN-1s or their downstream targets could substantially improve HSC function in individuals with chronic IFN-1s.

IFN administration has long been used therapeutically, albeit with mixed success, to treat a range of myeloproliferative neoplasms (MPN) originating from transformed HSCs with disease-propagating leukemic stem cell (LSC) activity, including CML- and JAK2^V617F-mediated essential thrombocytopenia (ET) and polycythemia vera (PV; Kiladjian et al., 2011). Our results may explain the limited efficacy of IFN-1 in inducing durable complete remission in CML patients because, during chronic IFN-1 exposure, the quiescent LSC compartment would remain largely insensitive to IFN-1–mediated apoptosis and would not be eliminated. Interestingly, treatment of mice with two sequential doses (up to 3 d) of poly I:C followed by 5-FU rendered proliferating IFN-exposed HSCs vulnerable to 5-FU–mediated killing (Essers et al., 2009), which raised the possibility that the in vivo pro-proliferative effects of IFN-1s could be used to sensitize quiescent LSCs to killing by tyrosine kinase inhibitors (TKI) such as Imatinib. These results provided a rationale for the success of combined Imatinib + IFN-1 therapy in curing some CML patients, and reignited interest in using IFNs to treat MPNs despite the severe toxicity associated with their use (Kiladjian et al., 2011). However, our findings would argue that the therapeutic window in which quiescent LSCs will be induced to proliferate by IFN-1s and thereby susceptible to TKI-mediated killing will be quite brief. Instead, they suggest that the poising of IFN-exposed LSCs toward apoptosis and their killing upon TKI treatment might be the mechanism underlying the therapeutic effects of IFN-1s, especially during the initial tumor debulking observed with TKI therapy. Interestingly, a recent work using a PV mouse model shows a preferential depletion of Jak2^V617F LSCs compared with WT HSCs in WT:Jak2^V617F BM chimeric mice after 4 wk treatment with IFN-α (Mullally et al., 2013). Although the authors propose that exaggerated cell cycle activity in Jak2^V617F LSCs underlies this effect, it cannot be ruled out that it solely reflects contamination by cycling progenitors in the CD150⁺ LSK gate used to define HSCs in this study, nor that differences in the IFN-mediated apoptosis response of normal versus transformed HSCs contribute to the specific elimination of Jak2^V617F LSCs. Future studies will be needed to determine whether transformed HSCs are more sensitive to the killing effect of IFN-1 treatments, and to carefully define for each MPN disease-type the kinetics of LSC cell cycle activity and activation of apoptotic mechanisms during IFN-1 administration. Such understanding will be critical for optimizing their therapeutic effect on LSCs and minimizing their toxicity to nonleukemic cells, particularly when used in combination with other drugs such as TKIs.

Collectively, our results clarify the link between the proliferative and suppressive effects of IFN-1s on HSC function by showing that IFN-1s only transiently induce HSC proliferation in vivo, which seems to be an acute ‘bystander’ effect related to a brief relaxation of quiescence-enforcing mechanisms. This occurs alongside a broader proapoptotic effect consistent with the antiviral role of IFN-1s. Importantly, induction of a p53-dependent IFN-mediated proapoptotic gene program occurs exclusively in proliferating HSCs, which prevents wholesale exhaustion of the quiescent HSC pool, but also degrades the regenerative potential of HSCs and adversely affects the maintenance of blood homeostasis during chronic IFN-1 exposure. Our findings therefore elucidate how canonically antiproliferative, proapoptotic cytokines induce HSC proliferation and alter HSC function in a manner consistent with their role in chronic inflammation and autoimmune diseases.

Congenic C57BL/6 WT mice were used as donors (CD45.2⁺) and recipients (CD45.1⁺) for all the experiments. Ifnar1^−/−, Trp53^−/−, Foxo3a^−/−, and BakBax^cKO mice have been described previously (Donehower et al., 1992; Müller et al., 1994; Castrillon et al., 2003; Warr et al., 2013). Transplantation of purified HSCs and generation of WT:Ifnar1^−/− BM chimeric mice were performed as previously described (Santaguida et al., 2009). For poly I:C treatment, 6–12 wk-old age- and gender-matched mice were injected i.p. with 10 µg/g body mass of poly I:C (GE Healthcare) in PBS at 2-d intervals for up to 30 d. For in vivo IFN-α treatment, mice were injected subcutaneously with 1 × 10⁴ U IFN-α4 (eBioscience) every 12 h until BM harvest. For in vivo HSC proliferation assays, mice were injected i.p. with 1 mg BrdU (Sigma-Aldrich) in d-PBS 3 h before BM harvest. For myeloablation treatment, mice were injected i.p. with 150 mg/kg of 5-FU (Sigma-Aldrich) in PBS, and PB was collected via the retroorbital vein into K₂EDTA-coated collection tubes (BD) for complete blood counts (CBC) performed on a Hemavet950 analyzer (DREW Scientific). All mouse experiments were performed in accordance with the Institutional Animal Care and Use Committee at the University of California, San Francisco.

Hematopoietic stem and progenitor cell surface staining and enrichment procedures were done as described previously (Santaguida et al., 2009). Cells were isolated on a FACS ARIAII (BD) using double sorting for maximum purity, and analyzed on a FACS LSRII (Becton Dickinson). Cellularity was determined on one tibia and one femur per mouse using a ViCell automated cell counter (Beckman-Coulter). Surface staining for IFNAR1 was performed with a PE-conjugated antibody (BD). BrdU staining and analysis of PB for CD45.1/CD45.2 chimerism were performed as previously described (Santaguida et al., 2009). For intracellular Ki67/DAPI staining, BM cells were stained with surface markers, fixed, and permeabilized using a Cytofix/Cytoperm kit according to the manufacturer’s instructions (BD) and stained with an anti–Ki67-FITC antibody (Vector Laboratories) for 1 h at room temperature. DAPI was added to cell samples and allowed to incubate for 20 min at RT before analysis. Intracellular phospho-AKT was performed as previously described (Chen et al., 2009).

For FoxO3a detection, HSCs (1,000/slide) were sorted directly onto PolyPrep l-lysine-coated slides (Sigma-Aldrich), fixed with 4% paraformaldehyde, blocked with 0.1% BSA in PBS, permeabilized with 0.2% Triton X-100, stained for 1 h at room temperature with an anti-FoxO3a antibody (Millipore), incubated for 1 h at room temperature with a goat anti–rat A549 antibody (Invitrogen), and mounted with coverslips in VectaShield (Vector Laboratories) containing 1 ng/ml DAPI. For EdU assays, HSCs (1,500/slide) were sorted onto slides and fixed in 4% PFA as described above, and EdU incorporation was detected using A594-labeled azide click chemistry according to the manufacturer’s instructions (Life Technologies). For cleaved caspase 3 (CC3) detection in BM sections, femurs were snap-frozen in OCT, and 10-µm sections were made on a CM1850 Cryostat using the CryoJane tape transfer system (Leica). Slides were subsequently air-dried for 3 h, fixed in 4% PFA, blocked in 0.1% Triton X-100 containing 10% goat serum, stained overnight at 4°C with an anti-CC3 antibody (Cell Signaling Technology), incubated for 1 h at room temperature with a donkey anti–rabbit A488 antibody (Invitrogen), stained with 1 ng/ml DAPI, and mounted with coverslips. Immunofluorescence images were acquired on a TCS-SP5 microscope (Leica) and analyzed using Velocity software (Perkin-Elmer). For EdU incorporation assays, at least 200 cells were scored per replicate slide.

For colony assays, HSCs (100/dish) were grown for 7 d in MethoCult M3231 methylcellulose media (StemCell Technologies) supplemented with SCF (10 ng/ml), IL-11 (10 ng/ml), Flt3L (10 ng/ml), IL-3 (10 ng/ml), GM-CSF (10 ng/ml), TPO (100 ng/ml), and EPO (10⁴ U/ml; all from PeproTech). For growth in liquid culture, HSCs (400/well) were cultured in StemPro34 medium (Invitrogen) supplemented with SCF, Flt3L, and TPO. For expansion analyses, cells were manually counted with a hemocytometer after 8 d. For apoptosis detection, Caspase-Glo (Promega) luminescent reagent was added at a 1:1 ratio after 12 h of culture and incubated according to the manufacturer’s instructions. Luminescence was read using a plate-based luminometer (BioTek). For in vitro BrdU incorporation, HSCs (2,500/well) were cultured for 12 h in StemPro34 medium supplemented with SCF, TPO, and 60 µM BrdU. For BM co-culture experiments, CFSE-labeled HSCs (2,500/well) were incubated with unfractionated BM cells (5 × 10⁵) for up to 48 h in StemPro34 medium supplemented with SCF and TPO. For OP9-DL1 co-culture experiments, CellTracker Violet-labeled HSCs (2,500/well) were added atop OP9-DL1 stromal cells (5 × 10³/well) in IMDM supplemented with 5% FBS, SCF, TPO, TGF-β (10 ng/ml; PeproTech) and 10 µM EdU as described (Santaguida et al., 2009). For CFSE labeling, HSCs were incubated for 10 min with 5 µM CFSE (Molecular Probes) in PBS, quenched with FBS, and washed twice before culture. Mouse recombinant IFN-α4 (eBioscience) was added to the culture at 100 ng/ml except when indicated.

For SA Biosciences PCR arrays, total RNA was isolated from 20,000 HSCs sorted directly into TRIzol-LS (Invitrogen) using the Arcturus PicoPure RNA kit and genomic DNA was removed using the QIAGEN RNase-free DNase kit following manufacturer’s protocols. Purified RNA was quantified on a Bioanalyzer (Agilent) and 4 ng were subjected to RT and 12 rounds of preamplification using the RT² Reverse transcription kit and target-specific primers for custom-designed RT² Profiler qRT-PCR arrays (SA Biosciences). Array plates were run on an Applied Biosystems HT7900 thermocycler and analyzed using the ΔC_T method. For standard quantitative RT-PCR, total RNA extraction and qRT-PCR analyses were performed as previously described (Mohrin et al., 2010).

Blood was harvested after euthanasia via cardiac puncture and serum was analyzed for cytokines using a Luminex xMAP 20-plex mouse cytokine array according to the manufacturer’s instructions. ELISA analyses of serum TNF, MCP-1, and IFN-α were performed according to the manufacturer’s instructions (eBioscience).

Measurements were performed in triplicate and n indicated the number of independent experiments. Data were expressed as mean ± SD. Statistical significance was determined using a two-tailed, unpaired Student’s t test. Significance of survival data were determined using a log-rank (Mantel-Cox) test. P-values of ≤0.05 were considered statistically significant.

Fig. S1 shows FACS gating strategy for analysis of mature and immature BM populations. Fig. S2 shows FACS gating scheme for analysis of BrdU staining, Ki67/DAPI staining, and phospho-AKT levels in HSCs. Fig. S3 shows FACS gating for identification and recovery of labeled HSCs in BM and OP9-DL1 co-culture experiments. Fig. S4 shows FACS gating scheme for analysis of HSCs in poly I:C– and 5-FU–injected mice.

We thank Dr. Jason Cyster for critical reading of this manuscript, Michael Kissner for management of the Broad Center Flow Cytometry Core Facility, and all members of the Passegué laboratory for helpful comments and discussions.

E.M.P. was supported by postdoctoral Ruth L. Kirschstein National Research Service Awards (NRSA) T32 CA108462, T32 AI007334 and F32 HL106989. R.L. was supported by a CIRM Bridges Fellowship. This work was supported by R01 HL092471 and CIRM New Faculty Award RN2-00934 to E.P.

The authors declare no financial conflicts of interest.