Central role of IP3R2-mediated Ca2+ oscillation in self-renewal of liver cancer stem cells elucidated by high-signal ER sensor

Ca2+ oscillation is a system-level property of the cellular Ca2+-handling machinery and encodes diverse physiological and pathological signals. The present study tests the hypothesis that Ca2+ oscillations play a vital role in maintaining the stemness of liver cancer stem cells (CSCs), which are postulated to be responsible for cancer initiation and progression. We found that niche factor-stimulated Ca2+ oscillation is a signature feature of CSC-enriched Hep-12 cells and purified α2δ1+ CSC fractions from hepatocellular carcinoma cell lines. In Hep-12 cells, the Ca2+ oscillation frequency positively correlated with the self-renewal potential. Using a newly developed high signal, endoplasmic reticulum (ER) localized Ca2+ sensor GCaMP-ER2, we demonstrated CSC-distinctive oscillatory ER Ca2+ release controlled by the type 2 inositol 1,4,5-trisphosphate receptor (IP3R2). Knockdown of IP3R2 severely suppressed the self-renewal capacity of liver CSCs. We propose that targeting the IP3R2-mediated Ca2+ oscillation in CSCs might afford a novel, physiologically inspired anti-tumor strategy for liver cancer.


Introduction
Ca 2+ signaling plays essential roles in the initiation and progression of diverse diseases including cancer. Recent studies have shown that dysregulation of Ca 2+ homeostasis emerges as an important hallmark of tumor cells because remodeling of Ca 2+ channels and transporters is common to tumor progression 1,2 and Ca 2+ oscillation orchestrates invadopodium formation and invasion during cancer metastasis 3 .
Cancer stem cells (CSCs) are thought to be the origin of cancer initiation and progression. It has been reported that Ca 2+ /calmodulin-dependent protein kinase IIγ prompts the stem-like properties of lung cancer cells 4 , while calcineurin, a Ca 2+ /calmodulin-dependent protein phosphatase, represses keratinocyte CSC potential 5 . Recently, we have shown that spontaneous Ca 2+ oscillation occurs in the CSC-enriched liver cell line Hep-12 6 , and modulates the efficiency of self-renewal 6,7 . Ca 2+ oscillation decodes the stimulations not only by its amplitude, but also by its frequency and duration, thereby expanding the capabilities of the signaling pathway [8][9][10] .
The endoplasmic reticulum (ER), which extends over the entire cytoplasm as an elaborated nanotubular network, acts as the most important intracellular Ca 2+ store. However, investigation of ER Ca 2+ signaling has been limited by a lack of ER-localized, high-signal sensors that can report store Ca 2+ dynamics in real time. Circularly permutated EGFP variants fused with calmodulin and M13 motif, termed GCaMPs or G-GECOs are robust Ca 2+ sensors that have undergone progressive improvements [11][12][13][14][15] , and for which the structural basis of Ca 2+ -dependent fluorescence has been determined 16 . While GCaMPs have been optimized for the detection of rapid cytosolic Ca 2+ signaling, there has been limited success in creating variants with high dynamic range within the micromolar Ca 2+ endoplasmic/sarcoplasmic environment. Recent progress has been made in this area using new classes of fluorescent protein sensors 17,18 , or GCaMP variants with mutated calmodulin moieties 18,19 . However, the brightness and dynamic range of these proteins, have been limited.
In the present study, we aimed to test the hypothesis that Ca 2+ oscillation might play a vital role in maintaining the stemness of CSCs. We investigated the Ca 2+ phenotypes, their subcellular and molecular mechanisms, their significance in cancer biology and potential strategies for intervening CSCs through manipulating Ca 2+ signaling in CSC-enriched Hep-12 cells. In addition, we developed GCaMP-ER2, a novel ER localized Ca 2+ sensor, to probe the store Ca 2+ dynamics simultaneously. Our comprehensive approaches allow us to propose that targeting the IP 3 R2mediated Ca 2+ oscillation in CSCs might afford a novel, physiologically inspired anti-tumor strategy for liver cancer.

Ca 2+ oscillation is a signature feature of liver CSCs
To test the hypothesis that Ca 2+ oscillation is a hallmark of CSCs, we applied a panel of potential niche factors, including ATP, epidermal growth factor (EGF)/basic fibroblast growth factor (FGFb)/B27, and interleukin (IL)-6, and assessed the Ca 2+ response in the liver CSC line Hep-12 and its matching hepatocellular cancer (HCC) line Hep-11. ATP is frequently released from cells to facilitate metastasis 20 . EGF is a well-accepted niche factor and EGF/FGFb/B27 are commonly used in media for in vitro spheroid formation assays, the most popular endpoint experiment for self-renewal determination in solid tumors 21,22 . IL-6 has been reported to elevate or maintain the self-renewal of liver CSCs 23 . Besides these, we opted to include methacholine, a muscarinic receptor agonist capable of inducing dynamic Ca 2+ changes in many types of cells 24 . All of these stimulants are linked to the inositol 1,4,5-trisphosphate (IP 3 ) signaling pathway and thereby trigger intracellular Ca 2+ release from the ER via IP 3 receptors (IP 3 Rs) [24][25][26][27] .
We found that these stimulants each induced a large, slow Ca 2+ transient that lasted several minutes in Hep-12 cells. Remarkably, Ca 2+ oscillations of spiky appearance overlaid the evoked Ca 2+ transient, and often continued even in the tail of the transient, when the cytosolic Ca 2+ level returned to baseline (Fig. 1a, Movie S1). In contrast, typical Hep-11 cells did not show any Ca 2+ change in response to most of the stimulants (EGF/FGFb/B27, IL-6, and methacholine), or exhibited only a smooth, monophasic Ca 2+ transient in response to ATP (Fig. 1a, b,  Fig. S1b, 2). That is, CSCs and HCCs exhibited distinct Ca 2+ responses upon stimulation by self-renewal-related factors or IP 3 pathway-related agonists.
We next extended the experiments to include multiple HCC lines, such as highly-metastatic LM3, poorly metastatic MHCC97-L, and Huh7 cells. ATP did elicit oscillatory Ca 2+ responses in these cells, but their frequencies were significantly lower than those in Hep-12 cells (Fig. 1a, b). Further, none of the other three stimulants induced any oscillatory Ca 2+ response in the vast majority of cells of each type (Fig. 1a, b). Similar results were found in the nontumorigenic immortalized liver cell line MIHA ( Fig. 1a, b, Fig. S1b).
We have previously shown that α2δ1, a subunit associated with L-, P-, N-, and R-type Ca 2+ channels, is a surface marker in liver CSCs 6,7 . We used the α2δ1 biomarker for sorting CSCs. More α2δ1 + Hep-11 cells than α2δ1 − cells displayed Ca 2+ oscillation and the ensembleaveraged oscillation frequency was 3.1-and 8.9-fold higher after stimulation with ATP and EGF/FGFb/B27, respectively (Fig. 1c, d). Likewise, α2δ1 + Huh7 cells also displayed higher Ca 2+ oscillation frequency (Fig. 1e, f). Taken together, we conclude that Ca 2+ oscillation is a signature that distinguishes liver CSCs from HCCs and nontumorigenic liver cells.

Development of high-signal ER-targeted Ca 2+ sensor GCaMP-ER2
To measure ER Ca 2+ dynamics, we developed GCaMP-ER2, a genetically encoded ER-targeting Ca 2+ indicator with low Ca 2+ affinity and high-dynamic range (K d = 388 μM, ΔF max /F 0 = 40). Briefly, we screened a series of random and structure-based mutations in the four calmodulin EF-hand loops of G-GECO1.2 13 , and identified a D129A mutant within the EF-hand of loop IV (GCaMP-L1, Fig. 2a), which has a dynamic range of 32 (ΔF max /F 0 ), and a K d for Ca 2+ of 106 μM. This variant was systematically mutated in EF-hand loops, resulting in an array of molecules with Ca 2+ affinities ranging from 106 μM to 5.9 mM (Table S1). Figure 2a, b show the most promising ER sensor GCaMP-L2, with a dynamic range (ΔF max /F 0 ) of 40, a marked improvement over the previously highest signal strength ER reporter, GCaMPer/GCaMP3 (10.19) 19 (ΔF max /F 0 = 14). As the K d of GCaMP-L2 (388 μM) is well below the resting level of ER Ca 2+ , the indicator would be expected to operate near maximum brightness prior to ER Ca 2+ release events.

Distinctive ER Ca 2+ dynamics in liver CSCs
Ca 2+ oscillation is an emergent property of the entire cellular Ca 2+ -handling system. To investigate the cellular mechanism underlying the distinctive Ca 2+ response of CSCs, we measured the resting cytosolic Ca 2+ concentration ([Ca 2+ ] c ) and the capacity of the releasable ER Ca 2+ store. The latter was indexed as the peak cytosolic Ca 2+ concentration during release stimulated by the Ca 2+ ionophore A23187 (10 μM) ([Ca 2+ ] release ) (Fig. 3a). We found that both [Ca 2+ ] c and [Ca 2+ ] release were similar in the two types of cells (Fig. 3b, c), excluding the possibility that Ca 2+ oscillation in Hep-12 cells is brought about by a cytosolic or ER Ca 2+ overload. Then we sought to measure ER Ca 2+ dynamics. With the newly developed GCaMP-ER2 and Rhod-4, a cytosol-retained Ca 2+ indicator, we simultaneously assessed the dynamic responses in both ER and cytosol (Fig. S2b). In Hep-11 cells, the ER Ca 2+ store declined precipitously upon ATP stimulation, mirroring the rapid rise of the cytosolic Ca 2+ transient (Fig. 3e). Subsequently, the GCaMP-ER2 fluorescence remained stable at a low level (nadir ΔF/F 0 = −0.77) ( Fig. 3f) even though the cytosolic Ca 2+ level was returning toward an elevated steady-state level (Fig. 3f). This result suggests that the ER Ca 2+ recycling mechanism is ineffective, presumably because of continued opening of the ER Ca 2+ release channels. In Hep-12 cells, however, the initial ER Ca 2+ depletion was smaller (nadir ΔF/F 0 = −0.34) (Fig. 3d, f), and strikingly, it was rapidly restored and often followed by a conspicuous overshoot (peak F/F 0 = 1.16) (Fig. 3d, g). The peak [Ca 2+ ] c , however, was comparable in both types of cells (F/F 0 = 3.68 ± 0.18 for Hep-11 and 3.23 ± 0.16 for Hep-12) (Fig. S2c, d). Close observation further revealed that each spiky cytosolic Ca 2+ oscillation was accompanied by an antiphasic ER Ca 2+ oscillation (Fig. 3d inset). This finding indicates that the cytosolic Ca 2+ oscillation originates from periodic ER release and recycling. ER store recovery and overshoot were also found in the subpopulation of MHCC97-L cells with stimulated Ca 2+ oscillation, whereas persistent ER Ca 2+ depletion was associated with the population of cells lacking the oscillatory behavior (Fig. S2e). Taken together, we concluded that distinctive regulation of ER Ca 2+ dynamics underlies the Ca 2+ oscillation phenotype characteristic of liver CSCs.

Role of ER Ca 2+ cycling in the genesis of Ca 2+ oscillation
Store-operated Ca 2+ entry (SOCE), the main Ca 2+ entry pathway in nonexcitable cells, is activated upon depletion of the ER Ca 2+ and plays pivotal roles in the timely replenishment of the internal Ca 2+ store. In both cell types, inhibition of ER Ca 2+ uptake by thapsigargin (4 μM) caused a biphasic cytosolic Ca 2+ increase, with the second peak reflecting the SOCE capacity (Fig. 4a). It was clear that the Hep-12 cells displayed a greater SOCE component (Fig. 4b). In addition, faster ER Ca 2+ uptake occurred in Hep-12 cells (Fig. 4c-e). To further delineate the specific role of SOCE, we measured cytosolic and ER Ca 2+ dynamics in the absence of extracellular Ca 2+ . We found that the ATP-stimulated initial Ca 2+ transients remained essentially unchanged; Ca 2+ oscillation occurred in the early phase, but was quickly damped and eventually disappeared in 200 s (Fig. 4g). Concomitantly, the ER Ca 2+ store showed a general trend of depletion without replenishment (Fig. 4g, l). Application of 100 μM 2-aminoethoxydiphenyl borate (2-APB), an inhibitor of both SOCE and IP 3 Rs at this concentration, greatly attenuated the Ca 2+ oscillation and prevented the ER store from recovering (Fig. 4h, j, l). The initial Ca 2+ transient was also markedly smaller and briefer (Fig. 4k), perhaps because of partial inhibition of release. Simultaneous inhibition of Ca 2+ entry and ER Ca 2+ recycling (0 Ca 2+ and 4 μM thapsigargin) had additive effects (Fig. 4i, j, l). These results indicate that the persistence of Ca 2+ oscillation is attributable to the rapid refilling and overshoot of the ER store due to enhanced SOCE and ER Ca 2+ uptake. Nonetheless, they are not obligatory for the genesis of stimulated Ca 2+ oscillation; rather, they play a permissive role in preventing it from damping.

IP 3 R2 as the intrinsic oscillator underlying Ca 2+ dynamics in Hep-12 cells
Next, we sought to identify the key molecular players underlying the Ca 2+ oscillation phenotype. By RNAsequencing analysis, we identified 1887 genes expressed differently in Hep-12 versus Hep-11 cells (Fig. 5a). Among them, 175 genes were found in Ca 2+ -related Gene Ontology (1093 genes included) (Fig. 5a). Interestingly, of the ten genes that directly participate in Ca 2+ transport, all were significantly upregulated in Hep-12 cells (Fig. 5a). These genes code proteins for pore subunits of the N-type (α1B) and T-type (α1G) Ca 2+ channels, subunits associated with most voltage-operated Ca 2+ channels (VOCCs) (α2δ1, α2δ2, and γ5), ER Ca 2+ release channels (IP 3 R1, IP 3 R2, and RyR2), and the ER Ca 2+ uptake pump (SERCA3), as well as transient receptor potential TRPC4 channels. Notably, α2δ1, which helps to anchor VOCCs in the plasma membrane, has been previously reported by us as a functional liver CSC marker 6,7 .
To identify key Ca 2+ regulators, we targeted the top five genes on the list, CACNA2D1, CACNA2D2, ITPR1, ITPR2, and ATP2A3, and determine the consequences of the Ca 2+ oscillation phenotype. We used α2δ1 as a positive control on some occasions, and included ITPR3 (IP 3 R3 gene) alongside ITPR1 (IP 3 R1 gene) and ITPR2 (IP 3 R2 gene) to acquire a comprehensive view of the IP 3 R family. We established stable cell lines harboring shRNAs against these genes (Fig. S3a). Functional analysis revealed that Ca 2+ oscillation frequency was severely suppressed after knockdown of α2δ1 and IP 3 R2, modestly decreased with knockdown of IP 3 R1 and SERCA3, and largely unchanged after knockdown of α2δ2 and IP 3 R3 (Fig. 5b, c,  Fig. S3b). The amplitude of the first peak, which reflects the ER Ca 2+ content as well as fractional release, was mildly decreased (knockdown of IP 3 R1, IP 3 R2, or IP 3 R3) or unchanged (knockdown of α2δ1, α2δ2, or SERCA3) (Fig. S3c). These results show that multiple Ca 2+ regulators orchestrate the phenotype of Ca 2+ oscillation. Nonetheless, other than α2δ1, IP 3 R2 emerged as the most important molecular determinants. Particularly, after IP 3 R2 knockdown, the ER Ca 2+ phenotype was drastically altered, with greater initial release and without any . Data are shown as mean ± SEM (***P < 0.001 versus Hep-11 b, e or control j-l). All data were acquired with three independent experiments recovery (Fig. 5d, e), resembling those in non-oscillatory HCCs. That is, IP 3 R2 may serve as an intrinsic oscillator essential for Ca 2+ oscillation. Previous studies have also shown that, unlike IP 3 R1 and IP 3 R3, IP 3 R2 is prone to generate the Ca 2+ oscillation behavior [28][29][30] .
Ca 2+ oscillation frequency positively correlates with spheroid-forming efficiency in Hep-12 cells To determine the functional significance of Ca 2+ oscillation in CSC biology, we assessed the efficiency of spheroid formation, which reflects the self-renewal capacity, in relation to Ca 2+ oscillation frequency over a broad range of experimental conditions. The experimental groups included: (i) knockdown of the VOCC subunits α2δ1 and α2δ2; (ii) knockdown of the ER release channels IP 3 R1, IP 3 R2, and IP 3 R3; (iii) removal of extracellular Ca 2+ ; (iv) inhibition of SOCE; and (v) treatment with BAPTA-AM or EGTA-AM to retard Ca 2+ changes. We found that, while inhibiting Ca 2+ oscillation, both α2δ1 and α2δ2 knockdown significantly compromised the selfrenewal (Fig. 6a, b, Fig. S4a), confirming and extending our previous report 6 . Spheroid formation was also reduced when extracellular Ca 2+ was removed, SOCE was inhibited or cytosolic Ca 2+ was buffered (Fig. 6a, c, Fig.  S4a). Notably, of the three IP 3 R subtypes, knockdown of IP 3 R2, but not IP 3 R1 and IP 3 R3, greatly curtailed the spheroid formation (Fig. 6a, b, Fig. S4a).
From these data, a general pattern emerged in which lowering the Hep-12 Ca 2+ oscillation frequency effectively decreased the efficiency of spheroid formation, regardless of whether the type of Ca 2+ manipulation was genetic, biophysical, or pharmacological. For quantitative analysis, we summarized our data in a scatter-plot (Fig. 6d). Linear regression revealed a strong positive correlation between spheroid-forming efficiency and Ca 2+ oscillation frequency (r = 0.76, p < 0.001) (Fig. 6d). This result strongly suggests that the signature Ca 2+ oscillation in liver CSCs plays an important functional role in promoting their self-renewal.

Targeting IP 3 R2-mediated Ca 2+ oscillation suppresses tumor formation of liver CSCs
In light of the newly established relationship between Ca 2+ oscillation and spheroid formation, we hypothesized that targeting the Ca 2+ oscillation machinery might provide an effective therapeutic strategy to limit CSC expansion and therefore retard or even prevent the genesis of tumors. Among the aforementioned molecular participants in the genesis and regulation of Ca 2+ oscillation, IP 3 R2 is particularly noteworthy, not only because it is greatly upregulated in liver CSCs, but also its cyclic opening may serve as the intrinsic oscillator [28][29][30] . To further evaluate the impact of IP 3 R2 on the self-renewal of Hep-12 cells, we injected 10 3 or 10 2 scrambled control cells and IP 3 R2-knockdown cells subcutaneously into nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice to test their tumorigenicity potential. Tumors were dissected and weighed 42 or 52 days after injection of 10 3 or 10 2 cells, respectively. As shown in Fig.  7a, b, the tumorigenicity of Hep-12 cells was significantly suppressed by IP 3 R2 knockdown. To further confirm the role of IP 3 R2, we purified α2δ1 + CSCs from Huh7 cells and assayed spheroid formation and tumorigenicity with IP 3 R2 knockdown. Both the efficiency of spheroid formation and the frequency of tumorigenic cells were markedly inhibited (Fig. 7c, d), substantiating the pivotal role of IP 3 R2 in the self-renewal of liver CSCs.

Discussion
In this study, we demonstrate that Ca 2+ oscillation is a functional signature of liver CSCs, and provide evidence that targeting Ca 2+ oscillation can effectively limit the self-renewal of liver CSCs and thereby tumor initiation and progression. In response to a panel of niche factors (ATP, EGF/FGFb/B27, and IL-6), robust Ca 2+ oscillation occurred in liver CSCs, including CSC-enriched Hep-12 cells, the small portion of α2δ1 + Hep-11 cells, and α2δ1 + Huh7 cells. This Ca 2+ phenotype is CSC-specific because the vast majority of matching Hep-11 and Huh7 cancer cells did not show similar Ca 2+ oscillation, and neither did HCC LM3, MHCC97-L, and the immortalized liver cell line MIHA. This finding is in general agreement with previous reports that Ca 2+ oscillation occurs in a portion of highly metastatic, but not in weakly/non-metastatic human prostate and breast cancer cells 31 . Such CSC Ca 2+ oscillation plays important roles in cancer biology. The efficiency of spheroid formation was significantly decreased by suppressing Ca 2+ oscillation by removing extracellular Ca 2+ , inhibiting SOCE, buffering cytosolic Ca 2+ , or knocking down key Ca 2+ regulators (IP 3 R2 or α2δ1). Strikingly, there was a positive correlation between the Ca 2+ oscillation frequency and spheroid formation efficiency under a broad range of conditions. More  : 100 μm). b, c Statistics of spheroid-forming efficiency. b Scrambled control group and shRNA knockdown groups. c Groups with and without drug treatment. One hundred cells per well were plated. Spheroids (R > 100 μm) were counted under a stereomicroscope (n = 6-12; ***P < 0.001 versus control). d Scatter-plot of oscillation frequency versus spheroid-forming efficiency. Linear regression (solid line) yielded a positive correlation coefficient (r) of 0.76 (P < 0.001). Data are expressed as mean ± SEM, n = 18. All data were acquired with at least three independent experiments importantly, we showed that CSCs with defective Ca 2+ oscillation, i.e., knockdown of IP 3 R2, exhibited a greatly compromised ability to form tumors in NOD/SCID mice.
Ca 2+ oscillation reflects a system-level, emergent behavior of the complex Ca 2+ -handling machinery. To sustain a persistent cyclic Ca 2+ oscillation, the ER store must be kept replenished. With our newly developed ER Ca 2+ sensor GCaMP-ER2, we disclosed a surprising new insight at the cellular level. In CSCs undergoing niche factor-stimulated Ca 2+ oscillation, initial ER depletion was followed by a complete restoration and oftentimes an overshoot of the ER Ca 2+ content. In HCCs displaying no Ca 2+ oscillation, however, stimulated ER Ca 2+ release was followed by a deep ER depletion without any sign of recovery over the period of observation (15 min). The distinctive ER Ca 2+ dynamics are in part attributable to higher activity of SOCE and ER Ca 2+ recycling in CSCs (this study) as well as Ca 2+ entry through VOCCs as we reported previously 6 . More importantly, these results strongly suggest that the ER release channels in CSCs undergo cyclic, brief openings, allowing for ER Ca 2+ retention and supporting the Ca 2+ oscillation behavior, Fig. 7 Crucial role of IP 3 R2 in self-renewal of liver cancer stem cells. a, b In vivo tumorigenicity of IP 3 R2-knockdown Hep-12 cells. One thousand (a) or one hundred cells (b) per mouse were injected (n = 7). Left: Images showing dissected tumors at termination of the experiment time (scale bars: 1 cm). Right: Tumor weight. Data are shown as mean ± SEM (*P < 0.05 versus control, ***P < 0.001). c IP 3 R2 knockdown suppresses self-renewal of α2δ1 + Huh7 cells. Left: Phase-contrast images of spheroid formation. Right: Statistics of spheroid-forming efficiency. One hundred cells per well were plated. Spheroids (R > 100 μm) were counted under a stereomicroscope (n = 9; **P < 0.01 versus control, ***P < 0.001. d Tumorigenic cell frequency in α2δ1 + Huh7 cells with IP 3 R2 knockdown in NOD/SCID mice whereas those in HCCs may be kept open after stimulation, preventing replenishment of the ER Ca 2+ . Through RNA-seq and bioinformatics analysis confirmed by functional validation, we identified the ER release channel IP 3 R2 as the intrinsic oscillator driving the CSC Ca 2+ phenotype, while the other Ca 2+ proteins examined serve largely permissive roles (e.g., α2δ1 and SERCA3).
The IP 3 R family consists of three subtypes, IP 3 R1, IP 3 R2, and IP 3 R3, which share 60-80% homology of amino acid sequence, and the subtype expression is differentially regulated in response to physiological and pathological stimuli. There is a bell-shaped Ca 2+ -dependence of the channel activity: activation via the CICR mechanism at low [Ca 2+ ] i , inhibition via Ca 2+ -dependent inactivation at higher [Ca 2+ ] i 32 . These properties lay the foundation for Ca 2+ oscillation in nonexcitable cells. All three IP 3 R subtypes are similar in terms of permeability, but their high-Ca 2+ inactivation properties are variable 33 . Functional experiments have shown that IP 3 R2 is required for long-lasting and regular Ca 2+ oscillation [28][29][30] , IP 3 R1 is responsible for short and irregular Ca 2+ oscillation 28 , and IP 3 R3 produces a monophasic Ca 2+ spike 28 and even has anti-oscillatory actions 28,34 . More interestingly, IP 3 R2 coexpression with IP 3 R1 or IP 3 R3 facilitates Ca 2+ oscillation 28 . In this regard, we found that IP 3 R2 was specifically expressed in liver CSCs but not HCCs; and knockdown of IP 3 R2, but not IP 3 R1 and IP 3 R3, repressed Ca 2+ oscillation (Fig. 5b, c, Fig. S3b), and hence the capacity for selfrenewal in CSCs (Fig. 6a, b, Fig. S4a). Knockdown of IP 3 R2 also caused ER Ca 2+ phenotypes similar to that in a typical Hep-11 cell. Collectively, these data indicate that IP 3 R2 is a target of choice for limiting CSC self-renewal through manipulating Ca 2+ oscillation. Indeed, CSCs with knockdown of IP 3 R2 showed depressed sphere formation (Figs. 6a, b, 7c) and in vivo tumorigenesis (Fig. 7a, b, d).
CSCs rely on niches to maintain their stemness and prompt metastasis 35 . In this regard, we found that all niche factors examined were able to elicit robust Ca 2+ oscillation in liver CSCs, supporting the hypothesis that Ca 2+ signaling is at the root of CSC maintenance and survival. Ca 2+ oscillation can mediate cellular signaling in a frequency-modulatory mode, with the frequency varying from tens of Hz to tens of mHz. Many protein kinases, protein phosphatases, and transcription factors are known to decipher Ca 2+ oscillation, and each has a specific dependence on oscillation frequency, interestingly with little overlap 36 . Future investigations to determine the involvement of these decoder proteins in CSC Ca 2+ oscillation to signal self-renewal are warranted.
In summary, we have shown that niche factorstimulated Ca 2+ oscillation is a signature feature pivotal to self-renewal of liver CSCs. Further, through the development of a new genetically coded ER Ca 2+ sensor GCaMP-ER2, we have identified IP 3 R2, which is enriched in CSCs, as the intrinsic oscillator driving such Ca 2+ oscillation. Targeting Ca 2+ oscillation in general and IP 3 R2 in particular might afford a physiologically inspired anti-tumor strategy by uprooting CSCs to limit tumor initiation, progression, recurrence, and drug resistance.

Materials and methods
Cell culture and establishment of stable target geneknockdown cell lines Hep-12, Hep-11, and Huh7 cells 37 were cultured in RPMI 1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS) (Invitrogen) at 37 ℃ under 5% CO 2 . The LM3 and MHCC97-L cell lines were gifted by Dongqin Yang (Huashan Hospital, Fudan University), and the MIHA cell line was gifted by Jinying Ning (Crownbio Co., Beijing, China). The HeLa cell line was kept in our labs. All four cell lines were cultured in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% FBS. To generate stable Hep-12 cell lines with knockdown of target genes, packaging plasmids pLP1, pLP2, and pVSVG (Invitrogen) were transfected with expression plasmids harboring shRNA targeting the specific genes into 293FT cells at the ratio 1:1:1:3, then the lentivirus was collected and cells were infected. The puromycin-resistant cells were cultured and stable target gene-knockdown cell lines were acquired. ShRNAs (Sigma) were gifted by Guoqiang Bi (University of Science and Technology of China) and their sequences are listed in Table S2.

Calcium measurement and imaging
Fluo-4 AM (Invitrogen) was used to monitor cytosolic Ca 2+ dynamics. When ER Ca 2+ was measured simultaneously using GCaMP-ER2, Rhod-4 AM (AAT Bioquest) was used for cytosolic Ca 2+ detection. Briefly, cells were incubated in the presence of Fluo-4 AM (5 μM) or Rhod-4 AM (5 μM) at room temperature for 30 min. After washing with Tyrode's solution consisting of (in mM) 137 NaCl, 5. The cytosolic Ca 2+ level was quantified using the equation [Ca 2+ ] = K d (F − F min )/(F max − F). After measurement of fluorescence at rest (F 0 ), cells were exposed to Tyrode's solution (zero Ca 2+ ) with 4 mM EGTA, 5 μM thapsigargin, and 10 μM A23187. Store Ca 2+ release was induced and cytosolic Ca 2+ increased abruptly and reached the peak in~10 s (F release ) followed by a gradual decline. When stabilized, 100 μM BAPTA-AM was added to further chelate cytosolic Ca 2+ to measure the minimum fluorescence level (F min ). Then, for measurement of the maximal level (F max ), cells were exposed to Tyrode's solution containing 10 mM Ca 2+ , 5 μM thapsigargin, 12 μM A23187, 3 μM FCCP, and 20 mM 2-DG. Assuming that K d = 1 μM for Fluo-4 in intact cells 38 , the resting Ca 2+ [Ca 2+ ] c , and the releasable store Ca 2+ indexed by [Ca 2+ ] release were obtained with the above equation.

Synthesis of low-affinity GCaMP (GCaMP-L) mutants
Mutations of CaM were introduced into the G-GECO1.2 sequence by overlap extension PCR. The amplified product was inserted into the NheI/HindIII sites of pRSET-A expression vector (Invitrogen). The mutant proteins were expressed in E. coli BL21 Star (DE3) pLysS cells and purified using Ni-charged resins as previously described 39 . After elution, the buffer was changed to 30 mM MOPS (pH 7.2) with 100 mM KCl using an Amicon Ultra-4 filter unit (Millipore). Protein concentration was measured using BCA Protein Assay (Pierce).

In vitro characterization of purified proteins
Calcium titration of G-GECO1. 2

Construction of ER-targeted GCaMP-ER2
The GCaMP-L2 was targeted to and retained in the ER via the N-terminal calreticulin ER targeting sequence MLLSVPLLLGLLGLAVA and the C-terminal ER retention signal KDEL, respectively, with a linker KL(AP) 6 between CaM and retention signal. The final construct was generated by PCR with primers containing described coding sequences and GCaMP-L2 template. The PCR product was cloned into the pEGFP-N1 mammalian expression vector (replacing EGFP) using BglII and NotI restriction sites.

Cell labeling and flow cytometry
The α2δ1 antibody (Abcam) was directly labeled with the Lightning-Link PE-Cy5 Labeling Kit following the vendor's protocol (Innova Biosciences). For flow cytometry, cells were digested, dispersed, labeled, and analyzed as previously described 40 .

Tumorigenicity assay in NOD/SCID mice
Cells were suspended in 50 μL of a 1:1 mixture of plain RPMI 1640 and Matrigel (BD Biosciences) and injected subcutaneously into the back of 4-to 6-week-old NOD/ SCID mice (Vital River). Tumor formation was monitored weekly.
To assess the tumorigenicity detection of α2δ1 + Huh7 cells with IP 3 R2 knockdown, sorted cells were first infected with shRNA-IP 3 R2 or scrambled control lentivirus for 4 h at 37 ℃.
All animals were treated in compliance with the Guide for the Care and Use of Laboratory animals published by the US National Institutes of Health (NIH Publication No. 85-23, revised 1996) and approved by the Animal Care and Use Committee of Peking University (accredited by AAALAC international).

RNA sequencing (RNA-seq) analysis
Using TRIzol reagent (Invitrogen), total RNA was isolated from cells according to the manufacturer's protocol. The quality of the extracted RNA was controlled using Agilent 2100. A sequencing library was prepared using the NEBNext ® Ultra TM RNA Library Prep Kit for Illumina ® (New England Biolabs) following the manufacturer's recommendations. Deep sequencing was then performed on an Illumina Hiseq 4000 platform and 150-bp paired-end reads were generated. To obtain the differentially-expressed genes, RNA-seq reads were first aligned to the human genome (hg19) with TopHat (v 1.2.0) and all the mapping results were evaluated as previously described 41 . Cuffdiff software (v 2.0.2) was used to obtain differentially expressed genes with a cutoff of fold-change > 2 and q value < 0.05. By searching Gene Ontology (http://www.geneontology.org/) we found Ca 2+ -related genes distributed in process, function, and component.

Statistics
The data are expressed as the mean ± SEM and, when appropriate, Student's t test was applied to determine statistical significance. P < 0.05 was considered statistically significant.