PI(4,5)P2 determines the threshold of mechanical force–induced B cell activation

B lymphocytes are responsible for the antibody responses arising from the recognition of pathological antigens by the surface-expressed B cell receptor (BCR; Pierce and Liu, 2010; Xu et al., 2014; Liu et al., 2016a). The BCR is composed of a membrane-bound immunoglobulin (mIg) and a noncovalently associated heterodimer of Igα and Igβ in a 1:1 mIg/Igα-Igβ heterodimer stoichiometry (Schamel and Reth, 2000; Tolar et al., 2005). BCR is an extraordinary receptor that can efficiently discriminate among a wide variety of chemical and physical features of antigens (Liu et al., 2016a) including antigen density (Fleire et al., 2006; Liu et al., 2010a; Tang et al., 2016; Wang et al., 2016), affinity (Fleire et al., 2006; Liu et al., 2010a), valency (Bachmann et al., 1993; Liu et al., 2004; Liu and Chen, 2005), Brownian mobility feature of antigen (Wan and Liu, 2012), the mechanical forces delivered to the BCRs by the antigens (Natkanski et al., 2013; Wan et al., 2015), and the stiffness feature of antigen-presenting substrates (Wan et al., 2013; Zeng et al., 2015; Shaheen et al., 2017). This discriminatory capacity plays a key role in B cell activation. Thus, elucidating the molecular mechanisms that enable B cells to discriminate among different antigens will provide important insights into how they develop the high-affinity antibodies crucial for an effective immune response.

Moreover, B cells exploit different BCR isotypes to recognize antigens and initiate transmembrane-activating signaling. Mature naive B cells use IgM- and IgD-BCRs, whereas memory B cells, which are responsible for the rapid antigen recall humoral responses upon vaccine immunization, mainly use isotype-switched IgG-BCRs (McHeyzer-Williams and McHeyzer-Williams, 2005; Pierce and Liu, 2010). Physical cues from the antigen can regulate B cell activation by applying a mechanical force on the BCR and have diverse effects on the different B cell subsets (Tolar, 2017). For example, compared with naive B cells, germinal center B cells apply more persistent and stronger tensile forces on the BCR. This negatively regulates antigen binding by using myosin II contractility to achieve more strict affinity discrimination during antigen extractions from immunological synapses (Nowosad et al., 2016). By using a double-stranded DNA (dsDNA)-based tension gauge tether (TGT) as a mechanical force sensor, we recently showed that IgM-BCR activation is highly dependent on mechanical forces, with the level of activation dependent on the amount of force (Wan et al., 2015). In contrast, the activation of isotype-switched IgG-BCR only requires a lower threshold of <12 pN (Wan et al., 2015). However, molecular mechanisms regulating these distinct thresholds of IgM-BCR versus IgG-BCR remain to be identified. In this study, we discover that the evolutionarily conserved cytoplasmic tail of the IgG-BCR heavy chain (IgG-tail) is responsible for the localized phosphatidylinositol (PI) (4,5)-biphosphate (PI(4,5)P2) enrichment by its PM-tethered and positively charged residues at resting stage, resulting in the low threshold of IgG-BCR activation by mechanical force.

To investigate the mechanical force–induced activation of different isotypes of BCRs, we constructed 4-hydroxy-3-nitrophenylacetyl (NP)-specific TGTs (NP-TGTs) to stimulate B1-8–specific BCRs with different amounts of mechanical force as described previously (Wan et al., 2015). In brief, each NP-TGT molecule is composed of two single-stranded DNA (ssDNA) molecules with specific modifications (Fig. 1 A). The first ssDNA molecule is biotin conjugated at three distinct positions to provide a defined range of rupture force (12, 43, and 56 pN), whereas the second ssDNA molecule is conjugated at the 3′ terminus with the B1-8 BCR–specific antigen NP. The activation of the BCRs is evaluated by quantifying the synaptic accumulation of both BCRs and phosphorylated spleen tyrosine kinase (Syk) in response to these NP-TGT mechanical force sensors (12, 43, and 56 pN) using total internal reflection fluorescence microscopy (TIRFM) imaging (Wan et al., 2015).

There are five isotypes of membrane-bound BCR: IgM-, IgD-, IgA-, IgG-, and IgE-BCR, which have distinctive constant regions and heavy chain cytoplasmic tails. We tested mechanical force–induced activation of all five isotypes of membrane-bound BCRs by constructing NP-specific human BCRs comprising the constant region of IgM-, IgD-, IgA-, IgG- and IgE-BCR combined with the B1-8 NP–specific variable region. IgM-, IgD-, and IgA-BCR shared the same multithreshold activation features: accumulation of BCRs was the strongest on high–mechanical force (56 pN) NP-TGTs, relatively moderate on medium–mechanical force NP-TGTs, and weak on low–mechanical force (12 pN) NP-TGTs. This indicates multiple mechanical force thresholds for the activation of IgM-, IgD-, and IgA-BCRs (Fig. 1, B–D). In marked contrast, IgG- and IgE-BCRs possess low-threshold characteristics as the recruitment of these BCRs was equally good on each of the 12-pN, 43-pN, and 56-pN NP-TGT surfaces (Fig. 1, E and F), consistent with our previous study (Wan et al., 2015). Further tests on downstream signaling of IgM- and IgG-BCR showed that the activation of MAPK (ERK) is positively correlated with the degree of BCR accumulation on the NP-TGT surfaces (Fig. 1, G–I).

Our previous study and the data in this study established a critical role for the conserved IgG-tail in lowering the threshold of mechanical force to induce IgG-BCR activation (Wan et al., 2015). There are two important motifs within the IgG-tail: the ITT motif and the PDZ-binding motif. When phosphorylated, the ITT motif (YKNM in IgG-tail) provides the binding site for growth factor receptor-bound protein 2 (GRB2), which together recruit BTK to enhance signaling transduction (Engels et al., 2009). The PDZ-binding motif (SSVV in the IgG-tail) provides the capacity to interact with SAP97, a PDZ motif–containing scaffold protein of the synapse-associated protein family, enhancing the formation of IgG-BCR microclusters (Liu et al., 2012). To investigate the contribution of these two motifs in lowering the threshold of mechanical force required to induce IgG-BCR activation, we constructed two mutant IgG-BCRs, YKNM to FKNM (Y/F) and SSVV/AAAA, to impair the function of the ITT motif and PDZ-binding motif, respectively (Fig. 1 J). These two mutant IgG-BCRs exhibited the same low threshold pattern as WT IgG-BCRs in response to NP-TGTs (Fig. 1, K and L). Thus, the low–mechanical force threshold of IgG-BCR activation is not dependent on the ITT or PDZ-binding motif within the IgG-tail.

Our previous study revealed that the IgG-tail in memory B cells interacted with the inner leaflet of the membrane at resting stage (Chen et al., 2015). Therefore, we investigated whether the PM-tethered status of the IgG-tail affects the threshold of mechanical force required to induce IgG-BCR activation. We rationally designed and constructed a series of IgG-BCR mutants with different PM-tethering status by modifying the sequence of the IgG-tail (Fig. S5). These mutants largely expressed similar BCRs on the PM as WT IgG-BCR (Fig. S1). We further examined their membrane-anchoring status in quiescent B cells by a fluorescence resonance energy transfer (FRET) assay following our published protocol (Chen et al., 2015). In brief, we fused the FRET donor, monomeric teal fluorescent protein (TFP; mTFP) to the C terminus of the membrane-bound IgG-tail (WT or mutant) and used the fluorescent membrane dye octadecyl rhodamine B (R18) as a FRET acceptor to stain the PM. As a positive control, the WT IgG-tail strongly associated with the inner leaflet of the PM as indicated by the high FRET efficiency between the IgG-tail–fused mTFP and R18 (Fig. 2 A), consistent with our previous research (Chen et al., 2015).

First, we detached the IgG-tail from the inner leaflet of the PM by inserting a flexible 25-aa linker between the transmembrane domain of mIgG and the N terminus of the IgG-tail (Fig. 2, A and B; Chen et al., 2015). We validated that both the 25 aa alone (Mut1) as a flexible linker control and the 25 aa–inserted IgG-tail (Mut2) dissociated from the PM by showing significantly lower FRET efficiency in comparison with the WT IgG-BCR (Fig. 2, C and D). When we analyzed the mechanical force–induced activation, we found that Mut1 and Mut2 now reacted to NP-TGTs in a multithreshold manner similar to that of WT IgM-BCR. This suggests that PM association of the cytoplasmic tail of IgG-BCR is required for the low–mechanical force threshold of activation (Fig. 2, E and F; see also Fig. 9).

The association of the IgG-tail with the inner leaflet of the PM is dependent on its positively charged and hydrophobic amino acid residues. The positively charged residues can interact with negatively charged acidic phospholipids, and the hydrophobic residues can be buried in the hydrophobic core of the membrane lipids. Considering these two criteria, we constructed another four mutant IgG-BCRs by mutating the positively charged and/or hydrophobic residues (Fig. 3 A). Both Mut3 and Mut4 were designed to decrease the proportion of positively charged amino acids as shown by the decreased isoelectric point (pI) and the reduced hydrophobic level of the IgG-tail of these two mutants in comparison with that of the WT IgG-tail. Mut5 was designed to keep the hydrophobic level intact but reduce the pI value of its cytoplasmic tail, whereas Mut6 was designed to keep the pI value intact but decrease the hydrophobicity compared with that of the WT IgG-tail (Fig. 3 A). All the cytoplasmic tails of these four mutant IgG-BCRs were dissociated from the PM as demonstrated by the drastically decreased FRET efficiencies between the IgG-tail–fused mTFP and R18 (Fig. 3, B–D). Thus, both the hydrophobic and positively charged residues on the IgG-tails are important for its PM association. Consistent with results for Mut1 and Mut2, when reacting to NP-TGTs, Mut3, Mut4, Mut5, and Mut6 also showed a typical multithreshold pattern similar to that of WT IgM-BCR, indicating the loss of the low–mechanical force threshold feature (Fig. 3, E–H; see also Fig. 9). Additionally, we found that the motilities of WT, Mut5–, and Mut6–IgG-BCRs were different in response to NP-TGTs. WT IgG-BCR moved faster at resting stage than Mut5–IgG-BCR and Mut6–IgG-BCR and significantly slowed down when encountering either 56-pN or 12-pN NP-TGTs. In contrast, Mut5– and Mut6–IgG-BCR only slowed down when encountering 56-pN NP-TGTs but not 12-pN NP-TGTs (Fig. 4). These results showed that the low–mechanical force threshold of IgG-BCR is dependent on the PM-tethered status of IgG-tail, the topology of which is mediated by both its hydrophobic and positively charged residues.

Furthermore, we designed several IgG-BCR with mutant IgG-tails by increasing positively charged residues (Mut7), reducing the hydrophobic and increasing the negatively charged residues (Mut8 as control for Mut7), or enhancing the hydrophobic level and keeping it almost electroneutral (Mut9 and Mut10). Mut7, Mut9, and Mut10 were theoretically and experimentally tethered to the PM in quiescent B cells as quantified by the abovementioned FRET assay (Fig. 5, A–D). Unexpectedly, only the positively charged mutant tail (Mut7) but not the other mutants (Mut9 and Mut10) displayed the low–mechanical force threshold pattern similar to WT IgG-BCR activation (Fig. 5, E–H; see also Fig. 9). Furthermore, IgM-BCRs equipped with the positively charged cytoplasmic tail of Mut7 were also able to lower the mechanical force threshold for IgM-BCR activation (Fig. S2, A–C). Thus, the presence of the positively charged residues of PM-tethered IgG-tail is mainly responsible for lowering the threshold on IgG-BCR activation by mechanical force.

Our previous research demonstrated that the positively charged residues of the IgG-tail interact with acidic phospholipids in the inner leaflet of the PM (Chen et al., 2015). The IgG-tail may recruit acidic phospholipids, which are known to be involved in signaling transduction in many types of cells (Falkenburger et al., 2010), through its positively charged residues. Therefore, we speculated that acidic phospholipids may be responsible for IgG-BCR activation by lowering the threshold of mechanical forces. To validate our hypothesis, we screened the enrichment of different types of acidic phospholipids, including phosphatidylserine (PS), PI(4,5)P2, PI (3,4,5)-trisphosphate (PI(3,4,5)P3), and phosphatidic acid (PA), into the WT IgG-BCR membrane microdomains on the PM by superresolution imaging of mEos3.2-based lipid-specific biosensors. We used the PH domain of phospholipase C δ (PLCδ) for tracking PI(4,5)P2 (Lemmon et al., 1995; Halet, 2005), Lact-C2 as a PS biosensor (Yeung et al., 2008), Grp1-PH binding domain to indicate PI(3,4,5)P3 (Knight and Falke, 2009), and the 51–91-aa region of Spo20 for probing PA (Kassas et al., 2012; Zhang et al., 2014a). As indicated in Fig. 6 A, only PI(4,5)P2 and not PS, PI(3,4,5)P3, nor PA was enriched in IgG-BCR membrane microdomains in quiescent B cells (Fig. 6, A and B). As further validation, we used a PI(4,5)P2-specific monoclonal antibody to stain PI(4,5)P2 and confirmed that PI(4,5)P2 was enriched in IgG-BCR membrane microdomains in quiescent B cells (Fig. S3 A). We also tested the recruitment of other lipid species such as diacylglycerol (DAG) by using an mEos3.2-fused C1 domain of PKCθ (Huse et al., 2007), and we found no significant enrichment (Fig. 6, A and B). All these data indicate that the IgG-tail enriches PI(4,5)P2 on the PM in quiescent B cells.

We next tested whether PI(4,5)P2 enrichment was dependent on the PM-tethered status of the IgG-tail by comparing enrichment of the acidic phospholipids within WT, Mut5–, or Mut6–IgG-BCRs. As mentioned above, both Mut5– and Mut6–IgG-BCRs did not tether their IgG-tail to the PM (Fig. 3 D). We found that PI(4,5)P2 enrichment was only obvious in the case of WT IgG-BCRs but not Mut5– nor Mut6–IgG-BCRs (Figs. 6 C and S3 B). We also validated this conclusion through superresolution imaging for both PI(4,5)P2 and IgG-BCR molecules by combining direct stochastic optical reconstruction microscopy (dSTORM) and photoactivated localization microscopy (PALM) technology (Fig. 6 D). Consistently, the results demonstrated that PI(4,5)P2 colocalized better with WT IgG-BCRs than with Mut5– or Mut6–IgG-BCRs (Fig. 6, E–G). All these data indicate that PI(4,5)P2 enrichment at IgG-BCRs is dependent on the PM-tethered status of the IgG-tail. Similarly, we tested the enrichment of PS, PI(3,4,5)P3, PA, and DAG into WT, Mut5–, and Mut6–IgG-BCRs and found no obvious difference (Fig. S3, C–F).

To further characterize the role of positively charged residues in enriching PI(4,5)P2 in the PM-tethered IgG-tail, we measured PI(4,5)P2 enrichment in Mut7 (IgG-tail highly positively charged and PM tethered) at resting stage. Mut7 enriched PI(4,5)P2 similarly to WT IgG-BCR (Fig. 6, H–J). As a negative control, no significant PI(4,5)P2 enrichment was observed in Mut9 (PM tethered but lacking positively charged residues; Fig. 6, K–M). These data indicate that positively charged residues in the PM-tethered IgG-tail are responsible for PI(4,5)P2 enrichment.

The aforementioned results drove us to speculate that PI(4,5)P2 enrichment might be responsible for lowering the mechanical force threshold during the initiation of IgG-BCR activation. To test this hypothesis, we designed a series of experiments to purposely manipulate the amount of PI(4,5)P2 on the PM and then investigated the subsequent effect on the mechanical force threshold of IgG-BCR activation.

First, we evaluated the effect of decreasing the general amount of PI(4,5)P2 on the PM by using a PI(4,5)P2 phosphatase, Ins54p, a 5′ phosphatase from yeast, which dephosphorylates PI(4,5)P2 to produce PI 4-phosphate (Stolz et al., 1998; Raucher et al., 2000; Suh et al., 2006). We also used an Ins54p phosphatase-dead mutant as a negative control. To maximize the efficiency of reducing the amount of PI(4,5)P2 on the PM through Ins54p, we fused the PM-anchoring sequence Lyn16 to the N terminus of Ins54p WT (Lyn16-Ins54p-WT) and the phosphatase-dead control (Lyn16-Ins54p phosphatase-dead control; Fig. 7 A). To quantitatively analyze the amount of PI(4,5)P2 on the PM, we analyzed the ratio of mean fluorescent intensity (MFI) of the PI(4,5)P2-specific fluorescent biosensor (GFP-PH) on the PM versus GFP-PH MFI in the cytosol (Halet, 2005). We observed a marked decrease of the PI(4,5)P2 sensor on the PM in the B cells transfected with Lyn16-Ins54p-WT in comparison with the Lyn16-Ins54p phosphatase-dead control (Fig. 7, B and C). We next tested NP-TGT–triggered activation of WT IgG-BCR–expressing B cells transfected with Lyn16-Ins54p-WT or the phosphatase-dead control. Consistent with our hypothesis, a profound difference was observed between these two types of IgG-B cells: IgG-B cells transfected with Lyn16-Ins54p-WT but not the ones expressing Lyn16-Ins54p phosphatase-dead control failed to maintain the low–mechanical force threshold of IgG-BCR activation (Fig. 7 D).

To exclude the possibility that the general manipulation of PI(4,5)P2 on the PM might indirectly affect the threshold of IgG-BCR activation by mechanical force, we designed a system to locally manipulate PI(4,5)P2 levels only within IgG-BCR membrane microdomains. Initially, we fused Ins54p to the C terminus of the IgG-tail; however, such chimeric BCRs did not localize to the PM (not depicted). To solve this problem, we took advantage of the rapamycin-based inducible system: rapamycin can simultaneously bind to the FK506 binding protein (FKBP) and FKBP–rapamycin binding (FRB) domains to form a ternary complex (Banaszynski et al., 2005; Suh et al., 2006). Therefore, we fused FKBP with WT Ins54p or Ins54p phosphatase-dead control and the FRB domain with the WT IgG-tail (Fig. 7 E). The WT IgG-tail fused with the FRB domain was well expressed on the PM. After preincubation with rapamycin for 5 min, PI(4,5)P2 enrichment within the WT IgG-BCR–FRB membrane microdomains of B cells expressing FKBP-Ins54p-WT was significantly reduced in comparison with B cells expressing the FKBP-Ins54p phosphatase-dead control (Fig. 7, F and G). Remarkably, WT IgG-BCR–FRB B cells expressing FKBP-Ins54p-WT but not the ones expressing the FKBP-Ins54p phosphatase-dead control failed to maintain the low–mechanical force threshold for IgG-BCR activation (Fig. 7, H and I). As the system control, without rapamycin preincubation treatment, these two cell lines showed no difference on both PI(4,5)P2 enrichment and BCR activation by mechanical force (Fig. S4, A and B). All these results clearly demonstrate that PI(4,5)P2 levels on the PM within IgG-BCR membrane microdomains are of crucial importance for the low–mechanical force threshold in IgG-BCR activation.

Reversely, we examined whether locally increased amounts of PI(4,5)P2 within the membrane microdomains of IgG-BCR mutants with PM-untethered IgG-tails would rescue the loss of the low–mechanical force threshold pattern of these IgG-BCR mutants. We overexpressed in B cells the Lyn16-fused PI 4-phosphate 5-kinase γ (Lyn16-PIP5K), which can potently phosphorylate PI 4-phosphate at the 5-position of the inositol ring to increase the amount of PI(4,5)P2 on the PM. To maximize the efficiency of increasing the amount of PI(4,5)P2 on the PM through Lyn16-PIP5K, we also preincubated B cells with Myo-inositol, a widely used and important component of PI and its various phosphates (Villalobos et al., 2011). Indeed, Myo-inositol preincubation and Lyn16-PIP5K overexpression significantly increased the amount of PI(4,5)P2 in comparison with Myo-inositol preincubation alone or Lyn16-PIP5K overexpression alone (Fig. 8, A–C; and Fig. S4, C–I). Next, we examined the effect of increasing PI(4,5)P2 on the activation of PM-untethered Mut6–IgG-BCR. The results showed that the preincubation of Myo-inositol in combination with the overexpression of Lyn16-PIP5K-WT but not the Lyn16-PIP5K kinase-dead control rescued the mechanical force threshold for the activation of Mut6–IgG-BCR (Fig. 8 D). As a further test, we also similarly increased the amount of PI(4,5)P2 on the PM of B cells expressing IgM-BCRs and found that such treatment lowered the mechanical force threshold of IgM-BCRs (Figs. 8 E and S4 H).

We again took advantage of the rapamycin-based inducible system to locally manipulate PI(4,5)P2 levels only within IgG-BCR membrane microdomains. In brief, FKBP was fused with PIP5K-WT or the PIP5K kinase-dead control, and the FRB domain was fused with Mut6–IgG-tail. After preincubation of Myo-inositol for 1 h and rapamycin for 5 min, PI(4,5)P2 enrichment within Mut6–IgG-BCR–FRB membrane microdomains of B cells expressing FKBP-PIP5K-WT was significantly increased in comparison with B cells expressing the FKBP-PIP5K kinase-dead control (Fig. 8, F-H). Mut6–IgG-BCR–FRB B cells expressing FKBP-PIP5K-WT but not the ones expressing the FKBP-PIP5K kinase-dead control significantly lowered the mechanical force threshold of IgG-BCR activation (Fig. 8, I and J). As the system control, without rapamycin preincubation treatment, these two cell lines showed no difference on both PI(4,5)P2 enrichment and BCR activation by mechanical force (Fig. S4, J and K). Thus, all these data indicate that enrichment of PI(4,5)P2 within IgG-BCR membrane microdomains is both required and sufficient to lower the mechanical force threshold during the initiation of IgG-BCR activation (see also Fig. 9).

This study reported that the low–mechanical force threshold of IgG-BCR activation depends on the PM-tethered state of the IgG-tail in quiescent B cells. Moreover, positively charged residues in PM-tethered IgG-tail are crucial for recruiting PI(4,5)P2 within IgG-BCR membrane microdomains and consequently maintain the low–mechanical force threshold of IgG-BCR activation. Manipulating the amounts of PI(4,5)P2 within IgG-BCR membrane microdomains drastically changes the thresholds of mechanical force–induced B cell activation. Our study sheds light on an IgG-tail–mediated PI(4,5)P2 enrichment–dependent mechanism that lowers the mechanical force threshold of IgG-BCR activation. All these findings highlight a novel explanation on the low force requirement to initiate the activation of IgG-BCR in memory B cells and imply a new mechanism to explain the rapid IgG antibody responses of memory B cells upon reencountering antigens.

Our study used the dsDNA-based TGT system that was originally developed by Wang and Ha (2013). Because the minimum rupture force value of TGTs is ∼12 pN, it will be interesting to take other approaches (Zhang et al., 2014b; Chowdhury et al., 2016) to test how the mechanical force <12 pN would induce IgG-BCR activation. Moreover, the TGT system cannot be applied to measure the dynamic changes of mechanical forces exerted through BCR structures during antigen recognition because its rupture is irreversible. Thus, future studies using the nonrupturable tension sensors will also be required (Liu et al., 2016b; Nowosad et al., 2016; Spillane and Tolar, 2017). In our experiments, we used the Ramos human B cell line, J558L/CH27 mouse B cell line, and DT40 chicken B cell line. We chose to use these different types of B cells for certain reasons: (A) it is required to ensure that the conclusion is of general meanings but is not limited to a certain type of B cells; (B) there is no single type of B cell that can fulfill the purpose of all the experiments in our manuscript. Moreover, we expressed the species-matched BCRs or the complete BCR complex in these cell lines to ensure the proper BCR expression and the following experiments. For example, we expressed the BCRs with human constant regions to Ramos human B cell lines, the BCRs with mouse constant regions to J558L/CH27 mouse B cell lines, and the whole complex of mouse BCRs to DT40 and 293T cell lines.

Our study demonstrates that PI(4,5)P2 enrichment in IgG-BCR membrane microdomains is vital for IgG-BCR activation by low mechanical force; however, we cannot rule out the contributions of other phospholipids such as PI, which has the largest number of negatively charged lipid species on the PM. According to our knowledge, there unfortunately is no well-developed PI biosensor with high specificity. Besides, our study shows that PI(3,4,5)P3 is not significantly enriched in IgG-BCR membrane microdomains compared with PI(4,5)P2 at resting stage. Such an observation is unexpected because PI(3,4,5)P3 is more negatively charged than PI(4,5)P2 at physiological conditions, and theoretically, PI(3,4,5)P3 would be more attracted by positively charged residues in IgG-tail. We suspect that such a phenomenon is because the total amount of PI(3,4,5)P3 is much less than that of PI(4,5)P2 on the PM. In this study, we found that at the physiological conditions of a B cell, PI(4,5)P2 enrichment by positively charged residues of PM-tethered cytoplasmic tail determines the mechanical force threshold of IgG-BCR activation. In marked contrast, at the nonphysiological condition, when B cells have artificially increased the total amount of PI(4,5)P2 on the PM by the overexpression of Lyn16-PIP5K and the preincubation with Myo-inositol, the amount of PI(4,5)P2 within Mut6–IgG-BCR membrane microdomains increased because there was overall more PI(4,5)P2 on the PM. In this nonphysiological condition, Mut6–IgG-BCR and IgM-BCR exhibited the low–mechanical force threshold in response to NP-TGTs, indicating that with too much PI(4,5)P2 on the PM, the function of PI(4,5)P2 in the force sensitivity profile is not necessary through direct interaction with IgG-tail.

PI(4,5)P2 may facilitate and direct BCR accumulation into immunological synapses when B cells are stimulated by low mechanical forces and also may provide directionality and contextual cues for downstream signaling to boost IgG B cell activation in several aspects: First of all, it is known that PI(4,5)P2 is one of the key lipid species managing the cytoskeleton system through regulating PI(4,5)P2-binding proteins such as N-WASP (Miki et al., 1996), Cdc42 (Burbage et al., 2015), the ezrin, radizin, and moesin family (Treanor et al., 2011), vinculin, talin, gelsolin, profilin, and other key regulators (Czech, 2000; Logan and Mandato, 2006). Second, the enrichment of PI(4,5)P2 within IgG-BCR membrane microdomains increases the PI(4,5)P2 accessibilities for other PI(4,5)P2-binding proteins, especially enzymes such as PI 3-kinase, PLC, and phospholipase D (McLaughlin et al., 2002; Hikida et al., 2009; Xu et al., 2015; Petersen et al., 2016), to amplify the weak signals initiated by the low mechanical force to further activate downstream signalosomes through PI(3,4,5)P3, IP3/Ca²⁺, or PA, respectively.

The cytoskeleton controls the mobility of BCRs and the formation of BCR microclusters during the initiation of B cell activation (Tolar et al., 2009; Liu et al., 2010a; Treanor et al., 2010, 2011; Schnyder et al., 2011; Davey and Pierce, 2012; Natkanski et al., 2013; Wang et al., 2016). When BCRs are activated by lipid bilayer–presenting antigens, the mobility of BCRs will be reduced concomitant with F-actin remodeling, BCR oligomerization, and the formation of BCR microclusters, all of which work concertedly to promote B cell activation (Tolar et al., 2009; Liu et al., 2010a,b). Indeed, our study showed that in conditions lacking PI(4,5)P2 enrichment within the PM-untethered mutant IgG-BCR membrane microdomains, the movement of these mutant IgG-BCRs cannot consistently slow down when activated by 12-pN NP-TGTs; however, PI(4,5)P2-enriched WT IgG-BCR can significantly reduce its mobility on 12-pN NP-TGTs, implying a sophisticated mobility regulation system through IgG-tail–enriched PI(4,5)P2. Additionally, related to BCR mobility, our study shows that WT IgG-BCRs diffuse faster than PM-untethered mutant IgG-BCRs at the resting stage but move slower than PM-untethered mutant IgG-BCRs when encountering 12-pN NP-TGTs. All these observations inspire us to hypothesize that the rupture forces that can be exerted by the energies from BCR mobility could be likely contributed as a sum by the factors including the oligomerization of BCRs, the heterogeneity of lipid species, membrane bending (curvature), cytoskeletons, motor proteins, et cetera. Moreover, the kinetic energy changes from fast mobility to slow movement might be also transformed to other types of energy such as elastic potential energy from BCR conformational change or cytoskeleton-shaping tension to initiate BCR activation signaling.

The enrichment of negatively charged lipid species by the positively charged juxtamembrane region at the resting stage has also been found in epidermal growth factor receptor, in which PI(4,5)P2 enrichment controls the clustering of epidermal growth factor receptor in quiescent state, further regulating downstream activation (Wang et al., 2014). This study showed that the value of PI(4,5)P2 enrichment within IgG-BCR membrane microdomains is ∼1.3, which is similar to the value of IgM-BCR within Syk microclusters (∼1.5; Mattila et al., 2013) but is not as dramatically high as the value of FcγRIIB enrichment in immune complex microclusters (∼3.0; Xu et al., 2016), indicating that potentially there are other membrane proteins or proteins located in membrane-proximal region in B cells regulating the PI(4,5)P2 distribution on the PM. We speculate that there might be a PI(4,5)P2-mediated orchestration system coordinately regulating different membrane-bound proteins and signal proteins to initiate activation of B cells that is yet to be discovered.

IgG mutants—which cannot recruit PI(4,5)P2 around IgG-BCR membrane microdomains at the resting stage—such as Mut5, get activated by mechanical force in the same multithreshold matter as IgM-BCRs. We speculate that IgM- and IgG-BCR may share the same force-transduction components. The antigenic signal needs to be delivered from the antigen-binding region on the extracellular domain of IgM/IgG-BCR to the signaling active immunoreceptor tyrosine–based activation motif on the cytoplasmic domain of both Igα and Igβ. However, the mechanism is still enigmatic. Because antigen binding–induced formation of BCR microclusters relies on conformational change of BCR heavy chain (Tolar et al., 2009), we speculate that mechanical force–induced BCR activation may also depend on conformational change to transduce force to chemical signals, similar to TCR (Kim et al., 2009; Huang et al., 2010; Liu et al., 2014; Lee et al., 2015; Dustin and Kam, 2016). Moreover, we speculate that Igα and Igβ may play different roles in the mechanical force transduction process, providing two distinct signaling pathways resulting in multiple mechanical force thresholds of activation as implied by Lee and Tolar (2013) that cytoplasmic domain of Igα increases membrane proximity after stimulation but not Igβ. All these speculations are under our further extensive investigation.

In summary, the findings presented in this study indicate a mechanism for mechanical force induced more sensitive activation of IgG-BCR–expressing memory B cells. Such a mechanism is achieved by the enrichment of the vital signaling transduction phospholipid PI(4,5)P2 through PM-tethered positively charged IgG-tail in quiescent B cells to boost IgG-BCR activation in a low–mechanical force threshold manner. Our study may have far-reaching implications on vaccine development and design and offers fresh prospects for explanation of rapid memory B cell activation.

Mouse B1-8–specific WT IgG-tail–mTFP plasmid was used as described previously (Chen et al., 2015). In brief, a monomeric version of TFP was attached to the C termini of mouse NP-specific B1-8 IgG heavy chains constructed in a pHAGE backbone. All the mutant cytoplasmic tail vectors were constructed based on the WT version in pHAGE backbone. Human B1-8–specific IgG-, IgM-, and IgE-heavy chain plasmids were constructed as described in our previous study (Wan et al., 2015). In short, human IgG-, IgM-, IgE-, IgA-, and IgD-constant regions of heavy chains were fused with a B1-8–specific variable region at the N terminus and with mTFP at the C terminus to construct B1-8–specific heavy chains. The plasmid of the human version of the B1-8–specific light chain was constructed by fusing the B1-8–variable region to the N terminus of the human Igλ-constant region. The cDNA mEos3.2 was a gift from P. Xu and T. Xu (Institute of Biophysics, Beijing, China); the plasmid PIP5Kγ87 was a gift from P. De Camilli (Yale University, New Haven, CT; 22300; Addgene); and CP-Ins54p, CP-Ins54p phosphatase-dead mutant (D281A), Lyn11-targeted FRB, and GFP-C1-PLCδ-PH were gifts from T. Meyer (Stanford University, Stanford, CA; 20155, 20156, 20147, and 21179, respectively; Addgene). PM-anchoring Ins54p-WT or phosphatase-dead mutant (D281A) and PIP5K-WT or kinase-dead mutant were constructed by fusing the membrane-anchoring sequence Lyn16 to the N terminus of these plasmids. BFP/mCherry-FKBP-Ins54p and BFP/mCherry-FKBP-Ins54p phosphatase-dead control plasmids were constructed based on CF-Ins54p and CF-Ins54p (D281A). FKBP fragment was fused to the PIP5K-WT or PIP5K kinase-dead mutant (D253A) to construct FKBP-PIP5K-WT or FKBP-PIP5K kinase-dead control. All PIP5K constructs used in this study were based on a mutant version of PIP5Kγ with R445E, K446E, and S264A substitutions to prevent its intrinsic PM localization (Suh et al., 2006). The FRB fragment was cloned from Lyn11-targeted FRB and fused to the C terminus of the WT– or Mut6–cytoplasmic tail of IgG-tail. All these plasmid construction, point mutations, and truncations were performed following Gibson assembly protocol (Gibson et al., 2009).

Mouse J558L cells, CH27, and human Ramos B cells were cultured in RPMI 1640 medium containing 10% FBS, penicillin, and streptomycin antibiotics (Invitrogen) as described previously (Liu et al., 2010c; Sohn et al., 2011). All these B cell lines were gifts from S.K. Pierce (National Institutes of Health, Bethesda, MD). DT40 B cells (Takata et al., 1995) were gifts for laboratory studies from T. Kurosaki (Institute of Physical and Chemical Research, Yokohama, Japan). DT40 B cells were cultured in RPMI 1640 medium containing 10% FBS, 1% chicken serum, 50 µM 2-mercaptoethanol, penicillin, and streptomycin antibiotics (Thermo Fisher Scientific) as described previously (Ishiai et al., 1999; Weber et al., 2008). 293T cells were cultured in DMEM containing 10% FBS, penicillin, and streptomycin antibiotics. Mouse J558L cells stably expressing Igα-YFP were constructed and maintained as previously described (Liu et al., 2010a) and further transfected with IgG–heavy chain of WT or mutant cytoplasmic tails. 293T or DT40 cells were transiently transfected with light chain, Igα, Igβ, and heavy chain of WT or each mutant cytoplasmic tail to assemble intact IgG-BCRs by calcium phosphate transfection or electroporation (Lonza), respectively. Ramos B cells were transfected with B1-8–specific human light chain and heavy chain of IgM-, IgD-, IgA-, IgG-, or IgE-BCR to test the mechanical force threshold of diverse types of BCR activation. Ramos B cells were transfected with WT or mutant IgG–heavy chain and mEos3.2-PH to detect the PI(4,5)P2 enrichments. CH27 B cells were transfected with WT or mutant IgG–heavy chain and mEos3.2-fused Lact-C2, Grp1-PH, and the 51–91-aa region of Spo20 for probing PA to detect the enrichments of PS, PI(3,4,5)P3, and PA in IgG-BCR membrane microdomains, respectively.

Alexa Fluor 647 or Cy3 AffiniPure Fab fragment goat anti–mouse IgG1 Fc fragment specific and Cy5-Fab fragment goat anti–mouse IgM µ chain specific (Jackson ImmunoResearch Laboratories, Inc.) were used for mouse IgG- and IgM-BCR cell surface staining, respectively. Alexa Fluor 647 AffiniPure Fab fragment goat anti–human IgG Fcγ fragment specific (Jackson ImmunoResearch Laboratories, Inc.) was used for human IgG-BCR cell surface staining. In short, the cells were stained with 100 nM antibody on ice for 10 min. After washing three times with PBS, the cells were ready for further experiments. Myo-inositol was purchased from Sigma-Aldrich. Rapamycin was a gift from Y. Zhang (Tsinghua University, Beijing, China).

J558L and DT40 B cell lines expressing WT or each mutant IgG-BCR were tested for both PM expression and total expression by flow cytometry (Accuri C6; BD). In detail, to test the BCR expression on the PM, intact B cells were stained with 100 nM Alexa Fluor 647 Fab fragment goat anti–mouse IgG1 Fc fragment specific on ice for 10 min; afterward, cells were washed twice with PBS and were ready to examine under flow cytometry. To examine the total expression of BCRs, B cells were fixed by 4% PFA for 30 min. After washing twice with PBS, the B cells were permeabilized with 0.1% Triton X-100 and then stained with Alexa Fluor 647 Fab fragment goat anti–mouse IgG1 Fc fragment specific for 30 min. After washing three times with PBS, cells were ready for flow cytometry examination.

NP-TGTs were prepared by following the published protocol (Wang and Ha, 2013) with some modifications. In detail, we conjugated NP hapten to one DNA strand (ssDNA) of the dsDNA of TGTs to specifically activate B cells. NP-e-aminocaproyl-OSu (Biosearch Technologies) were conjugated to ssDNA with NH2 group modification. The sequence used was 5′-CACAGCACGGAGGCACGACAC-NH2/-3′.

The other strand of the TGT had a biotin tag at a predesigned position for performing different rupture force points and binding to the coverslip through a biotin-neutravidin bond. The sequence used was 5′-GTGTCGTGCCTCCGTGCTGTG-3′ with the biotin label at the first, 11th, and 21st bases, which formed 12, 43, and 56 pN, respectively. NP-ssDNA and biotin-ssDNA were further annealed in the annealing buffer following the protocol from Invitrogen.

Coverslips (VWR International) were pretreated with stripping buffer (7:3 H₂SO₄/H₂O₂), washed, and dried. Then, dried coverslips were attached to the disposable eight-well chamber frame (Lab-Tek chambers; Nunc) or a home-designed microchamber with glue. Each well of microchamber had a diameter of 2.5 mm. After 20 min of curing at RT, chamber was ready to use for downstream experiment. 200 µg/ml neutravidin was added to the coverslip, and after incubation for 30 min at 37°C, extensive washing was performed. NP-TGTs were then loaded to the coverslip at the concentration of 50 nM for 30 min at RT for the purpose of tethering NP-TGTs to the coverslip. After careful washing with PBS, the coverslip was blocked with 1% casein (wt/vol) in PBS for 30 min at 37°C. After washing again with PBS, the NP-TGT–conjugated coverslip was ready for further use. The prestained cells were then loaded on the surface for reaction at 37°C for 10 min if there were no specific indications.

TIRFM images were acquired by an Olympus IX-81 microscope equipped with a TIRF port, an Andor iXon⁺ DU-897D electron-multiplying charge-coupled device camera, and an Olympus 100× 1.49 NA objective lens. The acquisition was controlled by MetaMorph software (Molecular Devices). For the imaging options, the exposure time was 100 ms for a 512 × 512–pixel image unless specially indicated. Confocal images were acquired by a FLUOVIEW FV1000 confocal laser scanning microscope (Olympus) with a 60× 1.42 NA oil objective lens. All the images were acquired while keeping the cells in PBS unless specially indicated and were confirmed not to overexpose by the software. All the images were analyzed and processed with ImageJ (National Institutes of Health). The MFI and the total fluorescence intensity (total FI) as AU of BCRs, lipid biosensors, and signaling molecules were calculated based on the intensity analysis as described previously (Lakadamyali et al., 2004; Liu et al., 2010b,c, 2012).

To perform the FRET experiments, 1 × 10⁶ cells were collected and washed once with 1× PBS and then suspended in 1 ml cell buffer. Before loading to the poly-l-lysine (Sigma-Aldrich)–coated chambered coverglass slides (Thermo Fisher Scientific), 300 µl aliquots of cells were stained with 300 nM octadecyl rhodamine B (R18; Invitrogen) on ice for 3 min. The chamber was then mounted onto an FV1000 microscope and maintained for 5 min to allow cell adhesion. Dequenching FRET images were captured (Xu et al., 2008; Shi et al., 2013; Chen et al., 2015). mTFP was excited with a 473-nm laser and visualized using a 520/40 bandpass filter. R18 was excited with a 559-nm laser line and visualized using the 625/100 bandpass filter. All the FRET experiments were performed in Ringer’s buffer with 2 mM Ca²⁺ and 1 mM Mg²⁺. The images of same cells on both mTFP and R18 channels were acquired before and after bleaching the acceptor (R18) by a 559-nm laser to dequench the donor (mTFP). All the images were processed with ImageJ, and the MFI of IgG-BCRs only expressed on the PM was quantified by using region of interest tools in ImageJ. FRET efficiency was calculated with the formula FRET efficiency = (DQ − Q)/DQ, with DQ and Q representing dequenched and quenched donor fluorescence intensity (mTFP), respectively.

The recruitment of signaling molecules into the immunological synapse of B cells stimulated by NP-TGTs was imaged by TIRFM by following our previously published protocol (Liu et al., 2010a,b,c, 2012). In brief, BCRs expressed on the PM were prestained with fluorescent dye labeled with Fab antibody as described in the Cells, antigens, antibodies, and reagents section. Then, B cells were loaded to the chambered coverglasses to react with NP-TGTs for 10 min followed by 4% PFA fixation for 30 min. After washing with 10 ml PBS, the B cells were permeabilized with 0.1% Triton X-100 and then blocked with 100 µg/ml goat nonspecific IgG (Jackson ImmunoResearch Laboratories, Inc.). Subsequently, cells were stained with phospho–Zap-70(Tyr319)/Syk (Tyr352) primary antibody (Cell Signaling Technology) at 37°C for 1 h. After washing with 10 ml PBS, B cells were stained with secondary antibody Alexa Fluor 568–conjugated F(ab’)₂ goat antibody specific for rabbit or mouse IgG (Invitrogen) as previously described (Liu et al., 2012). PI(4,5)P2 intracellular staining was performed by following a protocol provided by Echelon Biosciences Inc. In brief, 293T cells expressing human WT IgG-BCRs were fixed using 4% PFA (wt/vol), permeabilized with 0.5% saponin (Sigma-Aldrich), blocked by 10% goat serum and 2.5% BSA, and then incubated by 10 µg/ml anti-PI(4,5)P2 monoclonal antibody (Z-P045; Echelon Biosciences Inc.) at 37°C. After washing, the cells were stained with a second antibody, Cy3-conjugated F(ab’)₂ goat anti–mouse IgM, at a concentration of 2 µg/ml. In downstream activation tests, J558L cells expressing WT IgM- or WT IgG-BCRs were rested in RPMI 1640 culture media for 30 min before being loaded onto the NP-TGT–coated coverglasses. After 10 min of incubation, cells were fixed by 4% PFA, permeabilized with cold methanol, blocked with 10% goat serum, and sequentially stained with the anti-pERK antibody (AF1018; R&D Systems), an Alexa Fluor 647–conjugated F(ab’)₂ goat anti–rabbit IgG (Invitrogen), and a nucleic acid dye Hoechst. After washing with PBS, the cells were imaged by fluorescence microscopy (Su et al., 2016). Images were analyzed by ImageJ following our published protocols (Liu et al., 2010a,b,c, 2012).

For increasing PI(4,5)P2 studies, DT40, or J558L, B cells expressing either Lyn16-PIP5K kinase-dead control or Lyn16-PIP5K-WT were pretreated with Myo-inositol (Sigma-Aldrich) in 1 mM at cell culture conditions for 1 h before the imaging experiment (Villalobos et al., 2011).

To acquire the PALM images of PI(4,5)P2, PS, PI(3,4,5)P3, PA, and DAG at resting stage, specified cells respectively expressing mEos3.2-fused PLCδ-PH (Lemmon et al., 1995; Halet, 2005), Lact-C2 (Yeung et al., 2008), Grp1-PH (Knight and Falke, 2009), the 51–91-aa region of Spo20 (Kassas et al., 2012; Zhang et al., 2014a), or C1-PKCθ (Huse et al., 2007) were placed on poly-l-lysine– or fibronectin-coated coverglasses and fixed with 4% PFA at 10 min. Stream videos were acquired by 561-nm laser (10 mW) irradiation at a recording rate of 30 frames per second simultaneously with a continuous low-power 405-nm laser to convert a small portion of mEos3.2 from green to red to generate single molecular signals. Image registration and drift correction were achieved by the position of fluorescent microspheres (100 nm TetraSpek; Thermo Fisher Scientific). For analysis of PALM images, peaks of single points were first identified by using Insight3 software provided by X. Zhuang (Harvard University, Cambridge, MA), B. Huang (University of California, San Francisco, San Francisco, CA), and Y. Sun (Peking University, Beijing, China). Enrichment of PI(4,5)P2, PS, PI(3,4,5)P3, PA, and DAG inside BCRs was measured using the MATLAB (MathWorks) codes written according to the algorithms reported previously (Zhang et al., 2006; Owen et al., 2010; Mattila et al., 2013). In detail, the enrichment of lipid molecules within IgG-BCR membrane domains was analyzed in a rectangular window of 3 µm side length around the center of the cells, and we calculated the enrichment of lipid molecules within IgG-BCR membrane domains as the fraction of localized lipid molecules residing in IgG-BCR membrane domains divided by the fraction of the area occupied by IgG-BCR membrane microdomains.

dSTORM technology was used for superresolution imaging of IgG-BCRs. Alexa Fluor 647–conjugated AffiniPure Fab fragment goat anti–mouse IgG1 Fc fragment specific was used to stain IgG-BCRs for two-color superresolution imaging of BCR and PI(4,5)P2. Before imaging, fixation buffer was washed, and imaging buffer (containing 140 mM β-mercaptoethanol, 0.5 mg/ml glucose oxidase, and 40 µg/ml catalase) was loaded. dSTORM images were acquired by a commercially available Nikon system based on an inverted microscope (Nikon Ti-E) equipped with an oil-immersion TIRF objective (Apochromat TIRF100X, 1.49 NA; Nikon) as well as four lasers and a filter set suitable for multiple color imaging. L-function was used for examining the bivariate pair correlation between IgG-BCRs and PI(4,5)P2 in the two-color superresolution imaging experiments (Wiegand and Moloney, 2004; Barr et al., 2016). In detail, L-function is a transformation of K-function. The bivariate K-function K₁₂(r) was defined as the expected number of points of pattern 2 within a given distance r of an arbitrary point of pattern 1, divided by the intensity e₂ of points of pattern 2:

where # means ‘‘the number of’’ and E[ ] is the expectation operator under independence of the two-point patterns, K­₁₂­(r) = πr², without regard to the individual univariate point patterns. It can be difficult to interpret K₁₂(r) visually. Therefore, a square root transformation of K(r) called L-function was used instead:

This transformation (L-function) removes the scale dependence of K₁₂(r) for independent patterns and stabilizes the variance. Values of L₁₂(r) > 0 indicate that there were on average more points of pattern 2 within distance r of points of pattern 1 as one would expect under independence, thus indicating attraction between the two patterns up to distance r. The L-function analysis used in this study was computed using MATLAB. dSTORM technology was also used for capturing the superresolution images of monoclonal antibody–stained PI(4,5)P2 in 293T cells expressing human WT IgG-BCRs.

Single-molecule tracking was performed following our previously published protocol (Liu et al., 2010a; Xu et al., 2016). B cells were labeled with 1 nM Alexa Fluor 647–conjugated goat Fab anti–mouse IgG1 Fc region antibodies for the best resolution of individual fluorescence spots without the need for further extensive photobleaching. Labeled B cells were washed twice before imaging, and images were taken using TIRFM with a 647-nm laser. A subregion of ∼90 × 90 pixels of the available area of the electron-multiplying charge-coupled device chip was used and recorded at 30 frames per second. Single molecules were analyzed as described in our previous study (Liu et al., 2010a) using MATLAB code based on available positional fitting and tracking algorithms. Mean square displacement and diffusion coefficients for each molecule trajectory were calculated from positional coordinates (Liu et al., 2010a).

Fig. S1 shows IgG-BCR expression. Fig. S2 shows that Mut7 tail can lower the mechanical force threshold of IgM-BCR. Fig. S3 shows the enrichment of PI(4,5)P2, PS, PI(3,4,5)P3, PA, and DAG in WT, Mut5–, and Mut6–IgG-BCR. Fig. S4 shows the contribution of rapamycin, Myo-inositol, and PIP5K in PI(4,5)P2 manipulation system. Fig. S5 shows the sequence of cytoplasmic tails used in this article.

We thank Dr. Chenqi Xu (Institute of Biochemistry and Cell Biology, Shanghai, China), Dr. Mian Long (Chinese Academy of Sciences, Beijing, China), Dr. Hai Qi (Tsinghua University, Beijing, China), Dr. Yujie Sun (Peking University, Beijing, China), and Dr. Wei Chen (Zhejiang University, Hangzhou, China) for helpful discussions. We thank Dr. Hui Zhang (Tsinghua University, Beijing, China) for her technical support on dSTORM imaging experiments.

This work is supported by funds from National Science Foundation of China (81730043 and 81621002) and the Chinese Ministry of Science and Technology (2014CB542500-03).

The authors declare no competing financial interests.

Author contributions: Conceptualization: Z. Wan and W. Liu; Methodology: Z. Wan and W. Liu; Software: H. Xie; Investigation: Z. Wan and W. Liu; Writing, original draft: Z. Wan and W. Liu; Writing, review and editing: Z. Wan, S. Shaheen, Q. Ji, Y. Xu, J.-S. Zheng, F. Wang, J. Lou, W. Chen, and W. Liu; Funding acquisition: W. Liu; Resources: C. Xu, X. Chen, Z. Li, X. Ji, H. Chen, J. Wang, and Z. Tang; Supervision: W. Liu.