Long-term labelling and tracing of endodermal cells using a perpetual cycling Gal4-UAS system

Cell lineage tracing stands as a foundational technique in the study of embryonic development, capable of providing crucial insights into cell fate determination, lineage differentiation, migration, morphogenesis and the intricate processes of tissue formation (Kretzschmar and Watt, 2012; Kester and van Oudenaarden, 2018; Zhang et al., 2020). Traditional cell labelling, such as the use of vital tracer dyes (Filby et al., 2015) and photoconvertible fluorescent proteins (Adam et al., 2014), often faces issues with signal attenuation, making long-term tracing of labelled cells difficult. The Cre-loxP system is a widely used tool at present for lineage tracing, providing persistent labelling of the targeted cells through irreversible genetic recombination (Zong et al., 2005; Mosimann et al., 2011; He et al., 2017). However, its efficiency relies on the strength of specific promoter driving Cre (Ma et al., 2008), the activity of constitutive promoter controlling reporter cassettes (Lalonde et al., 2022) and the distance between loxP recombination sites (Vooijs et al., 2001). Therefore, expanding stable and long-term cell labelling strategies is of significant importance for deep comprehension of embryonic development and continuous mapping of cell lineages.

The endoderm is one of the earliest progenitor cell types to form during embryonic development and, in principle, retains the ability to generate nearly all cell lineages of respiratory and digestive organs. SRY-box transcription factor 17 (Sox17) is a representative gene for defining the endoderm lineage, critically contributing to the formation and maintenance of endoderm in fish, mice and other species (Ober et al., 2003; Viotti et al., 2014; Mukherjee et al., 2020). In zebrafish embryos, the injection of a photoactivatable fluorescein dextran conjugate into the Tg(sox17:GFP) transgenic line (Chung and Stainier, 2008 and Chung et al., 2008) or the direct generation of the Tg(sox17:hKikGR1) transgenic line that expressed a photoconvertible Kikume fluorescent protein (Yang et al., 2021) have been used to label and trace endodermal cells to identify progenitor regions of the digestive organ in the endoderm following gastrulation. Notably, these endoderm lineage studies focus on a relatively short stage of development, due to the temporal constraints of sox17 transcription (Alexander and Stainier, 1999), and are unable to mark subsequent differentiated cells constantly. Tracing the endodermal lineage in the long term remains an issue that needs to be addressed.

The Gal4-UAS system, initially developed in Drosophila, is a bipartite expression system consisting of the yeast transcription activator Gal4 and its upstream activating sequence (UAS). When Gal4 is expressed, it binds to UAS and activates the transcription of downstream genes (Brand and Perrimon, 1993; Phelps and Brand, 1998). This system has emerged as a powerful tool for controlling gene expression in model organisms, such as zebrafish (Goll et al., 2009; Kawakami et al., 2016; Zhang et al., 2019). The Gal4-VP16, a fusion of the Gal4 DNA-binding domain and the herpes simplex virion protein 16 (VP16) activation domain, has been shown to be a potent transcriptional enhancer that can increase the expression of UAS-driven downstream genes and amplify their output signals (Köster and Fraser, 2001; Asakawa et al., 2008), especially when a weak promoter is used (Iyer et al., 2001; Xiong et al., 2013). A pioneering study developed an effective strategy to achieve sustained expression of the Gal4 by an autoregulatory Gal4-UAS feedback loop (Distel et al., 2009). For these reasons, combined with fluorescent reporters, the Gal4-UAS system is well suited for tracing some progenitors whose marker gene expression is low and transient. In this Report, we have optimised and reconfigured a perpetual cycling Gal4-UAS system, utilising it to generate zebrafish transgenic lines with endoderm-specific and continuous fluorescent labelling, thereby enabling long-term tracing of endodermal cells and visualisation of their developmental process. This technique provides a valuable tool for future in-depth studies of endodermal differentiation and tissue formation.

Our perpetual cycling Gal4-UAS system is based upon the traditional Gal4-UAS system (Fig. 1A) with a series of optimisation improvements. In detail, Gal4FF is an attenuated version of the Gal4-VP16 (Asakawa and Kawakami, 2008), its protein coding sequence is linked with enhanced GFP (EGFP) via a Thosea asigna virus self-cleaving 2A peptide (T2A), placed the downstream of five tandem repeats of UAS (5×UAS). Once Gal4FF is driven to expression by a specific promoter, it binds to UAS and activates the transcription of Gal4FF-T2A-EGFP. The newly produced Gal4FF binds to UAS again, thus continuously cycling transcriptional activation of Gal4FF and EGFP (Fig. 1B). Considering the mutual exclusivity between the DNA-binding and nuclear-targeting activities of Gal4 (Chan et al., 1998; Chan and Jans, 2001), we incorporated a nuclear localisation signal (NLS) from simian virus 40 (SV40) large T-antigen to the 5′ end of Gal4FF protein-coding sequence, to improve its nuclear import efficiency. Meanwhile, a modified peptide sequence rich in proline, glutamic acid, serine and threonine (PEST) was added to the 3′ end (Li et al., 1998; Ravindran et al., 2022) to accelerate the degradation of Gal4FF, reducing its toxic accumulation during continuous transcriptional activation. This new NLS-Gal4FF-PEST fusion (Table S1), abbreviated as NP-Gal4FF, can increase the transcriptional potency of Gal4FF while minimising cytotoxicity.

To evaluate the above-described optimised scheme, we transiently transfected 293T cells with a tamoxifen-responsive Gal4FF-ER^T2 construct (Gerety et al., 2013) under the human cytomegalovirus (CMV) promoter to compare the expression levels of 5×UAS:EGFP, 5×UAS:EGFP-PEST and 5×UAS:NP-Gal4FF-T2A-EGFP reporters after tamoxifen exposure (Fig. S1A). The results showed that EGFP fused with the PEST domain degraded faster than unmodified EGFP, whereas NP-Gal4FF-T2A-EGFP markedly prolonged the duration of EGFP expression through Gal4-UAS cycling transcription (Fig. S1B,C). As part of the toxicity assay, different Gal4FF constructs carried by the pTol2-CMV backbone were separately injected into one-cell stage zebrafish embryos (Fig. S2A). By 48 h post fertilisation (hpf), through comparison at the same injection dosage, the ratios of dead and deformed embryos with the Gal4FF constructs containing the PEST sequence were lower than those without it (Fig. S2B,C). Furthermore, to minimise background toxicity, we generated a heat shock-inducible Gal4FF line Tg(hsp70l:NP-Gal4FF)^cq210 and crossed it with different reporter lines to examine the phenotype of their offspring after heat shock (Fig. S3A). The results showed that, under the Gal4-5×UAS cycling activation, NLS-Gal4FF-PEST caused less toxic damage to embryos than NLS-Gal4FF (Fig. S3B,C). Notably, under the Gal4-10×UAS cycling activation, the ratios of dead and deformed embryos were increased significantly.

Pursuing further assessment of the efficiency and persistence of cell labelling, a basic Hsp70l-Gal4FF transgenic line was crossed with the fluorescent reporter line Tg(5×UAS:NP-Gal4FF-T2A-EGFP)^cq187, which was selected through the above-optimised comparison. Their offspring were heat-shocked at 12 hpf to monitor EGFP expression during subsequent embryonic development. Within traditional Gal4-UAS systems, EGFP signal in transgenic line Tg(hsp70l:Gal4FF; 5×UAS:EGFP) had been depleted by 4 days post-fertilisation (dpf). By contrast, EGFP expression in the perpetual cycling Gal4-UAS system was maintained robustly for an extended period (Fig. 1C), even being able to continue into adulthood (Fig. 1D). These data validate the feasibility of this optimised scheme for the sustained expression of Gal4FF and EGFP, indicating that the perpetual cycling Gal4-UAS system can effectively accomplish stable and long-term cell labelling.

For applying the perpetual cycling Gal4-UAS system to endoderm lineage studies, we generated a transgenic line Tg(sox17:Gal4FF-T2A-EGFP)^cq186, in which Gal4FF-T2A-EGFP expression was driven by the endoderm-specific promoter sox17, and crossed it with the fluorescent reporter line Tg(5×UAS:NP-Gal4FF-T2A-EGFP) to enable continuous fluorescent labelling of the endoderm and its descendant cells (Fig. 2A). In the process of establishing double transgenic lines, we conducted extensive and multi-round screening to identify lines with relatively stable and consistent expression of reporter genes (Fig. S4A), as numerous different founders of the Tg(5×UAS:NP-Gal4FF-T2A-EGFP) produce offspring with suboptimal and variable expression patterns (Fig. S4B,C). Such variability is likely caused by the position effects from random integration into the genome (Clark et al., 1994), a phenomenon particularly pronounced in our perpetual cycling Gal4-UAS system.

Within a pre-screened and stabilised double transgenic genetic background, all endoderm-derived tissues along the anterior-posterior axis, from the oral cavity to the cloaca, could be visualised (Fig. 2B), including the major accessory organs such as the liver, pancreas and intestine, as shown in the magnified lateral views of larvae at 5 dpf (Fig. 2C). Tissue sections further displayed betaine-homocysteine methyltransferase (bhmt)-expressing hepatocytes (Gao et al., 2019), serine protease 1 (prss1, also known as trypsin)-expressing pancreatic acinar cells (Raby et al., 2021) and fatty acid binding protein 2 (fabp2)-expressing intestinal enterocytes (Yang et al., 2022), all featuring noticeable EGFP signals (Fig. 2D). These data suggest that our perpetual cycling Gal4-UAS system is capable of providing effective and reliable fluorescent labelling for endoderm lineage tracing, contributing to monitoring the developmental dynamics of endoderm-derived tissues and profiling their lineage differentiation.

Since sox17 expression begins in gastrulation and gradually disappears in somitogenesis (Alexander and Stainier, 1999), the Tg(sox17:GFP) transgenic line provides only short-term labelling of endodermal tissues within a particular temporal window. To further validate the long-term continuity and reliability of endoderm labelling in our perpetual cycling Gal4-UAS system, we compared it with the single transgenic line Tg(sox17:GFP) and the double transgenic line Tg(sox17:Gal4FF; 5×UAS:EGFP), which was established through the traditional Gal4-UAS system. The time-course analysis of embryonic development showed that, at 12 hpf, the post-gastrulation stage, sox17-positive endodermal cells in three comparative groups were dispersed across the embryo in a relatively sparse pattern (Fig. 3A). At 24 hpf, these cells formed the primitive gut tube, without detectable differences between groups. During the subsequent 48 hpf, multiple organ primordia, including the pharyngeal arches, liver, swim bladder, pancreas and intestine, had developed and extended from the primitive gut tube. Until 72 hpf, the fluorescence signal in gut endoderm of Tg(sox17:GFP) and Tg(sox17:Gal4FF; 5×UAS:EGFP) transgenic lines drastically decayed, except in the gallbladder and its surrounding tissues, because sox17 transcription was re-initiated there (Shin et al., 2012). In comparison, endodermal cells labelled using the perpetual cycling Gal4-UAS system maintained robust EGFP expression (Fig. 3B), corroborated by measuring the mean fluorescence intensity of endodermal tissue regions (Fig. 3C). These results demonstrate the effectiveness of a perpetual cycling Gal4-UAS system for labelling and tracing endoderm lineage.

In addition, we conducted comparative analyses between the conventional Cre-loxP system and the perpetual cycling Gal4-UAS system. The transgenic lines Tg(actb2:loxP-STOP-loxP-EGFP)^cq214 (actb2, also known as β-actin2) and Tg(5×UAS:NP-Gal4FF-T2A-EGFP) were individually crossed with the same heat-shock line Tg(hsp70l:Cre-P2A-Gal4FF)^cq211 for simultaneous activation of Cre and Gal4FF after heat shock. At 24 hpf, through analysis under different heat-shock durations (Fig. S5A), the perpetual cycling Gal4-UAS system had the advantage of rapidly amplifying fluorescent signals over a short time, whereas the Cre-loxP system displayed relatively latency (Fig. S5B,C). Within an endoderm-specific Tg(sox17:CreER^T2)^cq213 genetic background (Fabian et al., 2020), the Cre-loxP system following tamoxifen exposure irreversibly labelled endoderm-derived organs, including the liver, pancreas and intestine, when observed at 4 dpf, while the perpetual cycling Gal4-UAS system also demonstrated comparable capability (Fig. 3D). Notably, no obvious fluorescent signal was detected in the endoderm cells at 12 hpf, after tamoxifen treatment of Tg(sox17:CreER^T2; actb2:loxP-STOP-loxP-EGFP) lines (Fig. S5D), indicating that this system relies on the activity of the actb2 promoter as well as sox17. Due to the maternal contribution of actb2 during early embryogenesis (Pinto et al., 2020; Lalonde et al., 2022), its transcription initiation in endodermal cells may occur relatively late. In comparison, this issue was not present in Tg(sox17:Gal4FF; 5×UAS:NP-Gal4FF-T2A-EGFP) lines (Fig. S5D), suggesting the perpetual cycling Gal4-UAS system can timely label endodermal cells without being restricted by constitutive promoters, facilitating the visualisation and tracing of these cells at earlier embryonic stages.

Upon completing the analysis of the Tg(sox17:Gal4FF-T2A-EGFP; 5×UAS:NP-Gal4FF-T2A-EGFP) transgenic larvae, we next examined whether the perpetual cycling Gal4-UAS system can continuously label endodermal tissues and organs into adulthood. After laparotomy of adult fish at 4 months post-fertilisation (mpf), EGFP signals remained observable in the aero-digestive tract and its accessory organs derived from endoderm (Fig. 4A). Tissue sections further displayed that, although the fluorescent labelling of these organs was not as uniform as during the larval stage, a notable fraction of cells within them were continuously labelled by EGFP (Fig. 4B). Statistically, the gill and liver exhibited higher ratios of labelled cells, while the pancreas and intestine had relatively lower ratios (Fig. 4C). One possible reason for these discrepancies is, as recorded in previous studies, the faster rates of cellular turnover in the pancreas and intestine under natural conditions (Williams et al., 2015; Zhao et al., 2023), particularly in comparison to the liver (Duncan et al., 2009). This constant cell renewal can lead to increased susceptibility to DNA methylation, silencing UAS-driven transgene expression over time (Halpern et al., 2008; Goll et al., 2009). Despite the uneven labelling efficiency and evident tissue-specific differences in adult fish, these data still confirm that the perpetual cycling Gal4-UAS system has the ability to label endodermal tissues and organs into adulthood.

In summary, we showcase here a continuous fluorescent labelling strategy that utilises the optimised Gal4-UAS system to allow the perpetual cycling transcription of reporter genes, which is available for long-term cell lineage tracing. Within this strategy, the improved NP-Gal4FF fusion can not only enhance the transcriptional activation efficiency of Gal4FF in the nucleus, but also avoid the overaccumulation of Gal4FF during embryonic development, ensuring that its transgenic line grows healthily to adulthood (Fig. 1D), without encountering developmental abnormalities caused by potential cytotoxicity (Fig. S3B,C). Based on this continuous fluorescent labelling strategy, we further generate the transgenic lines, enabling long-term visualised labelling specific to endodermal tissues and organs, with labelling capability extending into adulthood (Fig. 4A). The susceptibility of UAS to methylation is a pervasive issue (Goll et al., 2009; Pang et al., 2015). As zebrafish age, the transgenic lines exhibited a partial loss of cell labelling (Fig. 4B). In future, introducing insulator elements upstream and downstream of UAS cassettes or designing methylation-resistant UAS sequences could serve as potential solutions worth exploring.

In terms of application, the perpetual cycling Gal4-UAS system can provide an additional option and supplement to the few currently available visualised lineage-tracing methods, and it can also be combined with the Cre-loxP system for intersectional control of reporter gene expression to enhance specificity and flexibility in cell labelling (Satou et al., 2013; Tabor et al., 2019). In addition to fundamental biological research, the zebrafish transgenic lines with endoderm-specific and continuous fluorescent labelling can function as a visualisation tool for preclinical drug development, such as antiviral or antibacterial drug screening in the respiratory and digestive systems (Rampes and Ma, 2023; Habjan et al., 2024).

All zebrafish were maintained and raised under standard laboratory conditions according to Institutional Animal Care and Use Committees (IACUCs) protocols. This work is approved by the Animal Ethics Committee of Southwest University, Chongqing, China (ETHICS CODE Permit NO. IACUC-20240612-01).

The Tg(hsp70l:Gal4FF)^cq185, Tg(sox17:Gal4FF-T2A-EGFP)^cq186 and Tg(5×UAS:NLS-Gal4FF-PEST-T2A-EGFP)^cq187, abbreviated as Tg(5×UAS:NP-Gal4FF-T2A-EGFP); Tg(hsp70l:NLS-Gal4FF-PEST)^cq210, abbreviated as Tg(hsp70l:NP-Gal4FF); Tg(hsp70l:Cre-P2A-Gal4FF)^cq211; Tg(sox17:Gal4FF)^cq212; Tg(sox17:CreER^T2)^cq213; Tg(actb2:loxP-STOP-loxP-EGFP)^cq214; Tg(5×UAS:EGFP)^nkuasgfp1a (Asakawa et al., 2008) and Tg(sox17:GFP)^s870 (Chung and Stainier, 2008 and Chung et al., 2008) transgenic lines were established and/or used in this study. Embryos were treated with 0.003% 1-phenyl-2-thiourea (PTU, Sigma) from 24 hpf to prevent pigmentation, as previously described (Yu et al., 2023). An equal number of male and female adult fish were used for experiments.

The NLS-Gal4FF-PEST-T2A-EGFP fragment was constructed using Golden Gate cloning, as previously described (Jiang et al., 2023). The Gal4FF sequence containing an NLS domain, modified PEST with S440A mutation (Li et al., 1998) and T2A-EGFP sequence were spliced together through a one-pot restriction-ligation procedure. The Cre-P2A-Gal4FF fragment was constructed by joining the Cre-P2A sequence to the Gal4FF sequence using overlap PCR. Within the pTol2 vector backbone: the Gal4FF, NLS-Gal4FF-PEST or Cre-P2A-Gal4FF fragment was subcloned the downstream of hsp70l promoter; the Gal4FF, CreER^T2 or Gal4FF-T2A-EGFP fragment was subcloned the downstream of sox17 (-5.1k) promoter (Sakaguchi et al., 2006); the NLS-Gal4FF-T2A-EGFP or NLS-Gal4FF-PEST-T2A-EGFP fragment was subcloned the downstream of CMV promoter; and the 5×UAS or 10×UAS element, and the loxP-STOP-loxP-EGFP fragment was subcloned the downstream of actb2 promoter. Within the pcDNA3.1 vector backbone: the Gal4FF sequence was joined with the ER^T2 sequence using overlap PCR and then subcloned the downstream of CMV promoter; the EGFP, EGFP-PEST or NLS-Gal4FF-PEST-T2A-EGFP fragment was subcloned the downstream of 5×UAS element. All primers and the component sequences used in constructs are detailed in Table S1.

The pTol2-hsp70l:Gal4FF, pTol2-hsp70l:NLS-Gal4FF-PEST, pTol2-hsp70l:Cre-P2A-Gal4FF, pTol2-sox17:Gal4FF, pTol2-sox17:Gal4FF-T2A-EGFP, pTol2-sox17:CreER^T2, pTol2-5×UAS:NLS-Gal4FF-PEST-T2A-EGFP and pTol2-actb2:loxP-STOP-loxP-EGFP constructs were used for transgenesis. Each construct was co-injected with 50 pg of capped Tol2 transposase mRNA into zebrafish embryos of the AB genetic background at the one-cell stage. The Tg(5×UAS:NLS-Gal4FF-PEST-T2A-EGFP) founders with germline transmission were identified through outcrossing with the Tg(sox17:Gal4FF-T2A-EGFP) lines. The identified founders were then crossed back to the AB genetic background for three consecutive generations to screen the optimal and stable lines.

Human embryonic kidney (HEK) 293T cells were purchased from Cell Bank, Chinese Academy of Sciences (Shanghai, China). The cells were pre-plated in six-well plates and cultured in DMEM medium (Gibco) containing 10% fetal bovine serum (Gibco) and 1% penicillin-streptomycin (Gibco). Upon reaching 70-80% confluency, pcDNA3.1-CMV:Gal4FF-ER^T2 plasmids were co-transfected with pcDNA3.1-5×UAS:EGFP, pcDNA3.1-5×UAS:EGFP-PEST or pcDNA3.1-5×UAS:NLS-Gal4FF-PEST-T2A-EGFP plasmids into the cells. Transfection was carried out using Lipofectamine 3000 (Invitrogen, L3000-001) according to the manufacturer's instructions. For conditional induction of Gal4FF-ER^T2 activity, post-transfection cells were treated with 1 μM tamoxifen (Sigma) for 24 h and then cultured until the time-points of analyses.

Different fluorescent reporter lines with Tg(hsp70l:Gal4FF) or Tg(hsp70l:NLS-Gal4FF-PEST) genetic background at 12 hpf were heat-shocked at 38.5°C for 30 min, and then incubated at 28.5°C until analyses at the larval stage or raised until adulthood. The fluorescent reporter lines with the Tg(hsp70l:Cre-P2A-Gal4FF) genetic background at 24 hpf were heat-shocked at 38.5°C for 10, 20, 30 or 40 min separately, and then incubated at 28.5°C until the collective analysis 180 min later.

For conditional induction of CreER^T2 activity, the double transgenic embryos Tg(sox17:CreER^T2; actb2:loxP-STOP-loxP-EGFP) were treated with 5 μM tamoxifen (Sigma). The treatment was initiated at 2 hpf, with embryos incubated at 28.5°C under either continuous exposure to tamoxifen until analysis at 12 hpf or exposure for 24 h and further incubation until analysis at 96 hpf.

Genomic DNA was isolated from Tg(5×UAS:NLS-Gal4FF-PEST-T2A-EGFP) embryos. 10 μg of genomic DNA were digested with 150 units of the EcoRI (NEB) restriction enzyme. Southern blotting was performed as previously described (Green and Sambrook, 2021). The digoxigenin-labelled RNA probe, synthesised from a 402 bp DNA fragment of Gal4FF, was used for hybridisation. Primers used for probe synthesis were as follows: forward, 5′-CTGTCTTCTATCGAACAAGCATGCG-3′; reverse, 5′-ATTTAGGTGACACTATAGACTACTCTCTTCCGATGATGATG-3′ (the SP6 promoter sequence is underlined).

Transgenic embryos were mounted in 35 mm glass bottom dishes containing 0.6-1% low melting point agarose as previously described (Yang et al., 2021). Images were captured using a 20× water immersion objective on the LSM880 confocal microscope (Zeiss). Three-dimensional images and stitched images were generated by z-stacks and Tile Scan, respectively, using ZEN2010 software (Zeiss).

Adult zebrafish imaging was performed as previously described (Gupta and Mullins, 2010). Anaesthesia was administered using 140 mg/l Tricaine (MS-222, sigma), followed by an abdominal dissection to remove the skin and body wall muscles, exposing the internal organs. Images were captured using a 1× plan apochromatic objective on the Leica M205 FCA stereo microscope.

Whole embryos or tissues were fixed overnight in 4% formaldehyde solution. Tissues were embedded in optimal cutting temperature (OCT) compound (Sakura) and frozen sectioned at 15 µm using a Leica CM 1950 cryostat. Fluorescence in situ hybridisation (FISH) was performed as previously described (He et al., 2020; Yang et al., 2022) using bhmt, prss1 and fabp2 antisense probes labelled with digoxigenin, anti-digoxigenin-POD (1:1000, Roche, 11633716001), and the tyramide signal amplification and fluorescence detection system (TSA, PerkinElmer).

Antibody staining was performed as previously described (Wu et al., 2023) using anti-GFP (1:1000, Invitrogen, A11122) primary antibody and Alexa 488-conjugated donkey anti-rabbit IgG (1:1000, Invitrogen, A-21206) secondary antibody. The nuclei were stained with the DNA fluorochrome 4′,6-diamidino-2-phenylindole (DAPI, Sigma).

Cell Counting, Co-localisation Threshold plug-ins, Look-up table (LUT)-based colour mapping and Fluorescence Image Analysis tool in ImageJ (version 1.50d) were used to quantify the ratios of EGFP-labelled cells and the mean fluorescence intensity of specific tissue regions. All sampling was carried out using randomly assigned siblings. Data were analysed for statistical significance in GraphPad Prism (version 9.0.0) using comparison of means, one-way ANOVA with Tukey's multiple comparisons test, a Chi-square test or an unpaired two-tailed Student's t-test. Variance for all groups of data are presented as ±s.d. No data were excluded from analyses. The exact sample size (n), P-value summarised with asterisks (*) or hashes (^#), and statistical tests are indicated in the figures and figure legends.

We thank the Laboratory Animal Center and Imaging Facilities at Southwest University, D. Y. Stainier, K. Kawakami and C. Betsholtz for the transgenic lines, J. Ma and J. Chen for helpful suggestions, and R. Ni and Q. Yang for technical assistance.