The glycosylation defect in solute carrier SLC35A2/SLC35A3 double knockout cells is rescued by SLC35A2–SLC35A3 and SLC35A3–SLC35A2 hybrids

SLC35A2 and SLC35A3 are homologous proteins with postulated nucleotide sugar transporting activities. Unlike SLC35A2, whose specificity for UDP‐Gal is well‐established, the UDP‐GlcNAc transporting activity initially attributed to SLC35A3 has recently been put into question. In this study, we constructed two hybrid proteins (SLC35A2–SLC35A3 and SLC35A3–SLC35A2) and expressed them in a previously generated SLC35A2/SLC35A3 double knockout HEK293T cell line. Our idea was to force equivalent stoichiometry of the two proteins in the cells in order to reproduce the behavior of the SLC35A2/SLC35A3 complexes in the Golgi membrane. The hybrid proteins were able to fully restore glycosylation in the double knockout. In contrast, the expression of SLC35A3 alone in these cells improved galactosylation only to a limited extent. Our study shows that the proper glycosylation requires a balanced cooperation between SLC35A2 and SLC35A3.

Glycosylation constitutes one of the most frequent post-translational modifications of proteins.Oligosaccharides are also attached to some lipids, yielding glycolipids.Moreover, they were even found to decorate certain RNA molecules [1].Glycosylated macromolecules play essential roles in numerous biological processes including the growth and development of eukaryotes, intrinsic and extrinsic recognition events, targeting of proteins to their final destinations, and the onset of immune responses.Most of glycosylation reactions occur in the lumen of the endoplasmic reticulum (ER) and the Golgi complex.The key players in this process are glycosyltransferases and the substrates they utilize are nucleotide sugars.
the ER/Golgi membranes in order to reach the luminally oriented catalytic domains of glycosyltransferases.This function was assigned to nucleotide sugar transporters (NSTs), which are members of the human solute carrier family 35 (SLC35) of proteins with the molecular weight of 30-45 kDa [3].Structurally, NSTs are type III membrane proteins with an even number of spans and the cytoplasmic orientation of N-and C termini.They are thought to act as antiporters that exchange a nucleotide sugar molecule for a corresponding nucleoside monophosphate moiety.
SLC35A2 is the only mammalian UDP-Gal transporter identified to date [4][5][6].Its detailed functional characterization was enabled by the development of mutant cell lines such as MDCK-RCA r [7] and Chinese hamster ovary (CHO)-Lec8 [6,8], which bear inactivating mutations in the SLC35A2 gene that result in Gal-deficient glycoconjugates.
SLC35A2 occurs in two splice variants: the Golgiresident UGT1 and UGT2, which localizes both to the ER and the Golgi apparatus [9].We showed that both UGT1 and UGT2 were able to restore galactosylation in MDCK-RCA r and CHO-Lec8 cells [10].We also generated and characterized HepG2 and HEK293T human cell lines deficient in SLC35A2 activity [11].Both studies revealed some residual galactosylation of N-glycans in SLC35A2-deficient mutants, which suggests that there might be another transporter involved in UDP-Gal delivery to the Golgi lumen.
Pathogenic mutations in the human SLC35A2 gene result in a subtype of a congenital disorder of glycosylation (CDG), namely SLC35A2-CDG.The affected individuals usually suffer from various degrees of neurological impairment, which in some cases is accompanied by skeletal abnormalities (e.g., [12]).
In contrast to SLC35A2, the substrate specificity of SLC35A3 has not been convincingly proven.In 1998, complementation of the Kluyveromyces lactis mutant unable to translocate UDP-GlcNAc across the Golgi membranes with the canine SLC35A3 gene was the basis for attributing UDP-GlcNAc transporting activity to this protein [13].The human SLC35A3 gene was identified based on nucleotide sequence similarity to the human SLC35A2 gene, and its protein product was shown to localize to the Golgi apparatus of CHO cells [14].In the same study, the specificity of the human SLC35A3 toward UDP-GlcNAc was demonstrated using the Saccharomyces cerevisiae heterologous system.
To verify whether SLC35A3 delivers UDP-GlcNAc for N-glycan biosynthesis, we silenced the SLC35A3 gene using siRNA technology in CHO, HeLa, and MDCK cell lines [15].We expected that the knockdown would result in a severe deficiency in the antennal GlcNAc of complex-type N-glycans.However, only the tri-and tetra-antennary species were depleted in SLC35A3 knockdown cells, while the biantennary structures remained unaffected.
We also studied the effect of SLC35A3 knockdown on the synthesis of keratan sulfate and heparan sulfate proteoglycans as these two glycosaminoglycans contain significant amounts of GlcNAc.The amount of keratan sulfate proteoglycans was significantly depleted in the SLC35A3-deficient MDCK cells, whereas heparan sulfate proteoglycans were not affected [15].
As a follow-up, we knocked out the SLC35A3 gene in HepG2, HEK293T, and CHO cell lines [11].Here, we also observed a decrease in the amount of tri-and tetraantennary complex-type N-glycans in the knockout cell lines with no effect on the biantennary structures.A more severe defect in N-glycan branching was observed in cell lines deficient in both SLC35A3 and SLC35A2.We also showed that despite the lack of SLC35A3 activity GlcNAc was still present in mucin-type O-glycan structures.These results suggest that SLC35A3 may play a redundant role in supplying UDP-GlcNAc for the biosynthesis of N-and O-glycans in mammalian cells.
A missense mutation in the bovine SLC35A3 gene causes complex vertebral malformation, a lethal inherited syndrome mainly characterized by abnormalities in the structure of the spine [16,17].However, structures made of connective tissue such as cloven hooves and horns were developed, which suggests that despite the lack of SLC35A3 activity GlcNAc-rich glycoconjugates including proteoglycans were synthesized in the affected calves.In a recent study, SLC35A3-deficient mice were generated and characterized [18].The knockout mice suffered from a severe chondrodysplasia resulting in perinatal lethality and synthesized significantly lower amounts of heparan sulfate, keratan sulfate, and chondroitin sulfate/ dermatan sulfate comparing with control animals.Therefore, the authors suggested that SLC35A3 regulates glycosaminoglycan biosynthesis.However, they found surprising that the lack of SLC35A3 negatively affected the level of chondroitin sulfate, as the latter glycosaminoglycan does not contain GlcNAc at all.Finally, pathogenic mutations in the human SLC35A3 gene are associated with skeletal abnormalities (dysplasia and arthrogryposis) and neurological symptoms (autism disorder, epilepsy, and encephalopathy) [19][20][21].
Some findings suggest that SLC35A3 may contribute to the delivery of UDP-Gal to the Golgi lumen.First of all, the galactosylation defect displayed by MDCK-RCA r and CHO-Lec8 cells was partially restored by the overexpression of SLC35A3 [22].Another study showed that silencing of the SLC35A3 gene in CHO cells affected the transport of the radiolabeled UDP-Gal into Golgi vesicles to a greater extent than the uptake of UDP-GlcNAc [15].Altogether, the UDP-GlcNAc transporting activity of SLC35A3 as its predominant specificity has been put into question by numerous studies [11,15,18,22], as on one hand its deficiency does not result in a complete disappearance of GlcNAc from glycoconjugates and on the other it appears to contribute into delivery of other nucleotide sugars with UDP-Gal being the most likely candidate.
Importantly, SLC35A2 and SLC35A3 were shown to associate by several different experimental approaches in multiple cell lines.Using coimmunoprecipitation and FLIM-FRET, we demonstrated the existence of the SLC35A2/SLC35A3 complex in MDCK-RCA r cells [23].This interaction was also visualized in HepG2 cells using in situ PLA [24].Finally, association between SLC35A2 and SLC35A3 was shown in HEK293T cells using the NanoBiT assay [25].
Additional pieces of evidence for the structural and functional link between SLC35A2 and SLC35A3 came from our studies involving generation of chimeric proteins composed of fragments of both NSTs.In one of these studies, we have generated a chimeric protein composed of amino acids 1-224 of SLC35A2 and 198-325 of SLC35A3 [26].Such a chimera was targeted to the Golgi apparatus and restored the wild-type galactosylation phenotype in SLC35A2-deficient MDCK-RCA r and Lec8 mutant cell lines.As a follow-up, four additional chimeras differing in the proportion between SLC35A2-and SLC35A3-derived polypeptide fragments were generated and investigated [27].In one of the constructs, the contribution of SLC35A2 was as little as 35 N-terminal amino acids.Here, again, all chimeras, including the one that contained only a short N-terminal fragment of SLC35A2, were properly localized to the Golgi apparatus and were fully capable of restoring the wild-type galactosylation phenotype in the cells deficient in SLC35A2 activity.
In our subsequent studies, we revealed that SLC35A2 and SLC35A3 have common interaction partners, which additionally support their existence and functional cooperation in a complex.As shown by FLIM-FRET and in situ PLA, both NSTs interacted with Mgat enzymes [28].In a more recent study, we showed that both SLC35A2 and SLC35A3 associated with Golgi pH regulator B (GPR89B) and ATPase 2 (ATP2A2) [29].The latter results were obtained in pull-down experiments and further confirmed by the NanoBiT assay.
Having been inspired by the remarkable functional connection between SLC35A2 and SLC35A3, here we constructed two hybrid proteins (SLC35A2-SLC35A3 and SLC35A3-SLC35A2) and overexpressed them in a knockout HEK293T cell line deficient in both SLC35A2 and SLC35A3 genes.The overexpressed hybrids localized to the Golgi apparatus and restored proper glycosylation in such a double knockout.Our findings support the tight interplay between SLC35A2 and SLC35A3 and confirm their cooperation in the supply of UDP-sugar substrates for the synthesis of complex-type N-glycans.

Generation of hybrids
Expression plasmids encoding hybrids composed of human SLC35A2 (NM_005660.3) and SLC35A3 (NM_012243.3) were generated using NEBuilder HiFi DNA Assembly (NEB, Ipswich, MA, USA) according to the manufacturer's protocol.Template cDNA was synthesized on total RNA isolated from wild-type HEK293T cells using Luna-ScriptÒ RT SuperMix Kit (NEB, Ipswich, MA, USA).The resulting plasmids and primers are listed in Table S1.

Flow cytometry analysis of cells labeled with lectins
All steps of sample preparation were carried out on ice.Cells were rinsed twice with cold PBS and gathered in 1 mL of PBS.Next, the cells were centrifuged (500 g, 5 min, 4 °C), resuspended in 2% BSA in PBS, and blocked on ice for 30 min.The cells were then centrifuged once more and incubated with lectins, diluted to 10 lgÁmL À1 in 2% BSA in PBS.Unattached lectins were eliminated by washing the cells two times with 2% BSA in PBS.Used lectins have been listed in Table S2.If a biotinylated lectin was employed, cells were incubated for an additional 30 min with streptavidin conjugated with Alexa Fluor 488 (Life Technologies, Carlsbad, CA, USA), diluted 1 : 100 in 2% BSA in PBS.Ultimately, cells were washed twice with 2% BSA in PBS and resuspended in the same solution.Flow cytometry analysis was conducted using a NovoCyte Flow Cytometer (ACEA Biosciences, Santa Clara, CA, USA), with 10 000 cells utilized for each measurement.For statistical data analysis, one-way ANOVA with the Tukey's post hoc test was employed.Analyses were performed with GRAPHPAD PRISM (GraphPad Software, San Diego, CA, USA).A P-value of <0.05 was considered statistically significant.

Western blotting and lectin dot blotting
Cells were lysed as described previously [30].For western blotting, lysates were separated in SDS/PAGE, electrotransferred onto nitrocellulose membranes, and probed with a mouse anti-HA antibody followed by anti-mouse IgG secondary antibody conjugated with HRP (Table S2).For lectin dot blotting, lysates were spotted on nitrocellulose membranes and subsequently probed with biotinylated lectins (Vector Laboratories, Burlingame, CA, USA) listed in Table S2, followed by avidin D conjugated with HRP diluted 1 : 50 000 (Vector Laboratories, Burlingame, CA, USA).

Immunofluorescence staining
Cells underwent immunostaining following a previously described protocol [31].Primary and secondary antibodies employed for immunostaining can be found in Table S2.The prepared samples were examined using a Leica TCS SP8 confocal microscope (Leica Microsystems, Wetzlar, Germany), and the captured images were processed with the IMAGEJ software version 1.48v (NIH, Bethesda, MD, USA).

Design and construction of hybrid proteins
Two hybrids were designed and constructed: The first (further referred to as 2H3) contained SLC35A2 as the N-terminal portion and SLC35A3 as the C-terminal portion and the second (further referred to as 3H2) contained SLC35A3 as the N-terminal portion and SLC35A2 as the C-terminal portion.In both hybrids, NSTs were separated by the HA fusion peptide.The design of hybrids is schematically shown in Fig. 1A.The resulting constructs were used to stably transfect HEK239T cell line deficient in both SLC35A2 and SLC35A3 activities.Next, we confirmed the integrity of both expressed hybrids via western blotting using an anti-HA antibody.As shown in Fig. 1B, both hybrids were expressed by the generated stable transfectants, whereas no signals were observed in nontransfected cells.The hybrids migrated somewhat faster with respect to their theoretical molecular weight (~55 vs ~78 kDa), which is a commonly observed phenomenon during the SDS/PAGE separation of membrane proteins [32].Therefore, we concluded that the constructed hybrids were expressed in the double knockout as intact, full-length fusion proteins.

Analysis of subcellular localization of hybrid proteins
Having confirmed the expression of 2H3 and 3H2 hybrids in double SLC35A2/SLC35A3 knockout cells, we next examined the subcellular localization of the fusion proteins.For this purpose, we coimmunostained the cells with an anti-HA antibody and antibodies specific for Golgi and ER marker proteins.As shown in Fig. 2A, the 2H3 hybrid colocalized with the cis Golgi marker GM130 and the trans Golgi network marker TGN46, while little or no colocalization was observed with the ER marker PDI.Similar results were obtained for the 3H2 hybrid (Fig. 2B).This allowed us to conclude that the expressed hybrids were properly folded in the ER and efficiently targeted to the Golgi apparatus, both of which are prerequisites for their functionality.

Analysis of reactivity of cellular glycoproteins and cell surface glycoconjugates with lectins
To make a preliminary evaluation of the functionality of our hybrids, we performed a series of dot blot analyses of whole-cell lysates with selected lectins.The results are shown in Fig. 3A.Glycoproteins synthesized by the double SLC35A2/SLC35A3 knockout are deficient in Gal and Sia, as shown by the decreased reactivity with Erythrina cristagalli lectin (ECL), MAL I, MAL II, and SNA (although in the case of the latter lectin, the decrease was less obvious).The deficiency of Gal results in the exposure of GlcNAc residues, which is reflected by an increased reactivity with GSL II.Moreover, complex-type N-glycans synthesized by these cells are less branched, as reflected by a lower reactivity with PHA-L and Datura stramonium lectin (DSL) specific for b1,6-GlcNAcand b1,4-GlcNAc-branched species, respectively.The lack of Gal in mucin-type O-glycans synthesized by the double knockout results in the exposure of GalNAc residues, which causes an increased reactivity with VVL and decreased reactivity with PNA.Importantly, the expression of both hybrids in the double knockout restored proper galactosylation and sialylation of glycoproteins as reflected by the increased reactivities with ECL, MAL I, MAL II, and SNA.This was accompanied by a decreased reactivity with GSL II due to a decreased exposure of GlcNAc residues.Moreover, the reactivity of cellular glycoproteins with PHA-L and DSL also increased upon the expression of hybrids, which demonstrates that the N-glycan branching was improved.A decrease in the reactivity of cellular glycoproteins with VVL accompanied by an increase in PNA binding shows that galactosylation of mucin-type O-glycans was also restored by the expression of hybrid proteins in the double knockout.
As a follow-up, we analyzed cell surface glycoconjugates using lectin-based flow cytometry.Here, we employed five lectins: SNA, MAL II, PNA, PHA-L, and HPA.As shown in Fig. 3B, the expression of both hybrids resulted in an elevated reactivity of cell surface glycoconjugates with Sia-specific lectins, SNA and MAL II, comparing with the double knockout.Interestingly, the reactivity of the wild-type and knockout glycoconjugates with SNA were comparable, which is in line with the results of dot blot analysis.Furthermore, the expression of both hybrids restored a2,6 sialylation to a level exceeding that of wild-type cells.The reactivity of cell surface glycoproteins with PHA-L was also restored upon the expression of both hybrids.Here, however, the 2H3 hybrid was significantly more efficient in improving the branching of complex-type N-glycans than the 3H2 one.The analysis with lectins specific for mucin-type O-glycans, that is PNA and HPA, also confirmed the restoration of galactosylation of these species upon the expression of both hybrids.Here, however, the 2H3 hybrid did not fully restore the wild-type level of reactivity with VVL.Altogether, we concluded that the constructed SLC35A2-SLC35A3 and SLC35A3-SLC35A2 hybrids are able to correct for the deficiencies in galactosylation and N-glycan branching displayed by the double knockout.

Analysis of N-glycans separated in normal phase chromatography
In the next step, we examined the effect of the expression of the hybrids on N-glycans.For this purpose, we isolated N-glycans from the cells of interest and treated them with a-mannosidase and sialidase prior to analysis on the GlycoSep N Plus chromatography column.Digestion with these glycosidases eliminates all mannose-terminating N-glycans as well as sialylated structures so that only neutral, complex-type species are left, which reduces the complexity of HPLC spectra and facilitates their evaluation.Apart from expressing the hybrids in the double knockout, we also stably expressed individual NSTs, that is either SLC35A2 or SLC35A3, both in the double knockout and in the wild-type cells.The results are shown in Fig. 4. The HPLC spectrum recorded for the double knockout contained significantly fewer peaks in the high-molecular-weight range than the wild-type spectrum, which is associated with the loss of Gal from complex-type N-glycans as well as their reduced branching.The expression of both hybrids resulted in HPLC spectra similar to the wild-type spectrum.Interestingly, the expression of single NSTs in the double knockout also improved its glycophenotype, although to a lesser extent than the expression of hybrids.Altogether, we concluded that SLC35A2 and SLC35A3 have partially overlapping roles in restoration of the wild-type phenotype in the double knockout.

MALDI-TOF-MS analysis of N-glycans
In order to extend and complete the HPLC data, we performed MALDI-TOF-MS analysis of 2-AB-labeled N-glycans treated with a-mannosidase and sialidase.The results are shown in Fig. 5.The obtained spectra demonstrate that the expression of both hybrids in the double knockout restored all galactosylated,  complex-type N-glycan species that were present in the wild-type sample but were missing from the knockout sample.Similar results were obtained upon the expression of SLC35A2 in the double knockout.Interestingly, the spectrum obtained for the knockout expressing SLC35A3 was strikingly more complex than all the other spectra as it contained a variety of both galactosylated and agalactosylated complex-type Nglycan species.In summary, we concluded that both hybrids fully restored the wild-type phenotype, whereas the expression of SLC35A3 alone resulted in a mixed population of corrected, partially corrected, and uncorrected complex-type N-glycan species.

Discussion
In this study, we constructed two hybrids composed of SLC35A2 and SLC35A3 full-length proteins via genetic engineering.The generated hybrids were stably expressed in a double knockout background, that is in HEK293T cells in which both the SLC35A2 and SLC35A3 genes have been disrupted.The idea of this approach was to force equivalent stoichiometry of SLC35A2 and SLC35A3 in the cells so that the behavior of the SLC35A2/SLC35A3 complexes in the Golgi membrane is more faithfully reproduced.In other words, when the hybrids are expressed in a double SLC35A2/SLC35A3 knockout, there are no "free" SLC35A2 and SLC35A3 molecules present in the cell and the only species is the hybrid in which both proteins are permanently associated with each other in an equimolar ratio.We decided to create such hybrids because we strongly believe that SLC35A2 and SLC35A3 exist and function in the cell mainly (or even exclusively) as a binary cooperative complex.
It was critical for us to use the double SLC35A2/ SLC35A3 knockout for the expression of the hybrids instead of wild-type or single knockout cells as only such a model provides a true zero-background conditions.Moreover, the HPLC spectrum obtained for desialylated complex-type N-glycans derived from the double knockout is relatively simple and contains only a few well-distinguishable peaks (Fig. 4).Therefore, it is easy to observe the appearance of more complex structures in such a simple (reduced) spectrum, which helps to identify the effects of phenotypic correction.
Upon expression in the cells, both hybrids were properly targeted to the Golgi apparatus.This clearly suggests that the hybrids are correctly folded in the ER and subsequently departed to the Golgi.Therefore, it appears that the molecular determinants responsible for Golgi targeting of SLC35A2 and SLC35A3 remain functional in the hybrids and are recognized by the appropriate components of the vesicular transport system.
Interestingly, the order in which the component proteins were aligned in a hybrid did not make any major difference as both constructs were similarly capable of correcting the phenotype of the double knockout.Importantly, the hybrids were much more efficient in restoring proper glycosylation of N-glycans that the component proteins expressed alone, although single expression of SLC35A2 and SLC35A3 also resulted in some extent of correction.When it comes to SLC35A3, its expression resulted in preferential restoration of galactosylation of the selected N-glycan species, suggesting some kind of disbalance.The partial correction of the double knockout by the expression of individual NSTs suggests that their functions partially overlap.On the contrary, however, the most efficient correction was provided by the hybrids, which suggests that there is tight functional cooperation between SLC35A2 and SLC35A3 and shows that these proteins fulfill some unique, nonreplaceable roles in N-glycan biosynthesis.
Our previous work [11] showed that the absence of SLC35A3 has no effect on mucin-type O-glycans, whereas SLC35A2 is critically important for the biosynthesis of core 1 species.Importantly, lectin analysis showed that both hybrids were able to restore galactosylation of O-glycans in the double knockout.This suggests that the complex composed of SLC35A2 and SLC35A3 (but not SLC35A3 alone) is important for galactosylation of mucin-type O-glycans.
To date, for both SLC35A2 and SLC35A3, numerous interaction partners have been identified.It is therefore tempting to speculate whether the SLC35A2/SLC35A3 hybrids are able to associate with the interaction partners of the individual component proteins.When it comes to N-glycans, both SLC35A2 and SLC35A3 were shown to interact with Mgat enzymes, which are key players in the biosynthesis of complex-type N-glycans.Therefore, it appears highly likely that the hybrids can also be part of the assemblies formed with the involvement of Mgats.In our recent study [33], we showed that SLC35A2 associates with C1GalT1 and its chaperone Cosmc, the two proteins strictly required for the biosynthesis of core 1 mucin-type O-glycans.However, it remains to be explored whether SLC35A3 is a part of the corresponding protein complexes.
One can imagine that the conformational changes required for the nucleotide sugar transport to occur might be somewhat constrained in such hybrids.We are not sure whether this is the case.However, we believe that another important function of NSTs is their involvement in the assembly and maintenance of multicomponent glycosylation-related complexes.It cannot be excluded that the formation of heterologous complexes composed of SLC35A2 and SLC35A3 is obligatory (or at least favorable) for both proteins to properly function.
In summary, the remarkable functionality of SLC35A2-SLC35A3 and SLC35A3-SLC35A2 hybrids proved in this study confirms that there is a tight interplay between SLC35A2 and SLC35A3 and suggests their functional cooperation in the supply of UDP-sugar substrates for the biosynthesis of complex-type N-glycans.Moreover, it appears that SLC35A3 is getting more and more distant from being considered a UDP-GlcNAc transporter.Our findings demonstrate that SLC35A2 and SLC35A3 do not act as independent transporters with single, strictly defined specificities but rather cooperate to fulfill more general or partially overlapping roles.

Fig. 3 .
Fig. 3. Effect of the expression of SLC35A2-SLC35A3 (2H3) and SLC35A3-SLC35A2 (3H2) hybrids in the double SLC35A2/SLC35A3 HEK293T knockout (A2A3KO) on cellular glycosylation.(A) The results of dot blot analysis performed on whole-cell lysates using selected lectins.SNA, Sambucus nigra agglutinin; MAL I, Maackia amurensis lectin I; GSL II, Griffonia simplicifolia lectin II, MAL II, Maackia amurensis lectin II; VVL, Vicia villosa lectin; PHA-L, phytohemagglutinin-L; PNA, peanut agglutinin.Ponceau S staining is presented alongside to demonstrate equal loading.WT, wild-type cells.(B) Analysis of reactivity of cell surface glycoconjugates with selected lectins using flow cytometry.SNA, Sambucus nigra agglutinin; MAL II, Maackia amurensis lectin II; PHA-L, phytohemagglutinin-L; PNA, peanut agglutinin; HPA, Helix pomatia agglutinin.Data are presented as mean fluorescence intensity of three independent biological replicates, each performed for 10 000 cells AE SEM.Data were analyzed using one-way ANOVA with Tukey's post hoc test.Black asterisks indicate statistically important differences between the wild-type sample and other samples, whereas red asterisks with underline indicate statistically important differences between individual hybrids.The significance level was set at P < 0.01 (**) and P < 0.0001 (****).