Retrograde Signaling by Syt 4 Induces Presynaptic Release and Synapse-Specific Growth

See allHide authors and affiliations

Science  04 Nov 2005:
Vol. 310, Issue 5749, pp. 858-863
DOI: 10.1126/science.1117541


The molecular pathways involved in retrograde signal transduction at synapses and the function of retrograde communication are poorly understood. Here, we demonstrate that postsynaptic calcium 2+ ion (Ca2+) influx through glutamate receptors and subsequent postsynaptic vesicle fusion trigger a robust induction of presynaptic miniature release after high-frequency stimulation at Drosophila neuromuscular junctions. An isoform of the synaptotagmin family, synaptotagmin 4 (Syt 4), serves as a postsynaptic Ca2+ sensor to release retrograde signals that stimulate enhanced presynaptic function through activation of the cyclic adenosine monophosphate (cAMP)–cAMP-dependent protein kinase pathway. Postsynaptic Ca2+ influx also stimulates local synaptic differentiation and growth through Syt 4–mediated retrograde signals in a synapse-specific manner.

Neuronal development requires coordinated signaling to orchestrate pre- and postsynaptic maturation of synaptic connections. Synapse-specific enhancement of synaptic strength as occurs during long-term potentiation (1), as well as compensatory homeostatic synaptic changes, have been suggested to require retrograde signals for their induction (2, 3). Although retrograde signaling has been implicated widely in synaptic plasticity, the molecular mechanisms that transduce postsynaptic Ca2+ signals during enhanced synaptic activity to alterations in presynaptic function are poorly characterized. Because postsynaptic Ca2+ is essential for synapse-specific potentiation (4), it is important to characterize how Ca2+ can regulate retrograde communication at synapses.

To dissect the mechanisms underlying activity-dependent synaptic plasticity, we tested whether newly formed Drosophila glutamatergic neuromuscular junctions (NMJs), which have ∼30 active zones, show physiological changes after 100-Hz stimulation (5). Within 1 min after stimulation, a gradual 100-fold increase in miniature excitatory postsynaptic current (miniature) frequency was observed (Fig. 1, A to C), from a baseline of 0.03 Hz to often more than 5 Hz. The high-frequency-stimulation-induced miniature release (termed HFMR) continued for a few minutes to as long as 20 min before subsiding to baseline levels. Perfusion of postsynaptic muscles with the Ca2+ chelator EGTA from the patch pipette caused a modest suppression of HFMR, whereas the fast Ca2+ chelator 1,2-bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA) induced strong suppression by 2.5 min of perfusion. Longer perfusion with BAPTA for 5 min before stimulation abolished HFMR (Fig. 1, A to C), indicating HFMR is induced after postsynaptic Ca2+ influx.

Fig. 1.

High-frequency stimulation triggers enhanced presynaptic miniature release that requires postsynaptic Ca2+ and postsynaptic vesicle trafficking. (A) A motor nerve innervating embryonic muscle fiber 6 at hatching stage (21 hours after egg laying) was stimulated with four trains of 1-s 100-Hz stimuli spaced 2 s apart in 0.5 mM extracellular Ca2+. Whereas spontaneous release is rarely seen without stimulation (top left), high-frequency stimulation (represented by arrows) induces a 100-fold increase in frequency of miniatures (top right). The lower image shows traces when BAPTA (5 mM) was included in the internal solution of the patch electrode. (B) Representative time courses of HFMR from control and BAPTA-treated muscle cells. At each 10-s interval, miniatures are displayed as mean frequency. The first time point represents averaged miniature frequency for 5 min before stimulation. (C) Miniature frequency at 1 min after 100-Hz stimulation (calculated mean between 50 and 70 s after stimulation), compared with miniature frequency before stimulation (mean for 5 min before stimulation). The number of samples analyzed were six for control and five for BAPTA perfusion. Double asterisks indicate P < 0.01 by paired t test. (D) High-frequency stimulation performed in Mhc-Gal4, UAS-shibirets1 animals at permissive (23°C) and restrictive (31°C) temperatures. Wild-type animals show normal HFMR at 31°C (bottom). (E) Temperature shift experiments from the restrictive (31°C) temperature to the permissive (23°C) temperature using Mhc-Gal4, UAS-shibirets1 animals. The scale in the top of (D) also applies to the middle and bottom of (D) and (E). (F) Quantification of miniature frequency at 1 min after high-frequency stimulation calculated as described above. Triple asterisks indicate Mhc-Gal4, UAS-shibirets1 at 31°C is significantly different both from Mhc-Gal4, UAS-shibirets1 at 23°C and from wild type at 31°C with the use of posthoc comparisons (Scheffe's multiple comparisons test; P < 0.001) after single-factor analysis of variance. The number of samples analyzed were six for Mhc-Gal4, UAS-shibirets1 at 23°C, 13 for Mhc-Gal4, UAS-shibirets1 at 31°C, and four for wild type at 31°C. (G) Evoked responses in Mhc-Gal4, UAS-shibirets1 animals at the permissive and restrictive temperatures. The numbers of samples analyzed were six at 23°C and six at 31°C. Error bars in (F) and (G) indicate SEM. (H) Miniature frequency in the temperature shift experiments (E) at 1 min after high-frequency stimulation calculated as described above. The number of samples analyzed was six. The asterisk indicates P < 0.05 (Wilcoxon paired-sample test).

Ca2+-induced vesicle fusion in presynaptic terminals provides a temporally controlled and spatially restricted signal essential for synaptic communication. Postsynaptic vesicles within dendrites have been visualized by transmission electron microscopy (6), and dendritic release of several neuromodulators has been reported (7). To test whether postsynaptic vesicle fusion might underlie the Ca2+-dependent release of retrograde signals, we blocked postsynaptic vesicle recycling by using the dominant negative shibirets1 mutation, which disrupts endocytosis at elevated temperatures (8). We expressed shibirets1 specifically in postsynaptic muscles by driving a UAS-shibirets1 transgene (9) with muscle-specific myosin heavy chain (Mhc)-Gal4, keeping presynaptic activity intact. At the permissive temperature (23°C), high-frequency stimulation induced normal HFMR (Fig. 1, D and F). However, raising the temperature to 31°C suppressed HFMR in the presence of postsynaptic shibirets1, whereas wild-type animals displayed normal HFMR at 31°C (Fig. 1, D and F). Basic synaptic properties in Mhc-Gal4, UAS-shibirets1 animals were not affected at either the permissive or the restrictive temperature (Fig. 1G). The suppression of HFMR is not due to irreversible damage induced by postsynaptic UAS-shibirets1 expression, because a second high-frequency stimulation after recovery to the permissive temperature induced normal HFMR (Fig. 1, E and H).

The synaptic vesicle protein synaptotagmin 1 (Syt 1) is the major Ca2+ sensor for vesicle fusion at presynaptic terminals (10, 11) but is not localized postsynaptically. We have recently shown that another isoform of the synaptotagmin family, synaptotagmin 4 (Syt 4), is present in the postsynaptic compartment (12), suggesting Syt 4 might function as a postsynaptic Ca2+ sensor. Syt 4 immunoreactivity is observed in a punctate pattern surrounding presynaptic terminals, suggesting Syt 4 is present on postsynaptic vesicles (Fig. 2B). We again blocked postsynaptic vesicle recycling by using the UAS-shibirets1 transgene driven with Mhc-Gal4. Without a temperature shift, Syt 4–containing vesicles showed their normal postsynaptic distribution surrounding presynaptic terminals (Fig. 2B). When the temperature was shifted to 37°C for 10 min in the presence of high-K+ saline containing 1.5 mM Ca2+ to drive synaptic activity, Syt 4–containing vesicles translocated to the plasma membrane (Fig. 2C). After recovery at 18°C for 20 min, postsynaptic vesicles returned to their normal position (Fig. 2D). Removing extracellular Ca2+ during the high-K+ stimulation resulted in vesicles that did not translocate to the postsynaptic membrane (Fig. 2E).

Fig. 2.

Ca2+-dependent translocation of Syt 4–positive vesicles to the postsynaptic membrane. (A to E) Mhc-Gal4, UAS-shibirets1 third instar larval NMJs were stained with antisera against Syt 4 (magenta) to visualize postsynaptic vesicles and anti-horseradish peroxidase (HRP) (green) to visualize the presynaptic plasma membrane. (A) Diagram of temperature shifts performed. (B) Larvae maintained at 21°C have abundant postsynaptic vesicles that form a cloud around presynaptic terminals. (C) Larvae shifted to 37°C for 10 min in high-K+ saline (60 mM K+ and 1.5 mM Ca2+) lose the Syt 4–positive vesicle halo surrounding synaptic terminals, with a shift in the signal to the muscle plasma membrane. (D) Recovery of the Syt 4 vesicle population was observed after lowering the temperature to 18°C for 20 min. (E) Temperature shifts in 60 mM K+ saline without Ca2+ resulted in no Syt 4 vesicle translocation. The scale bar in (D) is 5 μm and applies to all images in (B) to (E). Above each image are schematic diagrams depicting the distribution of Syt 4 signal at each time point. (F and H) Mhc-Gal4, UASSyt 4pHluorin localizes to the postsynaptic density. (F) Live images of Syt 4–pHluorin by confocal microscopy at the third instar larval NMJ, showing a projection of optical sections through the Z axis (Fig. 4G). Arrowheads point to varicosities with patches of Syt 4–pHluorin signal. Scale bar is 5 μm. (G) A Syt 4–pHluorin–expressing muscle (green, upper panel) costained with nc82 (magenta, middle panel) in an optical section of a fixed third instar NMJ. Syt 4–pHluorin–positive patches (arrowheads) are localized adjacent to presynaptic active zones (arrows). (H) Syt 4–pHluorin (green, top) colocalized with immunoreactivity against DPAK (magenta, middle; arrowheads). The scale bar in (H) is 5 μm and also applies to (G).

To further test whether the Syt 4 vesicle population undergoes fusion with the postsynaptic membrane as opposed to mediating fusion between intracellular compartments, we constructed transgenic animals expressing a pH-sensitive green fluorescent protein (GFP) variant (ecliptic pHluorin) (13) fused at the intravesicular N terminus of Syt 4. Ecliptic pHluorin increases its fluorescence 20-fold when exposed to the extracellular space from the acidic lumen of intracellular vesicles during fusion. Expression of Syt 4-pHluorin in postsynaptic muscles resulted in intense fluorescence at specific subdomains in the postsynaptic membrane, defining regions where Syt 4 vesicles undergo exocytosis (Fig. 2F). The fluorescence was not diffusely present over the postsynaptic membrane but directed to restricted compartments. We co-stained Mhc-Gal4, UASSyt 4pHluorin larvae with antibodies against the postsynaptic density protein, DPAK, and nc82, a monoclonal antibody against a presynaptic active zone protein (5). Syt 4–pHluorin colocalized with DPAK and localized adjacent to nc82, demonstrating that Syt 4–pHluorin translocates from postsynaptic vesicles to the plasma membrane at postsynaptic densities opposite presynaptic active zones (Fig. 2, G and H).

To examine the function of Syt 4–dependent postsynaptic vesicle fusion, we characterized the phenotype of a syt 4 null mutant (syt 4BA1) (12) and a syt 4 deficiency (rn16) (14). Mutants lacking Syt 4 hatch from the egg case 21 hours after egg laying at 25°C, similar to wild type, and grow to fully mature larvae that pupate and eclose with a normal time course. To determine whether postsynaptic vesicle fusion triggered by Ca2+ influx is required for HFMR, we analyzed the effects of high-frequency stimulation in syt 4 mutants. In contrast to controls (Fig. 3A), the increase of miniature release was eliminated in syt 4 mutants (Fig. 3B). Postsynaptic expression of a UASsyt 4 transgene (15) completely restored HFMR in the null mutant (Fig. 3C), demonstrating that postsynaptic Syt 4 is required for triggering enhanced presynaptic function. Presynaptic expression of a UASsyt 4 transgene did not restore HFMR (Fig. 3E). In addition, postsynaptic expression of a mutant Syt 4 with neutralized Ca2+-binding sites in both C2A and C2B domains did not rescue HFMR, indicating that retrograde signaling by Syt 4 requires Ca2+ binding (Fig. 3D).

Fig. 3.

HFMR is abolished in the absence of postsynaptic Syt 4. (A to F) Stimulation protocol was the same as in Fig. 1. Left images are traces without high-frequency stimulation. Middle images are representative traces when stimulated. Whereas spontaneous release is rarely seen without stimulation [left in (A)], high-frequency stimulation induces a robust HFMR response [middle in (A)]. The induction of presynaptic miniature release is abolished in the syt 4 null mutant [syt 4BA1 (B); rn16 showed an indistinguishable phenotype] but restored in postsynaptically rescued synapses by Mhc-Gal4, UASsyt 4 [rn16 background (C)]. A syt 4 transgene with mutations in the C2A and C2B Ca2+-binding sites, UASsyt 4 [C2A D4N, C2B D3,4N], did not rescue HFMR in the null mutant [syt 4BA1/rn16 background (D)]. Presynaptic expression by elav-Gal4, UASsyt 4 did not rescue the loss of HFMR (E). HFMR was not observed in a null mutant of PKA, DC0B3 (F). (Right graphs) Miniature frequency at 1 min after tetanic stimulation compared with miniature frequency without stimulation (mean for 2 min). Number of samples analyzed: (A) nine no stimulation and eight stimulation, (B) nine no stimulation and six stimulation, (C) five no stimulation and five stimulation, (D) 10 no stimulation and six simulation, (E) seven no stimulation and five stimulation, and (F) five no stimulation and five stimulation. Double asterisks indicate P < 0.01 by Mann-Whitney's U test. Error bars are SEM. (G and H) Left images show representative traces of recordings before application of forskolin (500 μM) to activate PKA, and middle images show traces at 20 min after application. Both wild type (G) and the syt 4 null mutant [syt 4BA1 (H)] showed 100-fold increases in miniature release. (Right graphs) Miniature frequency before stimulation and 20 min after application of forskolin. Number of samples analyzed: (G) four, before and after; (H) four, before and after. Single asterisks indicate P < 0.05 by paired t test. Error bars are SEM. The scale for trace in (A) applies to all traces in (A) to (H).

The large increase in miniature frequency observed during HFMR is similar to the enhancement of presynaptic release after activation of cyclic adenosine monophosphate (cAMP)–dependent protein kinase (PKA) described in Aplysia (2) and Drosophila (16). Bath application of forskolin, an activator of adenylyl cyclase, results in a robust enhancement of miniature frequency at Drosophila NMJs (Fig. 3G) similar to that observed during HFMR, suggesting retrograde signals may function to increase presynaptic cAMP. To test the role of the cAMP-PKA pathway in HFMR, we assayed DC0 mutants (17) for the presence of HFMR. DC0 encodes the major catalytic subunit of PKA in Drosophila and has been implicated in olfactory learning (18). Similar to the lack of forskolin-induced miniature induction (19), DC0 null mutants lacked HFMR (Fig. 3F). Bath application of forskolin in syt 4 mutants resulted in enhanced miniature frequency (Fig. 3H), suggesting activation of the cAMP pathway can bypass the requirement for Syt 4 in synaptic potentiation.

To further explore the role of retrograde signaling at Drosophila synapses, we characterized the role of activity in synapse differentiation and growth. During Drosophila embryonic development, presynaptic terminals undergo a stereotypical structural change from a flat path-finding growth cone into varicose synaptic terminals through dynamic reconstruction (20). Such developmental changes in synaptic structure may share common molecular mechanisms with morphological changes induced during activity-dependent plasticity. We eliminated synaptic transmission by using a deletion mutation that removes the postsynaptic glutamate receptors, DGluRIIA and DGluRIIB (21) (hereafter referred to as GluRs). Postsynaptic currents normally induced by nerve stimulation were completely absent in the mutants (gluR) (fig. S1, A and B). Miniatures were also eliminated, even at elevated extracellular Ca2+ concentrations of 4 mM. In the absence of GluRs, the presynaptic morphology of motor terminals is abnormal, even though GluRs are only expressed in postsynaptic muscles (22). GluR-deficient terminals maintain a flattened growth cone–like structure and fail to constrict into normal synaptic varicosities (Fig. 4, A and B, and fig. S1, C and D; see fig. S1N for quantification). We also assayed synaptic development in a null mutant of the presynaptic plasma membrane t-SNARE [SNAP (soluble N-ethylmaleimide–sensitive factor attachment protein) receptor], syntaxin (syx), which eliminates neurotransmitter release (23), providing an inactive synapse similar to that in the gluR mutant. syx null mutants also have abnormal growth cone–like presynaptic terminals with less varicose structure (fig. S1E).

Fig. 4.

Postsynaptic activity–dependent presynaptic development mediated through retrograde signaling by Syt 4 and presynaptic PKA. (A to F) Presynaptic morphology of embryonic NMJs at hatching stage (21 hours after egg laying). Presynaptic terminals innervating muscle fibers 1 and 9 [top images in (A) to (C)] and muscle fibers 6 and 7 [bottom images in (A) to (C) and (D) to (F)] were stained with the neural membrane marker anti-HRP. Confocal micrographs in top images in (A) to (C) are shown as stacked images parallel to the body wall (X-Y), projected along the Z axis (perpendicular to the body wall), as indicated in (G). In bottom images in (A) to (C) and (D) to (F), confocal micrographs are shown in three axes, parallel to the body wall (X-Y), perpendicular to the body wall and longitudinal (X-Z), and perpendicular to the body wall and across the longitudinal axis (Y-Z), as indicated in (G). Wild-type embryos (A) have synaptic terminals that have constricted into individual varicosities (arrowheads). In gluR (B), the syt 4 null mutant [syt 4BA1 (C); rn16 showed an indistinguishable phenotype] and the null mutant of PKA, DC0B3 (D), mutant terminals fail to constrict into normal varicosities and maintain a flat less-varicose appearance (arrowheads) spreading through the X-Y plane or through the X-Z plane. Presynaptic expression of a constitutively active PKA transgene using the motor neuron–specific driver, D42-Gal4, restored normally constricted varicosities (arrowheads) in the DC0B3 background (E) and in the syt 4BA1 background (F). The scale bar in (F) is 5 μm and applies to all images in (A) to (F). (G) Schematic diagrams to show the arrangement of muscles (labeled 1, 9, 12, 13, 6, and 7) and motor terminals on the body wall in a Drosophila embryo. (H) Schematic diagram of a proposed local feedback model to describe synaptic plasticity and growth at Drosophila NMJs.

Because activity is required for synapse development, we tested whether Syt 4–dependent vesicle fusion may be required, similar to its role in acute retrograde signaling during HFMR. Physiological analysis revealed that the amplitude of evoked currents in mutants lacking Syt 4 was moderately reduced compared with wild type (fig. S1, F and G), suggesting weaker synaptic function or development. Similar to the morphological phenotype of the gluR mutant (Fig. 4B and fig. S1, D and N), syt 4 null mutant embryos showed defective presynaptic differentiation (Fig. 4C and fig. S1, H and N). Nerve terminals lacking Syt 4 displayed reduced varicose structure, whereas wild-type terminals have already formed individual varicosities at this stage of development (Fig. 4A and fig. S1, C and N). Postsynaptic expression with a UASsyt 4 transgene rescued the physiological and morphological phenotypes (fig. S1, F, G, I, and N). Syt 4 Ca2+-binding deficient mutant transgenes did not rescue either the morphological immaturity or the reduced amplitude of evoked currents (fig. S1, F, G, J, and N), even though Syt 4 immunoreactivity at the postsynaptic compartment was restored by muscle-specific expression of the mutant syt 4 transgene (fig. S1M), similar to the wild-type syt 4 transgene (fig. S1L) and endogenous Syt 4 (Fig. 2B) immunoreactivity.

Mammalian syt 4 was originally identified as an immediate-early gene that is transcriptionally up-regulated by nerve activity in certain brain regions (24). We therefore analyzed gain-of-function phenotypes caused by postsynaptic Syt 4 overexpression specifically in muscle cells to increase the probability of postsynaptic vesicle fusion. Syt 4 overexpression induced overgrowth of presynaptic terminals in mature third instar larvae (fig. S2, A to C), in contrast to overexpression of Syt 1 (fig. S2C), which does not traffic to Syt 4–containing postsynaptic vesicles (25). In addition to synaptic overgrowth, Syt 4 overexpression occasionally induced the formation of abnormally large varicosities. Postsynaptic overexpression of the Syt 4 Ca2+-binding mutant did not induce synaptic overgrowth (fig. S2C), indicating that retrograde signaling by Syt 4 also requires Ca2+ binding to promote synaptic growth.

To determine whether the cAMP-PKA pathway is important in activity-dependent synaptic growth, we assayed the effects of PKA on synaptic morphology. Expression of constitutively active PKA (26) presynaptically using a motor neuron–specific Gal4 driver induced not only synaptic overgrowth but also larger individual varicosities in mature third instar larvae (fig. S2D), similar to those induced by postsynaptic overexpression of Syt 4. These observations are consistent with the presynaptic overgrowth observed in the learning mutant, dunce, which disrupts the enzyme that degrades cAMP (27), and with studies in Aplysia implicating PKA in synaptic varicosity formation (2). We next characterized the loss-of-function phenotype of PKA mutants (DC0B3) (17) at the embryonic NMJ (Fig. 4D) to compare with gluR and syt 4 mutants (Fig. 4, B and C). Presynaptic terminals in the DC0 mutant were morphologically aberrant, with abnormal growth cone–like features and less varicose structure. Presynaptic expression of a constitutively active PKA transgene in the DC0 or syt 4 mutant backgrounds rescued the immature morphology (Fig. 4, E and F), suggesting activation of PKA is downstream of Syt 4–dependent release of retrograde signals.

Similar to the role of Syt 1–dependent synaptic vesicle fusion in triggering synaptic transmission at individual synapses, Syt 4–dependent vesicle fusion might trigger synapse-specific plasticity and growth. To test synapse specificity, we took advantage of the specific properties of the Drosophila NMJ at muscle fibers 6 and 7, where two motorneurons innervate both muscle fibers 6 and 7 during development (Fig. 4G). We expressed Syt 4 specifically in embryonic muscle fiber 6 but not muscle fiber 7 by using the H94-Gal4 driver (28). If Syt 4–dependent retrograde signals induce general growth of the motorneuron, one would expect to see a proliferation of synapses on both muscle fibers. Alternatively, if Syt 4 promoted local synaptic growth, one would expect specific activation of synapse proliferation only on target muscle 6, releasing the Syt 4–dependent signal. UASsyt 4 driven by H94-Gal4 increased innervation on muscle fiber 6 compared with that on muscle fiber 7 in third instar larvae (fig. S2, F to H). Control experiments with Syt 4 Ca2+-binding deficient mutant transgenes, or a transgene encoding Syt 1, did not result in proliferation (fig. S2, E, G, and H). Thus, synaptic growth can be preferentially directed to specific postsynaptic targets where Syt 4–dependent retrograde signals predominate, allowing differential strengthening of active synapses via local rewiring.

On the basis of the results described above, we propose a local feedback model for activity-dependent synaptic plasticity and growth at Drosophila NMJs (Fig. 4H). Synapse-specific Ca2+ influx triggers postsynaptic vesicle fusion through Syt 4. Fusion of Syt 4–containing vesicles with the postsynaptic membrane releases locally acting retrograde signals that activate the presynaptic terminal, likely through the cAMP pathway. Active PKA then triggers cytoskeletal changes by unknown effectors to induce presynaptic growth and differentiation. Moreover, PKA is well known to facilitate neurotransmitter release directly, triggering a local synaptic enhancement of presynaptic release as shown in HFMR. Therefore, postsynaptic vesicular fusion might initiate a positive feedback loop, providing a localized activated synaptic state that can be maintained beyond the initial trigger.

As a general mechanism for memory storage, Hebb postulated that potentiated synapses maintain an activated state until structural changes occur to consolidate alterations in synaptic strength (29). Our results demonstrate that acute plasticity and synapse-specific growth require Syt 4–dependent retrograde signaling at Drosophila NMJs. The feedback mechanism described here could be a molecular basis for both input-specific postsynaptic tagging (30) and an output-specific presynaptic mark or tag (31) for long-lasting potentiation. The regenerative nature of a positive feedback signal allows individual synapses to be tagged in a discrete all-or-none manner until synaptic rewiring is completed. The synaptic tag is maintained as a large increase in miniature frequency at Drosophila NMJs, suggesting a previously unknown role for miniature release in neuronal function. The spatial resolution for input and output specificity would result from the accuracy insured by Ca2+-dependent vesicle fusion and subsequent diffusion, similar to the precision of presynaptic neurotransmitter release.

Supporting Online Material

Materials and Methods

Figs. S1 and S2

References and Notes

References and Notes

Correction (29 March 2019): “Postsynaptic” has been corrected to “Presynaptic” on page 861, column 3, line 34.

View Abstract

Stay Connected to Science

Navigate This Article