Glycolytic Specialization Shapes Neuronal Physiology and Function in vivo

Abstract

Neurons perform diverse functions that impose distinct energetic demands, but how energy-metabolic pathways are matched to these functions in vivo remains unknown. Here we show that two functionally divergent sister chemosensory neurons in C. elegans, ASEL and ASER, exhibit asymmetric glycolytic flux, with ASER exhibiting high and ASEL having low levels of glycolysis. Metabolic imaging, metabolic network modeling, and electrophysiology measurements reveal that ASER’s elevated glycolysis supports a hyperpolarized resting potential, low input resistance, and rapid repolarization that enable a distinct functional role compared to ASEL. Impairing glycolysis collapses these electrophysiological specializations without abolishing neuronal excitability, and selectively disrupts ASER’s calcium responses while leaving ASEL largely unaffected. These findings demonstrate that neuron-specific glycolytic programs shape core biophysical properties and are required for functional identity in vivo, establishing metabolism as an active determinant of neuronal physiology.

Introduction

Neurons show great diversity in function, and their specialized computational roles arise primarily from cell-intrinsic properties such as ion channel composition, membrane conductance, and firing patterns that all define their physiological nature^1–3. The elements that support the inter- and intracellular communications of neurons require substantial energetic input, and as a result the brain is the most energetically-demanding organ in the body^4–6. Because of this, impairment of brain energy metabolism can dramatically affect neuronal function or physiology^7,8. Accordingly, during increased neuronal activity, neurons adapt their metabolic outputs to meet these elevated energetic needs^9–11. These observations underscore a tight coupling between neuronal physiology and energy metabolism, raising the question of how metabolic programs are organized across neuron types to support their distinct functions.

Specific metabolic programs are linked to cellular functions in specific tissues. For example, M1 killer macrophages use glycolysis for energy and reprogram mitochondria for ROS generation¹²; type IIB fast twitch muscle fibers use glycolysis to support physiological demands of short bursts of rapid energy production¹³; and vertebrate photoreceptors depend on elevated glycolysis to sustain an energetically-costly ion dark current that underlies sensory transduction¹⁴. Whether metabolic programs differ across individual neurons to support their function is less understood. Recently, we observed that the model organism C. elegans displays a range of stable glycolytic states that are characteristic of specific neuron types^15,16, suggesting metabolic specialization may be an under-appreciated property of neurons. Yet how such metabolic states are established and whether they are causally linked to cell-specific physiological functions in vivo remain unclear.

To investigate how distinct neuronal functions associate with baseline metabolic states at single-cell resolution, we turned to the C. elegans ASE chemosensory neurons, ASEL and ASER. These lateralized sister cells provide sensory input into the same neuronal circuit but adopt asymmetric physiological properties and behavioral roles^17–19: ASEL is excited by increases (“ON” responses) in NaCl concentration, whereas ASER is excited by NaCl decreases (“OFF” responses), enabling animals to interpret various salt gradients^20,21. These functional asymmetries are reflected in divergent calcium dynamics, synaptic communication patterns, and contributions to chemotaxis^21–24. Their mirrored identities, common circuit context and divergent functional specializations make ASEL and ASER an ideal system for uncovering how neuron-specific metabolic profiles are set, and how they contribute to the physiological and computational properties of single neurons in vivo.

Results

Distinct baseline glycolytic states in the functionally asymmetric ASE neurons.

To examine whether the asymmetric physiology of ASEL and ASER is accompanied by differences in baseline patterns or regulation of metabolism, we established a single-cell metabolic profiling platform in the two functionally distinct ASE neurons (Fig. 1A). We achieved this via three genetically encoded biosensors that report distinct nodes of glycolytic activity in vivo (Fig. 1B): HYlight²⁵, which measures the central glycolytic intermediate fructose-1,6-bisphosphate (FBP); R-iLACCO1.2²⁶, which reports lactate production as the endpoint of glycolysis; and SoNar²⁷, which monitors the cytosolic NADH/NAD⁺ ratio, a metabolic cofactor tightly coupled to glycolytic flux. By visualizing animals expressing these sensors in ASEL and ASER, we characterized their resting states of glycolysis with single-cell resolution in living animals.

Figure 1: Distinct states of glycolysis in the functionally asymmetric ASE neurons.

(A) Left, cartoon diagram of the location of both ASE chemosensory neurons in C. elegans; right, schematic of lateralized ASE calcium responses to salt changes (B) Depiction of the glycolysis pathway showing metabolites detected by the three biosensors used in this study. (C) Three representative ratiometric images of HYlight expression in ASE cell somas; ratio of excitation by 488 nm over 405 nm is displayed. Scale bar is 3 μm. Right, paired mean soma values for ASER and ASEL; each dot pair represents measurements from a single animal. (D) Mean soma measurements of HYlight in wild type vs. CRISPR-deletion mutant pfk-1.1(ola458); each dot represents one animal. Horizontal line marks the median for all animals. (E) Mean soma measurements of R-iLACCO1.2 for wild-type and pfk-1.1(ola458) mutant animals. (F) Mean soma measurements for SoNar in ASEL and ASER in wild-type animals. (G) Diagram of the MERGE computational pipeline³² used to predict attainable fluxes via the iMAT++ algorithm and to estimate relative flux potentials using enhanced flux potential analysis (eFPA)³⁰ in C. elegans neurons and other cell types. (H) Predicted flux potentials for reactions in the glycolysis pathway. Bar graphs represent distribution for 103 neuron classes in C. elegans. Blue and orange dots represent values for ASER or ASEL, respectively. Enzymes in x-axis are colored by which ASE neuron is predicted to have higher flux potential (ASER, blue; ASEL, orange; if similar, black). See table S1.*P<0.5; ****P<0.0001; n.s., non-significant. Statistical details available in table S2.

Comparing measurements of the ratiometric FBP sensor HYLight in ASER and ASEL within the same animals revealed significantly higher HYlight signal in ASER as compared to ASEL under baseline conditions (Fig. 1C). This difference was absent when using HYlight-RA (Fig. S1A), a binding-deficient control sensor. To validate that these HYlight measurements track bona fide changes in glycolytic flux, we examined mutants that perturb FBP production (Fig. 1D, S1B). The genes pfk-1.1 and pgk-1 encode enzymes acting upstream and downstream of FBP, respectively. The allele pfk-1.1(ola458), a CRISPR deletion of the primary phosphofructokinase ortholog in C. elegans¹⁵, displayed a markedly reduced HYlight signal in ASER, but had minimal effect in ASEL (Fig. 1D). A loss-of-function allele of the sole phosphoglycerate kinase, pgk-1(tm5613), caused HYlight signal to increase in both ASEL and ASER, consistent with accumulation of upstream metabolites, including FBP (Fig. S1B). Together, these data: 1) demonstrate that HYlight accurately reports changes in glycolytic pathway activity in the ASE neurons; 2) provide a framework for interpreting the elevated FBP observed in ASER relative to ASEL; and 3) suggest that the glycolytic pathway is highly active in ASER compared to ASEL under basal conditions.

To further examine if the elevated HYlight signal in ASER reflects increased glycolytic flux, we used the reporter R-iLACCO1.2 to measure lactate, a terminal endpoint of glucose metabolism via glycolysis. We observed that ASER exhibited significantly higher lactate levels than ASEL (Fig. 1E), consistent with enhanced glycolytic output. As with FBP, we used genetic perturbations to validate the reporter’s specificity in vivo: pfk-1.1(ola458) markedly reduced R-iLACCO1.2 signal in both neurons. A CRISPR-engineered deletion mutant of the sole lactate dehydrogenase ortholog, ldh-1(ola514), also reduced the R-iLACCO1.2 signal (Fig. S1C). We then measured the cytosolic NADH/NAD⁺ ratio by using the sensor SoNar, and observed a significantly higher SoNar signal in ASER as compared to ASEL (Fig. 1F), consistent with elevated glycolysis. Together, measurements with these three different biosensors for glycolysis indicate that ASER exhibits higher utilization of glycolysis compared to ASEL.

To determine whether these metabolic differences are reflected in their transcriptional expression patterns of these neurons, we modeled flux potentials. Changes in metabolic gene expression are predictive of metabolic fluxes^28–30, so we integrated single-cell RNA-seq data for 103 neuron classes³¹ with the C. elegans iCEL1314 metabolic network using the MERGE computational pipeline³² to estimate pathway-specific flux potentials (Fig. 1G). This flux modeling predicts higher potential in ASER for all glycolytic reactions from phosphofructokinase (Pfk) to pyruvate kinase (Pyk), and lactate dehydrogenase (Ldh) as well as other glycolysis-adjacent pathways (Fig. 1H, S1D–F). The model also predicts elevated flux potentials for glycogen degradation, an observation consistent with our past report on the role of glycogen supporting metabolic plasticity in ASER¹⁶. Together, these modeled flux potentials support our biosensor measurements in defining ASER as a high-glycolysis neuron.

Intrinsic neuronal identity specifies distinct glycolytic states and responses

A canonical approach for assessing differences in glycolytic pathway activity across cell types is the glycolysis stress test³³, a metabolic phenotyping assay that reports cellular dependence on and flexibility of glycolytic energy production. To further examine how ASEL and ASER make differential use of glycolysis, we adapted key features of this in culture assay to an in vivo context by leveraging single-cell biosensors, applying transient hypoxic conditions using a microfluidic device³⁴ (Fig. 2A) and monitoring the responses of the two ASE neurons simultaneously. We determined three parameters in the response of these neurons: 1) basal glycolysis, as the HYlight signal before hypoxia is applied; 2) glycolytic capacity as the HYlight signal achieved upon hypoxia; and 3) glycolytic reserve as the difference between glycolytic capacity and basal glycolysis. Both ASEL and ASER responded to hypoxia by increasing glycolysis, while the binding deficient HYlight-RA showed no response (Fig. 2A). Consistent with their different basal glycolytic states, we also observed that ASEL had a significantly higher glycolytic reserve compared to ASER (Fig. 2B), though ASER had a significantly higher glycolytic capacity (Fig. 2C). Together, our data indicate that ASEL and ASER both adapt glycolysis upon energetic stress, yet differ in their basal utilizations of this pathway.

Figure 2: Basal usage of glycolysis is associated with neuronal identity.

(A) Left, diagram of the hypoxia microfluidics system³⁴ used in this experiment, with a hypoxia stress test curve and definition of key measurements shown below. Right, pooled soma measurements of HYlight (solid line) or negative control HYlight-RA (dashed line) in ASEL (orange) or ASER (blue) upon hypoxia (vertical dotted line at time 0 reflects induction of hypoxic condition). (B-C) Values for glycolytic reserve (B) and capacity (C) calculated from data in Fig. 2A. (D-E) Mean soma measurements of either the olaIs141; otIs204 (two ASEL-identity) animals (D) or olaIs141; lsy-6(ot71) (two ASER-identity) animals (E). Each dot represents a paired measurement of either the left-side or right-side neuron within the same animal. Shaded region of glycolysis stress tests represents standard deviation around the mean. ****P<0.0001; ns, non-significant. Statistical details available in table S2.

The ASEL and ASER cell fates are specified by a deterministic, transcriptionally encoded left-right differentiation program^17–19. To determine whether the metabolic differences we observed are cell-intrinsic consequences of this fate-specification program, we examined mutants that disrupt the ASE transcriptional cascade (Figs. 2D–E). In otIs204[ceh-36p::lsy-6] transgenic animals, in which both ASE neurons adopt an ASEL-like identity³⁵, the metabolic differences seen in wild-type animals were abolished. The two neurons exhibited low baseline HYlight levels akin to ASEL in wild-type animals, and both neurons displayed similar responses upon the glycolytic stress test (Fig. 2D). Conversely, in lsy-6(ot71) mutants, both neurons adopt ASER-like identities³⁶ and exhibited HYlight baseline levels and responses that matched the ASER wild-type characteristics (Fig. 2E). Together, these results indicate that the distinct glycolytic states of ASEL and ASER arise from cell-intrinsic properties of their transcriptionally defined neuronal identities.

Glycolysis responds to ion flux across the membrane

While glycolysis is rapidly modulated by neuronal activity^9,37–39, we observe stable, identity-dependent differences in basal glycolytic states between ASEL and ASER. Therefore, we asked whether sensory-driven neuronal activity contributes to these metabolic differences. To do so, we built a microfluidic system¹⁶ that allows rapid and precise modulation of a relevant sensory stimulus for ASE neurons (salt concentrations), while performing confocal imaging of calcium, using JGCaMP8m⁴⁰, or HYlight (Fig. 3A). We observed that ASEL and ASER exhibited robust calcium transients in response to increases (for ASEL) and decreases (for ASER) in NaCl concentration (Fig. 3B, S2A), consistent with the known functions of these cells²¹. Under the same stimulation paradigm, we observed that both neurons showed comparable changes in HYlight values upon stimulation (Figs. 3C, S2B). Our findings are consistent with prior work showing that neuronal activity can drive rapid elevation of glycolysis^9,16,37–39. Importantly, our results reveal that the ASE neurons use two layers of metabolic modulation: a transcriptionally encoded baseline state that differs between ASEL and ASER, and a shared capacity for adaptive, activity- or stress-dependent upregulation of glycolysis, likely mediated through post-transcriptional or biochemical mechanisms.

Figure 3: Glycolysis responds to changes in membrane ion currents.

(A) Diagram of the salt mixing microfluidics system used to stimulate the ASE neurons with changing salt concentrations. (B) Pooled mean soma calcium responses (as JGCaMP8m change to BFP fluorescence) for a salt pulse experiment. Both ASEL and ASER values were captured for the same individual worms. (C) A similar experiment as in (B), but measuring HYlight responses instead. (D) Left, diagram of ASE neurons, showing the sensory cilia relative to the soma and axonal regions. The genes unc-13 and unc-18 function in the axon to facilitate synaptic vesicle release (see Fig. S2C). The genes tax-2 and tax-4 function in the cilia to initiate sensory responses. Right, schematic of loss-of-function experiments of the tax-2/4 cation channels and gain-of-function experiments using heterologous expression of the hyperactive alleles unc-58(e665) and twk-18(e1913). (E) Mean soma measurements of HYlight in either wild-type, tax-2(gk117937), or tax-4(p678) animals. (F) Mean soma measurements comparing wild-type animals and animals expressing UNC-58(L428F) in both ASE neurons. (G) Pooled HYlight measurements from a glycolysis stress test using the same animals as in (F). (H) As in (F) but expressing TWK-18(G165D) in both ASE neurons. Shaded regions around each time course reflect the standard deviation around the mean. *P<0.5; ***P<0.001; ****P<0.0001; ns, non-significant. Statistical details available in table S2.

We next sought to determine whether this metabolic distinction of ASEL and ASER arises from differences in basal metabolic demands. Prior studies that model the energy budget of neurons have indicated that a majority of neuronal ATP is consumed either maintaining ionic gradients for membrane potentials or supporting synaptic transmission and its associated processes^4,5,41. Thus, we used genetic mutants to determine which of these processes might contribute to the distinct metabolic states of these neurons (Fig. 3D). We observed that loss-of-function mutations in unc-13(s69) and unc-18(e234), which impair synaptic vesicle priming and docking respectively, had no effect on the difference in baseline glycolysis between these neurons (Fig. S2C). However, loss-of-function mutations in tax-2 and tax-4, which encode the β and α subunits of the cGMP-gated cation channel required for sensory-evoked neuronal activity²¹, abolished the metabolic distinction between ASEL and ASER. (Fig. 3E). These results implicate ion channel-mediated membrane conductance, rather than synaptic transmission, as a key determinant of the basal glycolytic state of ASE neurons.

To further test whether membrane potential and ionic gradients are linked to the observed basal glycolytic states of the ASE neurons, we manipulated conductance of either sodium or potassium ions via ectopic expression of cation channels carrying hyperactive gain-of-function variants of two-pore domain channels, UNC-58 and TWK-18 (Fig. 3D, F–H). ASE-specific expression of UNC-58(L428F), which selectively increases inward sodium flux⁴², produced a significant elevation in HYlight signal in both ASE neurons (Fig. 3F). Interestingly, the differences observed for the two ASE neurons in the glycolytic stress test were abrogated in the UNC-58(L428F) animals (Fig. 3G), consistent with a role for membrane potential in setting these distinct baseline differences between ASEL and ASER. ASE-specific expression of TWK-18(G165D), which enables a constitutive outwards flow of potassium⁴³, also resulted in a significant increase in the baseline of ASER, as well as a modest though significant effect in ASEL (Fig. 3H). Taken together, these experiments show that increasing the conductance of either Na+ or K+ cations across the membrane elevates the glycolytic state of ASE neurons, suggesting that differences in ion channel composition and conductance could contribute to the identity-dependent differences in basal glycolysis observed for ASEL and ASER.

Glycolysis is necessary to maintain ASER electrophysiological and functional identity.

Because differences in ion channels play an important role in setting the electrophysiological properties of neurons, we next asked whether the elevated glycolysis of ASER is derived from distinct membrane properties relative to ASEL. Using either wild-type or pfk-1.1(ola458) animals in which glycolysis is impaired, we performed current-clamp recordings from both ASE neurons in situ and recorded membrane voltage responses to incremental current injections (Fig. 4A). We observed that the resting membrane potential (RMP) was significantly more hyperpolarized in ASER compared to ASEL in wild-type animals, while this difference was lost in pfk-1.1(ola458) animals (Figs. 4A,B). In contrast, ASEL showed no change in RMP when comparing wild-type and pfk-1.1(ola458) animals. The input resistance of ASER was also significantly lower than ASEL in wild type (Fig. 4C), while this difference was also lost in pfk-1.1(ola458) animals. Notably, no change in membrane capacitance was observed in ASER between wild-type and pfk-1.1(ola458) animals (Fig. S3A), showing that the change in input resistance between them was not due to a change in cell size.

Figure 4. Glycolysis is necessary to support physiology and sensory responses in ASER.

(A) Current injection protocol and the resulting representative traces of membrane voltage responses of ASER and ASEL in wild-type and pfk-1.1(ola458) animals. All subsequent analyses are shown for a sample size of five wild-type and six pfk-1.1 animals. (B-D) Statistical comparisons of the resting membrane potential (B), input resistance (C), and threshold for active responses (D) between neurons and across genotypes. (E) A voltage response trace illustrating the definitions of Peak 1 (P1) and Peak 2 (P2) amplitudes, activation time (ActT), 90% repolarization time (RepT90), and peak duration (PeakDur). See also Fig. S3. (F-G) Comparisons of P2 amplitude (F) and 90% repolarization time between neurons and across genotypes. Data for B-G are shown as mean ± SEM. (H) Left, pooled calcium traces in ASEL upon a salt pulse (0 mM NaCl to 50 mM NaCl, at the vertical dotted line). Right, comparison of the percent increase for the calcium peak for wild-type vs pfk-1.1(ola458) animals relative to stimulation. Each dot represents one animal. (I) Left, pooled calcium traces observed within ASER upon a salt pulse stimulation paradigm known to activate ASER (75 mM to 25 mM NaCl). Right, the percent change as in (H, right) but for ASER data. Shaded regions around each time course reflect the standard deviation around the mean. *P<0.5; **P<0.01; ***P<0.001; ****P<0.0001; ns, non-significant. Statistical details available in table S2.

Consistent with its hyperpolarized RMP and lower input resistance, the ASER neuron had a significantly higher threshold for active (regenerative) responses compared to ASEL, while in the pfk-1.1(ola458) mutant background, this distinction was lost (Fig 4D). In addition, other physiological characteristics of the ASER responses were selectively affected in the pfk-1.1 mutant (Figs. 4E–G, S3A–D). ASER in pfk-1.1(ola458) had significantly higher peak voltage amplitudes compared to ASER in wild type (Figs. 4F, S3B); this same trend was also observed for 90% repolarization time (Fig. 4G), though not observed for activation time nor peak duration (Figs. S3C–D). Notably, the threshold for regenerative responses, peak voltage amplitudes, and 90% repolarization time were unchanged in ASEL between wild-type and pfk-1.1 animals. Together, these findings suggest that the higher glycolytic state in wild-type ASER is necessary to maintain distinct electrophysiological properties compared to ASEL, including a more hyperpolarized RMP, lower input resistance, and higher threshold for active depolarization.

To determine whether these asymmetries in glycolysis and electrophysiology between ASEL and ASER underlie their roles in sensation, we next compared calcium responses of ASEL and ASER to stimulating changes in NaCl between wild-type and pfk-1.1(ola458) animals. ASEL calcium responses to NaCl were indistinguishable in wild-type and pfk-1.1(ola458) animals, indicating that disruption of glycolysis does not affect calcium responses in ASEL (Fig. 4H). In contrast, pfk-1.1(ola458) mutant worms exhibited a significantly reduced calcium response in ASER compared to wild-type worms (Fig. 4I). All together, these findings show that ASER, but not ASEL, requires glycolytic flux to support its characteristic sensory physiology, underscoring that elevated glycolysis is fundamental to the functional identity of this neuron.

Discussion

Our findings reveal that neurons exhibit cell type-specific metabolic programs that are integral to their physiological identities. Although the two ASE neurons together act as sensory inputs to the same behavioral circuit, they have distinct roles in the computation of this behavior. Here we have shown these two cells also maintain different metabolic states that support differentiated electrophysiological properties. ASER utilizes a metabolic paradigm organized around elevated glycolysis and depends upon this pathway to support its more hyperpolarized resting membrane potential, lower input resistance, and faster membrane repolarization kinetics. Disruption of glycolysis led to a loss of these features specifically in ASER, resulting in this neuron adopting properties of the less glycolytic ASEL, and dramatically affected sensation-induced calcium responses. Our findings thus establish metabolic state as a core parameter defining neuronal identity and function, akin to other properties such as ion channel composition, neurotransmitter use, and synaptic connectivity.

Neuronal energy demands are traditionally considered in the context of activity-driven consumption, yet neurons are also predicted to incur substantial metabolic costs at rest to maintain their membrane potentials^4,5,11. This principle is well illustrated by photoreceptors, where a dark current mediated by steady Na⁺ influx enables exquisite sensitivity but imposes one of the highest ATP burdens in the nervous system¹⁴. Here we have shown that increasing ion fluxes across the cell membrane by ectopic expression of hyperactive cation channels resulted in increased glycolysis, implicating passive currents in contributing to the different metabolic states. The hyperpolarized resting potential and reduced input resistance of ASER are likely indicative of greater outwards potassium flux at rest. Such elevated potassium efflux might incur higher demands on active ion pumps, such as the Na⁺/K⁺ ATPase, which is known to drive glycolysis upon neuronal stimulation⁴⁴. It is possible an absence of glycolysis limits the cycling rate of this pump, which would explain the shift in electrophysiological properties of ASER in the mutant. It is also possible that a glycolytically-derived metabolite could serve as a limiting factor in ASER function. For example, ASE neurons depend upon a conversion cycle between GTP and cGMP⁴⁵, and loss of glycolysis could limit GTP availability and disrupt cGMP signaling. While future experiments will be required to establish the precise mechanisms by which glycolysis regulates ASER physiology, our data indicate that glycolysis is required to sustain membrane conductance and shape neuron-specific electrophysiological properties.

Our findings indicate that metabolism in neurons serves as both a hardwired determinant of functional specialization and a flexible substrate for plasticity. The divergent metabolic programs of ASEL and ASER arise in part from their cell-intrinsic transcriptional identities, but can also shift within minutes in response to neuronal activity, transient hypoxia or other physiological cues. This dual nature of metabolism as both a fixed trait and a tunable module is reflected in the distinct glycolytic baselines of the ASE neurons, as well as in the fast metabolic responses we observe during sensory stimulation. Because ASER responsiveness is known to be modulated by feeding state⁴⁶, metabolic tuning provides a plausible node through which internal state may recalibrate sensory gain. Our findings support a model in which glycolysis not only dynamically adapts to changing metabolic demands as previously appreciated^9,16,37–39, but also may function to shape neuronal responsiveness, tuning ionic conductance and modulating information flow within circuits.

A metabolic dimension of neuronal identity expands our understanding of neural circuit organization and function. By demonstrating that metabolic pathways actively shape neuronal physiology in vivo, this study reframes metabolism as a determinant—rather than a mere consequence—of neuronal signaling. Because resources are spatially distributed across cells and circuits, metabolism may impose constraints on the connectome that shapes how information is processed. It may be that metabolic “landscapes” modulate information flow in ways analogous to how synaptic weights and positions affect circuit dynamics. Incorporating metabolism into our framework of neuronal diversity thus provides new insights into how neurons perform computations and how these computations may change when metabolic pathways are perturbed.

Methods

Molecular biology and constructs

Vectors for ASE expression of HYlight²⁵, R-iLACCO1.2²⁶, SoNar²⁷, and JGCaMP8m⁴⁰ were generated by inserting codon-optimized gene blocks of each sensor after a −3000 bp flp-6 promoter. R-iLACCO1.2 and JGCaMP8m was placed in frame with a subsequent T2A peptide and TagBFP2 construct to enable ratiometric measurements over BFP fluorescence; SoNar was placed in frame with a T2A-mScarlet-I3 construct so that the 405 excitation can be used ratiometrically over mScarlet fluorescence. To label ASER, we used either mCherry (alongside HYlight and JGCaMP8m), mTagBFP (alongside SoNar), or mStayGold (alongside R-iLACCO1.2) driven by a −308 bp gcy-5 promoter. Constructs for ectopic expression of hyperactive ion channels were made by inserting either UNC-58(L428F) or TWK-18(G165D) between the flp-6 promoter and a T2A-mCherry.

C. elegans maintenance and genetics

All worm strains were raised on nematode growth media at 20°C using the Escherichia coli strain OP50 as the sole food source on NGM/agar plates⁴⁷, which have approximately 51 mM of Na⁺ and 53 mM of Cl⁻. Analyses were performed on day-old adults by picking animals at the L4 larval stage to seeded NGM plates and performing experiments within 16–20 hours. The C. elegans Bristol strain N2 was used as wild-type controls. A list of strains used in this study can be found in table S3.

Transgenics

Transgenic strains were made by standard germline injection techniques; injection mixes of plasmids were normalized to 100 ng/uL of total DNA concentration using 100 kb ladder (New England Biolabs, Inc) as DNA filler. The strain expressing ASE::HYlight, DCR9228 (olaEx5471), was integrated using UV integration and outcrossed three generations to generate DCR9288 (olaIs141) and DCR9299 (olaIs142). CRISPR-Cas9 was used to generate genetic knockouts for ldh-1(ola514) based on published protocols⁴⁸. Cut sites adjacent to the 5’ and 3’ sites of the gene were ordered from Horizon Discovery as modified crRNA oligos (5’-GATATCTATTAAAAATGACA-3’ and 5’-GTTCGATGACTGAAGGAAAA-3’ respectively). Worms from the F1 generation were singled and sequenced for the ldh-1 deletion and the resulting homozygous progeny were outcrossed three times prior to use.

Microfluidics and microscopy

Imaging was done by placing an 8% imaging pad of agarose in water on top of a PDMS microfluidics device designed for controlling local atmospheric conditions of worms³⁴. Worms were placed within a 3 μL drop of imaging buffer (5 mM levamisole, 25 mM potassium phosphate pH 6, 1 mM calcium chloride, 1 mM magnesium sulfate, 50 mM NaCl, 102 mM glycerol). This buffer contained glycerol to increase the osmolality of the solution to approximately 260 mOsm to hold osmolality constant across NaCl concentrations ranging from 0 mM to 100 mM as required. 15–25 worms were placed in this drop without washing and allowed to sit for 10 minutes, covered with a plastic petri dish lid to prevent evaporation, before applying a No. 1.5 coverslip. This allowed the levamisole to take full effect and led to worms preferentially being on their backs when a coverslip is applied, allowing simultaneous visualization of both ASER and ASEL. For the identity swap experiments, the right and left neurons were identified based on their position in the left or right side of the animal, which was determined relative to the dorsal/ventral and anterior/posterior axes.

All imaging was performed on a Nikon Ti2 + CSU-W1 spinning disk confocal microscope and captured with a Hamamatsu Orca-Fusion BT CMOS camera at 16-bit pixel depth. After mounting, flow of compressed air was initiated immediately to prevent unwanted hypoxic conditions under the coverslip, and a holding period of 5 minutes was used to allow for thermal equilibration. During imaging, a manual valve was used to switch between air and pure compressed nitrogen gas at the same flow rate to induce hypoxic conditions.

Optimization of imaging parameters was done as previously described¹⁵. Conditions for each sensor were selected based on negative control experiments as available (e.g., HYlight-RA, or the ldh-1 mutant for R-iLACCO1.2) by adjusting laser power settings to generate a mean ratio of approximately 1.0–1.5x for the appropriate wavelengths, using approximately 50–200 ms exposure times to exceed an independent signal:noise ratio for each channel of approximately 20–25. Images were captured for experiments between 0.2–1 Hz.

For determining parameters of the glycolysis stress test, values for basal glycolysis were defined as the HYlight signal before hypoxia is applied. Each experiment consisted of animals mounted onto a microfluidics device, exposed to a 1 min period with normal air, followed by a 5 minute period under pure nitrogen gas. We first used animals expressing the negative control HYlight-RA construct, and calculated the baseline signal averaged over the last 30 seconds under hypoxia (4.5–5 minutes). We then repeated this experiment with animals expressing HYlight and averaged the HYlight response over the last 30 seconds of the experiment. Glycolytic capacity was calculated as the difference in the response for each neuron in the individual animals examined relative to the mean value of HYlight-RA in either ASER or ASEL. Glycolytic reserve was defined as the difference between this same 30 second window in hypoxia relative to the first 30 seconds of the experiment during normoxia calculated for each individual neuron.

Salt stimulation experiments

Experiments using salt flow microfluidics were performed by modifying the system and methods previously described¹⁶. Two NE-1002X syringe pumps (New Era Pump Systems, Inc., Farmingdale, NY) were loaded with variants of standard imaging buffer: a no-salt variant (NGM-0: 5 mM levamisole, 25 mM potassium phosphate pH 6, 1 mM calcium chloride, 1 mM magnesium sulfate, 202 mM glycerol) and a high-salt variant (NGM-100: 5 mM levamisole, 25 mM potassium phosphate pH 6, 1 mM calcium chloride, 1 mM magnesium sulfate, 100 mM NaCl, 2 mM glycerol). These were connected to a 600 μm passive microfluidics herringbone mixer (Darwin Microfluidics, Inc., Paris France; #CS-10001930) and flow-through conductivity electrode (Microelectrodes, Inc., Bedford NH; #16–900) in series. Tubing lengths were minimized to limit dead volume upon pump switching. This was connected to the input of a P10 PDMS microfluidics chamber⁴⁹ and subsequent waste line. Control of the pump system was via a custom Qt6/Python interface⁵⁰ written for this purpose; an improved version⁵¹ was subsequently built in C++, with GPT-4o (OpenAI, April 2025 version) and Claude Sonnet 4.5 (Anthropic, Sep 2025 version) used to assist with the code rewrite. Measurement of conductivity was achieved by continuous monitoring with a Orion Star A210 (Thermo Scientific, Inc) by using this custom interface software. The software is available on Github at https://github.com/adwolfe/PumpController.

Experiments for stimulating the ASE neurons were done by first anaesthetizing worms in imaging buffer containing either 0 mM NaCl or 50 mM NaCl, depending on chosen starting condition. Worms were washed in a 100 uL drop of imaging buffer on an unseeded NGM plate, then moved to a second 100 uL drop of imaging buffer for incubation for 10 minutes. A 1 mL syringe filled with the same imaging buffer was then used to aspirate the worms and inject them into the PDMS device worm chamber while flowing the starting buffer condition via the pumps. For comparing the change in glycolysis that occurred upon neuronal activation (Fig. 3B), a paradigm was selected that elicited a comparable calcium response in both cells. Worms were held at 0 mM NaCl, and then a 2 minute pulse of 50 mM NaCl was given, followed by a return to 0 mM NaCl. ASEL and ASER responded upon the increase and decrease of salt concentrations, respectively. For observing the effects of pfk-1.1(ola458), the paradigm was selected to maximize the change in calcium upon activation for either neuron. For ASEL, the same paradigm was used as before; for ASER, worms were held in 50 mM NaCl buffer, and then exposed to a 75 mM NaCl pulse for 2 minutes followed by a drop to 25 mM NaCl for 2 minutes.

Metabolic network modeling

Single-cell RNA-seq data from the CeNGEN project³¹ were aggregated into pseudo-bulk profiles using previously-described methods⁵², resulting in mRNA expression data in transcripts per million (TPM). All cell types represented by fewer than 100 single cells were excluded. For comparison, non-neuronal cell types with more than 500 sampled cells were included, with the exception of intestine, due to its unique metabolic role and direct access to degraded bacterial biomass⁵³, and body-wall muscle (anterior), which is physiologically similar to body-wall muscle and therefore redundant. The final dataset comprised 103 neuronal cell types, including 26 sensory neurons, and 10 non-neuronal cell types. These data were integrated with the C. elegans metabolic network model iCEL1314 through the MERGE computational pipeline³² as explained below.

Network-level optimized flux distributions were generated for each cell type using the iMAT++ algorithm with the following settings: (i) first, gene expression data were discretized into expression states using CatExp³². (ii) The iCEL1314 model was used with its default compartment structure. (iii) To avoid enforcing full biomass production in specialized cell types that synthesize only a subset of biomass components, the biomass assembly pathway was relaxed by introducing drain reactions consuming major biomass precursors (e.g., collagen and glycans). (iv) The model was constrained to allow uptake of nutrients that can be supplied by the intestine or are required for specific reactions to carry flux. Uptake of metabolites of non-bacterial origin and exchange of intermediate metabolites (e.g., TCA-cycle intermediates such as isocitrate) were discouraged by assigning the corresponding exchange reactions to the lowly expressed reaction set. Metabolites that are not part of the endogenous C. elegans metabolome⁵⁴ and may occur only as supplements, such as testosterone, were blocked using hard constraints. (v) Flux variability analysis (FVA) was performed where needed to determine the minimum and maximum feasible flux for each reaction, thereby characterizing the alternative solution space not captured by the optimized flux values.

Flux potential analysis was performed using the updated eFPA algorithm³⁰ with the following settings: (i) first, the modified iCEL1314 model used for iMAT++ analysis was retained. (ii) A default distance boundary of six reactions was used to define the local network neighborhood prioritized for expression integration³⁰. (iii) To discourage uptake and exchange of uncommon or intermediate metabolites (see above), the special penalties option of eFPA was used, penalizing the relevant boundary reactions by a factor of 25. (iv) For each cell type, reactions identified by FVA as incapable of carrying flux in one or both directions were constrained accordingly by adjusting reaction bounds, thereby preventing eFPA from using flux values outside the iMAT++ solution space.

Electrophysiology

Electrophysiological recordings were performed on Day 1 adult C. elegans hermaphrodites, using the strains DCR9288 for wild type and DCR9887 for pfk-1.1(ola458) experiments. For each experiment, an animal was immobilized by applying a small drop of 3M VetbondTM (1469SB) to the dorsal head region on a Sylgard-coated coverslip. A diamond dissecting tool (72028, Electron Microscopy Sciences, Hatfield, PA, USA) was used to make an incision through the glued cuticle. The resulting cuticle flap was pulled open and secured to the coverslip using GLUture (Zoetis Inc., Kalamazoo, MI, USA). The pharynx was then gently displaced to expose ASER and ASEL sensory neurons. ASE neurons were identified based on green HYlight fluorescence (flp-6 promoter) and their anatomical positions relative to the dorsal/ventral and anterior/posterior axes, and this identification was confirmed by the presence of ASER-specific mCherry fluorescence (gcy-5 promoter).

Whole-cell current-clamp recordings were obtained from both ASER and ASEL in each animal. A series of current injection steps (−1.0 pA to +6.0 pA, 0.5 pA increments, 600 ms duration) were applied to evoke changes in membrane potential. To mitigate potential time-dependent effects, the sequence of ASER and ASEL recordings was alternated across animals. ASER was recorded first in three of five wild-type animals and three of six pfk-1.1(ola458) mutant animals. Data from animals in which only one of the two neurons was successfully recorded were excluded from analysis.

Recordings were performed using borosilicate glass electrodes with tip resistance of approximately 20 MΩ. Signals were filtered at 2 kHz and sampled at 10 kHz using a dPatch Digital Patch Clamp Amplifier controlled by SutterPatch software (Sutter Instruments). The extracellular (bath) solution contained (in mM) 50 NaCl, 90 N-methyl-d-glucamine (NMDG), 5 KCl, 5 CaCl₂, 1 MgCl₂, 11 dextrose, and 5 HEPES (pH adjusted to 7.3 with HCl). The intracellular (pipette) solution contained (in mM) 107 potassium gluconate (K-Glu), 15.5 KOH, 13 KCl, 0.25 calcium gluconate (Ca-Glu), 4 MgCl₂, 5 Tris, 36 sucrose, 5 EGTA, and 4 Na₂ATP (pH adjusted to 7.2 with 8 mM HCl).

Electrophysiological recordings were exported as ABF files for subsequent quantification of the RMP, as well as the rise time and 90% decay time of membrane potential responses to current injection, using ClampFit (Molecular Devices). The RMP was calculated as the average membrane potential during the initial 100-ms pre-pulse interval across all 15 current injection steps. Rise time was defined as the interval between the onset of the current injection and the time to peak membrane potential. The 90% decay time was measured from the termination of the current injection to the point at which the membrane potential had decayed to 10% of its amplitude. Input resistance was determined from the slope of a linear fit to the relationship between injected current and the steady-state membrane potential. Current-voltage relationships were constructed from the steady-state membrane voltage changes (the average membrane voltage over the last 500 ms of the 600-ms current pulse) elicited by the first five current injection steps (ranging from −1.0 pA to +1.0 pA in 0.5-pA increments).