Lactate Attenuates Synaptic Transmission and Affects Brain Rhythms Featuring High Energy Expenditure

Summary Lactate shuttled from blood, astrocytes, and/or oligodendrocytes may serve as the major glucose alternative in brain energy metabolism. However, its effectiveness in fueling neuronal information processing underlying complex cortex functions like perception and memory is unclear. We show that sole lactate disturbs electrical gamma and theta-gamma oscillations in hippocampal networks by either attenuation or neural bursts. Bursting is suppressed by elevating the glucose fraction in substrate supply. By contrast, lactate does not affect electrical sharp wave-ripple activity featuring lower energy use. Lactate increases the oxygen consumption during the network states, reflecting enhanced oxidative ATP synthesis in mitochondria. Finally, lactate attenuates synaptic transmission in excitatory pyramidal cells and fast-spiking, inhibitory interneurons by reduced neurotransmitter release from presynaptic terminals, whereas action potential generation in the axon is regular. In conclusion, sole lactate is less effective and potentially harmful during gamma-band rhythms by omitting obligatory ATP delivery through fast glycolysis at the synapse.


INTRODUCTION
Lactate is a three-carbon, electron-rich metabolite that can be produced and released by various cell types of the body (Brooks, 2018). In utilizing cells, lactate is linked to oxidative ATP synthesis in mitochondria, which requires conversion back to pyruvate through the redox enzyme lactate dehydrogenase (LDH), the tricarboxylic acid cycle, and molecular oxygen serving as the final electron acceptor at the respiratory chain (Brooks, 2018;Dienel, 2019).
Neurons are generally capable of lactate uptake and utilization in mitochondria (Magistretti and Allaman, 2015;Dienel, 2019). Lactate can be released from glial cells, such as astrocytes and oligodendrocytes (Pellerin and Magistretti, 1994;Gandhi et al., 2009;Saab et al., 2016), and it can enter the brain parenchyma from the blood when physical activity increases plasma lactate to as high as 20 mM (Rasmussen et al., 2010;Dienel, 2019). High lactate levels also occur under pathological conditions, such as lactic acidosis, brain ischemia, and traumatic injury (Kraut and Madias, 2014;Glenn et al., 2015;Dienel, 2019). The shuttling of lactate between brain cells depends on various monocarboxylic acid transporters (MCTs) and follows the local concentration gradient (Barros, 2013;Mä chler et al., 2016).
A major issue is that lactate metabolism has been rarely linked to physiological neuronal activity that underlies information processing in the cortex. For example, the role of lactate utilization during different cortical network rhythms, which naturally occur during cognition and behavior in vivo, is widely unknown (Colgin, 2016;Dienel, 2019;Scheeringa and Fries, 2019). Similarly, lactate utilization in excitatory neurons and inhibitory interneurons, which can, based on their specific functions, substantially differ in electrophysiological and bioenergetic properties, has been barely explored Kann, 2016). Related to that, the necessity of fast glycolytic ATP supply during neuronal signaling, in particular the presynaptic vesicle filling with neurotransmitters, is not well established (Ikemoto et al., 2003;Hall et al., 2012;Ashrafi and Ryan, 2017;Lucas et al., 2018).
We addressed these fundamental issues by exploring lactate utilization at the neuronal network and single cell level in slice preparations of the hippocampus. We examined ex vivo slices from young adult rats and slice cultures from rat pups in interface (extracellular local field potential and oxygen recordings) or submerged (intracellular patch-clamp recordings) conditions (Discussion, Limitations of the Study, and Transparent Methods: Recording solution and drugs). We focused on two fast network rhythms: (1) gamma oscillations (30-70 Hz) that emerge in many cortical areas in awake mammals during perception, locomotion, and memory formation (Melloni et al., 2007;van Vugt et al., 2010;Colgin, 2016), and (2) sharp wave-ripples (>180 Hz) that arise during waking immobility and slow-wave sleep and likely assist in memory consolidation (Buzsá ki, 2015;Ramirez-Villegas et al., 2015). Both rhythms rely on precise mutual synaptic transmission between excitatory pyramidal cells and GABAergic interneurons (Há jos and Paulsen, 2009;Colgin, 2016).
In essence, we demonstrate that sole lactate is less effective than glucose in fueling gamma and thetagamma oscillations and identify attenuated synaptic transmission because of reduced neurotransmitter release as the main mechanistic cause.

Energetic Boundary Conditions during Gamma Oscillations in Ex Vivo Slices
In the normal brain, the concentrations of glucose and lactate in the extracellular space are approximately 2 and 3 mM, respectively (Zilberter et al., 2010). In brain slice preparations, glucose and lactate at 2-3 mM have been shown to maintain energy metabolism and thus synaptic function under special experimental conditions (Schurr et al., 1988;Ivanov et al., 2014;Díaz-García et al., 2017). We first tested whether fast neuronal network oscillations in the gamma band (30-70 Hz) tolerate energy substrate concentrations closer to the physiological range in ex vivo slices of the hippocampus. Notably, gamma oscillations associate with high energy expenditure (Niessing et al., 2005;Kann et al., 2011;Schneider et al., 2019).
Highly synchronized gamma oscillations were present in stratum pyramidale of the CA3 region under control conditions with standard glucose (10 mM). These oscillations share many properties with gamma oscillations in vivo (Há jos and Paulsen, 2009;Gulyá s et al., 2010;Kann et al., 2011). By contrast, 5 mM glucose or 10 mM lactate resulted in suppression of gamma oscillations, which was widely reversible . Specifically, there were clear decreases in frequency and power of the oscillations ( Figures  1E and 1F). These strong effects did not permit reliable analysis of synchronization and inner coherence of the oscillations (Transparent Methods: Data analysis). However, they likely reflect the large fall in the glucose concentration from the slice surface (10 mM) to the slice core (about 3 mM) in ex vivo slices (Lourenç o et al., 2019).
These data show that gamma oscillations in ex vivo slices require a large substrate concentration gradient from the ambient recording solution to the slice core because of high energy expenditure and longer diffusion distances inherent to slice preparations (Kann and Ková cs, 2007)

Lactate Evokes Neural Bursts during Highly Synchronized Gamma Oscillations
Previous studies suggested that lactate is an alternative -and even preferred -energy substrate of neurons to maintain survival and synaptic function (Schurr et al., 1988;Izumi et al., 1997;Bouzier-Sore et al., 2003;Wyss et al., 2011). We next tested whether highly synchronized gamma oscillations can be fueled with lactate at a concentration that mimics sufficient substrate supply in our experimental conditions. By approximation, we used the 2-fold concentration of lactate (two lactate molecules can be derived from one glucose molecule), emphasizing that 20 mM lactate and 10 mM glucose are not isocaloric (Limitations of the Study). 2 iScience 23, 101316, July 24, 2020 iScience Article Strikingly, lactate (20 mM) evoked recurrent neural bursts with an incidence of about 0.3/s that were superimposed onto gamma oscillations (Figures 2A and 2B). The amplitude of these bursts indicates moderate hyperexcitability rather than epileptiform discharges that can be evoked in these slices (Liotta et al., 2011). Indeed, moderate pharmacological disinhibition through GABA A -receptors also associated with neural bursting ( Figure S1). When increasing the fraction of glucose from 0 to 2 or 5 mM in energy substrate supply, the lactate effect was reversible for the number of slices that expressed neural bursts and for the burst amplitudes; the burst intervals were unchanged ( Figures 2C-2E). Further analysis did not reveal any signs of desynchronization or altered spatial propagation of gamma oscillations prior to the onset of the first neural burst ( Figures 2F, 2G, and S2).

OPEN ACCESS
These data show that sole lactate disturbs highly synchronized gamma oscillations, likely by an excitationinhibition imbalance in the local neuronal network. However, gamma oscillations can be fueled by supplemental lactate when a low amount of glucose is available.

Lactate Attenuates Less Synchronized Gamma Oscillations
Gamma oscillations can also be reliably induced in postnatal slice cultures of the hippocampus that permit experimental conditions with improved supply of oxygen and energy substrates (Kann et al., 2011;Huchzermeyer et al., 2013). We next tested how lactate affects gamma oscillations in slice cultures at ambient normoxia (20% oxygen fraction) as well as in the presence of theta oscillations that often occur simultaneously in vivo (Há jos and Paulsen, 2009;Colgin, 2016).
Gamma oscillations were induced in lactate (2-20 mM) that was later on replaced by standard glucose (10 mM) serving as control ( Figures 3A-3C). Lactate generally attenuated gamma oscillations. Specifically, power ( Figure 3E) and synchronization (Figures 3F and 3G) decreased stronger than frequency ( Figure 3D). Similarly, theta-gamma oscillations evoked by optogenetic tools (Figures 3H-3J) had a significantly lower power in lactate (20 mM) ( Figure 3L). The frequency was unaffected ( Figure 3K).  These data show that lactate attenuates less synchronized gamma and theta-gamma oscillations.

Lactate Does Not Affect Sharp Wave-Ripples
We next tested whether the aforementioned disturbances evoked by lactate were specific for gamma oscillations and focused on intermittent sharp wave-ripples that occur in vivo and in ex vivo slices Ramirez-Villegas et al., 2015).

OPEN ACCESS
iScience 23, 101316, July 24, 2020 5 iScience Article Sharp wave-ripples were stable in standard glucose (10 mM) ( Figures 4A and 4B). The sharp waves occurred with an incidence of about 12/min ( Figure 4D) and had a superimposed, fast periodic oscillatory component with frequencies of >180 Hz (''ripples''). These properties were similar to sharp wave-ripples in previous reports Hollnagel et al., 2014;Schlingloff et al., 2014). Remarkably, we did not detect any significant differences in the properties of sharp wave-ripples when fueled by sole lactate, even when lactate was present for 1 h ( Figures 4C-4F). This persistence of sharp wave-ripples also permitted to quantify adaptations in oxygen metabolism associated with lactate utilization at the same level of network activity (see below).
These data show that lactate specifically disturbs gamma and theta-gamma oscillations.

Lactate Increases the Oxygen Consumption
The energetic utilization of lactate in neurons requires lactate uptake through MCT-2 and conversion to pyruvate through LDH-1 (Chih et al., 2001;Magistretti and Allaman, 2015). The subsequent oxidative ATP synthesis in mitochondria consumes molecular oxygen (Kann and Ková cs, 2007;Dienel, 2019). We next investigated the oxygen metabolism associated with lactate supply. We used O 2 microsensors to determine local O 2 concentrations in the tissue (Huchzermeyer et al., 2013;Schneider et al., 2019).
The switch from sharp wave-ripples to highly synchronized gamma oscillations in ex vivo slices and in standard glucose (10 mM) markedly decreased the tissue O 2 concentration (Figures 5A, 5B, and 5D), indicative of increased oxidative ATP synthesis. To obtain a more precise estimate of the cerebral metabolic rate of oxygen (CMRO 2 ), we used oxygen depth profiles with high spatial resolution ( Figure 5C) and mathematical modeling of convective transport, diffusion, and activity-dependent consumption of oxygen (Schneider et al., 2019). Indeed, CMRO 2 was about 1.5-fold higher during highly synchronized gamma oscillations in the presence of glucose ( Figure 5E).
Lactate (20 mM) markedly decreased the tissue O 2 concentration and increased the CMRO 2 by about 9% during sharp wave-ripples (Figures 5F and 5G), with no change in the level of network activity ( Figure 4). Similar results were obtained with lactate (20 mM) during less synchronized gamma oscillations in slice cultures at ambient normoxia ( Figure 5J). Notably, the decrease in tissue O 2 concentration coincided with a clear attenuation of gamma oscillations (Figures 5H, 5I, and 3A-3G). This paradoxical effect of lactate hampered precise estimates on lactate-induced increases in CMRO 2 , however.
These data show that sharp wave-ripples feature lower energy expenditure than gamma oscillations and that lactate increases the oxygen consumption during unchanged or even attenuated network rhythms.

Lactate Attenuates Excitatory and Inhibitory Synaptic Transmission
Gamma oscillations and sharp wave-ripples are generated by precise mutual synaptic transmission between excitatory pyramidal cells and fast-spiking, GABAergic interneurons that inhibit the perisomatic region of pyramidal cells (  iScience 23, 101316, July 24, 2020 iScience Article interneurons show unique electrophysiological properties associated with high energy demand (Kann, 2016). To further examine the cellular mechanisms through which lactate disturbs fast network oscillations, we performed single and paired patch-clamp recordings in excitatory pyramidal cells and fast-spiking, inhibitory interneurons ( Figure 6A) in two regions of the hippocampus, i.e., CA3 and CA1 (Rozov et al., 2001;Valiullina et al., 2017).
We first characterized the excitatory drive that pyramidal cells (Figures 6B, top, and 6C) and fast-spiking interneurons (Figures 6B, middle, and 6D) receive at their dendrites in standard glucose (10 mM) and in lactate (20 mM), combining whole-cell, voltage-clamp recordings with extracellular electrical stimulation. To improve the ''space-clamp'' condition, we used Cs-containing intracellular solution. Lactate generally attenuated evoked excitatory postsynaptic currents (EPSCs) in pyramidal cells and in fast-spiking interneurons, which was reversible ( Figures 6C and 6D). Next, we determined the efficacy of synaptic transmission from fast-spiking interneurons to connected (postsynaptic) pyramidal cells using paired patch-clamp recordings (Figures 6B, bottom, and 6E). Although lactate did not affect the generation of action potentials in the axon of interneurons, it attenuated perisomatic inhibitory postsynaptic potentials (IPSPs) in pyramidal cells, which was reversible ( Figure 6E). These experiments were done with 10-Hz stimulation in either type of synapse.
Despite these clear effects on synaptic transmission, lactate affected neither the threshold for the generation of action potentials (''spiking'') ( Figures 6F and 6G) nor the number of action potentials ( Figure 6H) as revealed by intracellular injection of depolarizing electrical currents in pyramidal cells and fast-spiking, inhibitory interneurons (see also Figure S3). The slightly hyperpolarized resting membrane potentials (  The filling of neurotransmitters into vesicles in presynaptic terminals is an active process that requires substantial amounts of ATP to power the vacuolar H + -ATPase (Ashrafi and Ryan, 2017;Dienel, 2019). To further identify the mechanism by which lactate attenuates synaptic transmission, we explored the neurotransmitter content at active synapses. For this purpose, we investigated AMPA receptor-mediated glutamatergic transmission at excitatory synapses and GABA A receptor-mediated transmission at inhibitory synapses by combining extracellular stimulation, patch-clamp recordings and pharmacology to isolate and partially block the respective postsynaptic currents (Figure 7 and Transparent Methods: Electrophysiology).
We first explored the synaptic glutamate content by eliciting EPSCs in pyramidal cells with electrical synaptic stimulation (10 Hz) in the presence of the low-affinity competitive AMPA receptor antagonist g-D-glutamylglycine (gDGG) (Liu et al., 1999;Watanabe et al., 2005). Application of gDGG reduced the amplitudes of EPSCs during electrical stimulation in the presence of either glucose (10 mM) or lactate (20 mM) (Figure 7A). However, the effect of gDGG was significantly larger in lactate.
Next we explored the synaptic GABA content by eliciting inhibitory postsynaptic currents (IPSCs) in pyramidal cells with electrical extrasynaptic stimulation (10 and 40 Hz) in the presence of the low-affinity competitive GABA A receptor antagonist (1,2,5,6-Tetrahydropyridin-4-yl)methylphosphinic acid (TPMPA) (Jones et al., 2001). Application of TPMPA reduced the amplitudes of IPSCs during electrical stimulation in the presence of either glucose or lactate ( Figures 7B and 7C). Similar to the above results, the effect of TPMPA was significantly larger in lactate. The stimulation frequency had no effect on the amplitudes of IPSCs in each condition.
The larger suppression of postsynaptic responses in the presence of low-affinity competitive receptor antagonists suggests that sole lactate generally reduces neurotransmitter release from presynaptic terminals, most likely by omitting fast glycolytic ATP supply ( Figure 7D).

Gamma Oscillations and Sharp Wave-Ripples
Gamma oscillations emerge in many cortical areas and mainly in awake mammals, including humans. They support action potential timing and synaptic plasticity and associate with higher brain functions, such as sensory perception, attentional selection, motor activity, and memory formation (Melloni et al., 2007;van Vugt et al., 2010;Colgin, 2016). Sharp wave-ripples arise in the hippocampus during waking immobility, consummatory behavior, and slow-wave sleep (Buzsá ki, 2015;Ramirez-Villegas et al., 2015). They assist in transferring compressed hippocampal information to distributed neocortical circuits to support memory consolidation (Buzsá ki, 2015;Colgin, 2016). In the present study, we induced these cortical rhythms in  iScience Article (Schlingloff et al., 2014;Schö nberger et al., 2014;Bazelot et al., 2016) in stratum pyramidale that consists of densely packed somata of pyramidal cells. Therefore, local field potential recordings primarily reflect the strongly dominating inhibitory input to the perisomatic region of pyramidal cells by GABAergic interneurons (Freund and Buzsá ki, 1996;Megıas et al., 2001;Wittner et al., 2007;Schneider et al., 2015). Therefore, it is likely that the disturbances of fast network rhythms, especially of persistent gamma oscillations, are mainly caused by fast-spiking interneurons rather than pyramidal cells.
Using reliable estimations of CMRO 2 , we show that gamma oscillations associate with significantly higher energy expenditure than sharp wave-ripples; similar data were recently reported for the mouse (Schneider et al., 2019). Notably, gamma and theta-gamma oscillations require a hemodynamic response to match oxygen and energy substrate demands (

Lactate and Brain Energy Metabolism
Lactate is an alternative energy substrate in the brain ( Lactate can be generated and released from astrocytes and oligodendrocytes that primarily enwrap synapses and axons, respectively (Pellerin and Magistretti, 1994;Caesar et al., 2008;Gandhi et al., 2009;Saab et al., 2016). However, neurons can also release some lactate depending on the experimental condition (Waagepetersen et al., 2000;Dienel, 2019). Lactate uptake in neurons requires MCT-2 that has a lower Km value (0.7 mM) compared with the astrocytic transporters MCT-1 and MCT-4 (3.5 and 30 mM, respectively) (Hertz and Dienel, 2005). Lactate is also taken up from the blood during physical activity; exhaustive exercise, for example, can increase the arterial plasma lactate levels from about 1 mM to as high as 20 mM (Boumezbeur et al., 2010;Rasmussen et al., 2010;Dienel, 2012  iScience Article We note that the neuronal MCT-2 is proton-coupled (Halestrap, 2013). Thus, increased utilization of lactate may result in intracellular acidification (Dienel, 2017). Indeed, previous studies in neurons showed decreases in the intracellular pH by about 0.02 and 0.1 units at 5 and 20 mM lactate, respectively (Munsch and Pape, 1999;Ruusuvuori et al., 2010). Our data argue against a significant and functionally relevant intracellular acidification because of the presence of sharp wave-ripples and regular membrane excitability and spiking properties of different types of neurons in sole lactate (20 mM). However, quantification of local acidification in distinct subcellular compartments, such as presynaptic endings, in active excitatory and inhibitory neurons is technically challenging (Willoughby and Schwiening, 2002;Holmgren et al., 2010;Ruusuvuori et al., 2010).
During gamma oscillations, which feature particularly high energy expenditure, the consumption of extracellular lactate in neurons might be limited at the level of LDH-1 that converts lactate into pyruvate (Magistretti and Allaman, 2015;Dienel, 2019). Although the conversion from lactate to pyruvate is an equilibrative reaction (lactate + NAD + 4 pyruvate + NADH + H + ), it also depends on the regeneration of cytosolic NAD + . This putative rate-limiting step is governed by the malate-aspartate shuttle (Brand and Chappell, 1974;Mintun et al., 2004).

Effects on Fast Network Rhythms, Synaptic Transmission, and Oxygen Metabolism
Lactate disturbed gamma and theta-gamma oscillations in ex vivo slices and slice cultures under welldefined experimental conditions (Transparent Methods), similar to previous studies on metabolism and evoked synaptic activity (Cox and Bachelard, 1988;Kanatani et al., 1995). These findings do not support other reports suggesting that lactate is a full -and even preferred -energy substrate of active neurons in vitro and in vivo (Schurr et al., 1988;Izumi et al., 1997;Bouzier-Sore et al., 2003;Suzuki et al., 2011;Wyss et al., 2011). The main reasons for these diverging findings might be the use of immature neuronal cultures, the type of artificial electrical stimulation and/or anesthesia in the other reports, as well as the high energy expenditure of gamma rhythms (Niessing et al., 2005;Kann et al., 2011;Schneider et al., 2019). Remarkably, sole lactate induced recurrent bursts in ex vivo slices and attenuation of the oscillations in slice cultures, which might reflect general differences in the excitation-inhibition balance in both preparations. The neuronal disturbances might contribute to the (patho)physiological mechanisms underlying several (clinical) symptoms associated with high lactate levels, ranging from central fatigue during exhaustive exercise to epileptic seizures (Rasmussen et al., 2010;Kraut and Madias, 2014).
By contrast, lactate did not affect sharp wave-ripples. This finding is in line with previous reports suggesting that lactate can serve as an alternative energy substrate to support neuronal survival and basic forms of activity (Schurr et al., 1988;Izumi et al., 1997;Brown et al., 2001;Bouzier-Sore et al., 2003). The persistence of sharp wave-ripples might primarily reflect the lower energy expenditure and the intermittent nature of the events, both of which facilitate local diffusion of energy substrates and ATP and, therefore, metabolic recovery (Harris et al., 2012;Schlingloff et al., 2014;Dienel, 2019;Schneider et al., 2019). The persistence of sharp wave-ripples as well as the similar effects of low glucose and lactate on gamma oscillations argue against a significant action of the G i protein-coupled receptor for lactate (HCAR1) that decreases neuronal activity in vitro (Bozzo et al., 2013;Dienel, 2019).
Lactate evoked recurrent neural bursts during highly synchronized gamma oscillations in ex vivo slices, suggesting an excitation-inhibition imbalance. Indeed, lactate generally attenuated synaptic transmission at glutamatergic (pyramidal cell / pyramidal cell, pyramidal cell / fast-spiking interneuron; dendritic excitation) and inhibitory (fast-spiking interneuron / pyramidal cell; perisomatic inhibition) synapses because of reduced neurotransmitter release. During gamma oscillations pyramidal cells sparsely generate action potentials at 1-3 Hz, whereas fast-spiking, GABAergic interneurons fire much higher at >20 Hz (Gulyá s et al., 2010;Kann, 2016). Therefore, the synaptic effects of lactate that we observed during experimental electrical stimulation might differ more strongly at glutamatergic and GABAergic synapses during gamma oscillations. Gamma and theta-gamma oscillations in slice cultures maintained the excitation-inhibition balance, likely because of less synchronization and/or partial adaptation to uptake and utilization of lactate at this stage of tissue development (Limitations of the Study) (Schousboe et al., 1993;Wada et al., 1997;Dienel, 2019). We note that our slice cultures maturated in culture medium containing about 4 mM glucose (Transparent Methods: Preparation of slice cultures), which might have reduced functional alterations discussed for dissociated neuronal cultures grown in up to 30 mM glucose (Dienel, 2019). We limited the use of higher glucose and lactate concentrations to the short period of experimental recordings. However, lactate ll OPEN ACCESS iScience 23, 101316, July 24, 2020 13 iScience Article clearly attenuated gamma and theta-gamma oscillations, likely reflecting the attenuated synaptic transmission at excitatory and inhibitory synapses.
By contrast, sole lactate did not affect the intrinsic electrophysiological properties of neurons, including action potential generation. Similar findings were reported for stimulus-evoked axonal population responses (Walz and Harold, 1990;Yamane et al., 2000). These observations argue against the presence of general metabolic stress and/or acidification (see above), at least, in the soma, the proximal dendrites, and the axon. In fact, they support the concept that axons can be fueled with lactate from myelinating oligodendrocytes (Saab and Nave, 2017;Stedehouder et al., 2017).
The lactate-evoked disturbances of gamma-band rhythms are most likely caused by transient ATP shortages in presynaptic structures, particularly in fast-spiking inhibitory interneurons (Kann, 2016;Dienel, 2019). Presynaptic terminals of central neurons increase glucose uptake via GLUT-4 and upregulate glycolysis during sustained neuronal activity (stimulation at 10-20 Hz) (Ashrafi et al., 2017). This rapid way of glycolytic ATP supply has been proposed to significantly contribute to maintenance of the vesicle cycle, which is a major consumer of presynaptic ATP, especially when mitochondria are sparse or absent (Shepherd and Harris, 1998;Ikemoto et al., 2003;Ashrafi and Ryan, 2017). This notion is supported by our data demonstrating that adding low fractions of glucose (2 or 5 mM) to lactate supply (16 or 10 mM) progressively suppressed neural bursting. This is also in line with estimates that lactate can contribute up to 60% to oxidative brain metabolism, with glucose providing the rest (Boumezbeur et al., 2010;Dienel, 2019). By using pharmacological isolation and partial blockade of postsynaptic currents, we further identify the reduced neurotransmitter release from presynaptic endings of excitatory and inhibitory neurons in the presence of lactate (i.e., the lack of fast glycolytic ATP supply).
Lactate-evoked disturbances of gamma-band rhythms might also reflect limited neuronal uptake (MCT-2) and/or conversion (LDH-1) of lactate (Bak et al., 2006;Dienel, 2012), acidification of subcellular compartments (Walz and Harold, 1990;Dienel, 2019), and/or alterations of enzymes and ion channels caused by shutdown of the pentose phosphate pathway and concomitant changes in the redox state (Bolañ os, 2016). However, detailed knowledge about the bioenergetic properties of dendrites, axons, and presynaptic terminals in the different types of excitatory and inhibitory neurons is lacking Kann, 2016;Dienel, 2019).
Lactate increased the oxygen consumption during sharp wave-ripples and gamma oscillations. For sharp wave-ripples, we calculated an increase in CMRO 2 of about 9% under aerobic conditions; this range likely reflects enhanced oxidative metabolism in mitochondria to compensate for the shutdown of glycolytic ATP synthesis (Harris et al., 2012;Dienel, 2019). These data show that during network rhythms with lower energy expenditure, such as sharp wave-ripples, the cellular ATP synthesis can be reliably adapted, even when lactate replaces glucose.
Our finding that the lactate-induced increase in oxygen consumption can coincide with strikingly divergent effects on neuronal activity (no effects during sharp wave-ripples versus bursting or attenuation during gamma oscillations) is relevant to the interpretation of brain imaging data, such as fMRI (Magistretti and Allaman, 2015; Scheeringa and Fries, 2019).
The lactate-induced disturbances of gamma-band rhythms that we report here argue for careful therapeutic supplementation of exogenous lactate in neurologic (brain ischemia or traumatic injury) and other intensive care patients (Glenn et al., 2015;Magistretti and Allaman, 2015;Brooks, 2018).

Conclusions
Our data establish key principles regarding lactate and its effectiveness in fueling fast brain rhythms: (1) Lactate disturbs cortical gamma rhythms by neural bursting or attenuation, whereas sharp wave-ripples featuring lower energy expenditure resist.
(2) Lactate increases the oxygen consumption, whereas neuronal network activity can even decrease.
(3) Lactate attenuates synaptic transmission by reduced neurotransmitter release rather than altered intrinsic neuronal membrane properties, including action potential generation, in excitatory and inhibitory neurons. These principles are relevant to the general understanding of electrical and metabolic consequences of lactate fuel in neurons in health and disease, the interpretation of functional brain imaging, and the clinical application of lactate supplementation.

Limitations of the Study
We used ex vivo slices and slice cultures of the hippocampus that generally lack functional vasculature and blood supply. Therefore, slices are provided with external recording solution (artificial cerebrospinal fluid, aCSF). Slice cultures can tolerate relatively low fractions of oxygen (20%) and energy substrates (e.g., 5 mM glucose), whereas ex vivo slices require high oxygen (95%) and glucose (10 mM) fractions in the external recording solution to fuel gamma oscillations. However, the final concentrations of energy substrates depend on molecular properties, including weight, size, and charge, as well as diffusion, transport, and consumption within the tissue. In addition, the types of preparation (ex vivo slice with 400 mm thickness or slice culture with about 250 mm thickness) and the recording conditions (slice submerged in recording solution or slice kept at the interface of recording solution and ambient gas atmosphere) have a role. Because of technical limitations, we are currently unable to provide the exact concentrations of glucose and lactate for each depth of a slice. However, the large fall in the glucose concentration from the slice surface to the slice core to about one-third has been described for unstimulated hippocampal ex vivo slices, similar to the large falls in the O 2 concentration.
The concentrations of glucose (10 mM) and lactate (20 mM) used in our experiments are not isocaloric. This is because oxidation of one glucose molecule yields two more ATP than oxidation of two lactate molecules due to glycolytic metabolism. In addition, the rates of diffusion, membrane transport (MCTs), and conversion to pyruvate (LDH-1) might contribute to substantial differences between lactate and glucose metabolism.
We used ex vivo hippocampal slices from the young adult rat as well as organotypic hippocampal slice cultures from the rat pup after 10-15 days in vitro, which roughly corresponds to the third and fourth postnatal week in vivo. Whether our findings also apply to other species, cortical regions, and developmental stages needs to be addressed in further studies.
Our study on the effects of sole lactate fuel on cortical network rhythms and electrophysiological properties of central excitatory and inhibitory neurons is a complementary experimental approach to provide further insights into brain energy metabolism, without favoring either side of the controversy about the general role of lactate. Here, we demonstrate that sole lactate can fuel sharp wave-ripples. However, lactate is supplemental to some obligatory glucose in fueling gamma oscillations that feature high energy expenditure.  Note the much lower spiking rates of excitatory pyramidal cells. Data are given as median ± interquartile range (IQR = 75% percentile -25% percentile), error bars indicate minimal and maximal values.

Experimental model and subject details
Experiments were performed in organotypic hippocampal slice cultures (Kann et al., 2011) and in ex vivo (acute) hippocampal slice preparations from male Wistar rats (source: Charles-River Laboratories and in-house breeding facilities of the Kazan Federal University). All animal procedures were performed in accordance with the guidelines of the European Commission and were approved by the regional authorities of Baden-Württemberg (T46/14, T96/15, and T45/18) and the Kazan Federal University regulations on the use of laboratory animals (ethical approval by the Institutional Animal Care and Use Committee of Kazan State Medical University N9-2013).

Preparation of ex vivo slices
Adult Wistar rats (aged 6 -8 weeks, ~200 g) were decapitated during isoflurane anesthesia (1.5 vol% of isoflurane in a gas mixture of 70% N2O and 30% O2). Brains were rapidly removed and immediately transferred to aCSF (see below) at ~4°C, saturated with 95% O2 and 5% CO2. Horizontal hippocampal slices with 400 µm thickness were prepared at an angle of about 12.5° in the fronto-occipital direction (with the frontal portion up) using a Leica VT1000S Vibratome (Wetzlar, Germany) . This orientation preserves the connectivity within hippocampal regions as well as to the entorhinal cortex. After cutting, slices were immediately transferred to a Haas-type interface recording chamber, perfused with aCSF at a flow rate of 1.8 ml/min and maintained at 34 ± 1°C. Recordings were started after 2 h of recovery. For patch-clamp recordings, horizontal hippocampal slices with 300 µm thickness were prepared from 3 -4 week-old Wistar rats and stored until experiments at room temperature (22 -24°C).

Preparation of slice cultures
Organotypic slice cultures were prepared as follows (Kann et al., 2011;Huchzermeyer et al., 2013): hippocampal slices (400 μm) were cut with a McIlwain tissue chopper (Mickle Laboratory Engineering Company Ltd., Guildford, UK) from 9 -10 days-old Wistar rats (Charles-River, Sulzfeld, Germany) under sterile conditions. Slices were maintained on Biopore™ membranes (Millicell standing inserts, Merck Millipore, Darmstadt, Germany) between culture medium, consisting of 50% minimal essential medium, 25% Hank's balanced salt solution (Sigma-Aldrich, Taufkirchen, Germany), 25% horse serum (Life Technologies, Darmstadt, Germany), and 2 mM L-glutamine (Life Technologies), kept at pH 7.3, and humidified normal atmosphere (5% CO2, 36.5°C) in an incubator (Heracell, Thermoscientific, Dreieich, Germany). The calculated glucose concentration in the culture medium was about 4 mM. Using this glucose concentration in slice cultures, we aimed to reduce long-term adaptations in the expression of metabolic enzymes that have been discussed for cultures of primary neurons and astrocytes maintained in the presence of high glucose (Kann and Kovács, 2007;Dienel, 2017;Dienel, 2019).The culture medium (1 ml) was replaced three times a week. Slice cultures were used after 10 -15 days in vitro (DIV) (residual thickness of about 250 μm), when the tissue had recovered from the slice preparation and damaged cut surfaces were reorganized (Kann and Kovács, 2007). For recordings, the intact Biopore™ membrane carrying slice cultures was inserted into the interface type recording chamber (Huchzermeyer et al., 2013). Slice cultures were maintained at the interface between artificial cerebrospinal fluid (aCSF, flow rate 1.8 ml/min) and ambient gas mixture (75% N2, 20% O2 and 5% CO2, flow rate 1.5 l/min). Intact Biopore™ membrane inserts ensure constant supply of oxygen and energy substrates from the recording solution that flows underneath the Biopore™ membrane; the interface condition permits constant oxygen supply from the ambient gas mixture.
Notably, hippocampal slices mature during the culture period (Bahr et al., 1995;De Simoni et al., 2003). With respect to the animal's age at preparation, slice cultures at 10 -15 DIV feature complex networks of interconnected pyramidal cells and interneurons in the presence of glial cells (Schneider et al., 2015). The absence of hyperexcitable network states such as neural bursts indicates a well-balanced interplay between neuronal excitation and inhibition. This is crucial for reliable induction of network activities like gamma oscillations that are highly dependent on precise timing of action potentials.

Optogenetics
Optogenetical methods were used to evoke theta-gamma oscillations in slice cultures. At DIV 4 slice cultures were infected with an adeno-associated viral vector (AAV-CaMKIIα-hChR2(H134R)-mCherry, UNC Gene Therapy Center Vector Core, Chapel Hill, NC, USA). 1 µl virus was carefully applied onto the CA3 region of each slice. Slice cultures were then maintained in the incubator for at least 3 weeks for expression of humanized channelrhodopsin2 (hChr2) in pyramidal cells under the control of the CaMKIIα-promotor. For recordings, the intact Biopore™ membrane carrying slice cultures was inserted into the interface type recording chamber (see above). To evoke theta-gamma oscillations, slice cultures where excited with blue light (470 nm) from an LED. The light intensity of the LED was modulated with a sinusoidal intensity profile, with a frequency of 5 Hz (theta). Blue light was delivered to the entire slice culture, whereas LFPs were recorded in stratum pyramidale of the CA3 region.

Tissue oxygen concentration and CMRO2
The oxygen concentration was measured at different depths in stratum pyramidale of the CA3 region by using oxygen sensor microelectrodes (O2-sensor), i.e., standard OX-10 (Unisense A/S, Aarhus, Denmark). This modified polarographic Clark electrode consists of a glass-insulated Ag/AgCl reference anode and a guard cathode with the advantages of low sensitivity to motion artifact, minimal interaction with tissue, and low O2 consumption. The standard OX-10 has a tip diameter of 8 to 12 µm and a spatial resolution of the outside tip diameter. The O2-sensor was connected to a 4-channel microsensor multimeter (Unisense A/S) and polarized with -0.8 V overnight. For recordings, the O2-sensor was fixed in a mechanical micromanipulator at an angle of 60° and moved forward in steps of 23 µm (corresponding to a vertical depth of ~20 µm per step). Before and after each experiment, O2-sensors were individually calibrated using a two point calibration with aCSF saturated with 0% O2 + 100% N2 and 95% O2 + 5% CO2, respectively (Kann et al., 2011;Schneider et al., 2019). Changes in voltage were digitized on-line at 10 kHz and data were stored on a computer disk with Sensor Trace Basic (data rate: 10 samples/s, Unisense A/S) for offline analysis. The oxygen concentration at a given depth in the slice can be described by a reaction diffusion model (Hall et al., 2012;Huchzermeyer et al., 2013;Schneider et al., 2019) given by the following differential equation: • d 2 2 d 2 = • 2 2 + 2 : oxygen concentration [µM] at different depths in the slice : depth in the slice : diffusion constant 1.6 • 10 −3 µm/s : enzymatic properties of the respiratory chain 4.7 µM : oxygen consumption rate (CMRO2) [µM/s] The CMRO2 was estimated by fitting the solution of this differential equation to the experimentally measured oxygen depth profile, where CMRO2 was the fitting parameter. The differential equation was solved with MATLAB (MathWorks, Natick, MA, USA) using the bvp4c-function. Two boundary conditions were used: 1) the oxygen concentration was set to the measured value at the slice surface. 2) The gradient of the oxygen concentration was set to 0 at the minimal oxygen concentration measured. The model describes the oxygen dynamics, which depend on diffusive oxygen transport and the metabolic oxygen consumption rate within a slice. For each depth profile derived from the experiments, we determined the maximal CMRO2 by minimizing the square distance (R 2 -value) between values measured by the O2-sensor and the simulation of the reaction diffusion model. The model takes into account the physiological capillary pO2 and the saturation of the respiratory chain and was validated by oxygen depth profiles from dead hippocampal slices (Kann et al., 2011). For more detailed information on the mathematical model, see Schneider et al., (2019). Please note that the CMRO2 for a given brain tissue is also reported as µmol/g/min (Erecińska and Silver, 2001;Okada and Lipton, 2007;Zhu et al., 2007). To convert these values in mM/min (i.e. mmol/l/min) as provided in the present study, we assumed that the density of gray matter is roughly about 1.05 g/ml (Kasischke et al., 2011). Thus, the CMRO2 of about 1.8 µmol/g/min measured in anaesthetized rats at 37°C corresponds to approximately 1.9 mM/min (Zhu et al., 2007;Engl et al., 2017).

Data analysis
Offline analysis was performed in MATLAB (MathWorks) using custom written routines. For analysis of pharmacologically induced gamma oscillations, data segments of 5 min were subdivided into segments of 30 s, band-pass filtered (FFT filter, pass-band frequency: 5 -200 Hz) and processed with Welch's algorithm and a fast Fourier transformation (FFT size: 8192). The resulting power spectral density (PSD) plots had a resolution of 1.2207 Hz. Gamma oscillations were analyzed for various parameters, i.e., peak power spectral density (Power), peak frequency (f), area under the curve (AuC), full width at half maximum (FWHM) and TAU (Kann et al., 2011;Schneider et al., 2019). Power, FWHM and TAU primarily reflect number of activated synapses, synchronization and inner coherence, respectively. AuC reflects power and frequency width. Similarity and lag of gamma oscillations between CA1 and CA3 were calculated from cross correlation's 1 st peak amplitude and shift, respectively. Medians of subdivisions were calculated and used for further statistical analysis. Power and peak frequency of optogenetically induced gamma oscillations were calculated from wavelet transformations of data recorded during light stimulation. For statistical evaluation, we compared means from data segments of 1 min at an early (1.5 -2.5 min) and late stage (4 -5 min) of the experimental condition.
To analyze SPW-Rs, signals were first separated into their slow (sharp wave) and fast components (ripples). The slow component was obtained by low-pass filtering (FFT filter, cut frequency: 45 Hz) and used for event detection and calculation of amplitude and duration. The ripple component was isolated by a band-pass filter (FFT filter, pass-band frequency: 120 -400 Hz). Ripples were counted only when subsequent ripples crossed a threshold of 3 times the standard deviation (SD) of the band-pass filtered signal. For statistical evaluation, we compared 3 segments of 5 min.