SomeckpersName, LevipersName, SloinpersName, SpivakpersName, GattegnopersName, StarkpersName (2023date). “Positive and biphasic extracellular waveforms correspond to return currents and axonal spikes.” Communications Biologytitle·j.
Object of study — what non-negativeterm extracellularterm·mth waveformsterm are - which compartmentterm of which cell produces a positive or biphasicterm spiketerm
Research question — whether positive and biphasicterm spikesterm come from axonalterm forward propagation, dendritic backpropagationterm, or return currentsterm·def - three accounts that make different predictions about width and timing
Contribution — about a tenth of waveformsterm are non-negativeterm; near-somatic narrow positive potentials are return currentsterm·def and isolated non-negativeterm spikesterm are axonalterm potentials
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | non-negative extracellular spike waveforms — “Multiple biophysical mechanisms may generate non-negative extracellular waveforms during action potentials, but the origin and prevalence of positive spikes and biphasic spikes in the intact brain are unknown.” | |
| claim-type | mechanism — “Thus, near-somatic narrow positive extracellular potentials correspond to return currents, and isolated non-negative spikes correspond to axonal potentials.” | |
| method — evidence | — “Using high-density extracellular recordings from freely-moving mice, we investigated the possible origin of non-negative extracellular waveforms.” | |
| object of study — evidence | — “Taken together, the experimental findings and computational models support three hypotheses for the source of non-negative extracellular potentials during action potentials: axonal forward propagation, dendritic backpropagation, and return currents. Thus, the compartmental origin of non-negative extracellular waveforms is not established.” | |
| research question — evidence | — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | |
| contribution — evidence | — “Multiple biophysical mechanisms may generate non-negative extracellular waveforms during action potentials, but the origin and prevalence of positive spikes and biphasic spikes in the intact brain are unknown. Using extracellular recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveforms are non-negative.” | |
| transfer | from computational biophysics: the model prediction that somatic inward currents are balanced by outward return currents visible as positive spikes — “During an action potential, the recorded waveforms depend on neuronal morphology, spike initiation site, and the relative position of the electrodes and the subcellular compartments 3 – 7 . Extracellular potentials recorded next to different compartments are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somatic signals, recorded as negative deflections in an un-inverted extracellular record 11 – 14 , and non-negative potentials are rarely reported in vivo.” | transfers[0] |
| transfer | from in vitro slice physiology: the backpropagation and axonal accounts of non-negative spikes, brought into the intact freely-moving brain to be tested — “Taken together, the experimental findings and computational models support three hypotheses for the source of non-negative extracellular potentials during action potentials: axonal forward propagation, dendritic backpropagation, and return currents. Thus, the compartmental origin of non-negative extracellular waveforms is not established.” | transfers[1] |
| transfer | from hippocampal place coding: place fields, phase precession and ripple locking, used as the functional proof that these units are real neurons — “Positive phases of non-negative spikes occur in synchrony or just before wider same-unit negative spikes. Narrow positive spikes occur in isolation in the white matter. Isolated biphasic spikes are narrower than negative spikes, occurring right after spikes of verified inhibitory units. In CA1, units with dominant non-negative spikes exhibit place fields, phase precession, and phase-locking to ripples. Thus, near-somatic narrow positive extracellular potentials correspond to return currents, and isolated non-negative spikes correspond to axonal potentials.” | transfers[2] |
| transfer | from clinical neurology: the prospect of measuring axonal conduction velocity in the intact brain, for disorders with axonal pathology — “The analysis of P- and B-spikes may have direct implications for studying neural diseases. Presently, measurements of axonal propagation velocity are carried out ex vivo 49 , 64 , 87 – 89 or under anesthesia 90 . Future studies may measure axonal propagation velocity in the intact brain directly by computing time lags between N- and B-spikes in MM units with B-spikes. The approach may enable studying neurological disorders associated with axonal pathology, including Alzheimer’s disease 91 and multiple sclerosis 92 , 93 .” | transfers[3] |
| mechanism | A spike generated at the axon initial segment draws balancing outward return currents that appear as the positive phase in the extracellular record, and the spike then propagates to the soma, so the negative somatic phase arrives slightly later — “These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record.” | mechanisms[0] |
| mechanism | As the axonal spike propagates, proximal sodium channels are refractory so return currents lie distally; an electrode at one point therefore sees the positive wavefront arrive before the negative spike, producing the peak-then-trough biphasic shape — “When recorded at a single point in space near the axon over time, the wavefront of the positive peak occurs before the negative spike, yielding a peak-then-trough biphasic spike.” | mechanisms[1] |
| mechanism | Axons that myelinate closer to the soma stop producing recordable non-negative near-somatic potentials, which is why multi-modal interneurons are rare and multi-modal units are less frequent in CA1 — “The early myelination for neocortical INTs and CA1 PYRs may explain the scarcity of MM INTs and the lower prevalence of MM units in CA1.” | mechanisms[2] |
| mechanism | Non-negative spikes went largely unreported until now because low-density probes, restrictive detection settings and spike-sorting bias filtered them out of earlier datasets — “The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias.” | mechanisms[3] |
| mechanism | a spike generated at the axon initial segment produces balancing return currents that appear as the positive phase, and then reaches the soma slightly later as the negative spike — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currents. Up to charge movement times, return currents should occur at the same time as spike generation at the AIS 56 . Therefore, spikes representing return currents are expected to occur in sync or slightly before the somatic spike. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record. Then, the AIS spike propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike.” | mechanisms[4] |
| mechanism | and along an unmyelinated axon the return currents propagate ahead of the spike, so a single point in space records peak-then-trough — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currents. Up to charge movement times, return currents should occur at the same time as spike generation at the AIS 56 . Therefore, spikes representing return currents are expected to occur in sync or slightly before the somatic spike. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record. Then, the AIS spike propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike.” | mechanisms[5] |
| mechanism | with myelination distance from the soma - about 35 to 50 micrometers for neocortical pyramidal cells against 10 to 20 for CA1 - offered as the reason the classes differ in prevalence — “Isolated P- and B-spikes both corresponded to axonal spikes. Return currents in myelinated axons can flow via nodes of Ranvier, manifesting as isolated positive phase potentials that correspond to SM Punits. It is possible that the difference between the P- and B-spike waveforms originates from difference in the myelin sheath. In neocortical PYRs, the axon becomes myelinated 35–50 μm after the AIS 62 , 75–90 μm from the soma 63 . INT axons myelinate closer to the soma, 20–50 μm 64 . Axons of CA1 PYRs myelinate even closer, within 10–20 μm 65 . The early myelination for neocortical INTs and CA1 PYRs may explain the scarcity of MM INTs and the lower prevalence of MM units in CA1.” | mechanisms[6] |
| mechanism | and axonal potentials expected narrower than somatic ones because of channel density at the initial segment, backpropagating ones wider — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | mechanisms[7] |
| contradicts | the dendritic-backpropagation account of positive and biphasic spikes (cited by number as refs 17, 34, 35, 43) — “Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negative spikes are unlikely to be produced by backpropagation.” | claim_relations[0] |
| contradicts | the account that positive and biphasic spikes are backpropagating somatic spikes, on the grounds that they are narrower rather than wider — “It was suggested that P-spikes in vitro 17 and B-spikes in vivo 34 , 35 correspond to active backpropagation of somatic spikes into dendrites 43 . Previously-reported non-negative spikes were seen far from the somatic potentials 36 , whereas most of our P- and B-spikes were in close proximity to the N-spikes. Backpropagating spikes are wider than somatic signals 40 , 43 . Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negative spikes are unlikely to be produced by backpropagation.” | claim_relations[1] |
| supports | the return-current account of near-somatic positive spikes, from their synchrony with the negative spike — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currents. Up to charge movement times, return currents should occur at the same time as spike generation at the AIS 56 . Therefore, spikes representing return currents are expected to occur in sync or slightly before the somatic spike. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record. Then, the AIS spike propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike.” | claim_relations[2] |
| explains-away | the near-absence of non-negative spikes from the earlier literature, as an effect of recording density, detection parameters and sorting bias — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | claim_relations[3] |
| critiques | the practice of reporting spike timing without spike waveforms, which leaves the earlier place-cell literature unable to be checked — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” | claim_relations[4] |
| distinguishes-from | juxtacellular recording, gap-junctional coupling and potassium action potentials, each ruled out by its own signature — “A possible source of SM Punits is sporadic juxtacellular recording, established by electrode adhesion to the membrane 70 . However, the sizes of the electrodes employed (10 × 15 μm) and the diameter of a PYR soma (10–15 μm 71 ) are similar. Furthermore, SM Punit waveforms were recorded together with waveforms of other units by the same electrode. Thus, non-negative extracellular waveforms are unlikely to be due to juxtacellular recordings.” | claim_relations[5] |
| predicts | that blocking axonal sodium channels will convert the biphasic units into purely negative ones — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality. A causal link may be achieved using compartmental blockade. During recordings from the CA1 feedforward network, a specific blocker for axonal sodium channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockade will turn to SM PYRs with only N-spikes during the blockade, and that BIPs will disappear. Alternatively, if sodium channels in proximal dendrites are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes.” | claim_relations[6] |
| sense | A TENTH OF THE WAVEFORMS ARE A DIFFERENT PHYSICAL OBJECT: most studies focus on somatic signals recorded as negative deflections, and non-negative potentials are RARELY REPORTED IN VIVO — “During an action potential, the recorded waveforms depend on neuronal morphology, spike initiation site, and the relative position of the electrodes and the subcellular compartments 3 – 7 . Extracellular potentials recorded next to different compartments are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somatic signals, recorded as negative deflections in an un-inverted extracellular record 11 – 14 , and non-negative potentials are rarely reported in vivo.” | senses[0] |
| sense | AND THE PIPELINE IS WHY: the discrepancy may be due to low-density recordings, limited detection parameters, and SPIKE SORTING BIAS - with a very low high-pass corner used here to avoid distorting the waveform — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | senses[1] |
| sense | with the consequence stated as a prediction - the same signal processing should INCREASE the prevalence of non-negative spikes reported in future — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | senses[2] |
| sense | THE LITERATURE CANNOT BE RE-EXAMINED, BECAUSE THE DISCRIMINATING EVIDENCE WAS NEVER PUBLISHED: place-coding studies analyse spike timing and only rarely show waveforms, so it is unknown whether they counted these units as place cells — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” | senses[3] |
| sense | THREE RIVAL ACCOUNTS NAMED IN ADVANCE: axonal forward propagation, dendritic backpropagation, and return currents - with the compartmental origin declared NOT ESTABLISHED — “Taken together, the experimental findings and computational models support three hypotheses for the source of non-negative extracellular potentials during action potentials: axonal forward propagation, dendritic backpropagation, and return currents. Thus, the compartmental origin of non-negative extracellular waveforms is not established.” | senses[4] |
| sense | AND EACH IS SEPARATED BY A DIFFERENT MEASURABLE PROPERTY: axonal potentials narrower because of channel density, backpropagating ones wider, return currents narrow and synchronous or slightly early — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | senses[5] |
| sense | the backpropagation account is then killed by width - the observed spikes are NARROWER than the somatic ones, where backpropagating spikes are wider — “It was suggested that P-spikes in vitro 17 and B-spikes in vivo 34 , 35 correspond to active backpropagation of somatic spikes into dendrites 43 . Previously-reported non-negative spikes were seen far from the somatic potentials 36 , whereas most of our P- and B-spikes were in close proximity to the N-spikes. Backpropagating spikes are wider than somatic signals 40 , 43 . Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negative spikes are unlikely to be produced by backpropagation.” | senses[6] |
| sense | and the dendritic account by order - the positive phase is in synchrony with the negative spike and the biphasic trough comes after it, which forward propagation from the dendrites cannot give — “Sodium spikes can be generated not only in the soma/AIS but also in the dendrites 57 – 59 . In vitro 18 , 31 and in silico 32 , 33 , B-spikes were suggested to be generated in the dendrites by forward propagation of dendritic sodium spikes. In MM units, P-spikes were in sync with N-spikes, and the trough of the B-spikes occurred after the N-spike trough. Therefore, it is unlikely that P- and B-spikes were created by forward propagation of dendritic sodium spikes. Dendrites can also generate calcium spikes 60 , 61 . In a dendritic calcium spike, P- and B-spikes are expected to appear before the somatic spike, but calcium spikes are much wider. Thus, the most likely compartmental counterparts of MM P- and B-spikes are dendritic return currents and axonal potentials.” | senses[7] |
| sense | THREE ARTEFACTUAL EXPLANATIONS ELIMINATED TOO, EACH BY ITS OWN OBSERVATION: gap junctions would give two-sided cross-correlograms and these are unidirectional — “The short time lags of excitatory vINT-BIP pairs and the high-fidelity PYR-BIP transmission may suggest that B-spikes are generated by inter-neuronal gap junctions 66 – 68 . However, gap junction prevalence among neocortical PYRs is low 68 , inconsistent with the observation of neocortical MM PYRs with non-negative spikes. Moreover, neurons coupled by gap junctions typically yield CCHs with two-sided peaks 67 , 69 , whereas the CCHs of the excitatory vINT-BIP and PYR-BIP pairs are unidirectional. Therefore, it is unlikely that “excitatory” connections to BIP were caused by gap junctions.” | senses[8] |
| sense | juxtacellular contact is ruled out by the electrode being the same size as the soma AND by the waveforms appearing alongside other units on the same electrode — “A possible source of SM Punits is sporadic juxtacellular recording, established by electrode adhesion to the membrane 70 . However, the sizes of the electrodes employed (10 × 15 μm) and the diameter of a PYR soma (10–15 μm 71 ) are similar. Furthermore, SM Punit waveforms were recorded together with waveforms of other units by the same electrode. Thus, non-negative extracellular waveforms are unlikely to be due to juxtacellular recordings.” | senses[9] |
| sense | and potassium spikes by sharpness - previously reported ones are slow, these are much sharper — “In principle, P-spikes can be generated by somatic potassium action potentials. Potassium spikes were previously observed as slow “hyperpolarizing” responses in invertebrate muscles 72 , 73 , and may occur in neurons when extracellular potassium concentrations are very high or when potassium channels are inactivated 74 . However, the presently-reported P-spikes occur together with N-spikes and are much sharper than previously-reported potassium spikes.” | senses[10] |
| sense | THE GEOMETRY IS THE DISCRIMINATOR: the two regions have OPPOSITE axo-dendritic orientation, which is what makes the comparison between them informative — “The opposite axo-dendritic orientation of pyramidal cells in neocortex and CA1 makes the comparison between the two regions informative. Furthermore, the laminar organization of CA1, the distinct place coding in pyramidal cells in CA1 and in CA3 41 , and the local generation of high frequency ripples, allow differentiating between local and non-local generation of non-negative spikes.” | senses[11] |
| sense | and the relative orientation of the array and the axo-dendritic axis is shown by modeling to change the recorded waveform, so the apparent prevalence is partly a fact about the probe — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | senses[12] |
| sense | A RIVAL SOURCE ELIMINATED BY A PROPERTY PROFILE: CA3 place coding has a lower fraction of place cells, larger fields, higher spatial information and less phase precession, and these units do not match it — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | senses[13] |
| sense | THE FINDINGS ARE DECLARED CORRELATIVE AND THE CAUSAL TEST IS SPECIFIED: block axonal sodium channels and the biphasic units should lose their positive phase, block proximal dendritic ones and the two peaks should move apart — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality. A causal link may be achieved using compartmental blockade. During recordings from the CA1 feedforward network, a specific blocker for axonal sodium channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockade will turn to SM PYRs with only N-spikes during the blockade, and that BIPs will disappear. Alternatively, if sodium channels in proximal dendrites are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes.” | senses[14] |
| control | the waveform categorisation re-run under a symmetric definition of a B-spike (one SD at the peak and one at the trough) and equivalent definitions for P- and N-spikes — held — rules out that the three spike classes are an artefact of the particular SD thresholds chosen, rather than a property of the waveforms — “We verified that the categorization of waveforms was not sensitive to the specific parameter values noted above. In particular, when using a symmetric definition for B-spikes (one SD at the peak and one SD at the trough) and equivalent definitions for P-spikes and N-spikes (two SDs at the peak and trough, respectively), the number of B-/P-/N-spikes were 3123, 1097, and 32652.” | controls[0] |
| differs in | the account that positive and biphasic spikes are backpropagating somatic spikes, on the grounds that they are narrower rather than wider — THE AXIS IS DISTANCE FROM THE SOMATIC POTENTIAL. 'PREVIOUSLY-REPORTED NON-NEGATIVE SPIKES WERE SEEN FAR FROM THE SOMATIC POTENTIALS, WHEREAS MOST OF OUR P- AND B-SPIKES WERE IN CLOSE PROXIMITY TO THE N-SPIKES. Backpropagating spikes are wider than somatic signals.' The backpropagation account is not denied for the waveforms it was proposed for - it is denied for THESE, on a recording-geometry parameter and a width criterion. — “It was suggested that P-spikes in vitro 17 and B-spikes in vivo 34 , 35 correspond to active backpropagation of somatic spikes into dendrites 43 . Previously-reported non-negative spikes were seen far from the somatic potentials 36 , whereas most of our P- and B-spikes were in close proximity to the N-spikes. Backpropagating spikes are wider than somatic signals 40 , 43 . Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negative spikes are unlikely to be produced by backpropagation.” — “Previously-reported non-negative spikes were seen far from the somatic potentials” | claim_relations[1] |
| differs in | the practice of reporting spike timing without spike waveforms, which leaves the earlier place-cell literature unable to be checked — THE AXIS IS WHETHER THE MEASUREMENT NEEDED TO CHECK IS EVEN PUBLISHED. 'Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.' The earlier literature is not wrong and cannot be checked - the datum that would settle it was never reported. A defect that is unauditable in principle from the published record, which is the strongest form of this corpus's pipeline argument. — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells” | claim_relations[4] |
| differs in | juxtacellular recording, gap-junctional coupling and potassium action potentials, each ruled out by its own signature — THE AXIS IS THE ELECTRODE-TO-SOMA SIZE RATIO, GIVEN IN MICRONS. 'the sizes of the electrodes employed (10 x 15 um) and THE DIAMETER OF A PYR SOMA (10-15 um) ARE SIMILAR.' Juxtacellular recording is ruled out by a physical comparison rather than by an argument - the electrode cannot be inside what it is the size of. An alternative eliminated by geometry, and the numbers are in the quote. — “A possible source of SM Punits is sporadic juxtacellular recording, established by electrode adhesion to the membrane 70 . However, the sizes of the electrodes employed (10 × 15 μm) and the diameter of a PYR soma (10–15 μm 71 ) are similar. Furthermore, SM Punit waveforms were recorded together with waveforms of other units by the same electrode. Thus, non-negative extracellular waveforms are unlikely to be due to juxtacellular recordings.” — “the sizes of the electrodes employed (10 × 15 μm) and the diameter of a PYR soma” | claim_relations[5] |
| outcome (stated by the tagger, no quotation) | control — held | controls[0] |
| for which use | limitation — every compartmental attribution: ALL FINDINGS WERE CORRELATIVE AND WE DID NOT DEMONSTRATE CAUSALITY — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality.” | limitations[0] |
| for which use | limitation — the timing evidence specifically: THE TIMING ANALYSIS IS CORRELATIVE AND DOES NOT DEMONSTRATE CAUSALITY — “The timing analysis is correlative and does not demonstrate causality.” | limitations[1] |
| for which use | limitation — the spatial resolution of every waveform claim: THE EMPLOYED ELECTRODE SIZE averages over sources much smaller than an axon or proximal dendrite — “A second interpretational limitation is related to the employed electrode size.” | limitations[2] |
| for which use | limitation — the layer assignments: WE DID NOT RELY ON HISTOLOGICAL CONFIRMATION, BUT RATHER EMPLOYED AN ESTABLISHED PHYSIOLOGICAL ANCHOR — “A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1 pyramidal cell layer, high frequency ripples.” | limitations[3] |
| for which use | limitation — the prevalence figures: THE STRUCTURE OF THE MULTI-SHANK PROBES (VERTICAL SPAN OF 100-200 um) AND ELECTRODE SIZE ARE OPTIMIZED FOR high-density recording, not for these waveforms — “While the present dataset yielded many non-negative waveforms, the structure of the multi-shank probes (vertical span of 100-200 μm) and electrode size are optimized for high-density somatic recordings. The limited vertical span precludes monitoring distant potentials, and the prevalence of MM units may be even higher with higher-span probes 2 . A second interpretational limitation is related to the employed electrode size. A large electrode averages potentials from multiple small sources: the diameter of a neocortical PYR axon is 0.2-0.5 μm 78 , and the diameter of an apical proximal dendrite is 1-4 μm 79 . Smaller electrodes (e.g., 7 μm or smaller 80 ) may capture more potentials generated by non-somatic compartments, with the possible caveat of higher impedance and thermal noise. A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1 pyramidal cell layer, high frequency ripples.” | limitations[4] |
| for which use | limitation — the same scope as limitations[0], with the missing experiment named: A CAUSAL LINK MAY BE ACHIEVED U[sing a manipulation not performed here] — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality. A causal link may be achieved using compartmental blockade. During recordings from the CA1 feedforward network, a specific blocker for axonal sodium channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockade will turn to SM PYRs with only N-spikes during the blockade, and that BIPs will disappear. Alternatively, if sodium channels in proximal dendrites are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes.” | limitations[5] |
| for which use | limitation — any re-examination of the prior place-coding literature: BECAUSE STUDIES ... TYPICALLY ANALYZE SPIKE TIMING BUT SPIKE WAVEFORMS ARE ONLY RARELY SHOWN, IT IS UNKNOWN WHETHER PREVIOUS STUDIES CONSIDERED PUNITS AND B[IPs] — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” | limitations[6] |
| for which use | limitation — the same scope as limitations[4] - apparent prevalence depends on the region and on how the probe sits relative to the cells — “The fraction of P-spikes was higher in neocortex compared with CA1. The lower prevalence of P-spike units in CA1 may be due to a shadowing effect, caused by the high density of PYRs in str. pyramidale. Due to the layered organization of hippocampal region CA1, the shanks of a probe implanted vertically in the dorsal hippocampus and centered on str. pyramidale are parallel to the axo-dendritic tree of the PYRs. Therefore, B- and N-spikes of CA1 PYRs may be recorded simultaneously, yielding MM PYRs, consistent with the higher probability of observing MM units with B-spikes in CA1 compared with the neocortex.” | limitations[7] |
| replication | untested | |
| limitation | All the compartmental attributions are correlative; no causal manipulation was performed. — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality.” | limitations[0] |
| limitation | The timing evidence in particular does not establish causation. — “The timing analysis is correlative and does not demonstrate causality.” | limitations[1] |
| limitation | Electrode size averages over sources much smaller than an axon or proximal dendrite. — “A second interpretational limitation is related to the employed electrode size.” | limitations[2] |
| limitation | Layer identification relied on a physiological rather than a histological anchor. — “A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1 pyramidal cell layer, high frequency ripples.” | limitations[3] |
| limitation | three limitations are all about the probe - the vertical span misses distant potentials, the electrode averages over sources far smaller than itself, and the layer was identified physiologically rather than histologically — “While the present dataset yielded many non-negative waveforms, the structure of the multi-shank probes (vertical span of 100-200 μm) and electrode size are optimized for high-density somatic recordings. The limited vertical span precludes monitoring distant potentials, and the prevalence of MM units may be even higher with higher-span probes 2 . A second interpretational limitation is related to the employed electrode size. A large electrode averages potentials from multiple small sources: the diameter of a neocortical PYR axon is 0.2-0.5 μm 78 , and the diameter of an apical proximal dendrite is 1-4 μm 79 . Smaller electrodes (e.g., 7 μm or smaller 80 ) may capture more potentials generated by non-somatic compartments, with the possible caveat of higher impedance and thermal noise. A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1 pyramidal cell layer, high frequency ripples.” | limitations[4] |
| limitation | all findings were correlative, with the causal test named and its outcome predicted in both directions — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality. A causal link may be achieved using compartmental blockade. During recordings from the CA1 feedforward network, a specific blocker for axonal sodium channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockade will turn to SM PYRs with only N-spikes during the blockade, and that BIPs will disappear. Alternatively, if sodium channels in proximal dendrites are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes.” | limitations[5] |
| limitation | and it cannot be checked whether earlier place-cell studies included these units, because spike waveforms are only rarely shown — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” | limitations[6] |
| limitation | with the apparent prevalence itself depending on the region and on how the probe sits relative to the cells — “The fraction of P-spikes was higher in neocortex compared with CA1. The lower prevalence of P-spike units in CA1 may be due to a shadowing effect, caused by the high density of PYRs in str. pyramidale. Due to the layered organization of hippocampal region CA1, the shanks of a probe implanted vertically in the dorsal hippocampus and centered on str. pyramidale are parallel to the axo-dendritic tree of the PYRs. Therefore, B- and N-spikes of CA1 PYRs may be recorded simultaneously, yielding MM PYRs, consistent with the higher probability of observing MM units with B-spikes in CA1 compared with the neocortex.” | limitations[7] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | About a tenth of extracellular spike waveforms recorded in freely-moving mice are non-negative rather than the conventionally assumed negative deflection. — “Using extracellular recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveforms are non-negative.” | content_tags.findings[0] |
| finding | the waveform width and time lag analyses together suggest that extracellular P-spikes occurring in close proximity to the soma correspond to return currents — “Yet together, the waveform width and time lag analyses suggest that extracellular P-spikes that occur in close proximity to the soma correspond to return currents.” | content_tags.findings[1] |
| finding | since both dendritic return currents and backpropagation are unlikely sources of B-spikes, axonal potentials are most likely — “Since both dendritic return currents and backpropagation are unlikely sources of B-spikes, axonal potentials are most likely.” | content_tags.findings[2] |
| finding | Verified inhibitory interneurons showed apparently excitatory monosynaptic connections onto biphasic units, an apparent violation of Dale's law that is explained if the biphasic unit is a non-somatic compartment of the same interneuron. — “The apparent contradiction of Dale’s law is resolved if the BIP corresponds to non-somatic spikes of the vINT.” | content_tags.findings[3] |
| finding | Such apparently excitatory interneuron-to-biphasic-unit connections occurred far above chance (4% of 2208 pairs versus 0.1% expected). — “The number of excitatory connections (89/2208 pairs, 4%) was above chance (0.1%; p < 0.001, Binomial test).” | content_tags.findings[4] |
| finding | the similarity between the spatial coding properties of PYRs, Punits, and BIPs is consistent with non-negative potentials representing cellular compartments of local circuit CA1 PYRs — “In summary, the similarity between the spatial coding properties of PYRs, Punits, and BIPs is consistent with non-negative potentials representing cellular compartments of local circuit CA1 PYRs.” | content_tags.findings[5] |
| finding | precession was more prevalent among PYR fields (53.7%) compared to INT fields (3.9%), but not when compared with BIP (45.7%) or Punit fields (50%) — “Precession was more prevalent among PYR fields (440/820, 53.7%) compared to INT fields (6/155, 3.9%; p < 0.001), but not when compared with BIP (16/35, 45.7%; p = 0.6) or Punit fields (7/14, 50%; p = 0.88, G -test; Fig. 5l ).” | content_tags.findings[6] |
| finding | Punits and BIPs are phase-locked to high frequency ripples but spike earlier than PYRs in the ripple cycle, suggesting they correspond to non-somatic compartments of PYRs that reside in superficial CA1 layers — “Thus, Punits and BIPs are phase-locked to high frequency ripples but spike earlier than PYRs in the ripple cycle (Fig. 7o ), suggesting that Punits and BIPs correspond to non-somatic compartments of PYRs that reside in superficial CA1 layers.” | content_tags.findings[7] |
| finding | The two waveform types have opposite regional prevalence: positive-spike units are commoner in neocortex (6.4% vs 3.4%) while biphasic-spike units are commoner in CA1 (13.1% vs 10.2%). — “The fraction of units with P-spikes was higher in the neocortex (205/3189, 6.4%) compared with CA1 (207/5971, 3.4%; p < 0.001, G -test; Fig. 3d ), whereas the fraction of B-spike units was higher in CA1 (782/5971, 13.1%) compared with the neocortex (324/3189, 10.2%; p < 0.001, G -test; Fig. 3e ).” | content_tags.findings[8] |
| finding | in probes that span a vertical range of 300-620 μm, Punits appeared mainly in the white matter, 210 μm above str. pyramidale — “In probes that span a vertical range of 300–620 μm, Punits appeared mainly in the white matter, 210 [-40 260] μm ( n = 30) above str. pyramidale.” | content_tags.findings[9] |
| finding | the depth distribution of SM Punits and SM BIP differed from the depth distribution of SM PYRs (p < 0.001, Kolmogorov-Smirnov test) — “The depth distribution of SM Punits ( n = 150) and SM BIP ( n = 287) differed from the depth distribution of SM PYRs ( n = 4365; p < 0.001, Kolmogorov–Smirnov test; Fig. 6f ).” | content_tags.findings[10] |
| finding | Because a very low cutoff (0.1 Hz) was used and spikes were detrended, previously reported triphasic spikes likely correspond to the biphasic spikes reported here. — “Because we used a very low cutoff frequency (0.1 Hz) and detrended the individual spikes (Supplementary Fig. 2b ), previously-reported triphasic spikes likely correspond to B-spikes reported here.” | content_tags.findings[11] |
| finding | consistent with the predictions, in neocortex the cc was 0.23 for MM units with B-spikes, whereas in CA1 the cc was -0.66 — “Consistent with the predictions, in neocortex the cc was 0.23 ( n = 161 MM units with B-spikes; p = 0.005, permutation test; Fig. 4b ), whereas in CA1 the cc was -0.66 ( n = 472; p < 0.001; Fig. 4c ).” | content_tags.findings[12] |
| finding | A TENTH OF THE WAVEFORMS ARE NON-NEGATIVE, in densely-connected cortical networks of freely-moving mice — “Multiple biophysical mechanisms may generate non-negative extracellular waveforms during action potentials, but the origin and prevalence of positive spikes and biphasic spikes in the intact brain are unknown. Using extracellular recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveforms are non-negative.” | content_tags.findings[13] |
| finding | near-somatic narrow positive potentials correspond to return currents, and isolated non-negative spikes to axonal potentials — “Positive phases of non-negative spikes occur in synchrony or just before wider same-unit negative spikes. Narrow positive spikes occur in isolation in the white matter. Isolated biphasic spikes are narrower than negative spikes, occurring right after spikes of verified inhibitory units. In CA1, units with dominant non-negative spikes exhibit place fields, phase precession, and phase-locking to ripples. Thus, near-somatic narrow positive extracellular potentials correspond to return currents, and isolated non-negative spikes correspond to axonal potentials.” | content_tags.findings[14] |
| finding | with the implication stated for the field - how to interpret and understand neural signals captured with extracellular probes — “Identifying non-negative extracellular waveforms that correspond to non-somatic compartments during spikes can enhance the understanding of physiological and pathological neural mechanisms in intact animals. In-depth analysis of extracellular waveforms from freely-moving mice identifies their source and properties, offering widespread implications for how to interpret and understand neural signals captured with extracellular probes.” | content_tags.findings[15] |
| finding | MOST STUDIES FOCUS ON SOMATIC SIGNALS as negative deflections, and non-negative potentials are RARELY REPORTED IN VIVO — “During an action potential, the recorded waveforms depend on neuronal morphology, spike initiation site, and the relative position of the electrodes and the subcellular compartments 3 – 7 . Extracellular potentials recorded next to different compartments are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somatic signals, recorded as negative deflections in an un-inverted extracellular record 11 – 14 , and non-negative potentials are rarely reported in vivo.” | content_tags.findings[16] |
| finding | THREE RIVAL ACCOUNTS NAMED IN ADVANCE, with the compartmental origin declared NOT ESTABLISHED — “Taken together, the experimental findings and computational models support three hypotheses for the source of non-negative extracellular potentials during action potentials: axonal forward propagation, dendritic backpropagation, and return currents. Thus, the compartmental origin of non-negative extracellular waveforms is not established.” | content_tags.findings[17] |
| finding | AND EACH IS SEPARATED BY A DIFFERENT MEASURABLE PROPERTY: axonal narrower by channel density, backpropagating wider, return currents narrow and synchronous or slightly early — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | content_tags.findings[18] |
| finding | THE GEOMETRY IS THE DISCRIMINATOR: the two regions have opposite axo-dendritic orientation, which is what makes the comparison informative — “The opposite axo-dendritic orientation of pyramidal cells in neocortex and CA1 makes the comparison between the two regions informative. Furthermore, the laminar organization of CA1, the distinct place coding in pyramidal cells in CA1 and in CA3 41 , and the local generation of high frequency ripples, allow differentiating between local and non-local generation of non-negative spikes.” | content_tags.findings[19] |
| finding | the backpropagation account is killed by WIDTH - the observed spikes are narrower where backpropagating ones are wider — “It was suggested that P-spikes in vitro 17 and B-spikes in vivo 34 , 35 correspond to active backpropagation of somatic spikes into dendrites 43 . Previously-reported non-negative spikes were seen far from the somatic potentials 36 , whereas most of our P- and B-spikes were in close proximity to the N-spikes. Backpropagating spikes are wider than somatic signals 40 , 43 . Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negative spikes are unlikely to be produced by backpropagation.” | content_tags.findings[20] |
| finding | the positive phase is the balancing return current at the initial segment, with the negative spike arriving at the soma slightly later — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currents. Up to charge movement times, return currents should occur at the same time as spike generation at the AIS 56 . Therefore, spikes representing return currents are expected to occur in sync or slightly before the somatic spike. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record. Then, the AIS spike propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike.” | content_tags.findings[21] |
| finding | and the dendritic account is killed by ORDER, with calcium spikes additionally being much wider — “Sodium spikes can be generated not only in the soma/AIS but also in the dendrites 57 – 59 . In vitro 18 , 31 and in silico 32 , 33 , B-spikes were suggested to be generated in the dendrites by forward propagation of dendritic sodium spikes. In MM units, P-spikes were in sync with N-spikes, and the trough of the B-spikes occurred after the N-spike trough. Therefore, it is unlikely that P- and B-spikes were created by forward propagation of dendritic sodium spikes. Dendrites can also generate calcium spikes 60 , 61 . In a dendritic calcium spike, P- and B-spikes are expected to appear before the somatic spike, but calcium spikes are much wider. Thus, the most likely compartmental counterparts of MM P- and B-spikes are dendritic return currents and axonal potentials.” | content_tags.findings[22] |
| finding | myelination distance from the soma is offered as the reason the classes differ in prevalence — “Isolated P- and B-spikes both corresponded to axonal spikes. Return currents in myelinated axons can flow via nodes of Ranvier, manifesting as isolated positive phase potentials that correspond to SM Punits. It is possible that the difference between the P- and B-spike waveforms originates from difference in the myelin sheath. In neocortical PYRs, the axon becomes myelinated 35–50 μm after the AIS 62 , 75–90 μm from the soma 63 . INT axons myelinate closer to the soma, 20–50 μm 64 . Axons of CA1 PYRs myelinate even closer, within 10–20 μm 65 . The early myelination for neocortical INTs and CA1 PYRs may explain the scarcity of MM INTs and the lower prevalence of MM units in CA1.” | content_tags.findings[23] |
| finding | GAP JUNCTIONS RULED OUT BY CORRELOGRAM SHAPE: coupled neurons give two-sided peaks and these are unidirectional — “The short time lags of excitatory vINT-BIP pairs and the high-fidelity PYR-BIP transmission may suggest that B-spikes are generated by inter-neuronal gap junctions 66 – 68 . However, gap junction prevalence among neocortical PYRs is low 68 , inconsistent with the observation of neocortical MM PYRs with non-negative spikes. Moreover, neurons coupled by gap junctions typically yield CCHs with two-sided peaks 67 , 69 , whereas the CCHs of the excitatory vINT-BIP and PYR-BIP pairs are unidirectional. Therefore, it is unlikely that “excitatory” connections to BIP were caused by gap junctions.” | content_tags.findings[24] |
| finding | JUXTACELLULAR CONTACT RULED OUT TWICE: the electrode is the same size as the soma, and the waveforms appear alongside other units on the same electrode — “A possible source of SM Punits is sporadic juxtacellular recording, established by electrode adhesion to the membrane 70 . However, the sizes of the electrodes employed (10 × 15 μm) and the diameter of a PYR soma (10–15 μm 71 ) are similar. Furthermore, SM Punit waveforms were recorded together with waveforms of other units by the same electrode. Thus, non-negative extracellular waveforms are unlikely to be due to juxtacellular recordings.” | content_tags.findings[25] |
| finding | AND POTASSIUM SPIKES BY SHARPNESS: the previously reported ones are slow and these are much sharper — “In principle, P-spikes can be generated by somatic potassium action potentials. Potassium spikes were previously observed as slow “hyperpolarizing” responses in invertebrate muscles 72 , 73 , and may occur in neurons when extracellular potassium concentrations are very high or when potassium channels are inactivated 74 . However, the presently-reported P-spikes occur together with N-spikes and are much sharper than previously-reported potassium spikes.” | content_tags.findings[26] |
| finding | THE PIPELINE IS WHY THE LITERATURE HAS NONE: low-density recordings, limited detection parameters, and SPIKE SORTING BIAS — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | content_tags.findings[27] |
| finding | with the prediction that the reported prevalence will RISE as the same signal processing spreads — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | content_tags.findings[28] |
| finding | the fraction differs between regions, possibly through a shadowing effect from the density of cells in the pyramidal layer — “The fraction of P-spikes was higher in neocortex compared with CA1. The lower prevalence of P-spike units in CA1 may be due to a shadowing effect, caused by the high density of PYRs in str. pyramidale. Due to the layered organization of hippocampal region CA1, the shanks of a probe implanted vertically in the dorsal hippocampus and centered on str. pyramidale are parallel to the axo-dendritic tree of the PYRs. Therefore, B- and N-spikes of CA1 PYRs may be recorded simultaneously, yielding MM PYRs, consistent with the higher probability of observing MM units with B-spikes in CA1 compared with the neocortex.” | content_tags.findings[29] |
| finding | and modeling shows the relative orientation of array and axo-dendritic axis changes the recorded waveform — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | content_tags.findings[30] |
| finding | A RIVAL SOURCE ELIMINATED BY A PROPERTY PROFILE: these units match none of the four features of CA3 place coding — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | content_tags.findings[31] |
| finding | THE LITERATURE CANNOT BE RE-EXAMINED, because place-coding studies analyse spike timing and only rarely show waveforms — “Because studies of hippocampal place coding typically analyze spike timing but spike waveforms are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as place cells.” | content_tags.findings[32] |
| finding | THREE LIMITATIONS, ALL ABOUT THE PROBE: span, electrode size, and a physiological rather than histological anchor — “While the present dataset yielded many non-negative waveforms, the structure of the multi-shank probes (vertical span of 100-200 μm) and electrode size are optimized for high-density somatic recordings. The limited vertical span precludes monitoring distant potentials, and the prevalence of MM units may be even higher with higher-span probes 2 . A second interpretational limitation is related to the employed electrode size. A large electrode averages potentials from multiple small sources: the diameter of a neocortical PYR axon is 0.2-0.5 μm 78 , and the diameter of an apical proximal dendrite is 1-4 μm 79 . Smaller electrodes (e.g., 7 μm or smaller 80 ) may capture more potentials generated by non-somatic compartments, with the possible caveat of higher impedance and thermal noise. A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1 pyramidal cell layer, high frequency ripples.” | content_tags.findings[33] |
| finding | DECLARED CORRELATIVE, WITH THE CAUSAL TEST SPECIFIED and its outcome predicted in both directions — “Punits and BIPs were most consistent with return currents and axonal spikes, but all findings were correlative and we did not demonstrate causality. A causal link may be achieved using compartmental blockade. During recordings from the CA1 feedforward network, a specific blocker for axonal sodium channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockade will turn to SM PYRs with only N-spikes during the blockade, and that BIPs will disappear. Alternatively, if sodium channels in proximal dendrites are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes.” | content_tags.findings[34] |
| finding | and the next step named - classifiers AGNOSTIC to waveform shape or polarity may associate non-negative spikes with specific cell types — “Punits and BIPs may be further classified into different types. For instance, the present dataset already includes several relatively wide P-spikes, which may correspond to backpropagating spikes 34 , 35 . Because both PYR and INT can exhibit B-spikes, classification of single-modal units with non-negative spikes into distinct cell types 84 may be possible. Furthermore, classifiers agnostic to waveform shape or polarity 85 , 86 may associate non-negative spikes with specific cell types. The full dataset and relevant code for categorizing waveforms and classifying units are available.” | content_tags.findings[35] |
| finding | with a use proposed - axonal conduction velocity measured in the intact brain from the lag between the two phases — “The analysis of P- and B-spikes may have direct implications for studying neural diseases. Presently, measurements of axonal propagation velocity are carried out ex vivo 49 , 64 , 87 – 89 or under anesthesia 90 . Future studies may measure axonal propagation velocity in the intact brain directly by computing time lags between N- and B-spikes in MM units with B-spikes. The approach may enable studying neurological disorders associated with axonal pathology, including Alzheimer’s disease 91 and multiple sclerosis 92 , 93 .” | content_tags.findings[36] |
| quantity | 9160 — total single units recorded, from 17 mice across 197 sessions — “A total of 9160 units were recorded from the 17 mice during 197 sessions (Supplementary Table 1 ).” | content_tags.quantities[0] |
| quantity | 6959 (76%) ... 1334 (14.6%) ... 378 (4.1%) ... 489 (5.3%) — of the sorted units, PYRs, INTs, Punits, and BIPs respectively — “Of these, 6959 (76%) were PYRs, 1334 (14.6%) were INTs, 378 (4.1%) were Punits, and 489 (5.3%) were BIPs.” | content_tags.quantities[1] |
| quantity | 966/9160 (10.5%) — proportion of units that were multi-modal, higher in CA1 than in neocortex — “Pooling the neocortex and CA1 datasets together, 966/9160 (10.5%) of the units were MM, with higher prevalence of MM units in CA1 (689/5971, 11.5%) compared with the neocortex (277/3189, 8.7%; p < 0.001, G -test; Fig. 3c ).” | content_tags.quantities[2] |
| quantity | 38 [25 38] μs — median time lag between N- and P-spike in CA1 MM units, not consistently different from zero — “The median [95% confidence limits] time lag between N- and P-spike in CA1 MM units was 38 [25 38] μs, not consistently different from zero ( n = 25 MM units; p = 0.08, Wilcoxon’s test), and with an absolute value smaller than 50 μs ( p = 0.041, Wilcoxon’s test; Fig. 1g ).” | content_tags.quantities[3] |
| quantity | 110 [100 110] μs — median lag by which biphasic-spike troughs followed negative-spike troughs, in 99% of CA1 multi-modal units — “B-spike troughs followed the N-spike troughs (Fig. 2j ) in 466/472 (99%) of the CA1 MM units, with median [95% confidence limits] lags of 110 [100 110] μs ( p < 0.001, Wilcoxon’s test; Fig. 2j ).” | content_tags.quantities[4] |
| quantity | 0.98 [0.91 1.02] ms — width of CA1 biphasic spikes, narrower than the 1.07 ms same-unit negative spikes — “CA1 B-spikes had a width of 0.98 [0.91 1.02] ms, narrower than same-unit N-spikes (1.07 [1.02 1.11] ms; n = 469 MM units; p < 0.001, Wilcoxon’s test; Fig. 2k ).” | content_tags.quantities[5] |
| quantity | 0.306 — spike transmission gain of a pyramidal-biphasic pair recorded on opposite sides of one 30 micrometer shank, too high for a real synapse — “we found a PYR-BIP pair with extremely high spike transmission gain (STG; 0.306 postsynaptic spikes per presynaptic spike) and a CCH peak at a sub-millisecond, unidirectional time lag (Fig. 2i ).” | content_tags.quantities[6] |
| quantity | 0.53π ... 0.65π ... 1.01π ... 1.25π — every population fired at a different phase - earliest Punits, followed by BIPs, PYRs and finally INTs — “Every population fired at a different phase: earliest were Punits (0.53π [0.46π 0.87π]; n = 78), followed by BIPs (0.65π [0.46π 1.02π]; n = 155), PYRs (1.01π [0.89π 1.11π]; n = 3522) and finally INTs (1.25π [1.11π 1.48π]; n = 592; p < 0.001, Wheeler–Watson test; Fig. 7m ).” | content_tags.quantities[7] |
| quantity | 210 [-40 260] μm — height of white-matter positive units above the CA1 pyramidal layer — “In probes that span a vertical range of 300–620 μm, Punits appeared mainly in the white matter, 210 [-40 260] μm ( n = 30) above str. pyramidale.” | content_tags.quantities[8] |
| quantity | a tenth — the share of recorded waveforms that are non-negative - a class the field routinely does not report, and therefore a tenth of the data whose meaning was unexamined — “Multiple biophysical mechanisms may generate non-negative extracellular waveforms during action potentials, but the origin and prevalence of positive spikes and biphasic spikes in the intact brain are unknown. Using extracellular recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveforms are non-negative.” | content_tags.quantities[9] |
| quantity | about 10% — the same figure in the discussion, set directly against the observation that past studies rarely reported any — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | content_tags.quantities[10] |
| variable | spike category (independent) — “Every unit was then classified according to the spike category of the waveform in the main channel.” | content_tags.variables[1] |
| variable | time lag (dependent) — “To contrast the possibility that P-spikes recorded in close proximity to the N-spike correspond to axonal potentials with the possibility of dendritic return currents, we measured time lags between the trough of N-spikes and the peak of same-unit P-spikes (Fig. 1g ).” | content_tags.variables[3] |
| variable | spatial information (dependent) — “the spatial information carried by BIPs was 1.07 [0.35 2.76] bit/s ( n = 50 units), not consistently different from PYRs (1.18 [0.44 3.03] bit/s; n = 1042; p = 0.8; Kruskal–Wallis test; Fig. 5i )” | content_tags.variables[4] |
| variable | running speed (control) — “Trials with a mean running speed below 10 cm/s were excluded from analyses.” | content_tags.variables[5] |
| variable | prevalence — “Non-negative spikes comprise about 10% of the waveforms, but past studies rarely reported non-negative spikes. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sorting bias. To minimize waveform distortion due to filtering 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalence of non-negative spikes reported in the future.” | content_tags.variables[7] |
| variable | waveform — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | content_tags.variables[8] |
| variable | orientation — “Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientation of the electrode array and the axo-dendritic axis impacts the recorded waveforms. It is a-priori possible that some of the CA1 Punits and BIPs represent non-somatic (e.g., axonal) compartments of CA3 neurons projecting to CA1. Place coding in CA3 PYRs includes lower fraction of place cells, larger place fields, higher spatial information, and lower phase precession prevalence 41 . However, the prevalence and properties of place coding by SM CA1 BIPs and Punits were not consistently different from CA1 PYR place coding. Thus, the properties of Punits and BIPs recorded in CA1 are inconsistent with CA3 place coding.” | content_tags.variables[9] |
| definition | P-spike units — “Units for which the positive peak was larger than the trough on the electrode with the largest trough-to-peak magnitude were denoted P-spike units, Punits.” | content_tags.definitions[0] |
| definition | B-spike units — “Units for which the electrode with the largest trough-to-peak magnitude contained a B-spike are denoted as B-spike units, BIP (Fig. 2c ).” | content_tags.definitions[1] |
| definition | single-modal (SM) unit — “Units with negative-spike polarity were recorded by one or more electrodes on the same shank, denoted as single-modal (SM) units (Fig. 1b, c ).” | content_tags.definitions[2] |
| definition | multi-modal (MM) unit — “However, we also recorded units with spikes of distinct polarities on different electrodes, denoted as multi-modal (MM) units (Fig. 1b, c ).” | content_tags.definitions[3] |
| definition | biphasic index (BPI) — “Every spike was quantified using a biphasic index (BPI) that measures the ratio between the positive and negative phases (Eq. 1 ).” | content_tags.definitions[4] |
| definition | spatial span — “For a given unit, the spatial span is defined as the number of contiguous electrodes for which the waveforms are categorized, multiplied by the physical spacing between the electrodes.” | content_tags.definitions[5] |
| definition | spike transmission gain (STG) — “The spike transmission gain (STG) was defined as the area under the peak in the monosynaptic temporal region of interest (ROI; 0<t ≤ 5 ms), extended until the causal zero-crossing points.” | content_tags.definitions[6] |
| definition | verified inhibitory interneuron (vINT) — “We thus denote the INT as a verified inhibitory INT (vINT).” | content_tags.definitions[7] |
| definition | return currents — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential.” | content_tags.definitions[8] |
| definition | return current — “Return currents represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currents. Up to charge movement times, return currents should occur at the same time as spike generation at the AIS 56 . Therefore, spikes representing return currents are expected to occur in sync or slightly before the somatic spike. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spike is generated at the axon initial segment, the balancing return currents are evident as the positive B-spike phase in the extracellular record. Then, the AIS spike propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike.” | content_tags.definitions[9] |
| definition | non-negative — “During an action potential, the recorded waveforms depend on neuronal morphology, spike initiation site, and the relative position of the electrodes and the subcellular compartments 3 – 7 . Extracellular potentials recorded next to different compartments are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somatic signals, recorded as negative deflections in an un-inverted extracellular record 11 – 14 , and non-negative potentials are rarely reported in vivo.” | content_tags.definitions[10] |
| definition | biphasic — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | content_tags.definitions[11] |
| definition | AIS — “Sodium spikes generated at the axon initial segment (AIS) actively propagate forward along the axon and backwards into the soma and dendrites 37 , 38 . Therefore, in extracellular recordings near the soma, potentials generated by axons and backpropagation into the dendrites are expected to exhibit sub-millisecond time lags with respect to somatic spikes due to conductance time. Axonal potentials are expected to be narrower than the somatic spikes due to higher density of voltage-gated potassium (Kv1) channels in the AIS 39 , whereas backpropagating spikes are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currents in the proximal dendrites are expected to balance AIS spikes. Charge movement time, AIS-to-soma propagation delay, and the somatic charging add up to several dozens of μs, therefore return currents are expected to be narrow and occur in synchrony with or slightly before the somatic spikes.” | content_tags.definitions[12] |
| key-terms | waveform, spike, extracellular, non-negative, biphasic, axonal, dendrite, somatic, AIS, return current, backpropagation, compartment, myelin, unit, electrode, probe, prevalence, filtering, spike sorting, place, ripple, CA1, neocortex, CA3, pyramidal, orientation, causality, blockade, gap junction, juxtacellular, potassium, sodium, morphology |
| Mann-Whitney U test / Wilcoxon signed-rank test U | as written: Mann-Whitney’s U -test (unpaired samples) or Wilcoxon’s paired signed-rank test — “Differences between two group medians were tested with either Mann-Whitney’s U -test (unpaired samples) or Wilcoxon’s paired signed-rank test (two-tailed).” | content_tags.statistics[0] |
| Kruskal-Wallis test | as written: Kruskal–Wallis test, and corrected for multiple comparisons using Tukey’s procedure — “Differences between medians of four groups were tested with Kruskal–Wallis test, and corrected for multiple comparisons using Tukey’s procedure.” | content_tags.statistics[1] |
| likelihood-ratio (G) test G | as written: likelihood-ratio ( G -) test (two-tailed) — “Differences in the proportions of two categorical variables were tested with a likelihood-ratio ( G -) test (two-tailed).” | content_tags.statistics[2] |
| binomial test | as written: exact one-tailed Binomial test — “To estimate whether fractions were larger or smaller than expected by chance, an exact one-tailed Binomial test was used.” | content_tags.statistics[3] |
| Spearman rank correlation with permutation test cc | as written: Spearman’s rank correlation and tested with a permutation test — “Association between parameters was quantified using Spearman’s rank correlation and tested with a permutation test.” | content_tags.statistics[4] |
| Rayleigh test | as written: Rayleigh’s likelihood-ratio test of uniformity — “The statistical significance of unimodal phase locking was tested using Rayleigh’s likelihood-ratio test of uniformity.” | content_tags.statistics[5] |
| Wheeler-Watson two-sample test | as written: Wheeler–Watson’s nonparametric two-sample test (two-tailed) — “Differences between the circular medians of two groups were tested with Wheeler–Watson’s nonparametric two-sample test (two-tailed).” | content_tags.statistics[6] |
| Kolmogorov-Smirnov test | — “The depth distribution of SM Punits ( n = 150) and SM BIP ( n = 287) differed from the depth distribution of SM PYRs ( n = 4365; p < 0.001, Kolmogorov–Smirnov test; Fig. 6f ).” | content_tags.statistics[7] |
| Bonferroni correction | as written: Bonferroni’s correction — “Bonferroni’s correction was employed in cases of G -test multiple comparisons.” | content_tags.statistics[8] |
| median and interquartile range IQR | as written: median and interquartile range (IQR) — “Every box plot shows median and interquartile range (IQR), whiskers extend for 1.5 times the IQR in every direction, and a plus indicates an outlier.” | content_tags.statistics[9] |
| Gaussian mixture model | — “Finally, negative-spike units were classified into putative pyramidal cells (PYRs) or PV-like interneurons (INTs) using a Gaussian mixture model 42 (Supplementary Fig. 1a ).” | content_tags.statistics[10] |
| t-SNE dimensionality reduction | as written: t-SNE projection — “A t-SNE projection 54 based on five spatial coding features (spatial information, field size, in-field gain, precession effect size, and precession slope) showed that BIP and Punit fields are interspersed within PYR fields (Fig. 5o ).” | content_tags.statistics[11] |
| Poisson test | — “Units of all four populations exhibited consistent ( p < 0.05, Poisson test) firing rate increases during ripple events ( p < 0.001, Binomial test; Fig. 7e–h ).” | content_tags.statistics[12] |
| cross-correlograms | — “The short time lags of excitatory vINT-BIP pairs and the high-fidelity PYR-BIP transmission may suggest that B-spikes are generated by inter-neuronal gap junctions 66 – 68 . However, gap junction prevalence among neocortical PYRs is low 68 , inconsistent with the observation of neocortical MM PYRs with non-negative spikes. Moreover, neurons coupled by gap junctions typically yield CCHs with two-sided peaks 67 , 69 , whereas the CCHs of the excitatory vINT-BIP and PYR-BIP pairs are unidirectional. Therefore, it is unlikely that “excitatory” connections to BIP were caused by gap junctions.” | content_tags.statistics[13] |
| claim type | existence — a second reading DIFFERING from the recorded value; recorded: mechanism — “Multiple biophysical mechanisms may generate non-negative extracellular waveforms during action potentials, but the origin and prevalence of positive spikes and biphasic spikes in the intact brain are unknown. Using extracellular recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveforms are non-negative.” | alternatives.claim_type |
| method | high-density arrays in two regions whose pyramidal cells have opposite axo-dendritic orientation, so the geometry itself discriminates between the accounts — a second reading DIFFERING from the recorded value; recorded: electrophysiology — “The opposite axo-dendritic orientation of pyramidal cells in neocortex and CA1 makes the comparison between the two regions informative. Furthermore, the laminar organization of CA1, the distinct place coding in pyramidal cells in CA1 and in CA3 41 , and the local generation of high frequency ripples, allow differentiating between local and non-local generation of non-negative spikes.” | alternatives.method |
| phenomenon | positive and biphasic extracellular spikes, rarely reported in vivo and routinely treated as if they were not there — a second reading DIFFERING from the recorded value; recorded: non-negative extracellular spike waveforms — “During an action potential, the recorded waveforms depend on neuronal morphology, spike initiation site, and the relative position of the electrodes and the subcellular compartments 3 – 7 . Extracellular potentials recorded next to different compartments are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somatic signals, recorded as negative deflections in an un-inverted extracellular record 11 – 14 , and non-negative potentials are rarely reported in vivo.” | alternatives.phenomenon |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
Abstract. Multiple biophysical mechanisms may generate non-negativeterm extracellularterm·mth waveformsterm during action potentials, but the origin and prevalenceterm of positive spikesterm and biphasicterm spikesterm in the intact brain are unknown. Using extracellularterm·mth recordings from densely-connected cortical networks in freely-moving mice, we find that a tenth of the waveformsterm are non-negativeterm. Positive phases of non-negativeterm spikesterm occur in synchrony or just before wider same-unit negative spikesterm. Narrow positive spikesterm occur in isolation in the white matter. Isolated biphasicterm spikesterm are narrower than negative spikesterm, occurring right after spikesterm of verified inhibitory unitsterm. In CA1term, unitsterm with dominant non-negativeterm spikesterm exhibit placeterm·dom fields, phase precession, and phase-locking to ripplesterm. Thus, near-somatic narrow positive extracellularterm·mth potentials correspond to return currentsterm·def, and isolated non-negativeterm spikesterm correspond to axonalterm potentials. Identifying non-negativeterm extracellularterm·mth waveformsterm that correspond to non-somatic compartmentsterm during spikesterm can enhance the understanding of physiological and pathological neural mechanisms in intact animals. In-depth analysis of extracellularterm·mth waveformsterm from freely-moving mice identifies their source and properties, offering widespread implications for how to interpret and understand neural signals captured with extracellularterm·mth probesterm. Introduction Extracellularterm·mth recordings have been employed extensively during the last century to monitor the electrical activity of multiple neurons 1 , 2 . During an action potential, the recorded waveformsterm depend on neuronal morphologyterm, spiketerm initiation site, and the relative position of the electrodesterm and the subcellular compartmentsterm 3 – 7 . Extracellularterm·mth potentials recorded next to different compartmentsterm are expected to exhibit distinct polarities 8 – 10 . However, most studies focus on somaticterm signals, recorded as negative deflections in an un-inverted extracellularterm·mth record 11 – 14 , and non-negativeterm potentials are rarely reported in vivo. For simplified neuronal morphologies, computational models show that the somaticterm inward currents during a spiketerm are balanced by outward return currentsterm·def evident as positive extracellularterm·mth spikesterm (P-spikes), expected to be generated next to the proximal dendritesterm 5 , 9 , 15 , 16 . P-spikes were reported to correspond to backpropagationterm of somaticterm spikesterm into the dendritesterm 17 , 18 and to outward currents in the dendritesterm 19 , 20 . Others reported that P-spikes are generated by axonalterm potentials 21 , 22 . Near the axon, a propagating sodiumterm spiketerm is expected to yield inward currents with a negative phase, preceded by a wavefront of local return currentsterm·def yielding a positive phase 9 , 15 . Together, the two phases are expected to generate a biphasicterm spiketerm (B-spike). B-spikes were reported to correspond to axonalterm spikesterm 7 , 21 , 23 – 30 or dendritic spikesterm 18 , 31 . When sodiumterm spikesterm initiate at the distal dendritesterm, P-spikes are expected next to the soma, and B-spikes at the proximal dendritesterm 32 , 33 . Experimental studies suggested that B-spikes correspond to backpropagationterm of somaticterm spikesterm 34 – 36 . Taken together, the experimental findings and computational models support three hypotheses for the source of non-negativeterm extracellularterm·mth potentials during action potentials: axonalterm forward propagation, dendritic backpropagationterm, and return currentsterm·def. Thus, the compartmental origin of non-negativeterm extracellularterm·mth waveformsterm is not established. The three hypotheses may be contrasted in freely-moving animals using high-density extracellularterm·mth recordings and network activity analysis in regions where the axo-dendritic axis is parallel to the recording array, the neocortexterm and hippocampal region CA1term. Sodiumterm spikesterm generated at the axon initial segment (AISterm) actively propagate forward along the axon and backwards into the soma and dendritesterm 37 , 38 . Therefore, in extracellularterm·mth recordings near the soma, potentials generated by axons and backpropagationterm into the dendritesterm are expected to exhibit sub-millisecond time lagsterm·var with respect to somaticterm spikesterm due to conductance time. Axonalterm potentials are expected to be narrower than the somaticterm spikesterm due to higher density of voltage-gated potassiumterm (Kv1) channels in the AISterm 39 , whereas backpropagating spikesterm are expected to be wider due to lower dendritic density of KA channels 40 . On the other hand, return currentsterm·def in the proximal dendritesterm are expected to balance AISterm spikesterm. Charge movement time, AIS-to-soma propagation delay, and the somaticterm charging add up to several dozens of μs, therefore return currentsterm·def are expected to be narrow and occur in synchrony with or slightly before the somaticterm spikesterm. The opposite axo-dendritic orientationterm of pyramidalterm cells in neocortexterm and CA1term makes the comparison between the two regions informative. Furthermore, the laminar organization of CA1term, the distinct placeterm·dom coding in pyramidalterm cells in CA1term and in CA3term 41 , and the local generation of high frequency ripplesterm, allow differentiating between local and non-local generation of non-negativeterm spikesterm. Here, we show that in CA1term and neocortexterm, near-somatic narrow positive extracellularterm·mth potentials correspond to return currentsterm·def, and isolated non-negativeterm spikesterm correspond to axonalterm potentials. Results Positive extracellularterm·mth spikesterm correspond to return currentsterm·def To determine whether spikesterm of multiple polarities are observable in the intact brain, we recorded 5971 unitsterm from hippocampal region CA1term of freely-moving mice ( n = 9; Supplementary Fig. 1 ; Supplementary Table 1 ; Supplementary Table 2 ). Recordings were conducted using multi-shank arrays with vertical inter-electrode spacing of 15–20 μm (Fig. 1a ). Fig. 1 Positive extracellularterm·mth spikesterm correspond to return currentsterm·def. a Multi-shank recording array in hippocampal CA1term region. b Left, Schematic shank with 13 simultaneously-recorded unitsterm in CA1term of a freely-moving mouse. Str. pyramidale is in the top electrodeterm. X marks correspond to the unitsterm depicted in ( c ). Right, Wideband (0.1–7,500 Hz) traces recorded by four adjacent electrodesterm. acc, head acceleration. c Single-modal (SM) and multi-modal (MM) unitsterm·def recorded simultaneously from CA1term. Wideband spiketerm waveformsterm and auto-correlation histograms (ACHs) of the SM pyramidalterm cell (PYR), and of the MM PYR with a negative spiketerm (N-spike) and a positive spiketerm (P-spike). d Positive unitterm (Punit) in extracellularterm·mth recordings from CA1term. A Punit is defined as a unitterm with a dominant positive peak in the main channel waveformterm. e P-spikes are narrower than same-unit N-spikes. Spiketerm width, inverse of the dominant frequency of the waveformterm spectrum. MM PYRs with P-spikes and MM Punits are included. * p < 0.05, Wilcoxon’s test. f P-spikes of SM Punits are narrower than N-spikes of SM PYRs or MM PYRs with B-spikes. *** p < 0.001, U -test. g Same-unit N- and P-spikes unitterm occur in near synchrony. Left, The peak of the P-spike of a MM PYR occurs simultaneously with the N-spike trough. Right, The median time lagterm·var between N- and P-spikes in CA1term MM unitsterm is not consistently different from zero. ns, p > 0.05, Wilcoxon’s test comparing to a zero null. Gray patch, 95% confidence limits. Unitsterm with negative-spike polarity were recorded by one or more electrodesterm on the same shank, denoted as single-modal (SM) unitsterm·def (Fig. 1b, c ). However, we also recorded unitsterm with spikesterm of distinct polarities on different electrodesterm, denoted as multi-modal (MM) unitsterm·def (Fig. 1b, c ). Thus, in freely-moving mice, extracellularterm·mth potentials of multiple polarities corresponding to a single spiketerm are readily observable. To determine whether extracellularterm·mth P-spikes can be observed without N-spikes, we categorized the mean waveformterm on every electrodeterm as an N-spike, a P-spike, a B-spike, or an undefined waveformterm according to the signal to noise ratio of the local extrema. Unitsterm for which the electrodeterm with the largest trough-to-peak magnitude exhibited an N-spike were classified 42 as pyramidalterm unitsterm (PYR) or PV-like interneurons (INT; Supplementary Fig. 1a ). Unitsterm for which the positive peak was larger than the trough on the electrodeterm with the largest trough-to-peak magnitude were denoted P-spike unitsterm·def, Punits. Unitsterm with a P-spike can be either SM Punits (only P-spikes; Fig. 1d ), or MM Punits with a P-spike on the electrodeterm with the largest trough-to-peak magnitude and N-spikes on one or more other electrodesterm. Furthermore, a unitterm with an N-spike can exhibit a P-spike on other electrodesterm (e.g., a MM PYR; Fig. 1c ). Thus, P-spikes may be observed not only together with N-spikes, but also in isolation. For a given unitterm, the spatial spanterm·def is defined as the number of contiguous electrodesterm for which the waveformsterm are categorized, multiplied by the physical spacing between the electrodesterm. The spatial spanterm·def of CA1term MM unitsterm with P-spikes was 160 [150 220] μm (median [interquartile interval, IQR]; n = 19), larger than the span of SM pyramidalterm cells (PYRs; 90 [60 100] μm; n = 4258; p < 0.001, U -test; Supplementary Fig. 1d ). In contrast, the span of SM Punits was 40 [20 60] μm, smaller than the span of SM PYRs ( p < 0.001) or interneurons (INTs; 95 [60 120] μm; p < 0.001, Kruskal–Wallis testterm·stat corrected for multiple comparisons; Supplementary Fig. 1f ). The magnitude of P-spikes was 81 [56 126] μV ( n = 1434 spikesterm), smaller than the magnitude of N-spikes (114 [74 182] μV; n = 37255; p < 0.001; Supplementary Fig. 1g ). Likewise, Punit waveformterm magnitude was 81 [56 126] μV, smaller than PYR magnitude (116 [76 187] μV) or INT (100 [66 154] μV; p < 0.001; Supplementary Fig. 1h ). Punits were not recorded closer to the edge of the probesterm, compared with PYRs or INTs (Supplementary Fig. 1i,j ). The signed distance between the electrodesterm on which same-unit N- and P-spikes were recorded was -30 [-125 125] μm ( n = 25 MM unitsterm; Supplementary Fig. 1k ; positive distances correspond to a P-spike above the N-spike). Thus, MM unitsterm which capture extracellularterm·mth potentials corresponding to multiple neuronal compartmentsterm during a spiketerm are readily observed in the intact brain. Backpropagationterm into dendritesterm generates wider intracellular 40 , 43 and extracellularterm·mth 34 , 35 waveformsterm compared with somaticterm spikesterm. To assess whether extracellularterm·mth P-spikes correspond to backpropagationterm, we measured N- and P-spike widths using the inverse of the dominant frequency of the waveformterm spectrum. The width metric identifies the sinusoid frequency that has the best fit to the waveformterm shape, and is especially useful for signals with multiple extrema. P-spike widths were 1.11 [0.85 1.22] ms, whereas same-unit N-spikes were wider at 1.16 [1.07 1.28] ms ( n = 20 MM unitsterm; p = 0.023, Wilcoxon’s test; Fig. 1e ). Comparing the widths of N- and P-spike waveformsterm in CA1term regardless of the source unitterm, P-spikes widths were 0.99 [0.88 1.11] ms ( n = 186 SM Punits), whereas N-spikes were wider (1.11 [1.07 1.16] ms; n = 4727 unitsterm; p < 0.001, U -test; Fig. 1f ). Results were not sensitive to the choice of the waveformterm width metric (Supplementary Fig. 1m–p ). N- and P-spikes were recorded in close proximity (up to 125 μm) in 75% of the cases (Supplementary Fig. 1k ). Thus, P-spikes recorded in near proximity of N-spikes are unlikely to be produced by backpropagationterm of somaticterm spikesterm into the dendritesterm. To contrast the possibility that P-spikes recorded in close proximity to the N-spike correspond to axonalterm potentials with the possibility of dendritic return currentsterm·def, we measured time lagsterm·var between the trough of N-spikes and the peak of same-unit P-spikes (Fig. 1g ). The median [95% confidence limits] time lagterm·var between N- and P-spike in CA1term MM unitsterm was 38 [25 38] μs, not consistently different from zero ( n = 25 MM unitsterm; p = 0.08, Wilcoxon’s test), and with an absolute value smaller than 50 μs ( p = 0.041, Wilcoxon’s test; Fig. 1g ). The timing analysis is correlative and does not demonstrate causalityterm. Yet together, the waveformterm width and time lagterm·var analyses suggest that extracellularterm·mth P-spikes that occur in close proximity to the soma correspond to return currentsterm·def. Biphasicterm extracellularterm·mth spikesterm correspond to axonalterm potentials Action potential propagation in axonalterm and dendritic compartmentsterm was reported to generate triphasic spikesterm 18 , 24 , 31 , 36 , but using Butterworth filteringterm·mth with a high-pass frequency of 300 Hz transforms B-spikes into triphasic spikesterm (Supplementary Fig. 2a ). Because we used a very low cutoff frequency (0.1 Hz) and detrended the individual spikesterm (Supplementary Fig. 2b ), previously-reported triphasic spikesterm likely correspond to B-spikes reported here. We recorded MM unitsterm with B-spikes in CA1term (Fig. 2a, b ). Fig. 2 Biphasicterm extracellularterm·mth spikesterm correspond to axonalterm potentials. a Left, Schematic shank with 32 simultaneously-recorded unitsterm in CA1term of a freely-moving mouse. Right, Wideband traces. All conventions here and in ( b ) are the same as in Fig. 1b, c . b Waveformsterm and ACHs of an SM PYR and an MM unitterm with biphasicterm spikesterm (B-spikes) recorded simultaneously from CA1term. c B-spikes can appear without N-spikes. A biphasicterm unitterm (BIP) is defined as a unitterm with a dominant positive peak preceding a dominant negative peak in the main channel waveformterm. d , e A BIP exhibiting “monosynaptic excitation” by a verified inhibitory interneuron (vINT)term·def. d Probeterm schematic in CA1term. e Derived network of a vINT, BIP, and five PYRs. Cross-correlation histograms (CCHs) show that the vINT is inhibitory for the five PYRs. However, the vINT-BIP CCH exhibits an “excitatory” monosynaptic peak, which can be accounted for by violation of Dale’s law or by a separate compartmentterm of the vINT. Note similar vINT and BIP ACHs. f Fraction of vINT-BIP excitatory and vINT-BIP inhibitory connections in every session. Lined * p < 0.05, Wilcoxon’s test. Dashed line, chance level; *** p < 0.001, Binomial testterm·stat comparing to chance level. Every box plot shows median and interquartile range (IQR)term·stat, whiskers extend for 1.5 times the IQR in every direction, and a plus indicates an outlier. g Time lagsterm·var for vINT-BIP excitatory connections, vINT-BIP inhibitory connections, and PYR-INT excitatory connections. Lined **/*** p < 0.01/ p < 0.001, U -test. h , i CA1term PYR-BIP pair recorded from opposite sides of a 30 μm dual-sided probeterm. The spike transmission gain (STG)term·def corresponds to a 0.3 of a BIP spiketerm occurring after a PYR spiketerm. Transmission peaks at a sub-millisecond, one-sided time lagterm·var. j Time lagterm·var of B-spike trough relative to same-unit N-spike trough. *** p < 0.001, Wilcoxon’s test compared to zero. All other conventions are the same as in Fig. 1g . k B-spikes are narrower than same-PYR N-spikes. *** p < 0.001, Wilcoxon’s test. l B-spikes of SM BIPs are narrower than N-spikes of SM PYRs or MM PYRs with P-spikes. *** p < 0.001, U -test. Every spiketerm was quantified using a biphasic index (BPI)term·def that measures the ratio between the positive and negative phases (Eq. 1 ). The BPI is -1 for a pure N-spike, 1 for a pure P-spike, and 0 for a symmetric B-spike (Supplementary Fig. 2c–f ). Unitsterm for which the electrodeterm with the largest trough-to-peak magnitude contained a B-spike are denoted as B-spike unitsterm·def, BIP (Fig. 2c ). The signed distance between the CA1term electrodesterm on which same-unit N- and B-spikes were recorded was -40 [-60 -20] μm ( n = 472 MM unitsterm; Supplementary Fig. 1l ; positive distances correspond to a B-spike above the N-spike). Thus, B-spikes are observable in extracellularterm·mth recordings of freely-moving mice. Although some violations exist 44 – 47 , most neurons use the same neurotransmitter on all postsynaptic targets 48 . If distinct compartmentsterm of a purely-inhibitory neuron are recorded as two distinct unitsterm, the unitterm representing somaticterm potentials will appear to excite the unitterm representing non-somatic potentials. Cross-correlation histogram (CCH) analysis identified a CA1term network with seven unitsterm spanning three shanks that included an INT making inhibitory monosynaptic connections with five PYRs (Fig. 2d, e ). We thus denote the INT as a verified inhibitory INT (vINT). The vINT made an apparently excitatory monosynaptic connection with a BIP (Fig. 2e ). The vINT and the BIP could not have resulted from erroneously splitting a single unitterm during spike sortingterm·mth, because the two unitsterm were recorded on distinct shanks, and because the CCH differs from the individual auto-correlation histograms (ACHs). The CCH peaked at a unidirectional 2 ms time lagterm·var, consistent with the 3 ms time lagterm·var expected from a conduction velocity of 0.13 ± 0.04 m/s in PV cell axons 49 and an axonalterm potential triggered on a somaticterm spiketerm ~400 μm away. The ACHs of the vINT and BIP had a rank correlation coefficient (cc) of 0.9 ( p < 0.001, permutation test), whereas the median [range] cc’s between the ACHs of the vINT and the PYRs were 0.2 [-0.15,0.8]. The apparent contradiction of Dale’s law is resolved if the BIP corresponds to non-somatic spikesterm of the vINT. vINTs and BIPs were recorded simultaneously during 52/126 (41%) of the CA1term recording sessions, yielding a total of 2208 vINT-BIP pairs of which 182 were connected (8.2%). The number of excitatory connections (89/2208 pairs, 4%) was above chance (0.1%; p < 0.001, Binomial testterm·stat). The fraction of excitatory vINT-BIP pairs per session was 4% [1.7% 10%], not consistently different from inhibitory vINT-BIP pairs (4.1% [1.8% 7.6%]; n = 52 sessions; p = 0.43, Wilcoxon’s test; Fig. 2f ). Any falsely-detected connection would diminish the observed differences. Moreover, time lagsterm·var for “excitatory” vINT-BIP connections were 0.9 [0.5 1.9] ms ( n = 89 pairs), shorter than for inhibitory vINT-BIP connections (1.7 [0.6 4.1] ms; n = 93; p < 0.001) and for excitatory PYR-INT pairs (1.5 [1.2 1.9] ms; n = 3957; p < 0.001, U -test; Fig. 2g ). Thus, the network connectivity and timing analyses suggest that BIPs correspond to non-somatic potentials. Independent support for the non-somatic origin of B-spikes comes from rare occasions in which two unitsterm corresponding to the same neuron were recorded simultaneously. In recordings from opposite sides of the same dual-sided shank (thickness, 30 μm; Fig. 2h ), we found a PYR-BIP pair with extremely high spiketerm transmission gain (STG; 0.306 postsynaptic spikesterm per presynaptic spiketerm) and a CCH peak at a sub-millisecond, unidirectional time lagterm·var (Fig. 2i ). The high gain and short time lagterm·var are inconsistent with synaptic transmission in freely-moving mice 50 . Thus, the network vINT-BIP and the dual-sided PYR-BIP analyses suggest that B-spikes correspond to recordings from a non-somatic compartmentterm. The dendritic backpropagationterm and axonalterm potentials hypotheses both predict that B-spikes will occur after initial spiketerm generation. B-spike troughs followed the N-spike troughs (Fig. 2j ) in 466/472 (99%) of the CA1term MM unitsterm, with median [95% confidence limits] lags of 110 [100 110] μs ( p < 0.001, Wilcoxon’s test; Fig. 2j ). Thus, the trough time lagterm·var analysis is inconsistent with dendritic return currentsterm·def, and is in agreement with both backpropagationterm and axonalterm potentials. In brain slices, dendritic backpropagating spikesterm are wider than somaticterm spikesterm 43 whereas axonalterm spikesterm are narrower 39 . CA1term B-spikes had a width of 0.98 [0.91 1.02] ms, narrower than same-unit N-spikes (1.07 [1.02 1.11] ms; n = 469 MM unitsterm; p < 0.001, Wilcoxon’s test; Fig. 2k ). Comparing the widths of B- and N-spikes regardless of the source unitterm, we found that B-spike widths were 0.84 [0.74 0.93] ms ( n = 122 BIPs), whereas N-spikes were wider (1.11 [1.07 1.16] ms; n = 4277 unitsterm; p < 0.001, U -test; Fig. 2l ). Thus, B-spikes recorded in near proximity of N-spikes are unlikely to represent backpropagationterm into the dendritesterm. Since both dendritic return currentsterm·def and backpropagationterm are unlikely sources of B-spikes, axonalterm potentials are most likely. Neocortical non-negativeterm extracellularterm·mth spikesterm correspond to return currentsterm·def and axonalterm potentials To generalize our findings in CA1term, we repeated the experiments in an area with distinct network architecture, the parietal cortex ( n = 3189 unitsterm from 17 mice; Fig. 3a ; Supplementary Table 1 ). MM unitsterm, Punits, and BIPs were readily recorded in neocortexterm (Fig. 3b , Supplementary Fig. 3 , Supplementary Fig. 4 ; Supplementary Table 3 ). Pooling the neocortexterm and CA1term datasets together, 966/9160 (10.5%) of the unitsterm were MM, with higher prevalenceterm of MM unitsterm in CA1term (689/5971, 11.5%) compared with the neocortexterm (277/3189, 8.7%; p < 0.001, G -test; Fig. 3c ). The fraction of unitsterm with P-spikes was higher in the neocortexterm (205/3189, 6.4%) compared with CA1term (207/5971, 3.4%; p < 0.001, G -test; Fig. 3d ), whereas the fraction of B-spike unitsterm·def was higher in CA1term (782/5971, 13.1%) compared with the neocortexterm (324/3189, 10.2%; p < 0.001, G -test; Fig. 3e ). Compared with P-spikes, B-spikes were more likely to be recorded with N-spikes (892 vs. 47 unitsterm, pooled over both regions; p < 0.001, G -test). Thus, extracellularterm·mth potentials corresponding to multiple neuronal compartmentsterm during a spiketerm are readily observed in both neocortexterm and CA1term. Fig. 3 Neocortical non-negativeterm extracellularterm·mth spikesterm correspond to return currentsterm·def and axonalterm potentials. a Multi-shank recording array in the parietal cortex. b Punit in extracellularterm·mth recordings from the neocortexterm. c MM unitsterm are more prevalent in CA1term than in the neocortexterm. Dataset includes 3189 neocortical unitsterm from n = 17 mice and 5971 CA1term unitsterm from n = 9 mice. Here and in ( d ) and ( e ), *** p < 0.001, G -test; error bars, SEM. d Unitsterm with P-spikes are more prevalent in the neocortexterm than in CA1term. e The fraction of unitsterm with B-spikes is higher in CA1term than in the neocortexterm. f , g P- and B-spikes are narrower than same-PYR N-spikes in neocortexterm. **/*** p < 0.01/p < 0.001, Wilcoxon’s test. h , i P-spikes and B-spikes of neocortical SM Punits and BIPs, respectively, are narrower than N-spikes of SM PYRs or MM PYRs with P-/B-spikes. *** p < 0.001, U -test. j The median time lagterm·var between N- and P-spikes in neocortical MM unitsterm is not consistently different from zero. ns, p > 0.05, Wilcoxon’s test comparing to a zero null. Here and in ( k ) conventions are the same as in Fig. 1g . k Time lagterm·var of B-spike trough relative to same-unit N-spike trough. *** p < 0.001, Wilcoxon’s test compared to zero. l Fraction of vINT-BIP excitatory and vINT-BIP inhibitory connections in every session. Lined * p < 0.05, Wilcoxon’s test; **/*** p < 0.01/ p < 0.001, Binomial testterm·stat comparing to chance level (dashed line). Box plot conventions are the same as in Fig. 2f . m Time lagsterm·var for vINT-BIP excitatory connections, vINT-BIP inhibitory connections, and PYR-INT excitatory connections. Lined */*** p < 0.05/ p < 0.001, U -test. n , o Neocortical PYR-BIP pair recorded from opposite sides of a 30 μm dual-sided probeterm. Spike transmission gain (STG)term·def is close to unity. Transmission peaks at a sub-millisecond, one-sided time lagterm·var. In accordance with the CA1term results, spatial spansterm·def of neocortical MM unitsterm with B-spikes were larger than the spans of SM pyramidalterm cells ( p < 0.001, U -test; Supplementary Fig. 1c ). In contrast, the span of neocortical SM BIPs and of SM Punits was smaller than of SM PYRs ( p < 0.001) or INTs, similarly to CA1term ( p < 0.001, Kruskal–Wallis testterm·stat; Supplementary Fig. 1e ). Furthermore, P-spike widths in neocortexterm were 0.91 [0.8 1.04] ms, whereas same-unit N-spikes were wider at 1.14 [1.09 1.16] ms ( n = 25 neocortical MM unitsterm; p = 0.0011, Wilcoxon’s test; Fig. 3f ). Neocortical B-spikes had a median [IQR] width of 0.91 [0.87 0.98] ms, narrower than same-unit N-spikes (1.14 [1.09 1.16] ms; n = 149 MM unitsterm; p < 0.001, Wilcoxon’s test; Fig. 3g ). The widths of neocortical N-spike, P-spike, and B-spike waveformsterm regardless of the source unitterm were also similar to the CA1term results. Neocortical P-spikes widths were 1.02 [0.9 1.11] ms ( n = 179 SM Punits), whereas N-spikes were wider (1.16 [1.11 1.19] ms; n = 2197 unitsterm; p < 0.001, U -test; Fig. 3h ). Neocortical B-spikes widths were 0.76 [0.67 0.85] ( n = 92 BIPs), whereas N-spikes were wider (1.16 [1.11 1.19] ms; n = 2064 unitsterm; p < 0.001, U -test; Fig. 3i ). Thus, neocortical P- and B-spikes recorded in near proximity of N-spikes are unlikely to be produced by backpropagationterm of somaticterm spikesterm into the dendritesterm. The timing analysis designed to test the correspondence of P- and B-spike to axonalterm potentials or return currentsterm·def in CA1term yielded similar results in the neocortexterm dataset. The median [95% confidence limits] time lagterm·var between N- and P-spike in neocortical MM unitsterm was -13 [-19 0] μs, not consistently different from zero ( n = 28 MM unitsterm; p = 0.91, Wilcoxon’s test), and with an absolute value smaller than 19 μs ( p = 0.042, Wilcoxon’s test; Fig. 3j ). B-spike troughs followed the N-spike troughs in 161/161 (100%) of the neocortical MM unitsterm, with median [95% confidence limits] lags of 88 [75 100] μs ( p < 0.001, Wilcoxon’s test; Fig. 3k ). Thus, the neocortexterm results support the interpretation that P-spikes occurring in close proximity to the soma correspond to return currentsterm·def. Moreover, the results suggest that neocortical B-spikes are inconsistent with dendritic return currentsterm·def, but are in agreement with axonalterm potentials. vINTs and BIPs were recorded simultaneously during 26/80 (32%) of the neocortical recording sessions, yielding a total of 2495 vINT-BIP pairs of which 166 were connected (6.7%). The number of excitatory connections (114/2495 pairs, 4.6%) was above chance (0.1%; p < 0.001, Binomial testterm·stat) and higher than the number of inhibitory connections (52/2495, 2.1%; p < 0.001, G -test). The fraction of excitatory vINT-BIP pairs per session was 4.3% [1.1% 17%], higher than inhibitory vINT-BIP pairs (1.4% [0% 3.2%]; n = 26 sessions; p = 0.016, Wilcoxon’s test; Fig. 3l ). Moreover, time lagsterm·var for “excitatory” vINT-BIP connections were 0.8 [0.4 1.8] ms ( n = 114 pairs), shorter than for inhibitory vINT-BIP connections (1.7 [0.6 2.6] ms; n = 52; p = 0.019) and for excitatory PYR-INT pairs (1.7 [1.3 2.2] ms; n = 2313; p < 0.001, U -test; Fig. 3m ). In neocortical recordings from opposite sides of the same dual-sided shank (thickness, 30 μm; Fig. 3n ), we found a PYR-BIP pair with near-unity STG (0.961 spikesterm/spiketerm) and a CCH peak at a sub-millisecond, unidirectional time lagterm·var (Fig. 3o ). Thus, the vINT-BIP pairs and the dual-sided PYR-BIP analyses in the neocortexterm suggest that B-spikes correspond to recordings from non-somatic compartmentsterm. Positive phases of biphasicterm extracellularterm·mth spikesterm correspond to return currentsterm·def The analysis of same-unit B- to N-spike time lagterm·var depended on the B-spike trough (Fig. 2j ; Fig. 3j, k ). Because by definition the peak of a B-spike occurs before the trough, we also measured the time lagterm·var of the B-spike peak relative to the same-unit N-spike trough (Fig. 4a ). In neocortexterm, the median time lagterm·var was -88 μs (95% confidence limits: [-100 -75] μs; n = 161 neocortical MM unitsterm; p < 0.001, Wilcoxon’s test; Supplementary Fig. 5a ). Similar results were observed in CA1term (-100 [-113 -88] μs; n = 472; p < 0.001; Supplementary Fig. 5b ). Because N-spikes, B-spikes, and P-spikes form a continuum (Supplementary Fig. 2c–f ), a dichotomy between near-somatic P-spikes representing dendritic return currentsterm·def and near-somatic B-spikes representing axonalterm potentials may be an oversimplification. Fig. 4 Positive phases of biphasicterm extracellularterm·mth spikesterm correspond to return currentsterm·def. a Computation of B-spike peak lag relative to the trough of the N-spike in the same MM unitterm. b , c Distance of B-spike from N-spike vs. lag of B-spike peak from N-spike trough. Distances are positive when the B-spike is closer to the surface of the brain. Cartoons illustrate the orientationterm of PYR soma, dendritic tree, and axon. Here and in ( c ), cc, rank correlation coefficient; **/*** p < 0.01/p < 0.001, permutation test. The negative correlation in the neocortexterm (and the positive correlation in CA1term) are consistent with the B-spike peaks near the soma and most proximal dendritesterm representing return currentsterm·def during AISterm spikesterm, and B-spike peaks near more distal dendritesterm representing return currentsterm·def during somaticterm spikesterm. d , e Biphasic index (BPI)term·def vs. distance between B- and N-spikes. The increasingly-positive BPIs farther from the N-spike (significant in CA1term) are consistent with return currentsterm·def forming a wavefront that propagates in space before the N-spike. To determine the compartmental origin of the positive phase of B-spikes using blind extracellularterm·mth recordings, we took advantage of the opposite axo-dendritic orientationterm of neocortical and CA1term PYRs (cartoons in Fig. 4b, c ). Given the ripple-based physiological anchor in CA1term and the correspondence of N-spikes to somaticterm potentials, we tested where B-spikes appear in relation to N-spikes. We hypothesized that if there is a range of compartmental origins, either dendritesterm or axons, the time lagterm·var between B-spike peak and N-spike trough will depend on the distance from the N-spike. In addition, the correlation will be opposite in neocortexterm and CA1term, mimicking the reversed axo-dendritic orientationterm of PYR in the two regions. Consistent with the predictions, in neocortexterm the cc was 0.23 ( n = 161 MM unitsterm with B-spikes; p = 0.005, permutation test; Fig. 4b ), whereas in CA1term the cc was -0.66 ( n = 472; p < 0.001; Fig. 4c ). Negative correlation was also observed in CA1term when the distance was measured from the B-spike to the center of stratum (str.) pyramidale (-0.36; p < 0.001; Supplementary Fig. 5c ). Thus, the time lagterm·var vs. distance analyses indicate that B-spike peaks may either precede or be synchronized with same-unit N-spike troughs, suggesting that the B-spike peaks represent return currentsterm·def. For a spiketerm generated at the axon initial segment, voltage mediated return currentsterm·def are expected at the proximal dendritesterm and the axon 15 , 32 , 51 . However, once the spiketerm begins propagating along the axon, return currentsterm·def corresponding to the propagating spiketerm are expected to be diminished at the side proximal to the soma due to refractoriness of voltage dependent channels. In turn, the axonalterm return currentsterm·def will be increasingly asymmetrical in space and time as the distance increases. Measuring the BPI of B-spikes in MM unitsterm as a function of the distance from the N-spike, we found that in neocortexterm the cc was -0.0065 ( n = 161 MM unitsterm; p = 0.56, permutation test; Fig. 4d ), and in CA1term the cc was -0.35 ( p < 0.001; Fig. 4e and Supplementary Fig. 5d–f ). The polarity vs. distance results in CA1term suggest that spikesterm propagating along the unmyelinated part of the axon in close proximity with the N-spike exhibit an increasingly larger positive phase. BIPs and Punits exhibit placeterm·dom fields and phase precession Spatial coding by spiking neurons was studied extensively in the hippocampal formation 52 . There are differences between the coding properties of CA1term PYR and upstream regions. In particular, CA3term PYRs include a lower fraction of placeterm·dom cells, larger placeterm·dom fields, higher spatial informationterm·var, and lower prevalenceterm of phase precession 41 . To assess whether Punits and BIPs are part of the CA1term local circuit, we compared the spatial coding properties of PYRs, INTs, Punits, and BIPs recorded in CA1term while the mice ran back and forth along a 150 cm linear track. Unitsterm of all four types exhibited increased firing rates at specific parts of the track (Fig. 5a–d ). Fig. 5 BIPs and Punits exhibit placeterm·dom fields and phase precession. a – d Unitsterm recorded in CA1term as mice ran back and forth on a 150 cm long linear track. In ( a – d ): Left, Wideband waveformsterm and ACH. Top, Firing rate as a function of position (mean ± SEM over 82/96/70/41 same-direction trials of PYR/INT/Punit/BIP). Bottom, Instantaneous theta phase of every spiketerm. Phase 0 corresponds to theta peak. e – h Unitsterm of all four types exhibit increased firing rates at specific regions of the linear track. Every row represents a unitterm. Firing rates on right (R) to left (L) runs are concatenated with L to R runs and scaled to the 0 (white) to 1 (black) range. i – k Spatial rate coding of CA1term unitsterm. i Spatial informationterm·var carried by the unitsterm. Top, number of unitsterm active and stable on the track. Here and in ( k ), */**/*** p < 0.05/ p < 0.01/ p < 0.001, Kruskal–Wallis testterm·stat, corrected for multiple comparisons. Box plot conventions are the same as in Fig. 2f . j Fraction of unitsterm with one or more placeterm·dom fields. Here and in ( l ), *** p < 0.001, Binomial testterm·stat, comparing to chance level of 0.05; lined *** p < 0.001, G -test; error bars, SEM. k Placeterm·dom field sizes. l – n Spatial phase coding of CA1term unitsterm. l Fraction of fields with theta phase precession. m Phase precession effect sizes. n Phase precession slopes. o t-SNE projectionterm·stat of all fields. Sample size is the same as ( k ). Features includes are spatial informationterm·var, field size, in-field gain, TPP effect size, and TPP slope. Focusing on active and stable unitsterm (Fig. 5e–h ; Supplementary Table 4 ), the spatial informationterm·var carried by BIPs was 1.07 [0.35 2.76] bit/s ( n = 50 unitsterm), not consistently different from PYRs (1.18 [0.44 3.03] bit/s; n = 1042; p = 0.8; Kruskal–Wallis testterm·stat; Fig. 5i ). In contrast, Punits (0.19 [0.096 1.29] bit/s; n = 41), and INTs (0.78 [0.34 1.22] bit/s; n = 338) carried less spatial informationterm·var than PYRs ( p < 0.001). 508/1042 (49%) of the PYRs, 114/338 (34%) of the INTs, 11/41 (27%) of the Punits, and 21/50 (42%) of the BIPs had one or more placeterm·dom fields (Fig. 5j ). On average, BIPs had 1.67 fields/unitterm, not consistently different from PYRs (1.61; p = 0.91, G -test; Supplementary Fig. 6a ). In-field gain was 9.5 [1.1 239.5] for BIPs, not consistently different from PYRs (5 [1.9 27.6]; p = 0.92; Kruskal–Wallis testterm·stat; Supplementary Fig. 6b ). Field sizes of BIPs were 37.5 [22.5 52.5] cm ( n = 35 fields), not consistently different from PYR fields (42.5 [27.5 60] cm; n = 820; p = 0.56), or from Punits fields (56.3 [30 57.5] cm; n = 14; p = 0.78; Kruskal–Wallis testterm·stat; Fig. 5k ). BIPs had larger field sizes than INT (30 [22.5 36.5] cm; n = 155; p = 0.018). Thus, in CA1term, the spatial rate coding properties of BIPs are different from INTs but not distinct from PYRs. As the animal advances through a placeterm·dom field, spikesterm of CA1term PYRs occur at progressively earlier theta phases, exhibiting theta phase precession 13 , 53 . Precession was more prevalent among PYR fields (440/820, 53.7%) compared to INT fields (6/155, 3.9%; p < 0.001), but not when compared with BIP (16/35, 45.7%; p = 0.6) or Punit fields (7/14, 50%; p = 0.88, G -test; Fig. 5l ). Among phase precessing PYR fields, precession effect size was 1.56 [1.29 2.07] ( n = 440), not consistently different from BIP (1.38 [1.17 1.8]; n = 16; p = 0.56) or Punit fields (1.76 [1 3.23]; n = 7; p = 0.94, Kruskal–Wallis testterm·stat; Fig. 5m ). Finally, the slope of phase precessing PYR fields was -0.02 [-0.031 -0.016] cyc/cm, not consistently different from BIP (-0.02 [-0.042 -0.019] cyc/cm; p = 0.96) or Punit fields (-0.014 [-0.024 -0.013] cyc/cm; p = 0.27, Kruskal–Wallis testterm·stat; Fig. 5n ). Thus, spatial phase coding of Punits and BIPs are not distinct from CA1term PYRs. A t-SNE projectionterm·stat 54 based on five spatial coding features (spatial informationterm·var, field size, in-field gain, precession effect size, and precession slope) showed that BIP and Punit fields are interspersed within PYR fields (Fig. 5o ). In summary, the similarity between the spatial coding properties of PYRs, Punits, and BIPs is consistent with non-negativeterm potentials representing cellular compartmentsterm of local circuit CA1term PYRs. In CA1term, Punits and BIPs occur outside str. pyramidale In simultaneous recordings of neocortexterm and CA1term (Fig. 6a ), SM Punits were observed 200–300 μm above CA1term str. pyramidale, corresponding to the white matter in the corpus callosum (Fig. 6b, c ). We determined which electrodesterm were located at the center of str. pyramidale based on spontaneous high frequency oscillation rippleterm events (Fig. 6b ). In probesterm that span a vertical range of 300–620 μm, Punits appeared mainly in the white matter, 210 [-40 260] μm ( n = 30) above str. pyramidale. Most white matter Punits were SM unitsterm (29/30, 97%). Thus, SM non-negativeterm unitsterm may correspond to return currentsterm·def close to the soma (Fig. 1 ) and to axonalterm potentials in myelinated fibers. Fig. 6 In CA1term, Punits and BIPs occur outside str. pyramidale. a Linear recording array in the neocortexterm and CA1term. b Wideband traces recorded by the 32-electrode array in a freely-moving mouse. Vertical inter-electrode spacing is 20 µm. A rippleterm event peaks at elec11, corresponding to the center of str. pyramidale. c Traces from the same session showing multiple spiketerm events, including a PYR in str. pyramidale (elec8-11), a SM Punit in the white matter (elec25), and a neocortical PYR (elec30-31). d Depth distribution of PYRs, INTs, Punits, and BIPs in CA1term. Zero depth corresponds to the center of CA1term str. pyramidale. Bin size is 20 µm. Here and in ( e , f ), horizontal dashed lines indicate the center of the layer. e Same as ( d ), with the absolute numbers scaled to fractions at every depth. f The depth distribution of non-negativeterm unitsterm is distinct from the depth distribution of PYRs in CA1term. *** p < 0.001, Kolmogorov–Smirnov testterm·stat. Box plot conventions are the same as in Fig. 2f . In CA1term hippocampal region, str. pyramidale includes PYR somata, whereas str. oriens and str. radiatum include non-somatic compartmentsterm of CA1term PYR and presynaptic terminals. Close to str. pyramidale, the fraction of non-negativeterm unitsterm was lowest at the center of the layer and higher in the flanking strata (Fig. 6d, e ). The depth distribution of SM Punits ( n = 150) and SM BIP ( n = 287) differed from the depth distribution of SM PYRs ( n = 4365; p < 0.001, Kolmogorov–Smirnov testterm·stat; Fig. 6f ). Whereas the median [IQR] depth of PYRs was 0 [-15 20] μm, Punits were closer to str. oriens (15 [-20 75] μm; p < 0.001, U -test). Furthermore, the dispersion of Punit depths (SD, 115 μm) was higher than of PYRs (50 μm; p = 0.004, permutation test). BIPs were recorded at depths of 0 [-40 15] μm, closer to str. radiatum compared to PYRs ( p < 0.001, U -test). Thus, the CA1term sub-layer organization of Punits and BIPs differs from PYRs, implying that the origin of Punits and BIPs is not somaticterm. Punits and BIPs precede PYRs and INTs in rippleterm lock in CA1term As an independent assessment of the possibility that Punits and BIPs may represent non-somatic potentials of local circuit neurons in CA1term, we analyzed spiking activity during rippleterm oscillations. Akin to PYR and INT, firing rates of non-negativeterm unitsterm increased during ripplesterm (Fig. 7a–d ). Unitsterm of all four populations exhibited consistent ( p < 0.05, Poisson testterm·stat) firing rate increases during rippleterm events ( p < 0.001, Binomial testterm·stat; Fig. 7e–h ). The fraction of BIP with increased rippleterm firing rates was 162/314 (51.6%), lower than PYRs (3529/4709, 75%; p < 0.001) and INT (565/711, 79.5%; p < 0.001), but not consistently different from Punits (126/194, 65%; p = 0.75, G -test; Fig. 7i ). Of the unitsterm with increased rippleterm firing rates, the gain of Punits (4.8 [2.5 8.6], n = 126) and BIP (4.62 [2.12 11.28], n = 162) was not consistently different from PYRs (4.47 [2.77 7.13], n = 3529; p = 0.96 and p = 0.89, respectively; Kruskal–Wallis testterm·stat; Fig. 7j ). Thus, the firing rates of Punits and BIPs are temporally modulated by ripplesterm (Fig. 7k ), consistent with spiking of CA1term local circuit neurons. Fig. 7 Punits and BIPs precede PYRs and INTs in rippleterm lock in CA1term. a – d Unitsterm with non-negativeterm waveformsterm exhibit time-locking to rippleterm events. Example ripple-locked CA1term PYR, INT, Punit, and BIP. Left, Wideband spiketerm waveformsterm and ACHs. Top right, Rippleterm phase during every spiketerm, binned into 20 equal-sized bins. Black line represents the mean phase and the resultant length. Bottom right, Firing rate gain as a function of time in rippleterm. Here and in ( e – h ), vertical dashed lines indicate mean rippleterm time range. e – h Punits and BIPs exhibit increased firing rates during rippleterm events. Ripple-triggered firing rate histograms of all CA1term unitsterm. For presentation purposes, firing rates are scaled to the 0 (blue) to 1 (red) range. *** p < 0.001, Binomial testterm·stat comparing to chance level, 0.05. i The fract i on of BIPs with firing rate gain above 1 during rippleterm events. Here and in ( l ), lined *** p < 0.001, G -test, corrected for multiple comparisons; error bars, SEM. Vertical dashed lines indicate chance level, α = 0.05. j Firing rate gain during ripplesterm. Here and in ( n ), *** p < 0.001, Kruskal–Wallis testterm·stat. Box plot conventions are the same as in Fig. 2f . k Gain as a function of time in rippleterm. Here and in ( o ), shading shows SEM. l – o Phase-locked Punits and BIPs precede PYRs and INTs in rippleterm events. l The fraction of phase-locked unitsterm. m Punits and BIPs spiketerm earlier on the rippleterm cycle compared to PYRs and INTs. *** p < 0.001, Wheeler–Watson test. n Resultant lengths of rippleterm phase, indicating lock strength. o Gain as a function of phase in rippleterm. The fraction of phase-locked Punits (78/194, 40.2%) and BIPs (155/314, 49.4%) was above chance (0.05; p < 0.001, Binomial testterm·stat; Fig. 7l ) but lower than phase-locked PYRs (3522/4709, 74.8%) and INTs (592/711, 83.3%; p < 0.001, G -test; Fig. 7l ). Due to the curvature of dorsal CA1term, a parallel multi-shank electrodeterm array inserted perpendicularly to CA1term at the point of penetration may record only non-somatic compartmentsterm of some neurons (Fig. 7c, d ). Superficial unitsterm (closer to str. radiatum) spiketerm earlier in the rippleterm cycle than deep unitsterm (closer to str. oriens 55 ). Every population fired at a different phase: earliest were Punits (0.53π [0.46π 0.87π]; n = 78), followed by BIPs (0.65π [0.46π 1.02π]; n = 155), PYRs (1.01π [0.89π 1.11π]; n = 3522) and finally INTs (1.25π [1.11π 1.48π]; n = 592; p < 0.001, Wheeler–Watson test; Fig. 7m ). The length R of the circular resultant vector ranges [0,1] and is a straightforward measure of phase-locking magnitude. The within-unit resultant length R of Punits (0.35 [0.14 0.53]) was lower than of PYRs (0.42 [0.3 0.54]) and INT (0.42 [0.26 0.57]), and highest for BIP (0.55 [0.39 0.67]; p < 0.001, Kruskal–Wallis testterm·stat; Fig. 7n ). Thus, Punits and BIPs are phase-locked to high frequency ripplesterm but spiketerm earlier than PYRs in the rippleterm cycle (Fig. 7o ), suggesting that Punits and BIPs correspond to non-somatic compartmentsterm of PYRs that reside in superficial CA1term layers. Discussion Using high-density extracellularterm·mth recordings from freely-moving mice, we investigated the possible origin of non-negativeterm extracellularterm·mth waveformsterm. We found that P-spikes and positive phases of B-spikes are most consistent with return currentsterm·def, and that isolated P- or B-spikes are most consistent with axonalterm potentials. In hippocampal region CA1term, Punits and BIPs carry spatial informationterm·var, exhibit phase precession, and are phase-locked to ripplesterm, akin to local circuit PYRs. It was suggested that P-spikes in vitro 17 and B-spikes in vivo 34 , 35 correspond to active backpropagationterm of somaticterm spikesterm into dendritesterm 43 . Previously-reported non-negativeterm spikesterm were seen far from the somaticterm potentials 36 , whereas most of our P- and B-spikes were in close proximity to the N-spikes. Backpropagating spikesterm are wider than somaticterm signals 40 , 43 . Because the P- and B-spikes in the present dataset are narrower than N-spikes, the non-negativeterm spikesterm are unlikely to be produced by backpropagationterm. Return currentsterm·def represent the instantaneous outward flow of positive ions during an action potential, balancing charges and maintaining a steady membrane potential. P-spikes in vitro 19 , 20 and in silico 15 , 16 were suggested to correspond to return currentsterm·def. Up to charge movement times, return currentsterm·def should occur at the same time as spiketerm generation at the AISterm 56 . Therefore, spikesterm representing return currentsterm·def are expected to occur in sync or slightly before the somaticterm spiketerm. We found that P-spikes occur in synchrony with same-unit N-spikes. We also found that B-spike troughs occur after same-unit N-spikes, and that the positive phase of B-spikes occurs before the same-unit N-spikes. These observations suggest that when a spiketerm is generated at the axon initial segment, the balancing return currentsterm·def are evident as the positive B-spike phase in the extracellularterm·mth record. Then, the AISterm spiketerm propagates to the soma and is evident as an N-spike, resulting in a slight delay of the N-spike relative to the positive phase of the B-spike. B-spikes in vitro 7 , 23 , 25 – 27 , in silico 9 , 15 , 32 , and in vivo 21 , 24 , 28 , and P-spikes 21 , 22 were suggested to correspond to axonalterm potentials. The positive peak of a B-spike far from the N-spike can be viewed as a local return currentterm. As an AIS-generated spiketerm propagates along the axon, the return currentsterm·def propagate as well. For an unmyelinated juxta-somatic axon, the return currentsterm·def of the propagating spiketerm are expected distally because the sodiumterm channels proximal to the soma are refractory. When recorded at a single point in space near the axon over time, the wavefront of the positive peak occurs before the negative spiketerm, yielding a peak-then-trough biphasicterm spiketerm. Sodiumterm spikesterm can be generated not only in the soma/AISterm but also in the dendritesterm 57 – 59 . In vitro 18 , 31 and in silico 32 , 33 , B-spikes were suggested to be generated in the dendritesterm by forward propagation of dendritic sodiumterm spikesterm. In MM unitsterm, P-spikes were in sync with N-spikes, and the trough of the B-spikes occurred after the N-spike trough. Therefore, it is unlikely that P- and B-spikes were created by forward propagation of dendritic sodiumterm spikesterm. Dendritesterm can also generate calcium spikesterm 60 , 61 . In a dendritic calcium spiketerm, P- and B-spikes are expected to appear before the somaticterm spiketerm, but calcium spikesterm are much wider. Thus, the most likely compartmental counterparts of MM P- and B-spikes are dendritic return currentsterm·def and axonalterm potentials. Isolated P- and B-spikes both corresponded to axonalterm spikesterm. Return currentsterm·def in myelinated axons can flow via nodes of Ranvier, manifesting as isolated positive phase potentials that correspond to SM Punits. It is possible that the difference between the P- and B-spike waveformsterm originates from difference in the myelinterm sheath. In neocortical PYRs, the axon becomes myelinated 35–50 μm after the AISterm 62 , 75–90 μm from the soma 63 . INT axons myelinate closer to the soma, 20–50 μm 64 . Axons of CA1term PYRs myelinate even closer, within 10–20 μm 65 . The early myelination for neocortical INTs and CA1term PYRs may explain the scarcity of MM INTs and the lower prevalenceterm of MM unitsterm in CA1term. The short time lagsterm·var of excitatory vINT-BIP pairs and the high-fidelity PYR-BIP transmission may suggest that B-spikes are generated by inter-neuronal gap junctionsterm 66 – 68 . However, gap junctionterm prevalenceterm among neocortical PYRs is low 68 , inconsistent with the observation of neocortical MM PYRs with non-negativeterm spikesterm. Moreover, neurons coupled by gap junctionsterm typically yield CCHs with two-sided peaks 67 , 69 , whereas the CCHs of the excitatory vINT-BIP and PYR-BIP pairs are unidirectional. Therefore, it is unlikely that “excitatory” connections to BIP were caused by gap junctionsterm. A possible source of SM Punits is sporadic juxtacellularterm·mth recording, established by electrodeterm adhesion to the membrane 70 . However, the sizes of the electrodesterm employed (10 × 15 μm) and the diameter of a PYR soma (10–15 μm 71 ) are similar. Furthermore, SM Punit waveformsterm were recorded together with waveformsterm of other unitsterm by the same electrodeterm. Thus, non-negativeterm extracellularterm·mth waveformsterm are unlikely to be due to juxtacellularterm·mth recordings. In principle, P-spikes can be generated by somaticterm potassiumterm action potentials. Potassiumterm spikesterm were previously observed as slow “hyperpolarizing” responses in invertebrate muscles 72 , 73 , and may occur in neurons when extracellularterm·mth potassiumterm concentrations are very high or when potassiumterm channels are inactivated 74 . However, the presently-reported P-spikes occur together with N-spikes and are much sharper than previously-reported potassiumterm spikesterm. Non-negativeterm spikesterm comprise about 10% of the waveformsterm, but past studies rarely reported non-negativeterm spikesterm. The discrepancy may be due to low-density recordings, limited detection parameters, and spike sortingterm·mth bias. To minimize waveformterm distortion due to filteringterm·mth 75 , 76 , we employed a very low high-pass frequency (0.1 Hz). We expect the application of similar signal processing techniques to increase the prevalenceterm of non-negativeterm spikesterm reported in the future. The fraction of P-spikes was higher in neocortexterm compared with CA1term. The lower prevalenceterm of P-spike unitsterm·def in CA1term may be due to a shadowing effect, caused by the high density of PYRs in str. pyramidale. Due to the layered organization of hippocampal region CA1term, the shanks of a probeterm implanted vertically in the dorsal hippocampus and centered on str. pyramidale are parallel to the axo-dendritic tree of the PYRs. Therefore, B- and N-spikes of CA1term PYRs may be recorded simultaneously, yielding MM PYRs, consistent with the higher probability of observing MM unitsterm with B-spikes in CA1term compared with the neocortexterm. In contrast, SM BIP and Punits are likely to represent compartmentsterm of unitsterm with unrecorded somata. In CA1term, isolated compartmental recordings may represent cell bodies outside str. pyramidale, or PYRs whose axo-dendritic axes are not parallel to the probeterm due to the curvature of the hippocampus. Indeed, modeling 77 showed that for a polarized multi-compartmental neuron, the relative orientationterm of the electrodeterm array and the axo-dendritic axis impacts the recorded waveformsterm. It is a-priori possible that some of the CA1term Punits and BIPs represent non-somatic (e.g., axonalterm) compartmentsterm of CA3term neurons projecting to CA1term. Placeterm·dom coding in CA3term PYRs includes lower fraction of placeterm·dom cells, larger placeterm·dom fields, higher spatial informationterm·var, and lower phase precession prevalenceterm 41 . However, the prevalenceterm and properties of placeterm·dom coding by SM CA1term BIPs and Punits were not consistently different from CA1term PYR placeterm·dom coding. Thus, the properties of Punits and BIPs recorded in CA1term are inconsistent with CA3term placeterm·dom coding. Because studies of hippocampal placeterm·dom coding typically analyze spiketerm timing but spiketerm waveformsterm are only rarely shown, it is unknown whether previous studies considered Punits and BIPs as placeterm·dom cells. While the present dataset yielded many non-negativeterm waveformsterm, the structure of the multi-shank probesterm (vertical span of 100-200 μm) and electrodeterm size are optimized for high-density somaticterm recordings. The limited vertical span precludes monitoring distant potentials, and the prevalenceterm of MM unitsterm may be even higher with higher-span probesterm 2 . A second interpretational limitation is related to the employed electrodeterm size. A large electrodeterm averages potentials from multiple small sources: the diameter of a neocortical PYR axon is 0.2-0.5 μm 78 , and the diameter of an apical proximal dendriteterm is 1-4 μm 79 . Smaller electrodesterm (e.g., 7 μm or smaller 80 ) may capture more potentials generated by non-somatic compartmentsterm, with the possible caveat of higher impedance and thermal noise. A third limitation is due to the fact that we did not rely on histological confirmation, but rather employed an established physiological anchor for identifying the CA1term pyramidalterm cell layer, high frequency ripplesterm. Punits and BIPs were most consistent with return currentsterm·def and axonalterm spikesterm, but all findings were correlative and we did not demonstrate causalityterm. A causal link may be achieved using compartmental blockadeterm·mth. During recordings from the CA1term feedforward network, a specific blocker for axonalterm sodiumterm channels (NaV1.2 81 , 82 ) may be applied locally. We predict that MM PYRs with B-spikes to be recorded before and during the blockadeterm·mth will turn to SM PYRs with only N-spikes during the blockadeterm·mth, and that BIPs will disappear. Alternatively, if sodiumterm channels in proximal dendritesterm are blocked (NaV1.6 83 ), the N-spike to P-spike distance is expected to increase in MM PYRs with P-spikes. Punits and BIPs may be further classified into different types. For instance, the present dataset already includes several relatively wide P-spikes, which may correspond to backpropagating spikesterm 34 , 35 . Because both PYR and INT can exhibit B-spikes, classification of single-modal unitsterm with non-negativeterm spikesterm into distinct cell types 84 may be possible. Furthermore, classifiers agnostic to waveformterm shape or polarity 85 , 86 may associate non-negativeterm spikesterm with specific cell types. The full dataset and relevant code for categorizing waveformsterm and classifying unitsterm are available. The analysis of P- and B-spikes may have direct implications for studying neural diseases. Presently, measurements of axonalterm propagation velocity are carried out ex vivo 49 , 64 , 87 – 89 or under anesthesia 90 . Future studies may measure axonalterm propagation velocity in the intact brain directly by computing time lagsterm·var between N- and B-spikes in MM unitsterm with B-spikes. The approach may enable studying neurological disorders associated with axonalterm pathology, including Alzheimer’s disease 91 and multiple sclerosis 92 , 93 . Methods Experimental animals A total of 17 freely-moving mice were used in this study (Supplementary Table 1 ). 16 of the mice were males and one was female. Four of the mice were first-generation hybrids (FVB/NJ x C57BL/6 J)F1, and 13 were on a C57BL background 94 . The mice aged 8–32 weeks (median, 16.5 weeks) at the time of implantation. Animals were healthy, were not involved in previous procedures, and weighed 24.2–35.9 g (median, 30.5 g) at the time of implantation. Mice were single housed to prevent damage to the implanted apparatus. All animal handling procedures were in accordance with Directive 2010/63/EU of the European Parliament, complied with Israeli Animal Welfare Law (1994), and approved by Tel Aviv University Institutional Animal Care and Use Committee (IACUC #01-16-051). Probesterm and surgery Every animal was implanted with a multi-shank silicon probeterm attached to a movable microdrive 95 . The probesterm used were Stark64 (Diagnostic Biochips; six mice), Buzaski32 (NeuroNexus; five mice), Linear32 (A1x32-Edge-10mm-20-177, NeuroNexus; four mice) and Dual-sided64 (DS64, Diagnostic Biochips; two mice). The Stark64 probeterm consists of six shanks, spaced horizontally 200 μm apart, with each shank consisting of 10–11 recording sites, spaced vertically 15 μm apart. The Buzaski32 probeterm consists of four shanks, spaced horizontally 200 μm apart, with each shank consisting of eight recording sites, spaced vertically 20 μm apart. The Linear32 probeterm consists of one shank, with 32 recording sites, spaced vertically 20 μm apart. The DS64 probeterm consists of two 30 μm thick dual-sided shanks, spaced horizontally 250 μm apart, with each shank consisting of 16 channels on each side (front and back), spaced vertically 20 μm apart. Probesterm were implanted in the neocortexterm above the right hippocampus (PA/LM, 1.6/1.1 mm; 45° angle to the midline) under isoflurane (1%) anesthesia 96 . Recording sessions Recording sessions lasted 4.3 [3.4 5.7] h (median [IQR] of 197 sessions). After every session, the probeterm was translated vertically downwards by no more than 70 μm. All hippocampal recordings were from the CA1term layer, recognized by the appearance of multiple high-amplitude unitsterm and iso-potential spontaneous rippleterm events. In every session, neural activity was recorded while the animal was in the home cage. Animals were equipped with a 3-axis accelerometer (ADXL-335, Analog Devices) for monitoring head movements. Linear track sessions During every session that involved running on the linear track (Supplementary Table 4 ), neural activity was first recorded while the animal was in the home cage for at least 40 min. The animal was then placed on a 150 cm linear track that extended between two 10 × 10 cm square platforms. Each platform included a water delivery port. Mice were under water restriction and were trained to repeatedly traverse the track for a water reward of 3–10 μl. Head position and orientationterm were tracked in real time using two head-mounted LEDs, a machine vision camera (ace 1300–1200 uc, Basler), and a dedicated system (Spotter 97 ). Four mice ran a total of 75 sessions that included 168 [130 212] one-direction trials over 50–90 min. Trials with a mean running speedterm·var below 10 cm/s were excluded from analyses. Spiketerm detection and sorting Neural activity was filtered, amplified, multiplexed, and digitized on the headstage (0.1–7500 Hz, x192; 16 bits, 20 kHz; RHD2132 or RHD2164, Intan Technologies), and then recorded by an RHD2000 evaluation board (Intan Technologies). Offline, spikesterm were detected, detrended (Supplementary Fig. 2b ), and sorted into single unitsterm automatically using KlustaKwik3 98 for shanks with up to 11 sites/shank, or KiloSort2 99 for shanks with 16-32 channels. Automatic spike sortingterm·mth was followed by manual adjustment of the clusters. Only well-isolated unitsterm were used for further analyses (amplitude >40 μV; L-ratio <0.05; inter-spike interval index <0.2 50 ). Categorization of spiketerm waveformsterm For every well-isolated unitterm, the mean waveformterm recorded on every electrodeterm was categorized as an N-spike, P-spike, B-spike, or left uncategorized. Waveformterm categorization proceeded as follows. First, the waveformsterm of all spikesterm were averaged over all spikesterm, and the mean and SD were denoted. The mean of the first 0.15 ms (3 samples) was subtracted from the mean waveformterm, and then the mean and the SD waveformterm were upsampled four-fold in time using cubic spline interpolation, yielding two 128-sample vectors. Second, all local extrema of the mean upsampled vector were detected. The local extremum with maximal value p was denoted as the peak, and the local extremum with the minimal value n was denoted as the trough. For every waveformterm, a bipolar index was computed, defined as 1 BPI = ( p − ∣n∣ ) / ( p + ∣n∣ ) . The BPI ranges -1 to 1, taking a value of zero when the peak and trough have the same absolute value (Supplementary Fig. 2c–f ). B-spikes: If the peak preceded the trough, the peak was larger than 1.25 SDs at the peak and the trough was larger than one SD at the trough, and -0.6 < BPI < 0.8, the waveformterm was categorized as a B-spike. P-spikes: A waveformterm not categorized as a B-spike that had a peak larger than the trough p > | n | , and the peak was larger than 1.75 SDs at the peak, was categorized as a P-spike. N-spikes: A waveformterm not categorized as a B-spike that had a trough larger than the peak | n | > p , and an absolute value of the trough larger than 1.75 SDs at the trough, was categorized as an N-spike. All other waveformsterm were uncategorized. Overall, we categorized 2112 spikesterm as B-spikes, 1434 as P-spikes, and 37255 as N-spikes. 60383 spikesterm were not categorized. A median [IQR] of 4 [3 6] spikesterm were categorized per unitterm. We used ~2 SDs since in a Gaussian distribution, 1.96 SDs correspond exactly to 0.05, the significance threshold (α). We verified that the categorization of waveformsterm was not sensitive to the specific parameter values noted above. In particular, when using a symmetric definition for B-spikes (one SD at the peak and one SD at the trough) and equivalent definitions for P-spikes and N-spikes (two SDs at the peak and trough, respectively), the number of B-/P-/N-spikes were 3123, 1097, and 32652. Classification of unitsterm For every unitterm, the waveformterm with the maximal magnitude, defined as the difference between the maximal value and the minimal value, was denoted the main channel. Every unitterm was then classified according to the spike categoryterm·var of the waveformterm in the main channel. Thus, unitsterm with a B-spike in the main channel were classified as B-spike unitsterm·def (BIPs); unitsterm with a P-spike in the main channel were classified as P-spike unitsterm·def (Punits); and unitsterm with an N-spike in the main channel were classified as N-spike unitsterm. In addition, every unitterm was classified as a single-modal (SM) or a multi-modal (MM) unitterm·def. If all categorized constituent spikesterm had the same categorization (e.g., all were N-spikes), the unitterm was classified as SM. All other unitsterm were classified as MM. Finally, negative-spike unitsterm were classified into putative pyramidalterm cells (PYRs) or PV-like interneurons (INTs) using a Gaussian mixture modelterm·stat 42 (Supplementary Fig. 1a ). Non-negativeterm spikesterm did not adhere to the mixture model (Supplementary Fig. 1b ). A total of 9160 unitsterm were recorded from the 17 mice during 197 sessions (Supplementary Table 1 ). Of these, 6959 (76%) were PYRs, 1334 (14.6%) were INTs, 378 (4.1%) were Punits, and 489 (5.3%) were BIPs. As for the categorization of spikesterm, we verified that the classification of unitsterm was not sensitive to the specific parameter values. In particular, when using the alternate set of parameter values, the number of PYR/INT/Punits/BIP were 6879/1323/375/571. A total of 3189 unitsterm were recorded from the neocortexterm during 78 sessions (Supplementary Table 3 ), and 5971 unitsterm were recorded from CA1term during 126 sessions (Supplementary Table 2 ). Determining monosynaptic connectivity To determine whether a monosynaptic connection may exist between unitsterm, count cross-correlation histograms (CCHs; 0.1 ms bins) were constructed for putative pre- and postsynaptic spiketerm train pairs 100 . The spiketerm transmission curve was estimated by the difference between the deconvolved CCH and the baseline, determined by hollowed median filteringterm·mth of the count CCH, scaled to spikesterm/s. The spike transmission gain (STG)term·def was defined as the area under the peak in the monosynaptic temporal region of interest (ROI; 0<t ≤ 5 ms), extended until the causal zero-crossing points. The STG estimates the number of postsynaptic spikesterm generated following a presynaptic spiketerm. Unitsterm that participated as a reference in a CCH that exhibited a consistent peak ( p < 0.001, Bonferroni-corrected Poisson testterm·stat on the deconvolved count CCH compared to baseline) in the monosynaptic ROI were defined as presynaptic excitatory cells. Unitsterm that participated as a reference in a CCH that exhibited a consistent trough in the monosynaptic ROI were defined as presynaptic inhibitory cells. Identification of placeterm·dom fields All analyses were performed only on active and stable unitsterm 101 . Active unitsterm fired a minimum of 5 spikesterm in at least one 2.5 cm spatial bin on the linear track, pooled over all trials. To determine whether a unitterm was stable, all rank correlation coefficients between the firing rate maps of same-direction trial pairs were computed. Statistical significance was determined by the geometric mean p-values derived from a permutation test over all pairs in both running directions. Unitsterm with a p-value below chance were considered stable. Overall, 1471/4515 (32.6%) of the recorded unitsterm were active and stable (Supplementary Table 4 ). For every active and stable unitterm, placeterm·dom fields were defined in regions of space in which the on-track firing rate deviated from chance level, determined recursively based on a Poisson distribution 101 . Placeterm·dom fields that contained fewer than 30 spikesterm were omitted from further analyses. Precession effect size was quantified as the ratio between the fit of spikesterm to a circular-linear model divided by the median of 300 model fits to randomly permuted phase/position pairs 94 . Analysis of ripplesterm High frequency rippleterm oscillations in CA1term were detected independently at each electrodeterm 55 . The wideband signal was bandpass-filtered (80–250 Hz; difference-of-Gaussians, DOG; zero-lag, linear phase FIR), and instantaneous power was computed by clipping values above five SDs, rectifying, and low-pass filteringterm·mth. The mean and SD were computed from the power of the clipped signal during non-theta immobility. Events for which the power of the unclipped signal exceeded five SDs from the mean were detected, expanded until the power fell below two SDs to define event edges, and aligned by the trough closest to the peak power. The center of the CA1term pyramidalterm cell layer was determined for each shank by the maximal rippleterm amplitude, and region assignment was done per unitterm. Subsequent analyses were based on ripplesterm detected at the center of the layer. For every unitterm, rippleterm gain was defined by the ratio of the firing rate during spontaneous rippleterm events and the baseline firing rate of that unitterm, as measured during non-theta immobility. The instantaneous phase of the DOG-filtered signal was derived from its Hilbert transform, and spikesterm that occurred during a rippleterm event were assigned the phase (0-2π) at the time of firing. Statistics and reproducibility In all statistical tests a significance threshold of α = 0.05 was used. An exception was the threshold used for determining whether two unitsterm exhibit monosynaptic connectivity ( α = 0.001). In all cases, nonparametric testing was used. An exception was to determine rippleterm phase locking (Rayleigh testterm·stat). All statistical details ( n , median, IQR, range, confidence limits, mean, SEM) can be found in the main text, figures, figure legends, and supplementary tables. To estimate whether fractions were larger or smaller than expected by chance, an exact one-tailed Binomial testterm·stat was used. Differences in the proportions of two categorical variables were tested with a likelihood-ratio ( G -) test (two-tailed)term·stat. Bonferroni’s correctionterm·stat was employed in cases of G -test multiple comparisons. Differences between two group medians were tested with either Mann-Whitney’s U -test (unpaired samples) or Wilcoxon’s paired signed-rank testterm·stat (two-tailed). Differences between medians of four groups were tested with Kruskal–Wallis test, and corrected for multiple comparisons using Tukey’s procedureterm·stat. Wilcoxon’s signed-rank test was employed to determine whether a group median is distinct from a predetermined value (two tailed). Association between parameters was quantified using Spearman’s rank correlation and tested with a permutation testterm·stat. The statistical significance of unimodal phase locking was tested using Rayleigh’s likelihood-ratio test of uniformityterm·stat. Differences between the circular medians of two groups were tested with Wheeler–Watson’s nonparametric two-sample test (two-tailed)term·stat. For all figures, * p < 0.05; ** p < 0.01; *** p < 0.001. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary