Single neurons detect spatiotemporal activity transitions through STP and EI imbalance

Reviewer #1 (Public review):

Summary:

This study uses optogenetics to activate CA3 while recordings from CA1 neurons and characterizing the excitation/inhibition (E/I) balance. They observe use-dependent alterations in the E/I balance as a result of STP and they develop a model to describe these observations. This is a very ambitious paper that deals with many issues using both experimental and modeling approaches.

Strengths:

This paper examines important principles regarding the manner in which synaptic circuitry and use-dependent synaptic plasticity can transform inputs and perform computations.

Weaknesses:

(1) I am pleased that the authors revised their section on external calcium They now approach this issue by calculating the effect of low external calcium on plasticity. These calculations are based on studies of numerous types of synapses, which are unlikely to be directly comparable to hippocampal synapses. They focus on PPR rather than the overall response during more sustained activation that is likely to produce more complex responses. Obviously, it would have been preferable to simply measure the responses in the appropriate external calcium.

(2) The illumination intensity and the effect on fiber stimulation is a limitation that will alter apparent plasticity.

Reviewer #3 (Public review):

Summary:

This work shows experimentally and computationally that single CA1 neurons can perform mismatch detection on patterned CA3 inputs and that STP and EI balance underlie this detection.

Strengths:

It has been known that STP can enhance the EPSP when the corresponding presynaptic input exhibits abrupt changes in firing rate. This work provides experimental evidence and further computational support for the hypothesis that the basic computation through STP is useful for detecting abrupt changes in the spatial pattern of synaptic inputs at the Schaffer collaterals. Further, their results indicate the novel view that mismatch detection is most efficient when gamma-frequency bursting inputs exhibit mismatches between theta cycles. The authors included novel results in the revised manuscript to show that the effective frequency range of gamma oscillation is broad, including both slow and fast gamma bands.

In the initial submission, the dependence of mismatch detection performance on model parameters and experimental settings, such as pattern overlaps and other network parameters, was not sufficiently explored. In the revised manuscript, the authors extensively studied these points and summarized the novel results in Fig. 9. Furthermore, the authors clarified that jitters in input spikes can improve detection performance in some cases. These results show the robustness of their results against variations in external and internal conditions.

Weaknesses:

While this study shows an intriguing example of combined experimental and computational studies, the results of experiments and those of the computational model display relatively large deviations in some conditions. The cause of this discrepancy needs further clarification.

Author response:

The following is the authors’ response to the previous reviews.

In this revision we have addressed all the reviewer comments, most particularly we acknowledge that our calculations and understanding of calcium levels were incorrect, and have revisited all sections of the paper affected by this. We have performed numerous additional analyses on data, rerun many calculations with low-Calcium versions of models, and have restructured several figures and text. In brief, here is an overview of the changes in this revision:

Changes regarding Calcium levels.

(1) We have referred to the literature for short-term plasticity effects over a range of concentrations for bath calcium, both in glutamatergic and GABAergic synapses in the hippocampus. By interpolation, the increase of PPR when going from 2 to 1.35 mM is below ~20%.

(2) We used 1.35 mM external Calcium in revised presynaptic neurotransmitter release models, and show that both glu and GABA models follow the same direction of change as seen in experiments at lower Calcium.

(3) We have rerun the key calculations using these updated synaptic models and show that mismatch detection remains intact, but its strength is diminished, at 1.35 mM calcium.

Changes regarding use-dependent decrease in CA3 field responses:

(4) We have performed additional analysis on recordings to show that the decline in CA3 field response over the 32 pulses of the mismatch trials is under 30%, and that the mismatch responses are present even when this decline has saturated.

(5) Additional simulations without ChR2 desensitization show no effect on mismatch selectivity.

Changes regarding analysis of jitter:

(6) We have performed extensive additional simulations to better characterize and analyze the effects of jitter on mismatch selectivity.

Changes regarding reliability of spiking:

(7) We have added a section referring to our previous work using a similar protocol (Bhatia et al., eLife 2019), where we see a range of spiking probabilities from recorded CA3 neurons. We also refer specifically to our previous analysis which set tight limits to the number of spikes per light pulse. We present our interpretation of how this affects mismatch detection.

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study uses optogenetics to activate CA3 while recordings from CA1 neurons and characterizing the excitation/inhibition (E/I) balance. They observe use-dependent alterations in the E/I balance as a result of STP and they develop a model to describe these observations. This is a very ambitious paper that deals with many issues using both experimental and modeling approaches.

Strengths:

This paper examines important principles regarding the manner in which synaptic circuitry and use-dependent synaptic plasticity can transform inputs and perform computations.

Weaknesses:

There are three issues that cause concern regarding the applicability of their slice recordings to physiological conditions and that make some aspects of their results difficult to interpret. First, they state that 2 mM added external calcium mimics calcium levels in CSF, but this is not the case. This will influence the plasticity they observe.

We agree that our calculations and understanding of calcium levels were incorrect. We have addressed this issue by referring to the literature in which external Ca is varied while measuring PPR.. From experiments by Blatow et al and Kravachenko et al we estimate that the PPR is ~20% higher at 1.35 mM external Ca than at 2mM, both for Glu and GABA synapses, but note that their experiments used a different set of conditions than ours. Based on this, we redid our synaptic plasticity models for Glu and GABA, reran our calculations and confirmed our core prediction of mismatch detection. We have added the following text to the introduction:

“While most studies on synaptic plasticity utilize 2mM external Ca2+ (extCa) this is substantially higher than physiological Ca2+ (1.2 to 1.5 mM) and has been shown to significantly alter plasticity outcomes. Long-term plasticity is particularly sensitive to elevated extCa levels (Inglebert et al. 2020). There are significant but smaller changes in short-term plasticity in several systems, wherein lower extCa leads to a) higher paired pulse ratio (PPR) and b) lower EPSCs (Rat hippocampal culture: (Sippy et al. 2003), Calyx of Held: (Thanawala and Regehr 2013; Bernal-Correa et al. 2026; Lorteije et al. 2009), Mouse hippocampal slice: (Blatow et al. 2003)). Specifically, in hippocampal CA3->CA1 glutamatergic synapses at 1.5 mM extCa the paired-pulse ratio is 15% larger than at 2mM (Blatow et al. 2003). In primary hippocampal neuronal culture, inhibitory synapses also undergo a 50% increase in paired-pulse ratio at 0.5 mM compared to 2 mM extCa (Kravchenko et al. 2006). By linear extrapolation/interpolation both these results suggest a PPR increase of ~20% at 1.35 mM extCa compared to the conventional 2 mM. In the current study we extend our simulations into the physiological domain of extCa to check consistency of the predicted computational outcomes.”

Also, from the methods:

“Model tuning for physiological conditions

We implemented two sets of changes to build a distinct presynaptic model which more closely matched physiological conditions. First, we increased the rates of all the reaction steps in the E and I presynaptic model by 40%, with the assumption of a 3.5℃ rise in temperature and a doubling of rates every 10℃ (Q10 of 2). Second, we lowered the calcium influx during stimulus by a factor of extCa/defaultCa = 1.35mM/2.0mM, again both for E and I. This effectively halves synaptic release (Figure 4-figure supplement 8, compare panels A and B), which we compensated for by a simple doubling of synaptic weight, both for E and I.

It also causes an increase in STP levels compared to the reference 2.0 mM extCa model. We confirmed that these changes were qualitatively similar to those expected from experiments over a range of extCa levels (Kravchenko et al. 2006; Blatow et al. 2003). However, several features in these earlier experiments differed from ours: the inter-pulse intervals, the preparation, and the temperature. Accordingly, rather than paste in their numbers of ~20% increase in STP in E and I, we used the 37℃ and 1.35mM extCa variant of our model for additional simulations.”

Second, they indicate that there is a 2% decrease in activated fibers per stimulus and attribute this to ChR2 desensitization. Such use-dependent decreases in fiber activation are expected to build during their repetitive activation experiments and artifactually influence their results.

The reviewer correctly points out that there would be a substantial decrease in fiber activation if the 2% decrease per pulse extended linearly over the entire train of ~200 pulses in the Poisson train or 32 pulses in the mismatch experiments/models. Clearly this does not happen, as the EPSPs settle rapidly to a steady range in both cases (Figure 6A, E, Figure 7 E). We acknowledge that the original statement implied a large decrease, and have added the clarificatory line around Figure 1:

“We attribute this to ChR2 desensitization, however in longer pulse trains this levels out as we discuss below.”

To quantitatively address the concern about decline of activation, we have performed an additional analysis to estimate how fiber activation evolves over multiple pulses. For the Poisson train experiments, we used our measured field potentials over the entire train (Figure 6-figure supplement 2) and saw that the level flattened out after ~2 seconds (~40 pulses). For the mismatch experiments, the mean fEPSP flattens with a time course of ~0.8 seconds and even at 50Hz the decrement is only about 30% of the initial value (Figure 7-Figure Supplement 2). Thus the field readouts show that the saturating level of use-dependent decrease is small. From the text around Figure 7:

“We observed this mismatch selectivity despite the reduction in inputs due to ChR2 desensitization over the long 32-pulse train. To estimate this reduction, we measured the field EPSP in the CA3 during the optical stimulation, and fitted an exponential decay function to the peaks (Figure 7-figure supplement 2). We find that the largest decay was ~30% of the initial fEPSP. The fEPSP decayed over a wide range of time courses, with median around 800ms. We estimate that the recovery time from ChR2 desensitization was around 150ms, again highly different between slices (Figure 7-figure supplement 2). Thus the mismatch detection was largely robust to ChR2 desensitization effects over the pulse train.”

Thus, we do not expect any qualitative effect of ChR2 desensitization on our results for the following reasons. First, the mismatch experiments themselves take place under the conditions of this fibre activation decay, yet yield numerous cases of mismatch selectivity (Figure 7Eiii, Fi). Second, the simulations for mismatch (Figure 7,9) explicitly included ChR2 desensitization, and they too exhibit mismatch selectivity. Third, in this revision we have added a simulation for mismatch selectivity in the absence of ChR2 desensitization (Figure 9 panel E). There is no statistically significant difference from the reference model with ChR2 desensitization.

Third, they do not know the responses of individual CA3 cells to stimulation. They do not know if each cell fires reliably during repetitive activation and whether each cell only fires once.

We acknowledge that we collected limited data on this in the current dataset, because this analysis has already been published. In our previous work using a similar preparation we have characterized CA3 pyramidal neuron responses to grid optical stimuli in the CA3 (Figure 1b, e, g, Bhatia, Moza and Bhalla 2019) in response to the light pulse. These recordings show that cells fire with probabilities over a wide range from zero to one, but these firing probabilities are reliable over multiple trials. In this revision we have clarified this point:

“Notably, the distribution of field amplitudes was tight (Figure 1E, Figure 1-figure supplement 2F), more so than the distribution of the corresponding EPSPs (Figure 1H). Together with previous work using a similar optical stimulus system and intracellular CA3 recordings (Bhatia et al. 2019), we interpret this to say that the spiking responses from CA3 neurons to optical stimuli were consistent from trial to trial.”

Regarding number of action potentials: In the previous revision we had examined field data from our recordings to characterize the occurrence of multiple spikes (Figure 1-figure supplement 2). Second field peaks were common, at about 40% of the amplitude of the first, and within ~3.5 ms of the first peak. Third peaks were much smaller. From the text adjacent to Figure 1:

We observed a small amount of ‘ringing’ of the field response which we interpret as either CA3 spiking in a burst, or recurrent activation of the CA3 neurons (Figure 1-figure supplement 2). The second peak was ~40% of the first, within 4 ms of the first. The ringing was down to ~5% by the third peak which occurred within ~9 ms of the flash. This sets an upper bound to any contribution to patterns by recurrence or multiple spikes during a ChR2 flash.

On the basis of these field responses and previous reports (Bhatia et al. 2019), we infer that the CA3 neuron response to optical input was a brief event, with reliable probability of a spike on any given flash, and typically having one or occasionally two spikes within 6 ms of the light flash.

Reviewer #3 (Public review):

Summary:

This work shows experimentally and computationally that single CA1 neurons can perform mismatch detection on patterned CA3 inputs and that STP and EI balance underlie this detection.

Strengths:

It has been known that STP can enhance the EPSP when the corresponding presynaptic input exhibits abrupt changes in firing rate. This work provides experimental evidence and further computational support for the hypothesis that the basic computation through STP is useful for detecting abrupt changes in the spatial pattern of synaptic inputs at the Schaffer collaterals. Further, their results indicate the novel view that mismatch detection is most efficient when gamma-frequency bursting inputs exhibit mismatches between theta cycles. The authors included novel results in the revised manuscript to show that the effective frequency range of gamma oscillation is broad, including both slow and fast gamma bands.

In the initial submission, the dependence of mismatch detection performance on model parameters and experimental settings, such as pattern overlaps and other network parameters, was not sufficiently explored. In the revised manuscript, the authors extensively studied these points and summarized the novel results in Fig. 9. Furthermore, the authors clarified that jitters in input spikes can improve detection performance in some cases. These results show the robustness of their results against variations in external and internal conditions.

Weaknesses:

While this study shows an intriguing example of combined experimental and computational studies, some analytic results, for instance, regarding the complex contributions of jitters to detection performance, could have clarified the underlying mechanism deeper and further strengthened the manuscript.

We acknowledge that the outcome of the jitter calculations looked complex. We redid this with substantially more repeats (30 simulated networks projecting to a CA1 neuron, rather than the original 6) and performed a cell-wise bootstrap analysis for error estimates for mean selectivity for each condition. This larger sample resulted in a smoother distribution of mismatch selectivity over different frequencies (Figure 7). We have carried out a further set of simulations over a range of jitter widths (Figure 7-figure supplement 3). We have added these findings to the paper:

“We next asked how sensitive the mismatch detection was to spike jitter, since in-vivo neural activity has a distribution of spike timings. We reran our model with 30 distinct simulated networks projecting to a CA1 neuron, each for a range of jitter values from 0 to 10ms. We performed a cell-wise bootstrap analysis to obtain error estimates for mean selectivity for each condition. We find that for 5-square stimuli selectivity is high at 0 jitter for all frequencies, but surprisingly for 15-square, selectivity is higher when jitter is nonzero (Figure 7-figure supplement 3). We suggest that when synaptic inputs are synchronous their effects add sublinearly, so some degree of jitter causes larger EPSPs when there are many inputs, as in the 15-square case. Consistent with this, the centroid for selectivity was always at a higher jitter level for 15-square than for 5-square (Figure 7-figure supplement 3, bootstrap test, p < 0.001).

For the 50-Hz stimulus, a 10ms jitter is half the stimulus interval, so the onset of patterns are quite smeared. Thus, as expected, selectivity fell both in 5 (bootstrap p<0.001) and 15-square (bootstrap p=0.005) cases for 50 Hz when jitter was larger than 4 ms. (Figure 7-Figure supplement 3).”

While we appreciate that a further analytic examination of jitter would be interesting, we feel that it would have to be done on a much reduced system compared to ours, and is therefore outside the scope of the current paper.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

The authors primarily responded with words rather than experiments. Their response has been adequate for some issues, but I remain concerned about their methods and how they could impact their own results, and how they might lead others astray.

(1) The paper by Klyachko and Stevens (2006) does a good job of addressing the temperature issue.

(2) The issue of free calcium is important. They provide a calculation that says that 2 mM added calcium corresponds to 1.27 mM free calcium. This seems very low based on the low affinity buffers in their solution. I used a standard and widely used tool (MINTEQ) to calculate the free calcium in their external solution and it says that the levels are 1.6 mM free calcium. It should be noted that most reports of calcium in CSF are for total calcium rather than free, and the total values are in the 1.1-1.4 range. If buffering in CSF is comparable to ACSF than this corresponds to 0.88 to 1.12 free calcium. So either I am wrong (and MINTEQ), or the authors are wrong. If they are incorrect then it makes a big difference for their experiments. It would also be a very bad thing for slice electrophysiology to get this wrong. Also see the paper by Borst at the calyx of Held that says that 2 mM added external calcium typically used in ACSF (as used in the present study) is too high (PMID 2394036). For these reasons I am highly skeptical that their calcium levels match those in vivo.

We agree that our calculations and understanding of calcium levels were incorrect. We used only two of the constituents of our buffer solution for our calculation of free calcium, which gave our previous low estimate of 1.27mM, whereas MINTEQ is much more complete. We redid our calculations using MINTEQ and obtained the same values as the reviewer, and we thank the reviewer for pointing out this resource. We also looked at a broader range of studies including those suggested by the reviewer and agree that the literature has flagged 2mM total calcium as too high. We have added a new section to the Introduction to reflect this body of work. We have removed our earlier free calcium calculations.

Please see our detailed response to this point above, including reference to the literature and re-running of many simulations at physiological levels of calcium, to examine the effect on our conclusions. In brief, the mismatch selectivity is weakened at physiological calcium, but is still present.

(3) The issue of showing a 2% per pulse decrease that they attribute to ChR2 desensitization is important. This would be sufficient to have a major influence on their experiments in which they use repetitive activation. What is the time course of recovery of this inability to stimulate all of the fibers?

Please see our response to the same point above. In brief, we have looked at the experimental data for field epsp decline over extended stimuli, and have shown that it settles at about 70% of the starting level in the case with the steepest decline (Figure 7-figure supplement 2). The time-course of decline is about 0.8 seconds. From the frequency dependence of the decline, we estimate that the recovery time is around 150 ms (Figure 7 -figure supplement 2). We also point out that our experimental data as well as our simulations of mismatch selectivity were carried out in the presence of ChR2 desensitization, yet we observed clear mismatch responses.

Can they get around this problem by lowering the stimulus intensity?

From Mattis et al Figure 2a, ChR2 performance over pulse trains deteriorates with decreasing power. Hence we do not expect that lowering the stimulus intensity would help.

From the methods:

“We chose short light pulses for two reasons. First, to decrease jitter in the light-triggered CA3 neuron spike timings. Second, if one is below saturating light intensities, short pulses at higher power give more consistent spiking over multi-pulse trains (Mattis et al. 2012), Figure 2A, using a 2ms pulse width). Note that the Mattis study was done at 20-23℃ as compared to our ~33℃, and the authors state that ChR2 kinetics are strongly temperature dependent.”

Are their results altered if they lower the illumination intensity by a factor of 2 or increase it by a factor of 2?

We did not try this, as we were aiming for short light pulses to minimize spike timing jitter, and we needed high intensity stimulation to get strong EPSPs on the postsynaptic cells. Based on the Mattis study we expect that yes, the results would be altered, and we would get more reliable spiking at still higher intensities. However, we used 100% intensity (estimated 100mW/mm²) in our optical system, so we could not push it further.

We did perform one manipulation to mitigate desensitization: For the 1-square illumination we sequenced illumination from one spot to another, before returning to the first spot, so as to minimize desensitization. From the methods:

“In 1-square stimulation experiments, the CA3 cell layer was illuminated one spot at a time with an interstimulus time of 3 seconds to avoid desensitisation of channelrhodopsin.”

(4) They base their reliability of stimulation on field recordings. They need to do extracellular measurements from CA3 cells to see how reliably each cell fires and if they only fire once or more that once during stimulation.

We used prior published work reporting intracellular recordings from CA3 cells, and additional analysis of field data to address this point. In brief, the probability of a cell firing is reliable, but on individual trials it might not fire. Additionally, we estimate that following a light pulse, cells fire twice about 40% of the time but rarely 3 or more times. We provide a more detailed response to this concern above.

If each cell does not fire reliably once and only once for each stimulus pulse it will alter use-dependent plasticity and call into question the whole result.

We agree that, at the single synapse level, an accurate account of use-dependent plasticity depends on single reliable spikes. Our recordings from CA1 are a sum over many synapses, and hence we use population CA3 activity measures to characterize our inputs, with the stipulation that these are reliable over multiple trials. Bhatia et al, 2019 showed reliability at the action-potential level, and in the current study we show it at the field potential level (Figure 1-figure supplement 2 panel F). We use our models to bridge the gap between the single-synapse responses to individual trials, to whole-cell readouts (Figures 5, 6, 7), and apply several tests to show that this produces models that fit the data (Table 1).

Our results, including EI balance and mismatch selectivity, are direct measurements in the presence of distributions of input APs. We point out in the discussion that:

“In-vivo input activity to this circuit is complex, and on successive stimuli there will be a distribution of exact CA3 pyramidal cell AP timings, number of APs per cell, and number of activated axons. To the extent that model robustness (Figure 9, Figure 7-figure supplement 3) gives insights into the real system we suggest that our small model of network + single neuron model + presynaptic signaling chemistry captures some of the attributes of the biological system, which lead to such robustness. This robustness is almost a necessity in a system where the probability of release for individual synapses is in the range of 20% (Murthy et al 1997): any signal transmission or plasticity-type computation has to accommodate huge stochastic uncertainty on any given trial.”

I remain highly skeptical of their methods and their experimental results.

We hope that we have addressed the specific points of concern that have led to this scepticism.