Recent applications of genetically encoded calcium indicators and high-density silicon electrodes have revolutionized our understanding of the cellular and circuit basis of behavior; however, technological barriers preclude similar exploration of the molecular basis of these processes in vivo. To investigate the physiological function of complex molecular systems in vivo, we require techniques to visualize endogenous proteins. Modern proteomic and transcriptomic methods provide biologists with myriad candidate proteins, but in many cases, there are no tools available to effectively study these targets at the level of endogenous proteins in vivo. For example, we are far from having reliable antibodies for the entire proteome, and even when antibodies are available, there are concerns regarding their target specificity. Another approach is to fluorescently tag proteins to visualize their dynamic expression in living tissue. Combined with in vivo two-photon (2p) microscopy, this approach enables detailed investigation of the molecular mechanisms underlying complex physiological and pathological systems.

AMPA-type glutamate receptors (AMPARs) are crucial molecules to study to understand the function and dynamics of the nervous system. AMPARs mediate the majority of fast excitatory synaptic transmission in the mammalian brain, and their regulation is regarded as a key mechanism underlying long-lasting changes in synaptic efficacy that give rise to learning and memory (Huganir and Nicoll, 2013; Malinow and Malenka, 2002). Long-term potentiation (LTP) is characterized by increased AMPAR trafficking to the postsynaptic membrane and associated spine enlargement, which together result in a long-lasting increase in synaptic efficacy, whereas long-term depression (LTD) is characterized by removal of postsynaptic AMPARs, resulting in attenuated synaptic transmission (Anggono and Huganir, 2012; Nicoll, 2017). Impaired regulation of synaptic plasticity is associated with human neurological and psychiatric disease (Berryer et al., 2013; Henley and Wilkinson, 2016; Volk et al., 2015). Despite this clear link between synaptic plasticity and learning, as well as a thorough understanding of molecular mechanisms regulating AMPAR trafficking, very little is known regarding how changes in plasticity are distributed among trillions of synapses throughout the brain. In contrast to advanced strategies to observe and manipulate neuronal activity – using genetically encoded calcium indicators (Dombeck et al., 2010; Lin and Schnitzer, 2016; Xu et al., 2012) or optogenetics (Fenno et al., 2011), respectively – there are currently no methods to physiologically measure postsynaptic strength in vivo on a brain-wide scale.

To overcome this barrier, we developed a new knockin mouse line wherein the AMPAR GluA1 subunit is tagged with super ecliptic pHluorin (SEP), a pH-sensitive variant of GFP that fluoresces at neutral pH and is quenched at acidic pH (Miesenböck et al., 1998). When coupled to the extracellular N-terminal domain of the AMPAR, this SEP tag reports the concentration of functional receptors at the cell surface, as the fluorescence of receptors localized in acidic, internal compartments such as endosomes and Golgi is quenched. Our genetic labeling strategy also avoids confounds arising from manipulation of the AMPAR C-terminus, a region important for proper function and trafficking to the postsynaptic membrane (Sheng et al., 2018; Zhou et al., 2018). Many groups have used overexpression of SEP-tagged AMPARs in neuronal culture to study AMPAR trafficking in vitro (Araki et al., 2015; Ashby et al., 2004; Kopec et al., 2006; Makino and Malinow, 2009; Patterson et al., 2010; Roth et al., 2017). In addition, previous work using overexpression of SEP-tagged AMPARs in vivo has provided valuable insights regarding the molecular mechanisms of behaviorally relevant plasticity (Diering et al., 2017; El-Boustani et al., 2018; Makino and Malinow, 2009; Miyamoto et al., 2021; Roth et al., 2020; Suresh and Dunaevsky, 2017; Tan et al., 2020; Zhang et al., 2015), but these methods enable receptor visualization in only a sparse subset of cells and exogenous overexpression may result in protein mistargeting and dysregulation. The novel genetic labeling strategy presented here avoids these confounds, allowing visualization of endogenous AMPAR expression in a manner that does not impair synaptic function, plasticity, or behavior. Used in conjunction with in vivo 2p microscopy, this novel SEP-GluA1 knockin mouse is the first tool that enables longitudinal tracking of synaptic plasticity underlying behavior at brain-wide scale with single-synapse resolution. Finally, we present a suite of algorithms to automatically detect and segment hundreds of thousands of fluorescently labeled AMPARs in vivo, enabling longitudinal tracking of synaptic plasticity across entire brain regions in awake behaving mice.

A central goal of neuroscience is to understand high-order cognitive functions in terms of their constituent components. Over the past 50 years, we have learned a great deal regarding the general role of synaptic plasticity in learning and memory. While seminal experiments have clearly implicated regulation of AMPARs as a central mechanism to modify the strength of synaptic communication between neurons (Andersen et al., 1977; Frey and Morris, 1997; Huganir and Nicoll, 2013; Malinow and Malenka, 2002; McNaughton et al., 1978), there is currently a dearth of methodologies to investigate how these molecular dynamics are distributed within vast networks of billions of synapses throughout the brain. For instance, patch-clamp recordings provide excellent spatial and temporal resolution, enabling investigation of how integration of specific synaptic inputs is dynamically tuned by plasticity (Spruston, 2008), but the scale of these recordings is limited to single neurons and performing them in behaving animals is challenging (Bittner et al., 2017; Epsztein et al., 2011). In vivo calcium imaging and high-channel-count electrophysiology offer superb spatial coverage, enabling investigation of neuronal activity within circuits of hundreds of neurons during behavior (Juavinett et al., 2019; Jun et al., 2017; Sofroniew et al., 2016), but these techniques lack the spatial resolution to study plasticity of individual synapses. In vivo structural imaging has provided valuable insights regarding how spine formation and elimination has been shown to contribute to neuronal development and synaptic plasticity (Bhatt et al., 2009; Holtmaat and Svoboda, 2009; Trachtenberg et al., 2002; Xu et al., 2009; Yang et al., 2009), though this method does not measure plasticity at existing spines. In addition, spine size has been reported to be proportional to synaptic strength (Matsuzaki et al., 2001), but this is at best an indirect readout of synaptic strength, and in certain conditions spine size and synapse strength are completely dissociated (Lee et al., 2012; Sdrulla and Linden, 2007; Tan et al., 2020; Zhang et al., 2015). Thus, our understanding of how the brain represents learning, memory, and behavior is constrained by currently available methodologies.

Here, we present a suite of novel tools and approaches that break through these constraints, enabling visualization of synaptic plasticity with molecular resolution at brain-wide scale in living animals. At the heart of this suite lies the newly generated SEP-GluA1 knockin mouse, a novel line that fluorescently labels all endogenous GluA1-containing AMPARs throughout the entire brain. This line enables direct investigation of the molecular dynamics underlying synaptic plasticity at any scale, from super-resolution synaptic imaging in primary cultures to circuit-level analyses of plasticity in acute slices to brain region-wide imaging of synaptic strength in behaving animals. The sequence linking the SEP tag and AMPAR N-terminus is known to affect protein expression and proper postsynaptic targeting of the receptor as previous attempts to fluorescently tag AMPARs at the N-terminus have been reported to result in impaired synaptic function (Díaz-Alonso and Nicoll, 2021; Díaz-Alonso et al., 2017). However, in our experiments, N-terminal linkers were lengthened and optimized to increase the flexibility of the SEP tag, thereby limiting disruption of GluA1 function.

We conducted extensive validation of our novel knockin line, demonstrating that SEP-GluA1 mice exhibit normal synaptic physiology, AMPA receptor trafficking, and general behavior that is indistinguishable from WT littermates, strongly supporting that our endogenous labeling strategy does not impair synaptic function in any detectable manner. Using both primary cultures and acute slices, we showed that the SEP-GluA1 line reports both increases and decreases in synaptic GluA1 content, and it is an effective tool to study several forms of AMPAR-mediated plasticity, including homeostatic scaling and LTP. Using 2p glutamate uncaging, we clearly demonstrated that the intensity of SEP fluorescence directly correlates with functional synaptic strength, indicating that this line can be used as an effective tool to study synaptic plasticity in vivo. Using FRAP, we demonstrated that SEP-GluA1-containing synapses display normal synaptic mobility and dynamics. Finally, we developed a computer vision algorithm to automatically detect and segment extremely large numbers of endogenously labeled synapses across entire brain regions in living animals. Using these tools, we were able to longitudinally track synaptic plasticity encoding sensory stimulation with unprecedented spatial coverage and molecular resolution, producing the most detailed spatiotemporal map of behaviorally relevant synaptic plasticity to date. While the SEP-GluA1 line serves as an effective tool to investigate many forms of plasticity in vitro and in vivo, we are continuing to develop similar knockin and transgenic lines that similarly label GluA2-4, as well as other proteins of interest that may play key roles in synaptic transmission and plasticity. For example, it is important to consider that our approach to monitor SEP-GluA1 might favor the detection of changes in synapses preferentially undergoing plasticity rather than an absolute change in synaptic strength. Similar endogenous labeling strategies of other synaptic proteins would be useful to investigate the molecular mechanisms of nearly any behavior, including GluA1-independent forms of plasticity (Frey et al., 2009) or disease models that display synaptic pathologies, such as SynGAP haploinsufficiency or Alzheimer’s disease (Gamache et al., 2020; Sheng et al., 2012).

By fluorescently tagging endogenous GluA1-containing AMPARs and utilizing in vivo 2p microscopy, we were able to directly visualize the functional strength of endogenous synaptic networks and track how they change during sensory stimulation. We demonstrated that mechanical stimulation of a single mouse whisker leads to increased synaptic GluA1 specifically in the cortical region corresponding to the stimulated whisker. This is consistent with previous studies showing NMDA receptor-dependent LTP following whisker stimulation (Gambino et al., 2014; Holtmaat and Caroni, 2016; Zhang et al., 2015). While our proof-of-principle experiments clearly illustrate the power of this approach to image endogenous AMPARs and track widespread synaptic plasticity in vivo, these data represent only the tip of the iceberg. As our genetic labeling strategy illuminates all GluA1-containing synapses throughout the brain, this line enables investigation of synaptic plasticity from any brain region during any behavioral paradigm of interest. Further, this flexible tool is compatible with any electrophysiological method or other fluorescence-based imaging approaches, such as neuronal activity sensors (e.g., RCaMP) and genetic tagging of cell types of interest (e.g., engram cells, Cre-lines, etc.). For example, by crossing the SEP-GluA1 knockin line with the Ai9 tdTomato reporter mouse and expressing Cre in a cell-type or neuronal circuit of interest, it is possible to specifically quantify changes in spine GluA1 expression of these neurons, as demonstrated in Figure 6 and Figure 7c and d. Beyond visualizing fluorescently labeled cortical synapses through cranial windows, one could also use our tools to investigate synaptic dynamics in subcortical structures using cortical excavation, endoscopes, or fiber photometry.

More generally, this strategy to label endogenous synaptic receptors has several key advantages over previous approaches. Building upon insights gleaned by spine dynamics, in which the formation and elimination of dendritic spines has been shown to be involved in several forms of learning (Holtmaat and Svoboda, 2009; Xu et al., 2009; Yang et al., 2009), our approach additionally enables investigation of plasticity in existing synapses. Recently, similar genetic labeling strategies have been used to investigate other synaptic proteins, such as PSD-95 (Cane et al., 2014; Fortin et al., 2014; Gray et al., 2006; Zhu et al., 2018). These studies have revealed crucial details regarding how scaffolding and structural proteins contribute to dynamic synapse function. Here, we build upon these findings by directly imaging AMPARs, which are the principal functional unit of the synapse. While overexpression of SEP-labeled AMPARs has been previously used to investigate behaviorally relevant plasticity (Diering et al., 2017; El-Boustani et al., 2018; Miyamoto et al., 2021; Roth et al., 2020; Suresh and Dunaevsky, 2017; Tan et al., 2020; Zhang et al., 2015), the current study is the first to engage labeling of endogenous receptors, which much more faithfully reports the direct physiological mechanisms of plasticity.

Recently, deep learning-based systems have achieved state-of-the-art performance in analyzing microscopy images (Moen et al., 2019). Many different architectures have been used, including our work detecting tau tangles with sliding windows to annotate single pixels (Tward et al., 2020), UNET (Olaf Ronneberger and Brox, 2015) for annotating larger blocks that has been implemented in Fiji (Falk et al., 2019), and other elaborations such as VNET (Milletari et al., 2016). Typically, trained networks assign class probabilities to each pixel, which are collected into larger objects based on connected components or watershed approaches available in standard packages, such as Fiji (Schindelin et al., 2012). A caveat of the deep learning approach is that these methods require a large quantity of high-quality training data, which is not available for new image types such as those used in our work. Another approach to annotating microscopy images has been to employ interactive methods. For example, the framework developed by ilastic allows users to annotate small regions and constructs a random forest model based on simple features (brightness, edges, texture) to extend the per pixel annotations to large images (Berg et al., 2019). The annotated regions can be modified or extended until optimal performance is achieved. Importantly, this approach works well in situations where annotation is easy ‘by eye,’ but is challenging at scale. Ultimately, we found that our rules-based, template-matching approach performed well at automatically detecting and segmenting synapses imaged in vivo, with true-positive, false-positive, and false-negative rates on par with or better than comparisons between two expert human annotators.

To achieve the goal of understanding plasticity at individual synapses during complex behaviors and learning, it is crucial to track individual synapses over time. Here, we have tracked the same population of spines during whisker stimulation (Figure 8) and have also shown that by adding a cytosolic fluorescent protein it is possible to achieve longitudinal imaging with the ability to track individual synapses across days (Figure 6a and c). In future studies, we will expand on our computational approach to enable alignment, registration, and tracking of millions of individual synapses throughout the entire process of learning. While our current imaging experiments were performed in lightly anesthetized mice, our previous work has demonstrated the feasibility to visualize and track SEP-GluA1 in individual spines in head-fixed awake behaving mice (Tan et al., 2020) with the same reliability and resolution as in anesthetized mice, suggesting that imaging awake SEP-GluA1 knockin mice will be feasible.

In conclusion, we aim to fundamentally advance our understanding of the synaptic basis of behavior, moving beyond merely studying synaptic plasticity in single neurons, seeking instead to explore dynamic modulation of the complete synaptome during learning and memory. The tools presented here make this goal achievable.

We have provided Source data for our Figures and deposited the remaining data to Dryad with https://doi.org/10.5061/dryad.ttdz08m0b. All code used to analyze and process data is freely available on GitHub, with links specified in the manuscript.

1. 1. Huganir RL
   2. Graves A

   (2021) Dryad Digital Repository

   Raw data for Graves et. al 2021 eLife.

   https://doi.org/10.5061/dryad.ttdz08m0b

1. 1. Andersen P
   2. Sundberg SH
   3. Sveen O
   4. Wigström H

   (1977) Specific long-lasting potentiation of synaptic transmission in hippocampal slices

   Nature 266:266736a0.

   https://doi.org/10.1038/266736a0

   • PubMed
   • Google Scholar
2. 1. Anggono V
   2. Huganir RL

   (2012) Regulation of AMPA receptor trafficking and synaptic plasticity

   Current Opinion in Neurobiology 22:006.

   https://doi.org/10.1016/j.conb.2011.12.006

   • PubMed
   • Google Scholar
3. 1. Araki Y
   2. Zeng M
   3. Zhang M
   4. Huganir RL

   (2015) Rapid dispersion of SynGAP from synaptic spines triggers AMPA receptor insertion and spine enlargement during LTP

   Neuron 85:173–189.

   https://doi.org/10.1016/j.neuron.2014.12.023

   • PubMed
   • Google Scholar
4. 1. Ashby MC
   2. De La Rue SA
   3. Ralph GS
   4. Uney J
   5. Collingridge GL
   6. Henley JM

   (2004) Removal of AMPA receptors (AMPARs) from synapses is preceded by transient endocytosis of extrasynaptic AMPARs

   The Journal of Neuroscience 24:5172–5176.

   https://doi.org/10.1523/JNEUROSCI.1042-04.2004

   • PubMed
   • Google Scholar
5. 1. Ashby MC
   2. Maier SR
   3. Nishimune A
   4. Henley JM

   (2006) Lateral diffusion drives constitutive exchange of AMPA receptors at dendritic spines and is regulated by spine morphology

   The Journal of Neuroscience 26:7046–7055.

   https://doi.org/10.1523/JNEUROSCI.1235-06.2006

   • PubMed
   • Google Scholar
6. 1. Bannerman DM
   2. Deacon RMJ
   3. Brady S
   4. Bruce A
   5. Sprengel R
   6. Seeburg PH
   7. Rawlins JNP

   (2004) A comparison of GluR-A-deficient and wild-type mice on a test battery assessing sensorimotor, affective, and cognitive behaviors

   Behavioral Neuroscience 118:643–647.

   https://doi.org/10.1037/0735-7044.118.3.643

   • PubMed
   • Google Scholar
7. 1. Bats C
   2. Groc L
   3. Choquet D

   (2007) The interaction between Stargazin and PSD-95 regulates AMPA receptor surface trafficking

   Neuron 53:719–734.

   https://doi.org/10.1016/j.neuron.2007.01.030

   • PubMed
   • Google Scholar
8. 1. Berg S
   2. Kutra D
   3. Kroeger T
   4. Straehle CN
   5. Kausler BX
   6. Haubold C
   7. Schiegg M
   8. Ales J
   9. Beier T
   10. Rudy M
   11. Eren K
   12. Cervantes JI
   13. Xu B
   14. Beuttenmueller F
   15. Wolny A
   16. Zhang C
   17. Koethe U
   18. Hamprecht FA
   19. Kreshuk A

   (2019) ilastik: interactive machine learning for (bio)image analysis

   Nature Methods 16:1226–1232.

   https://doi.org/10.1038/s41592-019-0582-9

   • PubMed
   • Google Scholar
9. 1. Berryer MH
   2. Hamdan FF
   3. Klitten LL
   4. Møller RS
   5. Carmant L
   6. Schwartzentruber J
   7. Patry L
   8. Dobrzeniecka S
   9. Rochefort D
   10. Neugnot-Cerioli M
   11. Lacaille J-C
   12. Niu Z
   13. Eng CM
   14. Yang Y
   15. Palardy S
   16. Belhumeur C
   17. Rouleau GA
   18. Tommerup N
   19. Immken L
   20. Beauchamp MH
   21. Patel GS
   22. Majewski J
   23. Tarnopolsky MA
   24. Scheffzek K
   25. Hjalgrim H
   26. Michaud JL
   27. Di Cristo G

   (2013) Mutations in SYNGAP1 cause intellectual disability, autism, and a specific form of epilepsy by inducing haploinsufficiency

   Human Mutation 34:385–394.

   https://doi.org/10.1002/humu.22248

   • PubMed
   • Google Scholar
10. 1. Bhatt DH
    2. Zhang S
    3. Gan WB

    (2009) Dendritic spine dynamics

    Annual Review of Physiology 71:261–282.

    https://doi.org/10.1146/annurev.physiol.010908.163140

    • PubMed
    • Google Scholar
11. 1. Bittner KC
    2. Milstein AD
    3. Grienberger C
    4. Romani S
    5. Magee JC

    (2017) Behavioral time scale synaptic plasticity underlies CA1 place fields

    Science 357:1033–1036.

    https://doi.org/10.1126/science.aan3846

    • PubMed
    • Google Scholar
12. 1. Bliss TV
    2. Lomo T

    (1973) Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path

    The Journal of Physiology 232:sp010273.

    https://doi.org/10.1113/jphysiol.1973.sp010273

    • PubMed
    • Google Scholar
13. 1. Boerner T
    2. Bygrave AM
    3. Chen J
    4. Fernando A
    5. Jackson S
    6. Barkus C
    7. Sprengel R
    8. Seeburg PH
    9. Harrison PJ
    10. Gilmour G
    11. Bannerman DM
    12. Sanderson DJ

    (2017) The group II metabotropic glutamate receptor agonist LY354740 and the D2 receptor antagonist haloperidol reduce locomotor hyperactivity but fail to rescue spatial working memory in GluA1 knockout mice

    The European Journal of Neuroscience 45:912–921.

    https://doi.org/10.1111/ejn.13539

    • PubMed
    • Google Scholar
14. Book

    1. Brunelli R

    (2009) Template Matching Techniques in Computer Vision Signal Processing

    Wiley.

    https://doi.org/10.1002/9780470744055

    • Google Scholar
15. 1. Bygrave AM
    2. Jahans-Price T
    3. Wolff AR
    4. Sprengel R
    5. Kullmann DM
    6. Bannerman DM
    7. Kätzel D

    (2019) Hippocampal-prefrontal coherence mediates working memory and selective attention at distinct frequency bands and provides a causal link between schizophrenia and its risk gene GRIA1

    Translational Psychiatry 9:142.

    https://doi.org/10.1038/s41398-019-0471-0

    • PubMed
    • Google Scholar
16. 1. Cane M
    2. Maco B
    3. Knott G
    4. Holtmaat A

    (2014) The relationship between PSD-95 clustering and spine stability in vivo

    The Journal of Neuroscience 34:2075–2086.

    https://doi.org/10.1523/JNEUROSCI.3353-13.2014

    • PubMed
    • Google Scholar
17. 1. Chen H
    2. Roth RH
    3. Lopez-Ortega E
    4. Tan HL
    5. Huganir RL

    (2021) AMPA Receptors Exist in Tunable Mobile and Immobile Synaptic Fractions In Vivo

    ENeuro 8:ENEURO.0015-21.2021.

    https://doi.org/10.1523/ENEURO.0015-21.2021

    • PubMed
    • Google Scholar
18. 1. Díaz-Alonso J
    2. Sun YJ
    3. Granger AJ
    4. Levy JM
    5. Blankenship SM
    6. Nicoll RA

    (2017) Subunit-specific role for the amino-terminal domain of AMPA receptors in synaptic targeting

    PNAS 114:7136–7141.

    https://doi.org/10.1073/pnas.1707472114

    • PubMed
    • Google Scholar
19. 1. Díaz-Alonso J
    2. Nicoll RA

    (2021) AMPA receptor trafficking and LTP: Carboxy-termini, amino-termini and TARPs

    Neuropharmacology 197:108710.

    https://doi.org/10.1016/j.neuropharm.2021.108710

    • PubMed
    • Google Scholar
20. 1. Diering GH
    2. Nirujogi RS
    3. Roth RH
    4. Worley PF
    5. Pandey A
    6. Huganir RL

    (2017) Homer1a drives homeostatic scaling-down of excitatory synapses during sleep

    Science 355:511–515.

    https://doi.org/10.1126/science.aai8355

    • PubMed
    • Google Scholar
21. 1. Dombeck DA
    2. Harvey CD
    3. Tian L
    4. Looger LL
    5. Tank DW

    (2010) Functional imaging of hippocampal place cells at cellular resolution during virtual navigation

    Nature Neuroscience 13:1433–1440.

    https://doi.org/10.1038/nn.2648

    • PubMed
    • Google Scholar
22. 1. El-Boustani S
    2. Ip JPK
    3. Breton-Provencher V
    4. Knott GW
    5. Okuno H
    6. Bito H
    7. Sur M

    (2018) Locally coordinated synaptic plasticity of visual cortex neurons in vivo

    Science 360:1349–1354.

    https://doi.org/10.1126/science.aao0862

    • PubMed
    • Google Scholar
23. 1. Epsztein J
    2. Brecht M
    3. Lee AK

    (2011) tracellular determinants of hippocampal CA1 place and silent cell activity in a novel environment

    Neuron 70:109–120.

    https://doi.org/10.1016/j.neuron.2011.03.006

    • PubMed
    • Google Scholar
24. 1. Falk T
    2. Mai D
    3. Bensch R
    4. Çiçek Ö
    5. Abdulkadir A
    6. Marrakchi Y
    7. Böhm A
    8. Deubner J
    9. Jäckel Z
    10. Seiwald K
    11. Dovzhenko A
    12. Tietz O
    13. Dal Bosco C
    14. Walsh S
    15. Saltukoglu D
    16. Tay TL
    17. Prinz M
    18. Palme K
    19. Simons M
    20. Diester I
    21. Brox T
    22. Ronneberger O

    (2019) U-Net: deep learning for cell counting, detection, and morphometry

    Nature Methods 16:67–70.

    https://doi.org/10.1038/s41592-018-0261-2

    • PubMed
    • Google Scholar
25. 1. Fang H
    2. Bygrave AM
    3. Roth RH
    4. Johnson RC
    5. Huganir RL

    (2021) An optimized CRISPR/Cas9 approach for precise genome editing in neurons

    eLife 10:e65202.

    https://doi.org/10.7554/eLife.65202

    • PubMed
    • Google Scholar
26. 1. Fenno L
    2. Yizhar O
    3. Deisseroth K

    (2011) The development and application of optogenetics

    Annual Review of Neuroscience 34:389–412.

    https://doi.org/10.1146/annurev-neuro-061010-113817

    • PubMed
    • Google Scholar
27. 1. Fortin DA
    2. Tillo SE
    3. Yang G
    4. Rah JC
    5. Melander JB
    6. Bai S
    7. Soler-Cedeño O
    8. Qin M
    9. Zemelman BV
    10. Guo C
    11. Mao T
    12. Zhong H

    (2014) Live imaging of endogenous PSD-95 using ENABLED: a conditional strategy to fluorescently label endogenous proteins

    The Journal of Neuroscience 34:2014.

    https://doi.org/10.1523/JNEUROSCI.3888-14.2014

    • PubMed
    • Google Scholar
28. 1. Frey U
    2. Morris RG

    (1997) Synaptic tagging and long-term potentiation

    Nature 385:533–536.

    https://doi.org/10.1038/385533a0

    • PubMed
    • Google Scholar
29. 1. Frey MC
    2. Sprengel R
    3. Nevian T

    (2009) Activity pattern-dependent long-term potentiation in neocortex and hippocampus of GluA1 (GluR-A) subunit-deficient mice

    The Journal of Neuroscience 29:5587–5596.

    https://doi.org/10.1523/JNEUROSCI.5314-08.2009

    • PubMed
    • Google Scholar
30. 1. Frischknecht R
    2. Heine M
    3. Perrais D
    4. Seidenbecher CI
    5. Choquet D
    6. Gundelfinger ED

    (2009) Brain extracellular matrix affects AMPA receptor lateral mobility and short-term synaptic plasticity

    Nature Neuroscience 12:897–904.

    https://doi.org/10.1038/nn.2338

    • PubMed
    • Google Scholar
31. 1. Gamache TR
    2. Araki Y
    3. Huganir RL

    (2020) Twenty Years of SynGAP Research: From Synapses to Cognition

    The Journal of Neuroscience 40:1596–1605.

    https://doi.org/10.1523/JNEUROSCI.0420-19.2020

    • PubMed
    • Google Scholar
32. 1. Gambino F
    2. Pagès S
    3. Kehayas V
    4. Baptista D
    5. Tatti R
    6. Carleton A
    7. Holtmaat A

    (2014) Sensory-evoked LTP driven by dendritic plateau potentials in vivo

    Nature 515:116–119.

    https://doi.org/10.1038/nature13664

    • PubMed
    • Google Scholar
33. Software

    1. Graves A
    2. Roth R
    3. Tan H
    4. Zhu Q
    5. Bygrave A
    6. Lopez-Ortega E
    7. Spiegel A
    8. Hong I
    9. Johnson R
    10. Vogelstein J
    11. Tward D
    12. Miller M
    13. Huganir R

    (2020) Synapse Labeling, version 5a274f9

    GitHub.

    https://github.com/twardlab/synapse_labeling
34. 1. Gray NW
    2. Weimer RM
    3. Bureau I
    4. Svoboda K

    (2006) Rapid redistribution of synaptic PSD-95 in the neocortex in vivo

    PLOS Biology 4:e370.

    https://doi.org/10.1371/journal.pbio.0040370

    • PubMed
    • Google Scholar
35. 1. Henley JM
    2. Wilkinson KA

    (2016) Synaptic AMPA receptor composition in development, plasticity and disease

    Nature Reviews. Neuroscience 17:337–350.

    https://doi.org/10.1038/nrn.2016.37

    • PubMed
    • Google Scholar
36. 1. Hochbaum DS
    2. Shmoys DB

    (1985) A Best Possible Heuristic for the k -Center Problem

    Mathematics of Operations Research 10:180–184.

    https://doi.org/10.1287/moor.10.2.180

    • Google Scholar
37. 1. Holtmaat A
    2. Svoboda K

    (2009) Experience-dependent structural synaptic plasticity in the mammalian brain

    Nature Reviews. Neuroscience 10:647–658.

    https://doi.org/10.1038/nrn2699

    • PubMed
    • Google Scholar
38. 1. Holtmaat A
    2. Caroni P

    (2016) Functional and structural underpinnings of neuronal assembly formation in learning

    Nature Neuroscience 19:1553–1562.

    https://doi.org/10.1038/nn.4418

    • PubMed
    • Google Scholar
39. Software

    1. Hong I

    (2021) StackGPS: 3D registration on-the-fly for longitudinal two photon imaging with submicron precision, version v1.0

    Github.

    https://github.com/ingiehong/StackGPS
40. 1. Huganir RL
    2. Nicoll RA

    (2013) AMPARs and synaptic plasticity: the last 25 years

    Neuron 80:704–717.

    https://doi.org/10.1016/j.neuron.2013.10.025

    • PubMed
    • Google Scholar
41. 1. Juavinett AL
    2. Bekheet G
    3. Churchland AK

    (2019) Chronically implanted Neuropixels probes enable high-yield recordings in freely moving mice

    eLife 8:e47188.

    https://doi.org/10.7554/eLife.47188

    • PubMed
    • Google Scholar
42. 1. Jun JJ
    2. Steinmetz NA
    3. Siegle JH
    4. Denman DJ
    5. Bauza M
    6. Barbarits B
    7. Lee AK
    8. Anastassiou CA
    9. Andrei A
    10. Aydın Ç
    11. Barbic M
    12. Blanche TJ
    13. Bonin V
    14. Couto J
    15. Dutta B
    16. Gratiy SL
    17. Gutnisky DA
    18. Häusser M
    19. Karsh B
    20. Ledochowitsch P
    21. Lopez CM
    22. Mitelut C
    23. Musa S
    24. Okun M
    25. Pachitariu M
    26. Putzeys J
    27. Rich PD
    28. Rossant C
    29. Sun W-L
    30. Svoboda K
    31. Carandini M
    32. Harris KD
    33. Koch C
    34. O’Keefe J
    35. Harris TD

    (2017) Fully integrated silicon probes for high-density recording of neural activity

    Nature 551:232–236.

    https://doi.org/10.1038/nature24636

    • PubMed
    • Google Scholar
43. 1. Kopec CD
    2. Li B
    3. Wei W
    4. Boehm J
    5. Malinow R

    (2006) Glutamate receptor exocytosis and spine enlargement during chemically induced long-term potentiation

    The Journal of Neuroscience 26:2000–2009.

    https://doi.org/10.1523/JNEUROSCI.3918-05.2006

    • PubMed
    • Google Scholar
44. 1. Lee KFH
    2. Soares C
    3. Béïque J-C

    (2012) Examining form and function of dendritic spines

    Neural Plasticity 2012:704103.

    https://doi.org/10.1155/2012/704103

    • PubMed
    • Google Scholar
45. 1. Lee SH
    2. Jin C
    3. Cai E
    4. Ge P
    5. Ishitsuka Y
    6. Teng KW
    7. de Thomaz AA
    8. Nall D
    9. Baday M
    10. Jeyifous O
    11. Demonte D
    12. Dundas CM
    13. Park S
    14. Delgado JY
    15. Green WN
    16. Selvin PR

    (2017) Super-resolution imaging of synaptic and Extra-synaptic AMPA receptors with different-sized fluorescent probes

    eLife 6:e27744.

    https://doi.org/10.7554/eLife.27744

    • PubMed
    • Google Scholar
46. 1. Li TP
    2. Song Y
    3. MacGillavry HD
    4. Blanpied TA
    5. Raghavachari S

    (2016) Protein Crowding within the Postsynaptic Density Can Impede the Escape of Membrane Proteins

    The Journal of Neuroscience 36:4276–4295.

    https://doi.org/10.1523/JNEUROSCI.3154-15.2016

    • PubMed
    • Google Scholar
47. 1. Lin MZ
    2. Schnitzer MJ

    (2016) Genetically encoded indicators of neuronal activity

    Nature Neuroscience 19:1142–1153.

    https://doi.org/10.1038/nn.4359

    • PubMed
    • Google Scholar
48. 1. Makino H
    2. Malinow R

    (2009) AMPA receptor incorporation into synapses during LTP: the role of lateral movement and exocytosis

    Neuron 64:381–390.

    https://doi.org/10.1016/j.neuron.2009.08.035

    • PubMed
    • Google Scholar
49. 1. Malinow R
    2. Malenka RC

    (2002) AMPA receptor trafficking and synaptic plasticity

    Annual Review of Neuroscience 25:103–126.

    https://doi.org/10.1146/annurev.neuro.25.112701.142758

    • PubMed
    • Google Scholar
50. 1. Mammen AL
    2. Huganir RL
    3. O’Brien RJ

    (1997) Redistribution and stabilization of cell surface glutamate receptors during synapse formation

    The Journal of Neuroscience 17:7351–7358.

    https://doi.org/10.1523/JNEUROSCI.17-19-07351.1997

    • PubMed
    • Google Scholar
51. 1. Martin S
    2. Henley JM
    3. Holman D
    4. Zhou M
    5. Wiegert O
    6. van Spronsen M
    7. Joëls M
    8. Hoogenraad CC
    9. Krugers HJ
    10. Chapouthier G

    (2009) Corticosterone alters AMPAR mobility and facilitates bidirectional synaptic plasticity

    PLOS ONE 4:e4714.

    https://doi.org/10.1371/journal.pone.0004714

    • Google Scholar
52. 1. Matsuzaki M
    2. Ellis-Davies GC
    3. Nemoto T
    4. Miyashita Y
    5. Iino M
    6. Kasai H

    (2001) Dendritic spine geometry is critical for AMPA receptor expression in hippocampal CA1 pyramidal neurons

    Nature Neuroscience 4:1086–1092.

    https://doi.org/10.1038/nn736

    • PubMed
    • Google Scholar
53. 1. McNaughton BL
    2. Douglas RM
    3. Goddard GV

    (1978) Synaptic enhancement in fascia dentata: cooperativity among coactive afferents

    Brain Research 157:277–293.

    https://doi.org/10.1016/0006-8993(78)90030-6

    • PubMed
    • Google Scholar
54. 1. Miesenböck G
    2. De Angelis DA
    3. Rothman JE

    (1998) Visualizing secretion and synaptic transmission with pH-sensitive green fluorescent proteins

    Nature 394:192–195.

    https://doi.org/10.1038/28190

    • PubMed
    • Google Scholar
55. Conference

    1. Milletari F
    2. Navab N
    3. Ahmadi SA

    (2016) 2016 Fourth International Conference on 3D Vision IEEE

    V-net: Fully convolutional neural networks for volumetric medical image segmentation.

    https://doi.org/10.1109/3DV.2016.79

    • Google Scholar
56. 1. Miyamoto D
    2. Marshall W
    3. Tononi G
    4. Cirelli C

    (2021) Net decrease in spine-surface GluA1-containing AMPA receptors after post-learning sleep in the adult mouse cortex

    Nature Communications 12:2881.

    https://doi.org/10.1038/s41467-021-23156-2

    • PubMed
    • Google Scholar
57. 1. Moen E
    2. Bannon D
    3. Kudo T
    4. Graf W
    5. Covert M
    6. Van Valen D

    (2019) Deep learning for cellular image analysis

    Nature Methods 16:1233–1246.

    https://doi.org/10.1038/s41592-019-0403-1

    • PubMed
    • Google Scholar
58. 1. Nicoll RA

    (2017) A Brief History of Long-Term Potentiation

    Neuron 93:281–290.

    https://doi.org/10.1016/j.neuron.2016.12.015

    • PubMed
    • Google Scholar
59. 1. Oku Y
    2. Huganir RL

    (2013) AGAP3 and Arf6 regulate trafficking of AMPA receptors and synaptic plasticity

    The Journal of Neuroscience 33:12586–12598.

    https://doi.org/10.1523/JNEUROSCI.0341-13.2013

    • PubMed
    • Google Scholar
60. Preprint

    1. Olaf Ronneberger PF
    2. Brox T

    (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation

    arXiv.

    https://arxiv.org/abs/1505.04597

    • Google Scholar
61. 1. O’Brien RJ
    2. Kamboj S
    3. Ehlers MD
    4. Rosen KR
    5. Fischbach GD
    6. Huganir RL

    (1998) Activity-dependent modulation of synaptic AMPA receptor accumulation

    Neuron 21:1067–1078.

    https://doi.org/10.1016/s0896-6273(00)80624-8

    • PubMed
    • Google Scholar
62. 1. Padman BS
    2. Bach M
    3. Ramm G

    (2014) An Improved Procedure for Subcellular Spatial Alignment during Live-Cell CLEM

    PLOS ONE 9:e95967.

    https://doi.org/10.1371/journal.pone.0095967

    • PubMed
    • Google Scholar
63. 1. Patterson MA
    2. Szatmari EM
    3. Yasuda R

    (2010) AMPA receptors are exocytosed in stimulated spines and adjacent dendrites in a Ras-ERK-dependent manner during long-term potentiation

    PNAS 107:0913875107.

    https://doi.org/10.1073/pnas.0913875107

    • PubMed
    • Google Scholar
64. 1. Rosenkrantz DJ
    2. Stearns RE
    3. Lewis, II PM

    (1977) An Analysis of Several Heuristics for the Traveling Salesman Problem

    SIAM Journal on Computing 6:563–581.

    https://doi.org/10.1137/0206041

    • Google Scholar
65. 1. Roth RH
    2. Zhang Y
    3. Huganir RL

    (2017) Dynamic imaging of AMPA receptor trafficking in vitro and in vivo

    Current Opinion in Neurobiology 45:51–58.

    https://doi.org/10.1016/j.conb.2017.03.008

    • PubMed
    • Google Scholar
66. 1. Roth RH
    2. Cudmore RH
    3. Tan HL
    4. Hong I
    5. Zhang Y
    6. Huganir RL

    (2020) Cortical Synaptic AMPA Receptor Plasticity during Motor Learning

    Neuron 105:895–908.

    https://doi.org/10.1016/j.neuron.2019.12.005

    • PubMed
    • Google Scholar
67. 1. Sanderson DJ
    2. Gray A
    3. Simon A
    4. Taylor AM
    5. Deacon RMJ
    6. Seeburg PH
    7. Sprengel R
    8. Good MA
    9. Rawlins JNP
    10. Bannerman DM

    (2007) Deletion of glutamate receptor-A (GluR-A) AMPA receptor subunits impairs one-trial spatial memory

    Behavioral Neuroscience 121:559–569.

    https://doi.org/10.1037/0735-7044.121.3.559

    • PubMed
    • Google Scholar
68. 1. Santuy A
    2. Tomás-Roca L
    3. Rodríguez J-R
    4. González-Soriano J
    5. Zhu F
    6. Qiu Z
    7. Grant SGN
    8. DeFelipe J
    9. Merchan-Perez A

    (2020) Estimation of the number of synapses in the hippocampus and brain-wide by volume electron microscopy and genetic labeling

    Scientific Reports 10:14014.

    https://doi.org/10.1038/s41598-020-70859-5

    • PubMed
    • Google Scholar
69. 1. Schindelin J
    2. Arganda-Carreras I
    3. Frise E
    4. Kaynig V
    5. Longair M
    6. Pietzsch T
    7. Preibisch S
    8. Rueden C
    9. Saalfeld S
    10. Schmid B
    11. Tinevez J-Y
    12. White DJ
    13. Hartenstein V
    14. Eliceiri K
    15. Tomancak P
    16. Cardona A

    (2012) Fiji: an open-source platform for biological-image analysis

    Nature Methods 9:676–682.

    https://doi.org/10.1038/nmeth.2019

    • PubMed
    • Google Scholar
70. 1. Sdrulla AD
    2. Linden DJ

    (2007) Double dissociation between long-term depression and dendritic spine morphology in cerebellar Purkinje cells

    Nature Neuroscience 10:546–548.

    https://doi.org/10.1038/nn1889

    • PubMed
    • Google Scholar
71. 1. Sharma K
    2. Fong DK
    3. Craig AM

    (2006) Postsynaptic protein mobility in dendritic spines: long-term regulation by synaptic NMDA receptor activation

    Molecular and Cellular Neurosciences 31:702–712.

    https://doi.org/10.1016/j.mcn.2006.01.010

    • PubMed
    • Google Scholar
72. 1. Sheng M
    2. Sabatini BL
    3. Sudhof TC

    (2012) Synapses and Alzheimer’s Disease

    Cold Spring Harbor Perspectives in Biology 4:a005777.

    https://doi.org/10.1101/cshperspect.a005777

    • Google Scholar
73. 1. Sheng N
    2. Bemben MA
    3. Díaz-Alonso J
    4. Tao W
    5. Shi YS
    6. Nicoll RA

    (2018) LTP requires postsynaptic PDZ-domain interactions with glutamate receptor/auxiliary protein complexes

    PNAS 115:3948–3953.

    https://doi.org/10.1073/pnas.1800719115

    • PubMed
    • Google Scholar
74. 1. Sofroniew NJ
    2. Flickinger D
    3. King J
    4. Svoboda K

    (2016) A large field of view two-photon mesoscope with subcellular resolution for in vivo imaging

    eLife 5:e14472.

    https://doi.org/10.7554/eLife.14472

    • PubMed
    • Google Scholar
75. 1. Spruston N

    (2008) Pyramidal neurons: dendritic structure and synaptic integration

    Nature Reviews. Neuroscience 9:206–221.

    https://doi.org/10.1038/nrn2286

    • PubMed
    • Google Scholar
76. 1. Suresh A
    2. Dunaevsky A

    (2017) Relationship Between Synaptic AMPAR and Spine Dynamics: Impairments in the FXS Mouse

    Cerebral Cortex 27:4244–4256.

    https://doi.org/10.1093/cercor/bhx128

    • PubMed
    • Google Scholar
77. 1. Tan HL
    2. Roth RH
    3. Graves AR
    4. Cudmore RH
    5. Huganir RL

    (2020) Lamina-specific AMPA receptor dynamics following visual deprivation in vivo

    eLife 9:e52420.

    https://doi.org/10.7554/eLife.52420

    • PubMed
    • Google Scholar
78. 1. Trachtenberg JT
    2. Chen BE
    3. Knott GW
    4. Feng G
    5. Sanes JR
    6. Welker E
    7. Svoboda K

    (2002) Long-term in vivo imaging of experience-dependent synaptic plasticity in adult cortex

    Nature 420:788–794.

    https://doi.org/10.1038/nature01273

    • PubMed
    • Google Scholar
79. 1. Turrigiano GG
    2. Leslie KR
    3. Desai NS
    4. Rutherford LC
    5. Nelson SB

    (1998) Activity-dependent scaling of quantal amplitude in neocortical neurons

    Nature 391:892–896.

    https://doi.org/10.1038/36103

    • PubMed
    • Google Scholar
80. 1. Tward D
    2. Brown T
    3. Kageyama Y
    4. Patel J
    5. Hou Z
    6. Mori S
    7. Albert M
    8. Troncoso J
    9. Miller M

    (2020) Diffeomorphic Registration With Intensity Transformation and Missing Data: Application to 3D Digital Pathology of Alzheimer’s Disease

    Frontiers in Neuroscience 14:52.

    https://doi.org/10.3389/fnins.2020.00052

    • PubMed
    • Google Scholar
81. 1. Verveer PJ
    2. Sevilla-Sánchez D
    3. Bloembergen PC
    4. Adam N
    5. Schoonderwoert VTG
    6. van der Voort HTM

    (2020) Fast Analysis of 2D and 3D Single-Molecule Localization Microscopy Data with Huygens Localizer

    Microscopy Today 28:20–25.

    https://doi.org/10.1017/S1551929520000024

    • Google Scholar
82. 1. Volk L
    2. Chiu SL
    3. Sharma K
    4. Huganir RL

    (2015) Glutamate synapses in human cognitive disorders

    Annual Review of Neuroscience 38:127–149.

    https://doi.org/10.1146/annurev-neuro-071714-033821

    • PubMed
    • Google Scholar
83. 1. Xu T
    2. Yu X
    3. Perlik AJ
    4. Tobin WF
    5. Zweig JA
    6. Tennant K
    7. Jones T
    8. Zuo Y

    (2009) Rapid formation and selective stabilization of synapses for enduring motor memories

    Nature 462:915–919.

    https://doi.org/10.1038/nature08389

    • PubMed
    • Google Scholar
84. 1. Xu N
    2. Harnett MT
    3. Williams SR
    4. Huber D
    5. O’Connor DH
    6. Svoboda K
    7. Magee JC

    (2012) Nonlinear dendritic integration of sensory and motor input during an active sensing task

    Nature 492:247–251.

    https://doi.org/10.1038/nature11601

    • PubMed
    • Google Scholar
85. 1. Yang G
    2. Pan F
    3. Gan WB

    (2009) Stably maintained dendritic spines are associated with lifelong memories

    Nature 462:920–924.

    https://doi.org/10.1038/nature08577

    • PubMed
    • Google Scholar
86. 1. Zamanillo D
    2. Sprengel R
    3. Hvalby O
    4. Jensen V
    5. Burnashev N
    6. Rozov A
    7. Kaiser KM
    8. Köster HJ
    9. Borchardt T
    10. Worley P
    11. Lübke J
    12. Frotscher M
    13. Kelly PH
    14. Sommer B
    15. Andersen P
    16. Seeburg PH
    17. Sakmann B

    (1999) Importance of AMPA receptors for hippocampal synaptic plasticity but not for spatial learning

    Science 284:1805–1811.

    https://doi.org/10.1126/science.284.5421.1805

    • PubMed
    • Google Scholar
87. 1. Zhang Y
    2. Cudmore RH
    3. Lin DT
    4. Linden DJ
    5. Huganir RL

    (2015) Visualization of NMDA receptor-dependent AMPA receptor synaptic plasticity in vivo

    Nature Neuroscience 18:402–407.

    https://doi.org/10.1038/nn.3936

    • PubMed
    • Google Scholar
88. 1. Zhou Z
    2. Liu A
    3. Xia S
    4. Leung C
    5. Qi J
    6. Meng Y
    7. Xie W
    8. Park P
    9. Collingridge GL
    10. Jia Z

    (2018) The C-terminal tails of endogenous GluA1 and GluA2 differentially contribute to hippocampal synaptic plasticity and learning

    Nature Neuroscience 21:50–62.

    https://doi.org/10.1038/s41593-017-0030-z

    • PubMed
    • Google Scholar
89. 1. Zhu F
    2. Cizeron M
    3. Qiu Z
    4. Benavides-Piccione R
    5. Kopanitsa MV
    6. Skene NG
    7. Koniaris B
    8. DeFelipe J
    9. Fransén E
    10. Komiyama NH
    11. Grant SGN

    (2018) Architecture of the Mouse Brain Synaptome

    Neuron 99:781–799.

    https://doi.org/10.1016/j.neuron.2018.07.007

    • Google Scholar

Thank you for submitting your article "Visualizing synaptic plasticity in vivo by large-scale imaging of endogenous AMPA receptors" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Brice Bathellier as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Lu Chen as the Senior Editor.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) The authors should clarify how the synapse detection algorithm account for extrasynaptic GluA1 labeling. Increases in this pool may not directly underlie increases at this synapse. Testing these confounding factors could be performed using physiologically relevant plasticity protocols in acute brain slices rather than cell culture, a preparation in which the neuron type and connectivity may not be the same as the mostly intact circuit.

2) The extended z-resolution of the 2P point-spread-function (PSF) in vivo will likely falsely assign fluorescence from multiple synapses to a single punctum. How does this limited resolution impact the final interpretations? The study would benefit from an in vivo ground-truth estimate of synapse location to confirm that SEP-GluA1 puncta correspond to an increase in AMPARs at synapses. The use of Homer does not entirely circumvent the uncertainty generated by the poor resolution of 2P. Post-hoc confocal microscopy in fixed tissue using morphology and synaptic markers could be sufficient. If not, then super-resolution fluorescence or immunogold EM are certainly suitable ground-truth approaches.

3) The quantification of synaptic strength by SEP-GluA1 may be an overestimate because SEP-GluA1 containing AMARs represents a fraction of the total receptor population, and LTP is strongly biased towards GluA1 insertion. How can this potential error be overcome?

4) To verify that SEP-GluA1 can reliably report decreases in synaptic strength, the authors should examine the activity-dependent endocytosis of receptors following LTD pairing protocols, preferably from the same cell type in the brain slices.

5) The choice to fluorescently label GluA1 subunits of the AMPAR to mark plasticity is, on the one hand, an obvious choice because of its implication in LTP. However, evidence in the literature suggests that LTP in the somatosensory cortex can be GluA1-independent (e.g., Frey et al., 2009). The authors should discuss these studies' implications and the limits of their method for addressing synaptic mechanisms for learning and memory. Does SEP-GluA1 fluorescence reproduce previous estimates of the distribution of synaptic weights? The SEP-GluA1 method is undoubtedly more attractive than the post hoc patch-clamp-based approaches performed by Ko et al. 2011, 2013 (Hofer and Mrsic-Flogel laboratories), but SEP-GluA1 should be compared to spine intensity (volume) imaging for example.

6) Conditional expression of the SEP-GluA1 would allow the precise dissection of synaptic plasticity specific neuron types within microcircuits, thereby linking plasticity with altered circuit function and behavior. This limitation must be addressed, in detail, in the Discussion. The current data sets of whole-brain expression of SEP-GluA1 do not distinguish between layer 2/3 pyramidal neurons, layer 5 pyramidal neurons, or interneurons. It is interesting to consider the possibility that a cross between the SEP-GluA1 transgenic mouse and a PSD-95 YPF-Enabled (Fortin et al., 2014) mouse could be used to assess the GluA1 changes at specific synapse types.

7) Using Western Blot, the authors show a 57% reduction of GluA1 expression at the postsynaptic density relative to WT, however, the potential implications of this are not addressed. In figure 1f, they show relative AMPA levels resulting from their Western blot analysis. Still, they do not adequately describe how the Western blot and analysis were performed, and namely, what control they used to normalize protein levels between different fractions. It would be nice to see additional statistical analyses of GluA2 and GluA3 expression relative to WT as there seems to be an increase and, therefore, potential compensation by GluA3 subunits SEP-GluA1 KI.

8) In the FRAP experiment in Figure 6, it seems like there could be other interpretations for the time course of recovery of fluorescence. The expected timescales of fluorescence recovery and caveats of its interpretations should be discussed.

9) An essential aspect of plasticity in the brain is that it can occur on long time scale on precise synapses. Here the authors do not show that it is possible to do longitudinal studies in which individual synapses are tracked. The authors need to provide some data to allow evaluating if this this is actually possible with their method. e.g. with a cytoplasmic marker for spines.

10) Comparison of their synapse detection algorithm with concurrent methods is necessary to establish it as a new tool.

Reviewer #1 (Recommendations for the authors):

1) Better discuss the limits of synapse identification on images. The limits are related both to the algorithm used and to 2-phoron imaging resolution limits. Thus better identification could be obtained maybe with super resolution methods. Ideally providing ground truth data would be an asset.

2) An essential aspect of plasticity in the brain is that it can occur on long time scale on precise synapses. Here the authors do not show that it is possible to do longitudinal studies in which individual synapses are tracked. I recommend to show some data that it is actually possible with their method, e.g. with a cytoplasmic marker for spines.

3) Comparison with previous methods (e.g. spine volume imaging) is necessary to establish this tool as a useful to track synaptic plasticity. This could be done in the whisker-based plasticity assay.

Reviewer #2 (Recommendations for the authors):

This is an important paper with well carried-out experiments. The conclusions are well supported by data. I have however 3 concerns (that do not dampen my enthusiasm):

1) The global expression of GluA2 is twice as low as the expression of GluA1 in wild-type mice, although the level of GluA2 remains unchanged. How this would affect GluA1/Glu2 heteromers? In addition, can the authors exclude the possibility that the decrease of SEP-GluA1 is brain region specific? (e.g. different levels of SEP-GluA1 between the hippocampus and the cortex). if the decrease is indeed region-specific, could the author exclude that physiological deficit does not occur in other brain region? At best, the authors should perform a better comparison of GluA1 and SEP-GluA1 between brain regions. A least, these point should be discussed. In the same line, bidirectional changes after Bic and TTX treatment in culture are really convincing (Figure 3B). However, it would be fair to present also the non-normalized level of surface AMPAR after treatment as compared to WT.

2) Although the authors present strong and convincing evidence that the high density of SEP-GluA1 synapses does not preclude the report of synaptic plasticity in anesthetized mice, it is much less clear whether it would work also in awake mice.

3) How effective is their supervised detecting method as compared to existing computer-based detection methods ? (e.g. see figure 7 from: https://www.biorxiv.org/content/10.1101/061507v2.full.pdf). Also, how robust is the supervised annotation of spines when the same FOV is imaged chronically (between days)?

Reviewer #3 (Recommendations for the authors):

My specific concerns are as follows:

1) How does the synapse detection algorithm account for extrasynaptic GluA1 labeling? Increases in this pool may not directly underlie increases at this synapse. Testing these confounding factors could be performed using physiologically relevant plasticity protocols in acute brain slices rather than cell culture, a preparation in which the neuron type and connectivity may not be the same as the mostly intact circuit.

2) The extended z-resolution of the 2P point-spread-function (PSF) in vivo will likely falsely assign fluorescence from multiple synapses to a single punctum. How does this limited resolution impact the final interpretations? The study would benefit from an in vivo ground-truth estimate of synapse location to confirm that SEP-GluA1 puncta correspond to an increase in AMPARs at synapses. The use of Homer does not entirely circumvent the uncertainty generated by the poor resolution of 2P. Post-hoc confocal microscopy in fixed tissue using morphology and synaptic markers could be sufficient. If not, then super-resolution fluorescence or immunogold EM are certainly suitable ground-truth approaches.

3) The quantification of synaptic strength by SEP-GluA1 may be an overestimate because SEP-GluA1 containing AMARs represents a fraction of the total receptor population, and LTP is strongly biased towards GluA1 insertion. How can this potential error be overcome?

4) To verify that SEP-GluA1 can reliably report decreases in synaptic strength, the authors should examine the activity-dependent endocytosis of receptors following LTD pairing protocols, preferably from the same cell type in the brain slices.

5) The choice to fluorescently label GluA1 subunits of the AMPAR to mark plasticity is, on the one hand, an obvious choice because of its implication in LTP. However, evidence in the literature suggests that LTP in the somatosensory cortex can be GluA1-independent (e.g., Frey et al., 2009). The authors should discuss these studies' implications and the limits of their method for addressing synaptic mechanisms for learning and memory. Does SEP-GluA1 fluorescence reproduce previous estimates of the distribution of synaptic weights? The SEP-GluA1 method is undoubtedly more attractive than the post hoc patch-clamp-based approaches performed by Ko et al. 2011, 2013 (Hofer and Mrsic-Flogel laboratories).

6) Conditional expression of the SEP-GluA1 would allow the precise dissection of synaptic plasticity specific neuron types within microcircuits, thereby linking plasticity with altered circuit function and behavior. This limitation must be addressed, in detail, in the Discussion. The current data sets of whole-brain expression of SEP-GluA1 do not distinguish between layer 2/3 pyramidal neurons, layer 5 pyramidal neurons, or interneurons. It is interesting to consider the possibility that a cross between the SEP-GluA1 transgenic mouse and a PSD-95 YPF-Enabled (Fortin et al., 2014) mouse could be used to assess the GluA1 changes at specific synapse types.

7) Using Western Blot, the authors show a 57% reduction of GluA1 expression at the postsynaptic density relative to WT, however, the potential implications of this are not addressed. In figure 1f, they show relative AMPA levels resulting from their Western blot analysis. Still, they do not adequately describe how the Western blot and analysis were performed, and namely, what control they used to normalize protein levels between different fractions. It would be nice to see additional statistical analyses of GluA2 and GluA3 expression relative to WT as there seems to be an increase and, therefore, potential compensation by GluA3 subunits SEP-GluA1 KI.

8) In the FRAP experiment in Figure 6, it seems like there could be other interpretations for the time course of recovery of fluorescence. The expected timescales of fluorescence recovery and caveats of its interpretations should be discussed.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Visualizing synaptic plasticity in vivo by large-scale imaging of endogenous AMPA receptors" for further consideration by eLife. Your revised article has been reviewed by 3 reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Lu Chen as the Senior Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

1. There is still no convincing data that spines can be tracked in vivo across a long time interval (e.g. 1 day). The authors show a picture with sparse Td-tomato labelling, but only on a single time point. It is necessary to show the same picture at a later time point to appreciate the degree of consistency of SEP-GluA1 labelling across days. The sentence in the abstract, "we can longitudinally track how the strength of synapses changes during behavior," is currently not fully supported by the data.

2. The authors did not compare the efficacy of their automated synapse segmentation methods to other methods. The authors state in the discussion that their method outperforms other methods but this is not documented. This statement requires a quantification.

3. The authors do not convincingly demonstrate that changes in synaptic strength can be quantified by relative changes in GluR1 fluorescence since the fraction of GluR1 containing receptors is different at rest and after plasticity, thus overestimating changes in plasticity. Perhaps Figure 5d and e can be better leveraged to show similar relative changes in EPSC and SEP-GluA1. Adequate statistics should be shown for this point.

4. The following sentences in the abstract should be changed:

• "We used this approach to generate an unprecedentedly detailed spatiotemporal map of synaptic plasticity underlying sensory experience" Change this to "a map of synapses undergoing changes in synaptic strength."

• "More generally, these tools can be used as an optical probe capable of measuring functional synapse strength" This should be changed to identifying synapses undergoing increased synaptic strength.

5. How do the authors estimate the error in synaptic strength estimates contributed by multiple synapses within the extended PSF volume? It is unclear how to differentiate between multiple synapses and one synapse with a high GluA1 content. The PSF resolution is particularly concerning for the high density of synapses observed in the cortex in vivo. The author should state this issue and recommend post hoc immunocytochemical labeling of synaptic markers (e.g., Homer or PSD95).

6. The following statement (page 15) is subjective and should be quantified: "Presumably, this trade-off led to increased false-negative rates, but reduced false-positive rates that would have arisen from incorrectly merging two distinct synapses."

7. Was the PSF performed with 1 μm beads? It seems impossible to have measured a FWHM of 0.57 um.

Essential revisions:

1) The authors should clarify how the synapse detection algorithm account for extrasynaptic GluA1 labeling. Increases in this pool may not directly underlie increases at this synapse. Testing these confounding factors could be performed using physiologically relevant plasticity protocols in acute brain slices rather than cell culture, a preparation in which the neuron type and connectivity may not be the same as the mostly intact circuit.

In neurons, membrane-inserted AMPA receptors are mostly clustered at functional postsynaptic sites on dendritic spines or the dendritic shaft. It is true, however, that a minority population of extrasynaptic AMPA receptors are present along the plasma membrane outside of these synaptic zones. To reduce false-positive detections of extrasynaptic SEP-GluA1 as functional synapses, we employed an automated template-matching approach that was trained on manually annotated synaptic puncta. Detections were filtered based on size in both the XY and Z planes (see description in the methods section). These size exclusion thresholds were chosen based on the known size of AMPA-containing PSDs from EM. We found that this approach performed well in excluding the extrasynaptic SEP-GluA1 signal, which tended to be smaller, more diffuse, and less punctate than synaptically enriched GluA1. We thus believe that our analysis of synaptic SEP-GluA1 levels only has minimal contributions of extrasynaptic GluA1. Nevertheless, this is a crucial point and that we have more clearly emphasized in the manuscript (page 13), detailing how our algorithm was optimized to detect synaptic GluA1 and largely exclude extrasynaptic signals.

2) The extended z-resolution of the 2P point-spread-function (PSF) in vivo will likely falsely assign fluorescence from multiple synapses to a single punctum. How does this limited resolution impact the final interpretations? The study would benefit from an in vivo ground-truth estimate of synapse location to confirm that SEP-GluA1 puncta correspond to an increase in AMPARs at synapses. The use of Homer does not entirely circumvent the uncertainty generated by the poor resolution of 2P. Post-hoc confocal microscopy in fixed tissue using morphology and synaptic markers could be sufficient. If not, then super-resolution fluorescence or immunogold EM are certainly suitable ground-truth approaches.

We agree with the reviewers that due to inherent constraints of in vivo light microscopy, our ability to accurately detect and segment closely neighboring synapses is somewhat limited. This is especially true in the z-plane, where the point spread function (PSF) of 2p imaging limits resolution and detection, far more so than in the XY planes. Throughout our manuscript, we acknowledge that while our endogenous knockin strategy does fluorescently label all GluA1 subunits, we are not able to detect every single GluA1-containing synapse. To quantitatively assess the limitations of in vivo 2p imaging and subsequent synapse detection, we have included measurements of our 2P PSF (see new Figure 7 Supplement 3). We have updated the manuscript to explicitly state these resolution constraints (page 15), as well as the conclusion that we cannot accurately detect and segment synapses that are within these bounds (i.e., that we cannot accurately discern and segment two overlapping synapses in XY that are within 3 μm in Z). We have expanded discussion regarding how these limitations impact our analyses (page 15), since our algorithm would detect and segment these closely abutting synapses as one large synapse, and ultimately discard this “false positive” synapse for being too big (detections that span more than 6 adjacent z-planes are discarded).

3) The quantification of synaptic strength by SEP-GluA1 may be an overestimate because SEP-GluA1 containing AMARs represents a fraction of the total receptor population, and LTP is strongly biased towards GluA1 insertion. How can this potential error be overcome?

We believe that the abundance of GluA1 serves as a valuable proxy for synaptic strength and demonstrate this empirically with the correlation between SEP-GluA1 signal and EPSC amplitude with glutamate uncaging (Figure 5). Regarding a potentially privileged role of GluA1 in LTP, this is why we sought to focus on GluA1 in particular, to aid the identification of synapses where there are plasticity changes. While this lies outside of the scope of this study, we are also working on developing SEP-KI mouse lines for other AMPAR subunits which will ultimately allow us to study differences in subunit specific AMPAR plasticity and how each subunit contributes to changes in the total receptor population. However, we accept the reviewers’ concerns and have modified our discussion (page 17) to highlight that SEP-GluA1 changes might indicate synapses preferentially undergoing plasticity.

4) To verify that SEP-GluA1 can reliably report decreases in synaptic strength, the authors should examine the activity-dependent endocytosis of receptors following LTD pairing protocols, preferably from the same cell type in the brain slices.

AMPARs, including GluA1-containing receptors, mediate both increases and decreases in synapse strength, corresponding to LTP and LTD, respectively (Malenka and Nicoll, Science 1999; Carroll et al., Nat Neuroscience 1999; Shepherd and Huganir, Annu Rev Cell Dev Biol 2007; Huganir and Nicoll, Neuron 2013). We agree with the reviewers that it is important that our SEP-GluA1 mouse can track both increases and decreases in synaptic GluA1 content, since it is well-established that synaptic weakening, such as during LTD or homeostatic down-scaling, depends on removal of synaptic AMPARs. However, we do not believe there is reason to expect that our SEP-GluA1 reporter is not bidirectional. Indeed, in our manuscript, we include biochemical data directly demonstrating that SEP-GluA1 is capable of reporting both increased and decreased AMPAR expression (see homeostatic plasticity experiments in Figure 3a-b). We have expanded discussion of this crucial point (page 19). Additionally, we have included longitudinal imaging data from individual spines from our FRAP experiments which show that in the unbleached controls spines that our SEP-GluA1 KI line can report increases and decreases in spine GluA1 levels at individual synapses across imaging sessions (new Figure 6c). These results are additionally supported by previous studies where bidirectional changes in spine GluA1 levels were observed while imaging sparsely expressed SEP-GluA1 in vivo, (Roth et al., 2020; Tan et al., 2020). We performed the slice LTP experiments (Figure 3), as well as the 2p uncaging experiments (Figure 5) as a control, to demonstrate that Hebbian plasticity is not impaired in our knockin and to directly demonstrate that SEP intensity directly correlates with functional synaptic strength. We thus believe that the proposed slice LTD experiments would not add anything further to demonstrate the capacity of SEP-GluA1 to report bidirectional changes in synaptic AMPAR content.

5) The choice to fluorescently label GluA1 subunits of the AMPAR to mark plasticity is, on the one hand, an obvious choice because of its implication in LTP. However, evidence in the literature suggests that LTP in the somatosensory cortex can be GluA1-independent (e.g., Frey et al., 2009). The authors should discuss these studies' implications and the limits of their method for addressing synaptic mechanisms for learning and memory. Does SEP-GluA1 fluorescence reproduce previous estimates of the distribution of synaptic weights? The SEP-GluA1 method is undoubtedly more attractive than the post hoc patch-clamp-based approaches performed by Ko et al. 2011, 2013 (Hofer and Mrsic-Flogel laboratories), but SEP-GluA1 should be compared to spine intensity (volume) imaging for example.

We agree with the reviewer that our SEP-GluA1 KI mouse line can only track plasticity changes involving GluA1. We have now modified our discussion to highlight that SEP-GluA1 changes might indicate synapses preferentially undergoing plasticity and that different types of plasticity might involve GluA1 to a different degree (page 20). Throughout our manuscript, we are clear that our SEP-GluA1 knockin line enables tracking of GluA1 expression and not other AMPAR subunits. This is not to say that only GluA1 expression is changing during the behaviors and experimental manipulations in this study, nor that GluA1-mediated synaptic plasticity is the sole mechanism of behavioral learning, in the sensorimotor cortex or anywhere else in the brain. Indeed, different forms of plasticity can differentially rely on different AMPAR subunits and we are developing SEP-tagged mouse lines for other AMPAR subunits that can be used to distinguish these in the future. However, as mentioned above in Point 3, we believe that the abundance of GluA1 serves as a valuable proxy for synaptic strength and we demonstrate this here with the correlation between SEP-GluA1 signal and EPSC amplitude with glutamate uncaging (Figure 5). Frey et al. (2009) show that LTP in somatosensory cortex can be induced in the absence of GluA1, however this does not mean that GluA1 levels are not increased during LTP in WT mice. Our lab’s previous work has demonstrated that whisker stimulation can indeed induce NMDA-receptor-dependent LTP that involves insertion of GluA1 to dendritic spines in vivo (Zhang et al., 2015).

Regarding the comparison of SEP-GluA1 and spine volume, we believe that tracking SEP-GluA1 is a more accurate readout of synaptic strength than spine volume since the strength is mediated by receptors rather than spine size. Indeed, while spine size generally correlates with SEP-GluA1 intensity, spine size does not always change during plasticity (Figure 3A and B in Zhang et al., 2015) or changes to a lesser degree than SEP-GluA1 (Tan et al., 2020, Roth et al., 2020). We even observed a population of spines that exhibit a divergence of spine form and function, with spine SEP-GluA1 increasing but spine size decreasing and vice versa (Tan et al., 2020). The dissociation of spine size and synaptic strength has been reported many times (Lee et al., 2012). For instance, spine number or volume is not changed at all at cerebellar Purkinje cell synapses during LTD (Sdrulla and Linden, 2007) and Insulin-induced endocytosis of AMPARs is not accompanied by spine shrinkage (Wang et al., 2007). Thus, spine size, in certain conditions, is not a good indication of synaptic strength, whereas our imaging of synaptic AMPAR expression provides a direct and accurate way to monitor functional changes at synapses.

6) Conditional expression of the SEP-GluA1 would allow the precise dissection of synaptic plasticity specific neuron types within microcircuits, thereby linking plasticity with altered circuit function and behavior. This limitation must be addressed, in detail, in the Discussion. The current data sets of whole-brain expression of SEP-GluA1 do not distinguish between layer 2/3 pyramidal neurons, layer 5 pyramidal neurons, or interneurons. It is interesting to consider the possibility that a cross between the SEP-GluA1 transgenic mouse and a PSD-95 YPF-Enabled (Fortin et al., 2014) mouse could be used to assess the GluA1 changes at specific synapse types.

We agree that a Cre-conditional SEP-GluA1 line – and, for that matter, conditional versions of all AMPAR subunit lines – would be valuable tools to have and would enable important cell-type-specific investigations into synaptic plasticity in vivo. This is the first paper demonstrating that a SEP knockin mouse line can be used to track synaptic AMPARs in general. We are currently trying to generate these conditional lines and future studies by our lab and others will continue to advance this line of inquiry. However, designing a conditional knockin at the N-terminus of an endogenous protein is highly challenging. While it might be hard to separate YFP fluorescence of the PSD-95 ENABLED mouse from our SEP-GluA1 fluorescence, the idea of combining a conditionally expressed cell or synaptic marker in a different color is a possible approach to achieve cell-type specificity in our global SEP-GluA1. Such a marker could be a red PSD-95 intrabody (Gross et al., 2013) or a tdTomato cell fill. In this study we have combined the SEP-GluA1 mouse with the Ai9 tdTomato reporter mouse and injected AAV-CaMKII-cre virus in L2/3 pyramidal neurons. This approach allowed us to specifically associate SEP-GluA1 synapses to dendrites of cre expressing neurons (Figure 7 c-d) and target these for FRAP (Figure 6). These results demonstrate that it is feasible to combine our SEP-GluA1 knockin line with a mouse line expressing tdTomato in the cell-type of interest and specifically quantify changes in spine GluA1 expression of these neurons. We have expanded our discussion and emphasized that combining our global SEP-GluA1 knockin with conditionally expressed red cell or synaptic markers is a viable approach to achieve cell-type specificity (page 21).

7) Using Western Blot, the authors show a 57% reduction of GluA1 expression at the postsynaptic density relative to WT, however, the potential implications of this are not addressed. In figure 1f, they show relative AMPA levels resulting from their Western blot analysis. Still, they do not adequately describe how the Western blot and analysis were performed, and namely, what control they used to normalize protein levels between different fractions. It would be nice to see additional statistical analyses of GluA2 and GluA3 expression relative to WT as there seems to be an increase and, therefore, potential compensation by GluA3 subunits SEP-GluA1 KI.

We do not know why the mRNA and GluA1 is lower than WT in these mice. It is likely to be due to RNA instability of the KI transcript. We will make attempts in the future to limit this issue. However, this decrease in GluA1 expression does not affect synapse function and plasticity. We are sorry we did not describe the methods and quantitation of the western blots in more detail. Equal amounts of protein were run in each lane and PSD95 was used as internal control and GluA1 levels were normalized to PSD95 for both P2 and PSD analysis. There was no significant difference in PSD95 level between WT and SEP-GluA1 KI. In addition, we used the raw data without any normalization with PSD95 for analysis and observed similar changes: 43.45% reduction in GluA1 after normalization to PSD95 v.s. 43.31% reduction with normalization to the amount of protein loaded (PSD fraction, Author response image 1). Regarding GluA2 and GluA3 expression, we saw a trend of increase (15.77% increase in GluA2 and 25.61% increase in GluA3 at PSD after normalization to PSD95) although the increases were not statistically significant (Author response image 1) . It is possible that there is a compensatory increase in GluA2/GluA3 subunits in SEP-GluA1 KI animals, however, our statistical analysis does not support this claim. We have included the new statistical analysis in the revised manuscript and updated our discussion of these results accordingly (page 6).

Author response image 1

Download asset Open asset

8) In the FRAP experiment in Figure 6, it seems like there could be other interpretations for the time course of recovery of fluorescence. The expected timescales of fluorescence recovery and caveats of its interpretations should be discussed.

We have clarified our interpretation of our FRAP experiments and expanded our discussion about the time course of fluorescence recovery (page 11). Here is the new text:

“We also found that the SEP-GluA1 signal recovered in two phases after photobleaching (Figure 6b). […] Overall, these results confirm that SEP-GluA1-containing synapses are mobile and present similar dynamics as other in vitro and in vivo systems, supporting that our knockin labelling strategy does not perturb normal synaptic dynamics or function.”

9) An essential aspect of plasticity in the brain is that it can occur on long time scale on precise synapses. Here the authors do not show that it is possible to do longitudinal studies in which individual synapses are tracked. The authors need to provide some data to allow evaluating if this this is actually possible with their method. e.g. with a cytoplasmic marker for spines.

We agree with the reviewers regarding the importance of tracking individual synapses over time to understand mechanisms of synaptic plasticity. To address this point, we have included additional imaging data that shows the capacity of our approach to track longitudinal changes in spine GluA1 levels at individual synapses across multiple days (new Figure 6c). Moreover, representative images of individual spines across imaging sessions are shown in figure 6a. In these experiments, a tdTomato cell fill was expressed in a sparse population of neurons to facilitate identification of the same dendritic spines at different time points. This demonstrates that, at least with the addition of a cytoplasmic marker as the reviewer suggests, it is possible to track individual spines across imaging sessions that span days. Our updated discussion of these new data appears on page 11.

10) Comparison of their synapse detection algorithm with concurrent methods is necessary to establish it as a new tool.

We have greatly expanded discussion of our automatic synapse detection algorithm, both in terms of how it was constructed (explicitly listing the rules for synapse detection, on page 13) and how it is differentiated from other similar tools (page 22). Here is the new text:

“Recently, deep-learning-based systems have achieved state-of-the-art performance in analyzing microscopy images (Moen et al., 2019). […] We found that none of these approaches performed as well as our rules-based, template-matching approach to automatically detect SEP-GluA1-containing synapses.”

Reviewer #1 (Recommendations for the authors):

1) Better discuss the limits of synapse identification on images. The limits are related both to the algorithm used and to 2-phoron imaging resolution limits. Thus better identification could be obtained maybe with super resolution methods. Ideally providing ground truth data would be an asset.

We agree with the reviewer that accuracy of our synapse identification is largely limited by the resolution of two-photon microscopy, and by extension how our automatic synapse detection algorithm is similarly limited by these fundamental constraints. We have now quantified the limits of our synapse detection by measuring the point spread function (PSF) of our microscope and discuss the parameters of our detection algorithm in more detail. We further agree with the reviewer that in vivo super resolution imaging would improve synapse detection, especially in the X and Y dimension, and can be used as an approach in experiments where a more detailed look at individual synapses and subsynaptic structures is necessary. For further details please see our response to this point under “Essential Revisions Point 2”.

2) An essential aspect of plasticity in the brain is that it can occur on long time scale on precise synapses. Here the authors do not show that it is possible to do longitudinal studies in which individual synapses are tracked. I recommend to show some data that it is actually possible with their method, e.g. with a cytoplasmic marker for spines.

Please see our response to this point under “Essential Revisions Point 9”.

3) Comparison with previous methods (e.g. spine volume imaging) is necessary to establish this tool as a useful to track synaptic plasticity. This could be done in the whisker-based plasticity assay.

Please see our response to this point under “Essential Revisions Point 5”.

Reviewer #2 (Recommendations for the authors):

This is an important paper with well carried-out experiments. The conclusions are well supported by data. I have however 3 concerns (that do not dampen my enthusiasm):

1) The global expression of GluA2 is twice as low as the expression of GluA1 in wild-type mice, although the level of GluA2 remains unchanged. How would this affect GluA1/Glu2 heteromers? In addition, can the authors exclude the possibility that the decrease of SEP-GluA1 is brain region specific? (e.g. different levels of SEP-GluA1 between the hippocampus and the cortex). if the decrease is indeed region-specific, could the author exclude that physiological deficit does not occur in other brain region? At best, the authors should perform a better comparison of GluA1 and SEP-GluA1 between brain regions. A least, these point should be discussed. In the same line, bidirectional changes after Bic and TTX treatment in culture are really convincing (Figure 3B). However, it would be fair to present also the non-normalized level of surface AMPAR after treatment as compared to WT.

We agree with the reviewer that it is important to understand region-specific levels of SEP-GluA1 expression in our knockin mouse and how the reduction in GluA1 can affect AMPAR subunit compositions. Our Western blot data unfortunately cannot distinguish GluA1/GluA2 heteromers from GluA1 homomers. Indeed, it is technically challenging to specifically examine GluA1/GluA2 heteromers. In our experiments, we saw a trend of an increase in GluA2 and GluA3 levels at PSD of SEP-GluA1 KI mice compared with WT mice although the increases were not statistically significant (see our response to this point under “Essential Revisions Point 7”). Therefore, it is possible that there is a reduction in GluA1/GluA2 heteromers accompanied by an elevation of GluA2/GluA3 heteromers in SEP-GluA1 KI mice.

We have included new comparisons of SEP-GluA expression to patterns of GluA1 mRNA level in WT mice. These atlases compare fluorescent SEP-GluA1 expression throughout the brain, with aligned references from Allen Brain Atlas (new Figure 1 Supplement 1).

Regarding the region-specific levels of SEP-GluA1 expression, we don’t believe that there are any differences between brain regions. We observed a reduction in GluA1 level in hippocampus (Figure 1e-f). We observed a similar reduction in surface SEP-GluA1 level in cultured cortical neurons of KI pups compared to WT (Figure 3a). Additionally, we performed new experiments using Western blots from either whole brain or specific brain regions and observed a similar pattern of decrease of total GluA1 in KI mice, irrespective of brain region (new Figure 1 Supplement 2). Therefore, we believe that the expression level of GluA1 in SEP-GluA1 mice is conserved across brain regions.

Regarding Figure 3b, we think it is more visually representative of the similarity of the homeostatic plasticity between WT and the KI by normalizing the data.

2) Although the authors present strong and convincing evidence that the high density of SEP-GluA1 synapses does not preclude the report of synaptic plasticity in anesthetized mice, it is much less clear whether it would work also in awake mice.

Our previous studies have shown that we can longitudinally image dendritic/spine SEP-GluA1 to study AMPAR dynamics both in anesthetized and head-fixed awake mice during different kinds of behaviors with the same set-up (Zhang et al., 2015; Roth et al., 2020; Tan et al., 2020). Our ability to resolve and track individual dendritic spines was identical between these two groups. Therefore, we are confident that imaging awake SEP-GluA1 mice is no different from imaging anesthetized mice.

3) How effective is their supervised detecting method as compared to existing computer-based detection methods ? (e.g. see figure 7 from: https://www.biorxiv.org/content/10.1101/061507v2.full.pdf). Also, how robust is the supervised annotation of spines when the same FOV is imaged chronically (between days)?

We agree with the reviewer that chronic imaging of the same FOV across days is essential for understanding synaptic plasticity during learning. While currently our machine-learning-based detection algorithm is focused on detecting synapses on a single imaging session, in future studies, we will expand on our approach to align, register, and track millions of individual synapses across days. Currently, we can achieve reliable longitudinal imaging by including cytosolic tdTomato in a sparse population of neurons to facilitate the identification of the same dendritic spines at the different time points. We have now included additional imaging data that shows the capacity of our approach to track longitudinal changes in spine GluA1 levels at individual synapses across multiple imaging sessions (Author response image 1 and new Figure 6c). Moreover, representative images of individual spines across imaging sessions are shown in figure 6a. This demonstrates that, at least with the addition of a cytoplasmic marker, it is possible to track individual spines across imaging sessions that span days. As suggested by the reviewer, we have also greatly expanded discussion of our automatic synapse detection algorithm, both in terms of how it was constructed and how it is differentiated from other similar tools. Please see our response to this point under “Essential Revisions Point 10”.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

1. There is still no convincing data that spines can be tracked in vivo across a long time interval (e.g. 1 day). The authors show a picture with sparse Td-tomato labelling, but only on a single time point. It is necessary to show the same picture at a later time point to appreciate the degree of consistency of SEP-GluA1 labelling across days. The sentence in the abstract, "we can longitudinally track how the strength of synapses changes during behavior," is currently not fully supported by the data.

We agree with the reviewer that our data do not support a claim that we are able to track individual, registered synapses over time. Rather, our data clearly support that we can track how populations of synapses change during sensory stimulation. While we are able to detect hundreds of thousands of individual synapses, longitudinal registration and tracking of each of these synapses is beyond the scope of this study. We have updated the abstract, results, and discussion to explain our approach and the efficacy of our tools more clearly.

2. The authors did not compare the efficacy of their automated synapse segmentation methods to other methods. The authors state in the discussion that their method outperforms other methods but this is not documented. This statement requires a quantification.

We have updated our discussion of other automated synapse detections methods, removing any claims that our tools outperform other approaches. The reviewers are correct that we do not have the data to make this claim.

3. The authors do not convincingly demonstrate that changes in synaptic strength can be quantified by relative changes in GluR1 fluorescence since the fraction of GluR1 containing receptors is different at rest and after plasticity, thus overestimating changes in plasticity. Perhaps Figure 5d and e can be better leveraged to show similar relative changes in EPSC and SEP-GluA1. Adequate statistics should be shown for this point.

It is not clear why this reviewer states that the fraction of GluA1 containing receptors change during plasticity. This concept is not accepted by many people in the field and has certainly has not been shown in vivo. Our data support that: (1) SEP-GluA1 intensity is strongly correlated with functional synaptic strength in vitro (Figure 5c), (2) LTP in vitro is associated with increased SEP-GluA1 intensity only in spines that received the plasticity-inducing stimulus (Figure 5d-f), and (3) that SEP-GluA1 intensity and EPSC amplitude are also strongly correlated after plasticity (Figure 5 Supp. 1). To more fully support this final point, we include new plots showing this correlation after LTP in Figure 5 Supplement 1. We feel that these data and statistical analyses fully address this point.

4. the following sentences in the abstract should be changed:

• "We used this approach to generate an unprecedentedly detailed spatiotemporal map of synaptic plasticity underlying sensory experience" Change this to "a map of synapses undergoing changes in synaptic strength."

• "More generally, these tools can be used as an optical probe capable of measuring functional synapse strength" This should be changed to identifying synapses undergoing increased synaptic strength.

We have changed the wording of the abstract based on the first point as suggested. However, we disagree with the second point. Firstly, as we said in the first round of reviewer comments (Essential Revision #4), our methods allow us to directly observe both increases and decreases in synaptic strength in vivo, not just increased synaptic strength, as this comment suggests. Indeed, we present data that demonstrate we can track both increases and decreases in synaptic strength over time (Figure 6). Further, given our glutamate uncaging data in Figure 5c, our results strongly support the claim that in vivo imaging of this SEP-GluA1 mouse “can be used as an optical probe capable of measuring functional synapse strength”. Thus, we feel strongly that this sentence should remain, unchanged.

5. How do the authors estimate the error in synaptic strength estimates contributed by multiple synapses within the extended PSF volume? It is unclear how to differentiate between multiple synapses and one synapse with a high GluA1 content. The PSF resolution is particularly concerning for the high density of synapses observed in the cortex in vivo. The author should state this issue and recommend post hoc immunocytochemical labeling of synaptic markers (e.g., Homer or PSD95).

We discuss this issue and describe our rules for defining synapses in a numbered list are clearly stated on page 13. Fluorescent puncta that obey all these rules are considered synapses. These parameters were carefully chosen, based on published ranges of synapse size, as well as our experience imaging labeled synaptic proteins in vitro and in vivo. The reviewer is correct that inherent limitations of light microscopy—especially the comparatively broad axial PSF—complicates detection and segmentation of densely labeled synapses, and could theoretically lead to cases where two closely abutting synapses are erroneously detected and segmented as one large synapse. To minimize this confound, we excluded synapses that spanned more than 6 adjacent z-sections (rule #8). We state this in the text (pages 13 and 16).

Based on these comments, we have also substantially expanded (pages 16-17) discussion of the potential pitfalls of detecting densely labeled synapses, and how we think our data support that we’re not making a significant number of merge errors. Briefly, if we were making a significant number of merge errors – erroneously merging two or more closely abutting synapses into a single automatic detection – the we should see additional peaks in histograms of synapse intensity (e.g., peaks at 2x and 3x of max intensity, arising from detections containing 2 or 3 synapses). We don’t see any evidence of this in our data. Rather, distributions of automatically detected synapses are strikingly normal, supporting that an overwhelming majority of our detected synapses correspond to single synapses, despite their high density and the poor axial PSF inherent to 2p imaging. However, we have added a statement in the updated manuscript that in future studies using post hoc immunocytochemical labeling of synaptic markers (e.g., Homer or PSD95) could be useful.

6. The following statement (page 15) is subjective and should be quantified: "Presumably, this trade-off led to increased false-negative rates, but reduced false-positive rates that would have arisen from incorrectly merging two distinct synapses."

We have updated the manuscript to more clearly state our rationale for excluding two (or more) closely opposed synapses that could not be adequately segmented. We also removed the subjectively worded sentence flagged by the reviewers. Many of these points are addressed in Point #5, above.

7. Was the PSF performed with 1 μm beads? It seems impossible to have measured a FWHM of 0.57 um.

The PSF we present is a "Reconstructed PSF from averaged two-photon images of 1μm TetraSpeck microspheres", which is computed through an inverse deconvolution of raw images following established methods (Padman, 2014; Schmied et al. 2016; Verveer et al. 2020). This process is necessary to extract the PSF from finite (non-zero) sized beads, as using the raw bead images directly will give an overestimation of the PSF. This point is clarified in the Results section of the updated manuscript (pages 15-16).

Author details

1. Austin R Graves

   1. Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States
   2. Kavli Neuroscience Discovery Institute, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Validation, Visualization, Writing - original draft, Writing - review and editing

   Contributed equally with

   Richard H Roth, Han L Tan, Qianwen Zhu, Alexei M Bygrave and Elena Lopez-Ortega

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1087-3684

2. Richard H Roth

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Validation, Visualization, Writing - review and editing

   Contributed equally with

   Austin R Graves, Han L Tan, Qianwen Zhu, Alexei M Bygrave and Elena Lopez-Ortega

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-6855-999X

3. Han L Tan

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Validation, Visualization, Writing - review and editing

   Contributed equally with

   Austin R Graves, Richard H Roth, Qianwen Zhu, Alexei M Bygrave and Elena Lopez-Ortega

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-5163-7720

4. Qianwen Zhu

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Validation, Visualization, Writing - review and editing

   Contributed equally with

   Austin R Graves, Richard H Roth, Han L Tan, Alexei M Bygrave and Elena Lopez-Ortega

   Competing interests

   No competing interests declared

5. Alexei M Bygrave

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Validation, Visualization, Writing - review and editing

   Contributed equally with

   Austin R Graves, Richard H Roth, Han L Tan, Qianwen Zhu and Elena Lopez-Ortega

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-2291-923X

6. Elena Lopez-Ortega

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Validation, Visualization, Writing - review and editing

   Contributed equally with

   Austin R Graves, Richard H Roth, Han L Tan, Qianwen Zhu and Alexei M Bygrave

   Competing interests

   No competing interests declared

7. Ingie Hong

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Resources, Software, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-7246-9233

8. Alina C Spiegel

   1. Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States
   2. Kavli Neuroscience Discovery Institute, Baltimore, United States

   Contribution

   Formal analysis, Investigation, Visualization, Writing - review and editing

   Competing interests

   No competing interests declared

9. Richard C Johnson

   Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States

   Contribution

   Methodology, Resources, Validation

   Competing interests

   No competing interests declared

10. Joshua T Vogelstein

    1. Kavli Neuroscience Discovery Institute, Baltimore, United States
    2. Center for Imaging Science, Johns Hopkins University School of Engineering, Baltimore, United States
    3. Department of Biomedical Engineering, Johns Hopkins University, Baltimore, United States

    Contribution

    Funding acquisition, Supervision

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0003-2487-6237

11. Daniel J Tward

    1. Kavli Neuroscience Discovery Institute, Baltimore, United States
    2. Center for Imaging Science, Johns Hopkins University School of Engineering, Baltimore, United States
    3. Department of Biomedical Engineering, Johns Hopkins University, Baltimore, United States

    Contribution

    Formal analysis, Methodology, Resources, Software, Visualization, Writing - review and editing

    Competing interests

    No competing interests declared

12. Michael I Miller

    1. Kavli Neuroscience Discovery Institute, Baltimore, United States
    2. Center for Imaging Science, Johns Hopkins University School of Engineering, Baltimore, United States
    3. Department of Biomedical Engineering, Johns Hopkins University, Baltimore, United States

    Contribution

    Funding acquisition, Supervision

    Competing interests

    Dr. Miller is a joint owner of AnatomyWorks. Dr. Miller's relationship with Anatomy-Works is being handled under full disclosure by the Johns Hopkins University

13. Richard L Huganir

    1. Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, United States
    2. Kavli Neuroscience Discovery Institute, Baltimore, United States

    Contribution

    Conceptualization, Funding acquisition, Project administration, Supervision, Writing - review and editing

    For correspondence

    rhuganir@jhmi.edu

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0001-9783-5183

• 6,922

  Page views

• 937

  Downloads

• 16

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.