Relationship between bound Ca²⁺ and cellular structure

To examine the relationship between bound Ca²⁺ and synaptic function, we prepared samples of hippocampal neurons for correlative fluorescence imaging and NanoSIMS experiments. These samples were obtained using primary hippocampal cultures, following a classical protocol (Banker & Cowan, 1977) that dissociates the hippocampi of newborn rats. The resulting culture is a mixture of glia and neurons, with the majority of the neurons being glutamatergic (>90%) (Benson et al, 1994). Before fluorescent labeling, available coverslips were divided into two groups: one group to examine synaptic size and another to examine synaptic activity. Neurons from both groups were fixed and immunostained after 18 d in culture (DIV18), followed by plastic embedding and ultrathin sectioning.

Because the ions measured during a single NanoSIMS run are dependent on the source used (Nuñez et al, 2017; Bonnin & Rizzoli, 2020) and because ⁴⁰Ca does not always clearly show cell morphology, samples were first analyzed using the Cs⁺ ion source to image ¹⁴N, ³¹P, and ³²S, which represent the organic molecules composing the cells (e.g., proteins and nucleic acids). After this initial imaging step, the instrument was then switched to the O⁻ ion source to image ⁴⁰Ca (Fig 1A). ³⁹K and ²³Na were also imaged using the O⁻ ion source, as they are ubiquitous where cellular material is present (Fig S1A and B).

Comparing ⁴⁰Ca with ¹⁴N and ³²S in cellular regions (defined by their high ¹⁴N and ³²S signals) reveal two different groups, both positively correlated to ⁴⁰Ca (Fig 1B and C). One group, characterized by average ⁴⁰Ca counts >2 per ROI, makes up roughly 18.3% of all data points analyzed. A second group, characterized by average ⁴⁰Ca counts <2 per ROI, makes up the remaining ∼81.6% of our samples. Although ⁴⁰Ca is correlated to ¹⁴N and ³²S in both groups, this correlation has a much steeper slope in the former group than in the latter. The existence of these two distinct groups of signals is expected, as Ca²⁺-binding proteins are not thought to be present at equal levels in all cellular regions. Most cellular areas are basically devoid of Ca²⁺, whereas a subset, presumably the areas containing Ca²⁺-binding proteins, shows a strong correlation between this ion and the local protein amounts.

No significant correlation between bound and free Ca²⁺

To test whether a relationship exists between the bound and free Ca²⁺ pools, we performed live imaging of the free Ca²⁺ pool using Neuroburst (Sartorius), a genetically encoded calcium indicator. A separate batch of hippocampal neurons were transfected with the Neuroburst label at DIV10 and transferred to an imaging chamber at DIV17. The fluorescence intensity of the Neuroburst label was recorded for 2 min to capture spontaneous activity. After this, cells underwent field stimulation at 20 Hz for 3 s, followed by a recovery phase and renewed stimulation, for 30 s. Cells were then treated with a 5-µM ionomycin solution to measure the maximum possible fluorescence intensity, F_max (Fig 2A).

The resulting movies show the normal neuronal activity of the cultures, in the form of bursts of Ca²⁺ fluorescence. These represent high-frequency trains of action potentials that occur spontaneously at intervals of a few seconds. We recorded the following parameters from each observed neuron: peak intensity, or average peak amplitude during the 5-min recording; peak interval, or the frequency of subcomponent peaks; burst length, or the average width of burst events; burst interval, the frequency of bursts over the 5-min recording period; total intensity, or the summed amplitude of all peaks over the course of a single burst; maximum intensity normalized to initial intensity, and the average intensities of both stimulus peaks (Fig 2K).

Each imaging location was marked, and samples were embedded and processed for NanoSIMS analysis (Fig 2B and E). After NanoSIMS analysis, which was conducted in the same manner as in the previously described experiment (Fig 1), NanoSIMS images and live images were matched. The average intensity of the ⁴⁰Ca peak was recorded for each cell. The resulting bound Ca²⁺ intensity was then compared with the free Ca²⁺ parameters described above (Fig 2C–L). There did not appear to be any significant correlation between the amount of bound Ca²⁺ and any parameter of the Ca²⁺ fluorescence (Fig 2). This suggests that the free Ca²⁺ fluorescence-derived measurements are not strongly correlated to the bound Ca²⁺, at the whole-cell level.

Bound Ca²⁺ turnover is slow in both neuronal and nonneuronal cells

Having found that bound Ca²⁺ does not correlate strongly to the dynamics of free Ca²⁺, we wondered whether this lack of a relationship is due to bound Ca²⁺ changing very rapidly. In principle, bound Ca²⁺ could be a transient parameter. Neuronal activity, which takes place every few seconds in culture, results in abundant Ca²⁺ entry into the cells, conceivably leading to gradual changes in the bound Ca²⁺ pool. To tackle this issue, we decided to measure the bound Ca²⁺ turnover, meaning the rate of exchange of these ions.

We pulsed cells with isotopic calcium, ⁴⁶Ca²⁺, and measured the appearance of this isotope in the bound Ca²⁺ pool. This process was very slow for fibroblasts (Human embryonic kidney-293 [HEK293] cells), only reaching, after 3 d of pulsing, ∼4% of the maximum possible isotopic labeling (Fig S2A–C). This is remarkable, because the fibroblasts also divided during this time interval, and therefore needed to produce new calcium-binding proteins.

Ca²⁺ turnover was also slow in neurons, reaching ∼20% of the maximum labeling after 10 d of pulsing (Fig S3A–C). This implies that, in spite of rapid dynamics for free Ca²⁺, the bound population is remarkably stable, both in neurons and in rapidly growing cells, as fibroblasts. This is all the more surprising, because the lifetimes of Ca²⁺ buffer proteins as calbindin, calretinin, and parvalbumin are not particularly long (close to the median of all neuronal proteins, [Fornasiero et al, 2018]).

The intensities of markers for inhibitory pre-synapses and excitatory post-synapses correlate to bound Ca²⁺

To examine the relationship between bound Ca²⁺ and pre- and post-synaptic parameters in neurons, we immunostained cultured hippocampal neurons for protein markers of the pre- and post-synapses. To analyze the pre-synapse, we used the vesicular glutamate transporter-1 (VGLUT1) and the vesicular GABA transporter (VGAT) as markers which are located in the pre-synapses of glutamatergic (excitatory) and GABAergic (inhibitory) neurons, respectively. As these markers label the pre-synaptic vesicle pool, their fluorescence intensity can be used to estimate the size of the respective synapses (Chaudhry et al, 1998; Vigneault et al, 2015). To image the post-synapse, we used post-synaptic density protein-95 (PSD95) as a marker of excitatory synapses (Ehrlich and Malinow, 2004). Immunostaining was conducted on fixed cells, and NanoSIMS analyses followed fluorescent imaging (Fig 3A and B, additional images in Fig S4). The same image regions were analyzed in the two techniques. As the axial resolution of the NanoSIMS instrument is around 10–20 nm in our experiments (Saka et al, 2014), this procedure does not lead to “mixed” measurements of multiple overlapping cells. In other words, when measuring from a region marked by neuronal or synaptic proteins, we are certain to provide ionic values from the respective neuronal cell. In total, ∼60 neurons were imaged using these techniques. From these 60 neurons, 130 regions of interest were selected for VGLUT and Ca²⁺ analysis (Fig 3C), 104 regions of interest were selected for PSD95 and Ca²⁺ analysis (Fig 3D), and 126 regions of interest were selected for VGAT and Ca²⁺ analysis (Fig 3E).

Comparing the intensities of these fluorescent markers in our samples with the intensity of the ⁴⁰Ca peak showed no significant correlation between bound Ca²⁺ and the VGLUT intensity (Fig 3C). A positive correlation was present between PSD95 intensity and bound Ca²⁺, albeit with a very limited R² value (Fig 3D). This correlation is likely artificially weakened by signal mixing, as the physical distance between the post-synapse and the glutamatergic pre-synapse is smaller than our image resolution, and the glutamatergic pre-synapse does not appear to correlate with bound Ca². The VGAT intensity is somewhat more strongly correlated to the amount of bound Ca²⁺ (Fig 3E). This implies that the synapse size may correlate to the amount of Ca-buffering proteins, but only for the inhibitory pre-synapse and, to a lower degree, for the excitatory post-synapse. No significant correlations were found between these parameters and either ²³Na or ³⁹K (Fig S5A–C).

Synaptic activity is inversely correlated to bound Ca²⁺

The experiments presented above do not provide any functional parameters that can be correlated to the size of the bound Ca²⁺ pool. To test this, we analyzed synaptic vesicle recycling. The behavior of synaptic vesicles was analyzed by conducting an initial 1-h immunostaining of live cells with antibodies targeting synaptotagmin 1 (Syt1), a vesicular calcium sensor (Kraszewski et al, 1995). During synaptic recycling, these antibodies are taken up by synaptic vesicles. This effectively labels any vesicle active during this 1-h time period (Sertel et al, 2021b), thereby measuring the size of the “recycling pool” of synaptic vesicles, which is a measure of the long-term activity capacity of the respective synapses (Truckenbrodt et al, 2018). After this staining was completed, cells were labeled for an additional 15 min with a fluorescent anti-mouse nanobody (NB), which identifies the Syt1 antibodies within the vesicles that are active during this shorter period, and thus provides a measure of the short-term activity of the respective synapses. The neurons were then fixed and further immunostained using a marker for synaptophysin (Syph), which labels the entire vesicle pool (Takamori et al, 2006) (Fig 4A). As with samples shown in Fig 3, cells were then embedded in resin and analyzed using both fluorescent microscopy and NanoSIMS (Fig 4B, additional images in Fig S6). In total, ∼171 neurons were imaged using these techniques. From these neurons, 761 regions of interest were selected. The intensity of all three fluorescent markers were analyzed at each region of interest and compared with Ca²⁺.

Comparisons between the intensity of these fluorescent markers and the intensity of the ⁴⁰Ca peak show that Syt1 labeling appears inversely correlated to bound Ca²⁺, suggesting that synaptic activity decreases when bound Ca²⁺ is high (Fig 4C). This result is initially surprising because Syt1 is a Ca²⁺ sensor, which enables exocytosis in the presence of high Ca²⁺ concentrations (Radhakrishnan et al, 2009; Jackman & Regehr, 2017). However, once free Ca²⁺ becomes bound, it is no longer available to participate in biological reactions, implying that increasing the size of the bound Ca²⁺ pool would decrease the amount of free Ca²⁺ available for synaptic reactions, thereby reducing the long-term activity of the synapses.

We similarly observed a negative correlation between the amount of bound Ca²⁺ at a given location and the intensity of the fluorescent nanobody (Fig 4D), suggesting that the short-term activity of the synapses is also regulated by bound Ca²⁺. We did not observe any significant correlations between Syt1, the fluorescent nanobody and either ²³Na or ³⁹K (Fig S7A and B). Similarly, no correlation was observed between bound Ca²⁺ and synaptophysin (Fig 4E), implying that these relationships are not reflective of the size of the vesicle pool. Some correlations could appear in this type of correlative imaging, simply because of biological material containing both proteins and metal ions. However, such correlations would be positive, because higher amounts of biological material would imply higher immunostaining levels and higher ion concentrations at the respective location (see Fig S7C). The negative correlations observed for Syt1 are difficult to explain by such effects.

Although the correlations between the Syt1 and NB labeling and bound Ca²⁺ have low R² values (Fig 4C and D), they are highly significant statistically, and are visible even on small subsets of the analyzed data. Thus, although variation in bound Ca²⁺ does not account for a substantial proportion of the dynamics of synaptic activity, it is nonetheless a factor that probably influences the synapse, as also explored in the following section.

Computational model supports relation between size of bound Ca²⁺ pools and synaptic activity

To investigate the inverse correlation between vesicle activity and bound Ca²⁺, we devised a computational model of calcium binding and coupled it with the canonical model of short-term plasticity (Tsodyks et al, 1998) (Fig 5A and B). In this model, the normalized size of the readily releasable pool of vesicles (x) and the release probability (u) are activity-dependent, as described by the following equations:dx(t)dt=1−xτd−uxR,(1)du(t)dt=U−uτf+U(1−u)R,(2)w(t)=u(t)x(t).(3)

Here, R denotes the stimulation frequency, and τ_f and τ_d specify the synaptic facilitation and depression timescales, respectively. Equation (3) defines the time-dependent normalized synaptic weight. A schematic description is provided in Fig 5A (right). In the standard model of short-term plasticity (Tsodyks et al, 1998), the basal exocytosis probability U enters as a free parameter that encapsulates the dependence of vesicle exocytosis rate on the intracellular calcium concentration. Thus, intracellular changes in free Ca²⁺ abundance because of, for example, transitions between bound and free states modulate U. In extreme cases, lack and surplus of free Ca²⁺ both saturate U to 0 or 1, respectively. Therefore, we modeled the link between basal exocytosis probability of vesicles and free Ca²⁺ by a sigmoid that reflects such plateaus:U(c)=21+exp(−c/λ)−1.(4)Here, λ is a scaling factor and c denotes the abundance of intracellular free Ca²⁺. Within the pre-synapse, free Ca²⁺ can become bound, as described in the following transitional model:dc(t)dt=−cτc−λcb(B−b)c+λbcb+μR,(5)db(t)dt=−bτb+λcb(B−b)c−λbcb.(6)In these equations, b specifies the amount of bound calcium in the pre-synapse and B describes the total binding capacity in the pre-synapse or buffer size. These equations read as follows: in general, both the amount of free and bound calcium decays, respectively, with timescales τ_c and τ_b. Furthermore, free Ca²⁺ can bind to B–b available free buffer slots with a rate of λ_cb. Similarly, bound Ca²⁺ may separate from buffer proteins, becoming free Ca²⁺ proportional to a rate of λ_bc. In addition, the number of input spikes R is assumed to linearly increase the level of free Ca²⁺ by the rate μ. Note that even though free Ca²⁺ is rinsed away during NanoSIMS preparation, its interaction with bound Ca²⁺ before fixation influences the size of the bound Ca²⁺ pool and, thus, must be incorporated in the model, Fig 5A (left). A numerical solution of this coupled system for an arbitrary spike train is shown in Fig 5B.

To replicate the measurements used for Fig 4D, we analytically solved Equations (1), (2), (3), (4), (5), and (6) for the steady state condition with constant input rate R and computed the (stable) size of the readily releasable pool (x) for different levels of bound Ca²⁺ b. We are interested in the size of readily releasable pool because it accounts for the immediate and short-term vesicle releases which were labeled by NB during immediate synaptic activity. Despite the multitude of parameters in the model, the overall decrease in the readily releasable vesicle pool size for an increasing number of bound Ca²⁺ emerges for a wide range of parameters matching our experimental findings (Fig 5C).

In addition to matching experimental results, our model provides several predictions (Fig 5D). First, the model predicts a higher maximal synaptic weight w_max for synapses with smaller buffer size. Second, our model yields the prediction that the maximum weight is reached by a higher input frequency R_max for synapses with larger calcium buffer size than synapses with smaller size. Third, once all binding sites in the buffer are filled, synapses of any buffer size will behave similarly. As expected, this occurs at a high input rate regime, where continuous calcium injection and availability of free Ca²⁺ promote binding even further.

Experimental design

The goal of the study was to examine the relationship of bound Ca²⁺ to synaptic function in neurons using a combination of NanoSIMS and fluorescence imaging. Because all free Ca²⁺ is washed out after the embedding procedure outlined below, all Ca²⁺ imaged using NanoSIMS belongs to the bound Ca²⁺ pool. Cultured cells were divided into two groups for correlated NanoSIMS and fluorescence analyses. The first group, intended to examine synaptic size, was immunostained for the glutamatergic pre-synaptic protein VGLUT1, the GABAergic pre-synaptic protein VGAT, and the post-synaptic protein PSD95. The second group was set aside to examine synaptic activity and was immunostained live for synaptotagmin to label the entire active vesicle pool, and for a secondary fluorescent nanobody to sample immediate activity. After fixation, these samples were immunostained for synaptophysin to label the entire vesicle pool.

Hippocampal cultures

Rats (Rattus norvegicus, wild type, Wistar) for the primary neuron cultures were handled according to the specifications of the University of Göttingen and of the local authority, the State of Lower Saxony (Landesamt für Verbraucherschutz, LAVES, Braunschweig, Germany). All experiments and procedures were approved by the local authority, the Lower Saxony State Office for Consumer Protection and Food Safety (Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit). All methods are reported in accordance with ARRIVE guidelines and were carried out in accordance with relevant guidelines and regulations. The experimental procedure was approved by the relevant institutional entity, the Tierschutzbüro of the University Medical Center Göttingen (approval number T 09/08).

Neuronal hippocampal cultures were obtained by collecting the dissociated hippocampi of newborn rat pups, as described in previous studies (Banker & Cowan, 1977; Kaech & Banker, 2006; Truckenbrodt et al, 2018; Yousefi et al, 2021). In summary, brains from rat pups were extracted on post-natal day 2, and the hippocampi were separated. Tissue debris was removed from dissected hippocampi by washing with HBSS (Invitrogen). After this wash, the hippocampi were incubated for 1 h in a sterile-filtered enzyme solution made up of 10 ml DMEM (Thermo Fisher Scientific), 2 mg cysteine, 100 mM CaCl₂, 50 mM EDTA, and 25 U papain, equilibrated with carbogen for 10 min. Hippocampi were then washed with HBSS and incubated for 15 min in an inactivating solution made up of 2 mg albumin and 2 mg trypsin inhibitor in DMEM containing 10 ml FCS.

Coverslips were prepared by treatment with nitric acid, thorough washing with ddH₂O, sterilization and overnight coating with 1 mg/ml poly-l-lysine. Coverslips were then washed once more with ddH₂O before being incubated with a solution of MEM (Thermo Fisher Scientific) supplemented with 10% horse serum, 3.3 mM glucose, and 2 mM glutamine. Neurons from dissected hippocampi were then plated at a concentration of ∼30,000/cm². Cells were left to adhere for 1–4 h at 37°C in a cell incubator, with 5% CO₂. After adhesion, cellular medium was replaced with Neurobasal-A medium (Gibco, Life Technologies), containing 1:50 B27 supplement (Gibco) and 1:100 GlutaMAX (Gibco). 5-fluoro-2′-deoxyuridine was added to the cells after 2–3 d in culture to avoid glial proliferation.

Neurons were kept in culture at 5% CO₂ and 37°C for 18 d before fluorescent labeling. On day in vitro (DIV) 10, the calcium concentration of the culture medium was increased by 10 mM through the addition of dissolved calcium carbonate (CaCO₃). Cells were then returned to the incubator and allowed to incubate for the remaining 8 d.

Immunostaining

Prior to fixation (DIV18), neurons were divided into two groups, one group to be labeled for VGAT, PSD95, and VGLUT simultaneously, and one group to be labeled for synaptotagmin, synaptophysin, and the fluorescent nanobody. Cells in the synaptotagmin group were immunostained live for synaptotagmin with a 1:500 dilution of the primary antibody (synaptotagmin, Synaptic Systems, Catalog #105 311 C2) and incubated for 1 h. After incubation, cells were washed three times with Tyrode buffer (124 mM NaCl, 5 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 30 mM D-glucose, and 25 mM HEPES) and incubated in medium labeled with a 1:500 dilution of the anti-mouse nanobody (FluoTagx2 anti-mouse with Atto542, Catalog #N1202-At54; NanoTag) for 15 min.

After live staining, all neurons from all groups were washed three times in Tyrode buffer, and were fixed for 30 min at room temperature with 4% PFA (Sigma-Aldrich) as performed in previous studies (Sertel et al, 2021a). Samples were quenched by a 30-min incubation at room temperature with 100 mM NH₄Cl in PBS. Cells were then washed three times with a permeabilization solution (3% BSA, 1:10,000 Triton-X-100 in PBS). This was followed by another 1 h incubation step with primary antibodies (Table S1) in the permeabilization solution. Samples were once more washed three times in permeabilization solution and then incubated for 1 h with secondary antibodies (Table S1), also in the permeabilization solution. Samples were stored in PBS overnight at 4°C before embedding.

Sample preparation for NanoSIMS

For correlated NanoSIMS and fluorescence imaging, samples were embedded in LR white resin (London Resin Company) as described in a previous study (Jähne et al, 2021). Samples were first partially dehydrated through a 5 min incubation step with 30% ethanol (in ddH₂O), followed by three 5 min incubation steps in 50% ethanol (in ddH₂O) with rotation at 150 rpm using a Rotamax 120 platform shaker (Heidolph Instruments). After this, samples were incubated for 1 h in a 1:1 solution of pure LR white and 50% ethanol (in ddH₂O). The coverslips were then transferred to a new plate to avoid sticking and incubated in pure LR white for 1 h at room temperature. After incubation, coverslips were removed from the plate. Beem capsules (BEEM Inc.) were placed on top of the coverslips. One drop of LR white accelerator (London Resin Company) was mixed with 10 ml LR white. This mixture was then added to the capsules to create a seal. After hardening for 30 min, capsules were topped with a fresh mixture of LR white and LR white accelerator. Samples were then cured at 60°C until the LR white resin had fully hardened (at least 90 min). After cooling, the coverslips were removed from the capsules. An ultramicrotome (EM UC6; Leica Microsystems) was used to cut 200-nm-thin sections from the samples, which were placed onto silicon (Si) wafers (Siegert Wafer GmbH) for analysis.

Fluorescence imaging

For fluorescence imaging, Si wafers were mounted onto glass slides and imaged at 60x using an inverted epifluorescence microscope (Nikon Eclipse Ti-E; Nikon). After imaging, oil was carefully removed from samples by repeated additions of isopropanol, which were subsequently removed from the sample using compressed air. As a final cleaning step, a small amount of a chloroform and isopropanol solution (∼1:20) was added to the sample surface. After the oil was removed, samples were transferred to the NanoSIMS for further analysis.

NanoSIMS analysis

SIMS images were obtained using a NanoSIMS 50L instrument (Cameca). Embedded samples were initially imaged using a Cs+ positive ion source to capture cell morphology. Secondary ions were generated using a primary current of ∼30 nA (primary aperture D1 = 2). Before each measurement, Cs+ ions were implanted at high current (L1 = 24,000, D1 = 1) on an area larger than the raster size used for analysis until steady-state conditions were achieved. Images were then taken at a raster size of 40 × 40 μm, and a resolution of 256 × 256 pixels, with 5,000 cts/px. The estimated resolution from these images is ∼156.25 nm/px. To minimize sample degradation, one image was collected with these settings at each location. The following peaks were collected for each run: ¹²C¹⁴N (referred to as ¹⁴N in this report), ³¹P, and ³²S.

After images were obtained from all samples using the Cs+ source, the instrument was switched to a radio frequency (RF) O⁻ source for calcium measurements. Secondary ions were generated using a primary current of ∼200 nA (primary apertures D1 = 2, D0 = 2). Similar to the images in the paragraph above, O⁻ ions were implanted at high current (L1 = 24,000, D1 = 1) on an area larger than the raster size. The same locations as were imaged using the Cs+ source were imaged using the O⁻ source at a raster size of 50 × 50 μm, and a resolution of 256 × 256 pixels, with 5,000 cts/px. The estimated resolution from these images is ∼195.31 nm/px. A larger raster size was used to ensure that the entire area imaged with Cs+ was visible in the O⁻ images. Three images were obtained at each location with the settings above. The following masses were collected during each run: ²³Na, ³⁹K, ⁴⁰Ca, and ⁴⁸Ca. Each image shown in this article is a result of the summation of all three image layers taken during analysis.

Calcium imaging

Neuroburst (Sartorius) was used as a genetically encoded calcium indicator, following a method used in a previous study (Sertel et al, 2021a). We used the Incucyte Neuroburst Orange Lentivirus, which is a third-generation, based on HIV, VSV-G pseudotyped lentivirus particle, provided in a suspension form, in DMEM. Virus transduction was performed on DIV10, with 3 μl of Neuroburst added to each well plate. The medium was replaced every few days, removing one third of the neuronal medium and adding a similar amount of fresh medium to maintain the cultures at a volume of ∼1 ml. On DIV17, coverslips were placed in Tyrode buffer and imaged at 37°C using the same Nikon Eclipse Ti-E inverted epifluorescence microscope as used for fluorescence imaging. Coverslips were imaged for 2 min in 1 s intervals (1 Hz). After 2 min of imaging, neurons underwent field stimulation for 3 s at 20 Hz, and then for 30 s at 20 Hz. After neurons resumed spontaneous activity, the Tyrode buffer in the imaging chamber was replaced with 400 μl of a 5 µM solution of ionomycin (Sigma-Aldrich) in Tyrode. Cells were allowed to acclimatize to the ionomycin addition, after which cells were imaged for 15 s in 1 s intervals.

The location of the imaged cells on the coverslip was marked after each recording. After recordings were completed, cells were fixed and embedded according to the sample preparation procedure described earlier in this article, with the exception that the coverslips were not removed from the plate between the 1 h incubation step with LR white and 50% ethanol and the 1 h incubation step with pure LR white. The Beem capsule was placed over the marked location of the cell, and coverslips were not moved during the hardening and sealing steps, to ensure that the same cells as imaged were embedded. NanoSIMS analysis proceeded as previously described. Cells were chosen by comparing the arrangement of cells visible in NanoSIMS with those of cells imaged during live imaging.

Ca²⁺ turnover in HEK293 cells and neurons

HEK293 cells were maintained in culture in DMEM with 5% FCS, 2 mM L-glutamine, 60 U/ml penicillin, and 60 U/ml streptomycin. In preparation for the experiment, cells were seeded into a six-well plate containing coverslips coated with poly-l-lysine. Five out of six wells contained DMEM Ca-free medium (Thermo Fisher Scientific) with 5% FCS, 2 mM L-glutamine, 60 U/ml penicillin, and 60 U/ml streptomycin. The calcium concentration of the medium was increased to 2 mM by the addition of a solution of calcium carbonate enriched in ⁴⁶Ca (5% ⁴⁶CaCO₃, 95% ⁴⁰CaCO₃; Cambridge Isotope Laboratories). The final well contained normal DMEM and was used as a control experiment.

Coverslips remained in culture for a set amount of experimental time, namely 1 and 6 h, 1 or 3 d. Control cells remained in culture for 3 d. After each experimental period was completed, cells were fixed for 30 min at room temperature with 4% PFA (Sigma-Aldrich) and embedded in LR white medium as previously described in the article.

Hippocampal neurons were prepared in the same way as all other neuronal cultures used here. On DIV 4, ⁴⁶Ca-enriched calcium carbonate solution was added to each well-plate, raising the Ca²⁺ concentration of the buffer up to ∼16 mM. Half of the coverslips were fixed and prepared for NanoSIMS on DIV8, after 4 d in the ⁴⁶Ca experiment. The remaining coverslips were fixed and prepared for NanoSIMS on DIV14 after 10 d in the ⁴⁶Ca experiment.

NanoSIMS analyses used a raster size of 70 μm (∼273 nm/px) for all images made using HEK293 cells, and a raster size of 20 μm (∼78 nm/px) for all images made using hippocampal neurons labeled with ⁴⁶Ca. Analyses otherwise proceeded using the same method as for other samples used in this study, with the additional collection of the ⁴⁶Ca peak using the radio frequency (RF) O⁻ source. After analyses, mean ⁴⁶Ca/⁴⁰Ca ratios of each cell were collected by averaging the intensity of the ⁴⁶Ca peak across the entire area of the cell and dividing that average by the average intensity of ⁴⁰Ca across the same area.

In all experiments, the ⁴⁶Ca/⁴⁰Ca ratio of the controls is slightly higher than expected for the natural abundance (∼0.0413‰) of this isotope. This is expected, because of counting statistics, as the NanoSIMS instrument collects data in units of counts (Fitzsimons et al, 2000). Because the natural abundance of ⁴⁶Ca is so low (0.004% of all Ca atoms), the collection of a single ⁴⁶Ca count in an image can artificially increase the resulting ⁴⁶Ca/⁴⁰Ca ratio.

Image processing & correlation

NanoSIMS images were initially processed using a custom Matlab script (Supplemental Data 1) (the Mathworks Inc.) adapted from the analysis script used in previous studies (Gagnon et al, 2012; Bonnin et al, 2021). Image layers were aligned relative to the second image from each element before summation. The resulting matrices were then saved as text files, and images were generated using FIJI/Image J (NIH, Bethesda, MD, USA).

Fluorescence images were extracted using the Bio-Formats plug-in in FIJI/ImageJ. Fluorescence images and ⁴⁰Ca SIMS images were then aligned by hand. A second Matlab processing script was used to select ROIs from the aligned images (Fig S8A–C). ROIs were selected by hand based on high signal in fluorescence. For comparison between ¹⁴N, ³²S, and ⁴⁰Ca (Fig 1), ROIs were selected based on high signal in either ¹⁴N or ³²S. For the VGAT/PSD95/VGLUT1 image set (Fig 3), ROIs were selected based on high signal either in VGAT or in VGLUT, denoting either an inhibitory or excitatory synapse, respectively, or based on high signal in PSD95, denoting the location of the excitatory post-synapse. For the Syt1/NB/Syph image set (Fig 4), ROIs were selected based on high intensity in any of the three fluorescent markers. Locations that did not correspond to cellular material were excluded from analysis.

The ROI selection script works by asking the user to select background points at each image. Once background levels are recorded, the user can select ROIs. The script then records the corresponding intensity value at each selected location for each associated image, subtracted from the background. ROIs are defined as circles with a radius of 6 pixels, leading to an interior area of ∼113 pixels. Intensity values for each pixel within an ROI are averaged, so that a single average value per ROI is recorded. Before further analysis, data points with ⁴⁰Ca intensities at or below background level were removed. This was done by assuming a normal distribution of background ⁴⁰Ca intensities with a mean of 0 and filtering out points within one SD of the mean.

Statistical analysis

The significance of the correlations between bound Ca²⁺ and either ¹⁴N and ³²S (Fig 1) or fluorescence intensities (Figs 3, 4, S1, S5, and S7) was determined using a linear regression test, with N and P-values as reported in the associated figures. A Bonferroni multiple comparison correction was applied to the data displayed in Figs 4, S1, S5, and S7, to correct for multiple parameters being compared against each other. For the free Ca²⁺ imaging experiment, significance between free Ca²⁺ parameters and bound Ca²⁺ was determined using a two-tailed Pearson correlation test, with N and P-values as reported in the associated figure (Fig 2). For Ca²⁺ turnover experiments (Figs S2 and S3), experimental groups were compared against each other using the Kruskal–Wallis one-way analysis of variance to determine whether statistically significant differences existed within the groups. After this, Dunn’s test for multiple comparisons was used to perform pairwise comparisons between each independent group.

Analytical solution of computational model

The steady-state solution of the dynamical system discussed in the main text can be found by setting all time derivatives to zero.dxtdt=1−xτd−uxR=0(7)du(t)dt=U−uτf+U(1−u)R=0(8)db(t)dt=−bτb+λcb(B−b)c−λbcb=0(9)dc(t)dt=−cτc−λcb(B−b)c+λbcb+μR=0(10)

Denoting the fixed point by subscript *, the first pair of equations yieldu*=U*1+Rτf1+U*Rτf(11)x*=11+u*Rτd(12)

The third equation leads tob*=Bτbλcb c*1+τb(λbc+λcbc*)(13)which is the relation we used to plot Fig 5B. Substituting this relation to the last equation above results in the following parabolic equation for c:c2[τbλcb]+c[(1+kc)+kb(B−μRτb)]−μRτc(1+kc)=0(14)where we have introduced k_b = τ_cλ_cb and k_c = τ_bλ_bc for convenience. This parabolic equation has two roots. Yet, only the stable one is of our interest. We numerically check the stability by increasing (or decreasing) the root and asserting that the time derivative of b is negative (or positive), thus exerting an attracting force on c toward the stable root. The stable (and positive) solution, c_*, is then substituted in Equations (11), (12), and (13) for different parameter values to plot Fig 5C.

Parameter selection

For τ_d and τ_f we followed the values reported in Mongillo et al (2008) (200 and 1,500 ms), respectively. The bound and free time scales τ_c and τ_b are related to the calcium lifetimes. Because of the high stability of the bound Ca²⁺ pool we choose a value of 10,000 s (corresponding to a time scale of order of days) for τ_b. In contrast, free calcium has a faster turnover. Therefore, we set τ_b to 100 s. The transition rates λ_cb and λ_bc are unknown. However, because of the abundance of bound Ca²⁺ in cells, we assumed that transition to bound states is more likely, and thus promoted binding over unbinding by setting λ_cb≫λ_bc. For Fig 5C, values of 200 and 30 are used. Interestingly, as long as the relative ratio of these transition rates is similar, the model is insensitive to the exact values. The stimulation rate R was set to 20 Hz to match the experimental value. Other parameters (B, μ, λ) were varied either systematically (Fig 5C and D) or manually to align to experimental data.