A Dynamic Balance between Neuronal Death and Clearance in an in Vitro Model of Acute Brain Injury

Cell death stains and fluorescent protein expression

We considered two possible explanations for the apparent over-reporting of programmed neuronal death by PI staining in our prior study (Berdichevsky et al., 2012). First, PI might also be staining healthy neurons. Healthy neurons can be assayed by robust emission of transgenically expressed fluorescent proteins (Steff et al., 2001; Strebel et al., 2001; Arrasate et al., 2004; Arrasate and Finkbeiner, 2005; Linsley et al., 2019). To address potential differences in labeling by the PI versus fluorescent protein biomarkers, slices were used from WT mice in which neurons expressed GFP following infection of slices with the adeno-associated virus (AAV) AAV9-hSyn-eGFP vector. Slices were also used from Clm mice in which neurons transgenically expressed both CFP and YFP under control of the Thy1 promoter (Kuner and Augustine, 2000); henceforth, Clm will be used to refer collectively to the fluorescent proteins expressed in slices from Clm mice. These slices were stained with PI on DIV 5-7 to assess the degree of overlap (Fig. 2A). Neurons with bright emission from transgenic fluorophores had normal morphology (Fig. 2B) and minimal nuclear PI staining (Fig. 2C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Overlap of transgenic fluorescent protein emission and PI nuclear staining. A, Visualization of the lack of overlap between PI positivity and FP positivity. In slices made from Clm mice imaged at DIV 7, neither the CFP emission (left) nor the YFP emission (middle) of Clm was present in cells with PI⁺ nuclei. The same was true of cells in WT slices that expressed GFP (right), imaged at DIV 5. Scale bars, 50 μm. B, Morphology of CA1 neurons in organotypic hippocampal slice cultures imaged at DIV 11. Neurons with bright emission from expressed TurboRFP (FP⁺; see Materials and Methods) demonstrate normal somata, nuclei, and dendritic arbors. Scale bar, 50 μm. C, FP⁺ neurons are PI^–. Across 23,697 PI⁺ neurons and 3488 GFP⁺ neurons, only 6 of the GFP⁺ neurons (or 0.17%) also exhibited PI positivity. n = 12 slices. For additional data, see Extended Data Figure 2-1.

Between 23,697 PI⁺ cells and 3488 GFP⁺ neurons identified by the ImageJ plugin TrackMate (Extended Data Fig. 2-1A), only 6 instances of overlap (or 0.17% of GFP⁺ neurons) were seen (n = 12 slices). In less healthy slices, overlapping fluorescence in the PI and fluorescent protein emission bands were more common because of the wide-spectrum emission properties of autofluorescent cellular debris; still, between 265,382 PI⁺ cells and 53,750 GFP- or Clm⁺ neurons identified by ImageJ, only 2434 instances of overlap (or 4.53% of FP⁺ neurons) were seen when these slices were included (n = 99 slices). Less healthy slices could be identified using Cellprofiler by the presence of autofluorescent debris, which was significantly smaller than healthy neurons (Extended Data Fig. 2-1B,C). This reduced the average diameter of cells identified by CellProfiler (18.71 ± 0.26 μm vs 16.98 ± 0.14 μm, p < 0.0001, n = 12 vs 87 slices) as well as the area (341.50 ± 9.08 μm² vs 275.30 ± 4.48 μm², p < 0.0001, n = 12 vs 87 slices) and perimeter (87.62 ± 1.12 μm vs 76.22 ± 0.68 μm, p < 0.0001, n = 12 vs 87 slices) in less healthy slices compared with healthy slices.

The number of FP⁺ neurons was 15%-20% of the number of PI⁺ neurons in these preparations, which is to be expected since PI stains the vast majority of neurons in late stages of programmed cell death, but fluorescent protein expression (be it viral or transgenic) may only be detected in a fraction of healthy neurons (Lau et al., 2022). These results indicate that FP⁺ neurons are healthy, with intact cellular and nuclear membranes that do not admit PI. They also confirm that a considerable number of neurons are undergoing programmed cell death in these organotypic slice cultures days or weeks after the initial brain slicing injury.

There was also minimal overlap between neurons with robust emission from transgenic fluorescent proteins and neurons exhibiting elevated caspase activity. Slices in which neurons expressed the RFP TurboRFP were loaded with FLICA to visualize caspase activity. Starving these slice cultures of nutrients and oxygen by stopping ACSF perfusion resulted in quenching of neuronal TurboRFP fluorescence emission after 1-2 h of starvation. FLICA signals were elevated in many neurons within the same time frame (Fig. 3A). In some cases, TurboRFP quenching and increases in FLICA occurred in the same neurons with a period of overlap of <1 h (the interval between sequential imaging sessions). Subsequently, FLICA fluorescence also decreased, indicating that the elevated caspase activity is transient (Fig. 3B) (Namura et al., 1998; Nakajima et al., 2000). Indeed, instances of overlap between FLICA and fluorescent protein emission were rare; in a 590 × 590 × 140 (X:Y:Z) μm region of a normally perfused slice, 107.7 ± 49.0 TurboRFP⁺ neurons, 41.6 ± 16.8 FLICA⁺ cells, and 12.9 ± 8.6 cells that were positive for both could be seen (Fig. 3C), with only 6.5 ± 3.3% of TurboRFP⁺ cells also being FLICA⁺, and only 17.0 ± 9.6% of FLICA⁺ cells also being TurboRFP⁺ (Fig. 3D) (n = 754 TurboRFP⁺ neurons and 291 FLICA⁺ neurons in 7 FOVs from 3 slices). This same lack of overlap was also seen when TurboRFP- and FLICA⁺ cells were imaged serially for several hours, with TurboRFP positivity dropping and FLICA positivity increasing shortly after programmed cell death was induced by stopping perfusion. After 6 h of starvation, both fluorescent protein emission and FLICA fluorescence had diminished, although some regions required a longer period of starvation than others (Fig. 3E). Most FLICA⁺ cells were neurons because the fraction of newly FLICA⁺ cells that were currently, or had been previously, TurboRFP⁺ was comparable to the typical fraction of neurons that label with AAV9-synapsin-driven fluorescent proteins in this preparation (Fig. 3F) (Lau et al., 2022). These experiments indicate that neurons with bright emission from transgenic fluorescent proteins rarely stain for caspase activity or PI. We conclude that neurons expressing fluorescent proteins are healthy in this preparation, and that PI is not falsely labeling healthy neurons, so that the over-reporting of neuronal death by PI is not because of false positive PI staining of healthy neurons.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Overlap of transgenic fluorescent protein expression and FLICA. A, Quenching of fluorescent proteins and an increase in caspase activity occur concurrently. Depriving CA1 neurons in an organotypic slice culture on DIV 7 of nutrients and oxygen by stopping perfusion at Hour 0 caused TurboRFP⁺ neurons with little FLICA intensity (left) to become largely TurboRFP^– and FLICA⁺ (right) by Hour 1. By Hour 2, the TurboRFP positivity has decreased even further and the transient caspase activity has largely ceased. Scale bars, 50 μm. B, Example of transient overlap of fluorescent protein expression and increased caspase activity. On DIV 7, the neuron had strong TurboRFP expression and negligible FLICA signal at Hour 0 when perfusion was shut off. By Hour 1, both fluorophores could be seen in the same neuron; TurboRFP expression was substantially decreased but still visible, and FLICA emission was now strong. By Hour 2, TurboRFP expression was nearly gone and FLICA remained bright. But by Hour 3, the transient caspase activity had begun to decrease and the neuron was barely visible with either fluorophore. Scale bar, 10 μm. C, FP + FLICA overlap is rare. Under control conditions with adequate perfusion, a 590 × 590 × 140 section of a slice contained 107.71 ± 49.01 TurboRFP⁺ neurons, 41.57 ± 16.79 FLICA⁺ cells, and 12.86 ± 8.60 cells that were positive for both. n = 754 TurboRFP⁺ neurons and 291 FLICA⁺ neurons in 7 FOVs from 3 slices. D, Few FP⁺ cells exhibit elevated caspase activity. In the same slices as in C, only 6.52 ± 3.32% of TurboRFP⁺ cells were also FLICA⁺. Conversely, 16.95 ± 9.63% of FLICA⁺ cells were also TurboRFP⁺. E, Examples of FP positivity and FLICA positivity, and overlap over time. In two regions of a slice, imaged sequentially, TurboRFP positivity dropped and FLICA positivity increased once the perfusion was shut off, but the amount of overlap at any given time remained low. After several hours, the transient caspase activity decreased as well, lowering the FLICA positivity. F, Most FLICA⁺ cells are neurons. Using TrackMate analysis of a 4 h time series experiment, 44% of 188 FLICA⁺ neurons were either currently also TurboRFP⁺, or had previously been TurboRFP⁺, which is very close to the rate of expression of TurboRFP in neurons in this preparation (Lau et al., 2022).

Rates of programmed cell death and efferocytosis

An alternate explanation for the over-reporting of programmed cell death by PI staining is illustrated in Extended Data Figure 4-1. The number of PI⁺ neurons in the preparation at any time point depends on the difference in rates of entry and exit into the pool of PI-receptive neurons, as well as the duration of that difference in rates. We therefore sought to characterize these rates. We first assayed the rate at which neurons exited the PI-receptive pool. Slices were prepared from Clm mice (Kuner and Augustine, 2000); and on DIV 7, PI was added to the slices' culture media. Following this one-time PI exposure, both PI and fluorescent proteins were visualized with two-photon microscopy on DIVs 7, 10, and 13, that is, on the day of PI exposure and 3 and 6 d later (Extended Data Fig. 4-2). In the 12 slices, the number of PI⁺ neurons declined continuously. A basic monoexponential curve had a high coefficient of determination (R² = 0.9478), so a best-fit monoexponential curve was derived and fit to the data. The decay curve had a time constant (τ) of 2.16 d (initial PI count = 967.0 ± 106.1, n = 12 slices). These findings indicate that, in hippocampal organotypic slice preparations on DIV 6-13, PI⁺ neurons waited an average of 1.5 d to be phagocytosed (Fig. 4A). After 3 d, the percentage of PI⁺ neurons remaining was negatively correlated with the number of visible FP⁺ neurons (R² = 0.5123, p = 0.0089, n = 12) (Fig. 4B). The number of FP⁺ neurons should be larger in healthier slices (Linsley et al., 2019). The monoexponential decline of PI⁺ neurons would be consistent with a first-order process (Espenson, 1995) that is limited by the number of PI⁺ neurons (Fig. 4A). In other words, under these experimental conditions, the rate of efferocytosis of PI⁺ neurons is proportional to the number of PI⁺ neurons in the programmed cell death pathway.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Survival of PI⁺ and fluorescent protein⁺ neurons. A, PI⁺ neurons can survive for days. Organotypic hippocampal slices that expressed the fluorescent protein Clm were given a single dose of the nuclear cell death indicator PI on DIV 7, then imaged, and subsequently imaged again on DIVs 10 and 13. PI⁺ cells were counted using the ImageJ plugin TrackMate. The initial PI count was 967 ± 106.1, and the time constant of the progressive loss of PI⁺ cells was 2.16 d (n = 12 slices). B, PI⁺ neurons survive for longer in less healthy slices. In the same experiment as in A, there was a correlation between the percentage of the initial PI⁺ population that remained on DIV 10, and the number of Clm⁺ neurons present on DIV 10 (R² = 0.5123, p = 0089, n = 12). TrackMate's Quality threshold for the Clm⁺ cells in each slice (see color bar) also correlated with these measurements. C, The entry rate into the PI-receptive pool does not increase in slices with higher PI counts. Slices were imaged for 72 h, and TrackMate was used to identify new PI⁺ neurons on each day. For each slice, the percentage of PI⁺ neurons that was newly PI⁺ on each day was found, and then the average of all days for that slice was taken. This average was then plotted against the total number of PI⁺ neurons counted over the entire 72 h period. n = 17 slices. D, PI⁺ neurons survive for longer in slices with higher PI counts. Slices were imaged for 72 h, and TrackMate was used to track the initial PI⁺ neuron population. The time constant was determined from the progressive loss of the initial PI⁺ cells for each slice, and plotted against the total number of PI⁺ neurons counted over the entire 72 h period. n = 7 slices. E, Example image of plentiful jRCaMP1a expression in an organotypic slice imaged at DIV 4. Scale bar, 100 μm. F, FP⁺ neurons can survive for weeks, or longer, during seizures. Organotypic slices infected with AAV-syn-jRCaMP1a and AAV-syn-cre-nls-GFP were imaged every 4 h from DIV 7 to DIV 17. GFP⁺ cells were counted and tracked using ImageJ and TrackMate. After 10 d of imaging, 84.03 ± 2.01% of neurons (7134 of 8486, n = 9 slices) were still visibly fluorescent and morphologically intact, indicating that the vast majority of neurons can survive for a considerable length of time despite the presence of seizures. For additional data, see Extended Data Figures 2-1, 4-2, 4-3, 4-4 and Extended Data Table 4-1.

To measure the rate at which neurons became PI⁺, the rate of entry into the PI-receptive pool was measured using serial applications of PI to another group of organotypic hippocampal slice cultures at DIV 7-10. The number of PI⁺ neurons was counted on each day (Extended Data Table 4-1), and neurons were tracked across multiple days (Extended Data Fig. 4-3). By determining the number of newly stained PI⁺ neurons on each day, we found the average daily entry rate of neurons into the PI-receptive pool to be 81% of the extant PI-receptive pool (Fig. 4C). Additionally, when the initial PI⁺ population in these slices is tracked and the loss of these cells is used to derive the time constants of monoexponential decay, those τ values are in line with those seen in Figure 4A with the average τ being 33.52 ± 5.77 h, or 1.40 ± 0.24 d (Fig. 4D). There was a strong, nonlinear correlation between the time constant for clearance of PI⁺ neurons and the total number of PI⁺ neurons counted over the entire 72 h period (Fig. 4D). This correlation suggests that the ratio of dying neurons to active microglia varies from slice to slice, and that slices with more dying neurons have longer time constants. We conclude that the maximum rate of microglial efferocytosis of dying neurons can be exceeded when too many neurons undergo programmed cell death at the same time. This is the condition represented by the system in Extended Data Figure 4-1C. Preparations in which efferocytosis rates were saturated (i.e., preparations with high ratios of dying neurons to active microglia) would be expected to have linear rates of PI clearance. However, the development of seizures in the first 2 weeks in vitro (Berdichevsky et al., 2012; Lau et al., 2022) may have slowed efferocytosis and distorted this linearity.

The average rate of entry into the PI⁺ pool was 81% of the PI⁺ pool size (Fig. 4C). From this, it might be concluded that single-time point staining is a reasonable measure of the rate of programmed cell death. However, there was no positive correlation between entry rates and size of the PI⁺ pool (Fig. 4C). Although the pool size variance was large, even if outliers are excluded, there is still no correlation between PI pool size and the rate of entry into the PI⁺ pool. We had previously assumed (Berdichevsky et al., 2012) that the rate of entry into the PI⁺ pool would be the primary determinant of the size of the PI⁺ pool: that is, in single-time point analyses, the number of PI⁺ neurons is a biomarker for the rate of neuronal death. But this is not the case, at least in the organotypic slice preparation (Fig. 4C). Rather, the rate of exit from the pool is the primary determinant of the size of the pool of PI⁺ neurons. This is because PI⁺ cells remain in the pool longer when the size of the pool itself is larger (Fig. 4D).

The PI results demonstrate wide variation in the rates of exit from the PI receptive pool, and the corresponding changes in pool size (Fig. 4D). This is schematized by Figure 1. The rates of entry into the PI⁺ pool appear to be the best measure of ongoing cell death because these neurons have most recently committed to programmed cell death. However, even the rate of entry into the PI⁺ pools comes with a caveat because the time between commitment to programmed cell death and entry into the PI-receptive pool is not known. It is possible that all these new PI⁺ cells committed to programmed cell death at the time of slice preparation, or it may be that the newly PI⁺ cells represent only neurons that had survived slicing unscathed but subsequently committed to programmed cell death, or a combination of these two populations.

We therefore used loss of fluorescent protein emission as another assay for neuronal death in serially imaged hippocampal organotypic slice cultures (Steff et al., 2001; Strebel et al., 2001; Arrasate et al., 2004; Arrasate and Finkbeiner, 2005; Linsley et al., 2019). Slice cultures infected with AAV-syn-jRCaMP1a and AAV-syn-cre-nls-GFP were imaged every 4 h from DIV 7 to DIV 17 using a single-photon microscope with a robotic 6-well stage built into a tissue culture incubator (Jacob et al., 2019). GFP⁺ neurons were counted and tracked using ImageJ and TrackMate. Spontaneous recurrent seizure activity was present in all slices, assayed by jRCaMP activity (Fig. 4E) (Jacob et al., 2019; Lau et al., 2022). By DIV 17, 84.03 ± 2.01% of neurons (7134 of 8486, n = 9 slices) were still visibly fluorescent (Fig. 4F). Thus, longitudinal assessment of neuronal viability by fluorescence quenching indicated a low rate of neuronal death, ∼1% of the healthy fluorescent pool per day, despite the ongoing presence of seizures.

Ideally, we could compare the rates of death from fluorescence extinction directly to the rate of death measured by newly PI⁺ neurons. However, comparing the newly PI⁺ rate to the number of FP⁺ neurons directly is not possible because we do not know the fraction of healthy neurons that are FP⁺ in these experiments. Our recent published experience with this preparation is that 23% of neurons whose health was established by NeuN and phosphoS6 staining are FP⁺ after AAV-mediated expression (Lau et al., 2022). Thus, if 1% of healthy cells enter the programmed cell death pathway every day, but only 1 in 4 healthy cells are FP⁺, then we would expect the total number of neurons entering the programmed cell death pathway to be 4 times 1% (i.e., 4% of FP⁺ neurons). Using Figure 2C, 4% of 3488 FP⁺ neurons in 12 slices is 140 neurons entering the programmed cell death pathway per day, or 12 neurons per slice per day. Yet if 81% of the PI⁺ pool of 23,697 PI⁺ neurons become newly PI⁺ each day, then 19,195 neurons per day became newly PI⁺ in 12 slices, so that the newly PI⁺ rate per day per slice is 1600. Of course, one must allow for variability in slice health, fluorescent protein expression, and PI staining, as well as the fact that the number of PI⁺ cells is strongly negatively correlated to time after slicing (Berdichevsky et al., 2012), and so this number will not remain high for the entire lifetime of a slice. But even allowing for this variability, this mismatch between the low rate of programmed cell death initiation (12 neurons per slice per day) and the higher rate of newly PI⁺ measures of neuronal death (1600 neurons per slice per day) indicates that there is a large pool of FP^– neurons that are not healthy (i.e., would not express fluorescent protein) but not yet PI⁺. These neurons may have entered the programmed cell death pathway, but not yet become PI⁺. We therefore examined this phase of programmed cell death.

AM dye uptake relative to other biomarkers of programmed cell death

To address the interval between the time that neurons undergoing programmed cell death lose fluorescence emission and the time they become PI⁺, we exploited a serendipitous finding that in organotypic slices, organic AM dyes selectively stain neurons undergoing programmed cell death. AM dyes are esterified to an acetoxymethyl moiety to enhance membrane permeability. Hydrolysis of the ester bond by intracellular esterases renders the dye fluorescent as well as less membrane permeable, “trapping” the fluorescent dye in the neuronal cytoplasm (Tsien, 1981). This technique was originally described for cells in suspension. Techniques incorporating pressure, solvents, and detergents have been described to load neurons and astrocytes in situ with AM dyes (Stosiek et al., 2003; Hamad et al., 2015). We found that, in hippocampal organotypic slice cultures, overnight incubation of slices with AM dyes selectively stained neurons undergoing apoptosis (see Materials and Methods). This staining pattern after overnight incubation was consistently observed with both SBFI-AM and FuraRed-AM across a range of slice conditions and experimental protocols.

First, we confirmed that AM dye⁺ cells were neurons by infecting slices with GFAP-based virus GFAP-eGFP, which is exclusively expressed in astrocytes. SBFI-AM and GFP⁺ cell populations were clearly distinct (Extended Data Fig. 5-1), confirming that neurons load preferentially with AM dyes with the protocols we used in the organotypic hippocampal slice preparation.

Next, we determined that FP⁺ neurons quench in vivo after acute hypoxic-ischemic injury. Acute ischemic stroke in anesthetized Clm mice induced by Rose Bengal photothrombosis (Maxwell and Dyck, 2005; J. K. Lee et al., 2007; Labat-gest and Tomasi, 2013) resulted in widespread fluorescent protein quenching (Extended Data Fig. 5-2). CFP and YFP emission intensities decreased to 45.26 ± 4.53% and 51.64 ± 2.98% of their initial values, respectively, after 75 min (n = 12 neurons from 3 mice), whereas in a control experiment where photothrombosis was not performed CFP and YFP remained at 97.20 ± 1.61% and 87.10 ± 1.70% of their initial values, respectively, after 90 min (n = 12 neurons from 1 mouse) (Fig. 5A). This affirmed the utility of fluorescent protein quenching as a biomarker for neuronal death after acute as well as degenerative (Steff et al., 2001; Strebel et al., 2001; Arrasate et al., 2004; Arrasate and Finkbeiner, 2005; Linsley et al., 2019) injury.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

AM dye⁺ neurons are undergoing programmed cell death. A, Quantification of CFP and YFP quenching in vivo after photothrombosis. Photothrombosis in vivo reduced CFP and YFP emission intensities to 45.26 ± 4.53% and 51.64 ± 2.98% of their initial values, respectively, after 75 min (n = 12 neurons from 3 mice), whereas in a control experiment where photothrombosis was not performed, CFP and YFP remained at 97.20 ± 1.61% and 87.10 ± 1.70% of their initial values, respectively, after 90 min (n = 12 neurons from 1 mouse). B, Quenching of fluorescent proteins and uptake of AM dyes occur concurrently. Depriving an organotypic slice of fresh culture media beyond the normal media change interval caused widespread quenching of TurboRFP over a ∼24 h period, at which point many neurons also began to take up SBFI-AM. Inset, Example neurons (*), which began as TurboRFP⁺ and SBFI^–, briefly became positive for both, and then rapidly became SBFI⁺ and TurboRFP^–. Scale bars, 50 μm. C, AM dye⁺ neurons do not express fluorescent proteins. Example image shows that SBFI⁺ neurons and TurboRFP⁺ neurons are two separate populations (left). Quantification of this lack of overlap, showing that, across 14,543 SBFI⁺ neurons and 15,749 neurons positive for either TurboRFP or Clm, only 199 of the FP⁺ neurons (or 1.26%) also exhibited SBFI positivity (right). n = 22 SBFI + TurboRFP slices, n = 3 SBFI + Clm slices. Scale bar, 50 μm. D, AM dye⁺ neurons are largely FLICA^–. Neurons that take up SBFI-AM rarely stained positive for FLICA. Slice imaged on DIV 7. Scale bar, 25 μm. E, AM dye⁺ neurons exhibit progressive cell shrinkage. FLICA⁺ neurons had the same somatic area as TurboRFP⁺ neurons (410.06 ± 18.37 μm² vs 397.56 ± 8.31 μm², n = 80 and 80, p = 0.5361). GFAP-GFP⁺ astrocytes had smaller soma than TurboRFP⁺ neurons (303.75 ± 14.43 μm² vs 397.56 ± 8.31 μm², n = 80 and 80, p < 0.0001). Neurons that had been SBFI⁺ for less than 24 h had a reduced somatic area compared with TurboRFP neurons (229.72 ± 7.14 μm² vs 397.56 ± 8.31 μm², n = 80 and 80, p < 0.0001), and neurons that had been SBFI⁺ for longer than 24 h had an even smaller somatic area (135.28 ± 7.72 μm² vs 397.56 ± 8.31 μm², n = 60 and 80, p < 0.0001). F, AM dye⁺ neurons exhibit distinctive chromatin morphology. Example image showing that the nuclei of FuraRed⁺ neurons, stained with NucBlue, are visually distinct from the nuclei of FuraRed^– neurons. Scale bar, 50 μm. G, AM-dye⁺ neurons exhibit chromatin condensation. Quantification of NucBlue staining shown in F, wherein FuraRed⁺ neurons had greater NucBlue intensity than FuraRed^– neurons (68.61 ± 1.67 vs 37.04 ± 1.80, n = 25 vs 25, p < 0.0001). H, Nuclear area of neurons staining with FuraRed AM is smaller than both FuraRed^– neurons (34.51 ± 0.59 μm² vs 90.70 ± 1.84 μm², n = 157 vs 157, p < 0.0001) and neurons whose nuclei expressed GFPcre (34.51 ± 0.59 μm² vs 97.47 ± 1.60 μm², n = 159 vs 113, p < 0.0001). For additional data, see Extended Data Figures 5-1, 5-2, 5-3, 5-4.

We next determined the extent of AM dye uptake over the course of fluorescent protein quenching in organotypic hippocampal slice cultures. Healthy TurboRFP⁺ neurons exhibited no SBFI-AM staining (Fig. 5B). However, if the same neurons were induced to initiate programmed cell death by media starvation (slice cultures kept in the same media for 24 h beyond the typical 3 d interval for exchanging spent for fresh media), TurboRFP fluorescence began to quench and scattered neurons began to take up SBFI (Fig. 5B, middle). After just 5 additional hours, many neurons now stained with SBFI-AM (Fig. 5B, right). Additionally, individual neurons could be observed quenching TurboRFP and taking up SBFI-AM simultaneously, and thus visibly transitioning from red to yellow to green (Fig. 5B, insets). This indicates that fluorescent protein quenching and an increased propensity to load with cell-permeant AM dyes are concurrent processes. At this early stage, SBFI-AM⁺ neurons retained their dendritic processes (Fig. 5B).

In control slices not deprived of fresh media, healthy neurons that expressed robust emission from transgenically expressed fluorescent proteins TurboRFP or Clm only rarely stained with SBFI-AM: only 199 FP⁺ neurons (or 1.26%) exhibited both SBFI-AM and TurboRFP/Clm simultaneously, whereas 15,749 neurons were observed that were FP⁺ but SBFI-AM negative, and 14,543 neurons were observed that were SBFI⁺ but FP^– (n = 22 SBFI + TurboRFP slices, n = 3 SBFI + Clm slices) (Fig. 5C). This small amount of overlap indicates that the AM dye⁺ population is comprised of neurons whose fluorescent proteins have quenched at the onset of programmed cell death.

SBFI-AM⁺ neurons were rarely FLICA⁺ (Fig. 5D). This suggests that increased caspase activity is not only an early step in neuronal programmed cell death (as seen in Fig. 3A), but also a transient one; otherwise, SBFI-AM⁺ neurons would continue to be FLICA⁺ throughout their remaining lifetime. The lack of FLICA staining of SBFI-AM⁺ neurons also indicated that SBFI-AM is not toxic under these conditions; that is, it was not inducing healthy neurons to initiate programmed cell death and thereby become both SBFI⁺ and FLICA⁺ (Smith et al., 2018).

Morphologic evidence of programmed cell death in AM dye⁺ neurons

Programmed cell death is accompanied by somatic shrinkage (Kerr et al., 1972). AM dye⁺ neurons exhibited progressive cell shrinkage during serial imaging experiments (Fig. 5E), with neurons that had been SBFI-AM⁺ for <24 h having a reduced somatic area compared with TurboRFP⁺ neurons (229.72 ± 7.14 μm² vs 397.56 ± 8.31 μm², n = 80 and 80 neurons, p < 0.0001). Neurons that had been SBFI-AM⁺ for longer than 24 h had an even smaller somatic area (135.28 ± 7.72 μm² vs 397.56 ± 8.31 μm², n = 60 and 80 neurons, p < 0.0001). FLICA⁺ neurons had the same somatic area as TurboRFP⁺ neurons (410.06 ± 18.37 μm² vs 397.56 ± 8.31 μm², n = 80 and 80, p = 0.5361), whereas GFAP-GFP⁺ astrocytes had smaller soma than TurboRFP⁺ neurons (303.75 ± 14.43 μm² vs 397.56 ± 8.31 μm², n = 80 and 80, p < 0.0001). This supports the idea that most FLICA⁺ cells are neurons, since their area matched that of TurboRFP⁺ neurons so closely. However, the distribution of somatic area values for FLICA⁺ cells was broader than that of TurboRFP⁺ neurons (Extended Data Fig. 5-3). This could mean that some FLICA⁺ cells were, for example, astrocytes. However, the vast majority of cells undergoing programmed cell death after acute brain injury are neurons (Li et al., 1995a, b), and neurons are also known to be much more vulnerable than astrocytes following oxygen-glucose deprivation (Lyden et al., 2021). Thus, it is more likely in our case that some FLICA⁺ neurons had already started to shrink, and as a result they skewed closer to the area values of newly SBFI⁺ cells.

Chromatin condensation is another morphologic marker of programmed cell death (Kerr et al., 1972). The viability of AM dye⁺ neurons was further confirmed via NucBlue staining. After overnight incubation of slices loaded with FuraRed-AM with four drops of NucBlue, the nuclei of FuraRed-AM⁺ neurons appeared markedly smaller and stained more intensely with NucBlue than did FuraRed^– neurons (Fig. 5F). This strongly suggests that AM dye⁺ neurons are undergoing chromatin condensation, another hallmark of programmed cell death (Galluzzi et al., 2009). This was quantified by comparing both the NucBlue emission intensity and the area of the nuclei. The nuclei of FuraRed⁺ neurons had significantly greater NucBlue emission than FuraRed^– neurons (68.61 ± 1.67 a.u. vs 37.04 ± 1.80 a.u., n = 25 vs 25, p < 0.0001) (Fig. 5G), and also had significantly smaller nuclear area (34.51 ± 0.59 μm² vs 90.70 ± 1.84 μm², n = 157 vs 157, p < 0.0001) (Fig. 5H). As an additional confirmation, the nuclear area of neurons positive for both FuraRed and NucBlue was compared with those of neurons expressing nuclear GFP because of infection with AAV9.hSyn.HI.eGFP-Cre.WPRE.SV40. The FuraRed⁺ neurons were again found to have a smaller nuclear area than the healthier GFPcre⁺ neurons (34.51 ± 0.59 μm² vs 97.47 ± 1.60 μm², n = 159 vs 113, p < 0.0001) (Fig. 5H). Based on these morphologic features, we can also conclude that these neurons are not undergoing ferroptosis. Ferroptosis is another form of programmed cell death characterized by lipid peroxidation and the presence of iron. However, ferroptosis does not involve cell shrinkage or chromatin condensation (Dixon et al., 2012; Chen et al., 2021; Qin et al., 2021).

We compared the time course of AM dye-staining and quenching of the emission of fluorescent proteins as indicators of neuronal death. As in Figure 5B, depriving slices of fresh media past when they would normally be fed resulted in widespread AM dye uptake and fluorescent protein quenching. Fifteen neurons from 5 such slices were selected randomly from the pool of neurons which underwent the FP-to-AM transition and could be visualized and tracked for the duration of the experiment. Using ImageJ, SBFI-AM and TurboRFP fluorescence intensity were both quantified (Extended Data Fig. 5-4A-C). The ratio of SBFI-AM to TurboRFP intensity for a given neuron was calculated, and the time interval at which this ratio exhibited its largest increase was taken to be time = 0 for the observable initiation of cell death. All other time points for that neuron were aligned accordingly. The largest drop in TurboRFP fluorescence (90.28 ± 3.84% of maximum to 58.78 ± 5.85%) and the largest increase in SBFI fluorescence (12.95 ± 6.22% of maximum to 51.08 ± 5.83%) were found to occur during the same time interval (Extended Data Fig. 5-4A). Graphing the SBFI/TurboRFP ratio itself over time further highlights how synchronous the two processes are, with the ratio experiencing its largest increase (0.13 ± 0.05 to 1.13 ± 0.13) at t = 0 (Extended Data Fig. 5-4B). Shrinkage of the soma, another indicator of programmed cell death, aligned very closely with these metrics, with largest drop in 2-D soma area (90.07 ± 5.08% of maximum to 70.03 ± 3.06%) occurring during the same time interval as did maximum SBFI uptake and maximum TurboRFP quenching (Extended Data Fig. 5-4C). This suggests that all three processes (SBFI staining, fluorescent protein quenching, and cell shrinkage) occur concurrently as programmed cell death begins. This is visualized in Extended Data Figure 5-4D, where a slice with plentiful TurboRFP expression and sparse SBFI uptake at Hour 0 is contrasted with the same slice at Hour 96, where 89% of TurboRFP⁺ neurons have quenched and SBFI uptake is pervasive. Thus, AM dye uptake can be a useful complement to enhance the temporal resolution of fluorescent protein quenching as an assay for cell death after acute injury.

Figure 5 and Extended Data Figure 5-4 establish that AM dye staining is an early biomarker of programmed cell death in the hippocampal organotypic slice preparation. To use AM dyes as a biomarker of programmed cell death, it is necessary to know how well the biomarker performs throughout the entire process of programmed cell death. Neuronal programmed cell death ends in microglial efferocytosis (Stolzing and Grune, 2004; Fricker et al., 2018; Mike and Ferriero, 2021). To test whether neurons undergoing programmed cell death remained AM-dye⁺ from the onset of programmed cell death through the final stage of efferocytosis, microglial efferocytosis of neurons was evaluated in the organotypic hippocampal slice preparation. Isolectin GS-IB4 conjugated to AlexaFluor-594 was used to visualize activated microglia. Slices were incubated with 10 μg/ml of isolectin for a minimum of 150 min, and were incubated overnight with 27 μm SBFI.

Survival duration of AM dye⁺ neurons

Isolectin⁺ microglia were frequently observed in close proximity to, and in direct contact with, SBFI⁺ cells (Fig. 6A). When imaged over several days, the entire timeline of the microglial-neuronal interaction could be observed, wherein an SBFI-AM⁺ cell is completely engulfed and consumed by microglia (Fig. 6B). Perhaps owing to the large number of neurons in the programmed cell death pathway waiting to be efferocytosed relative to the number of microglia available to consume those neurons, the duration of microglial efferocytosis was highly variable. Two slices were observed via serial imaging, and 18 SBFI⁺ neurons were selected randomly from the pool of neurons which could be observed undergoing the entire engulfment process. The amount of time those neurons spent in contact with microglia before being fully engulfed (Contact phase), the time spent being engulfed before the neurons underwent terminal cell shrinkage (Engulfment phase), and the time spent undergoing terminal neuronal shrinkage (Shrinkage phase) were all quantified (Fig. 6C). The Contact phase lasted an average of 10.17 ± 4.71 h, with the longest observed Contact taking 37 h and the shortest taking only 1 h. The Engulfment phase lasted an average of 16.43 ± 5.74 h, with the longest Engulfment taking 55.5 h and the shortest taking 1 h. Last, the Shrinkage phase lasted an average of 27.08 ± 5.19 h, with the longest taking 59 h and the shortest taking 4.5 h. For the neurons that could be observed undergoing all three phases, the average total duration of the entire engulfment process was 46.06 ± 7.86 h (n = 8). This, of course, does not take into account the additional time required for microglia to find their target neurons and move into contact with them. This would be expected to take longer in regions that are dense with dying neurons, and thus would be a source of additional variability. Notably, in a slice with TurboRFP⁺ neurons that was imaged over 3 d, no instances of microglial engulfment of these healthy neurons could be seen, despite the number of microglia present in the area increasing over the course of the experiment (Extended Data Fig. 6-1). This further supports the idea that microglia selectively target neurons in the programmed cell death pathway.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Microglial engulfment is the final stage in delayed neuronal death. A, Example images of active microglia stained with isolectin surrounding and engulfing SBFI⁺ neurons. Scale bars, 10 μm. B, Sequential images represent an SBFI⁺ neuron coming into contact with, and ultimately being destroyed by, microglia over 78 h. Scale bars, 10 μm. C, The duration of microglial engulfment is highly variable. The time during which microglia make contact with SBFI⁺ neurons but before they engulf them (Contact), the time during which the neurons are engulfed by the microglia but are not yet shrinking (Engulfment), and the time during which the neurons are undergoing terminal cell shrinkage while being engulfed (Shrinkage) all varied widely (n = 18 cells from 2 slices). For additional data, see Extended Data Figure 6-1.

These data indicate that AM dyes permeate neurons undergoing programmed cell death from the time of fluorescence quenching (Fig. 5B) through terminal efferocytosis (Fig. 6). AM dyes therefore can be used to estimate the time course of neuronal programmed cell death after acute, widespread neuronal injury in the organotypic slice preparation. Slices expressing TurboRFP and incubated with SBFI in standard culture conditions were imaged every 3-5 d for an 18 d period (Fig. 7A). A subset of SBFI-AM⁺ neurons was manually selected based on the following criteria: neurons that (1) were newly SBFI-AM⁺, (2) underwent terminal cell shrinkage and died over the course of the experiment, and (3) were located in regions with sparse SBFI staining, to aid in their precise visual identification over time. These criteria yielded 41 neurons from 4 slices, which were manually tracked for the duration of the experiment. The continued presence of TurboRFP⁺ neurons was used to verify that the slices remained healthy overall, and indeed TurboRFP positivity persisted even as the SBFI-AM⁺ population dwindled (Fig. 7B). The SBFI⁺ neurons were found to have a τ of 4.69 d and a half-life of 3.66 d, with some neurons surviving for up to 12 d (Fig. 7C). These data indicate that neurons undergoing programmed cell death can persist for up to 12 d as they await microglial efferocytosis, accounting for the delay referred to in the term Delayed Neuronal Death (Kirino, 1982). In a separate experiment, slices were loaded with isolectin and SBFI-AM and imaged serially over 5 d. τ was determined from the loss of the initial SBFI-AM⁺ cells for each slice, and τ was then plotted against the ratio of the ratio of the number of isolectin⁺ microglia to SBFI⁺ neurons, averaged over the first 24 h of the experiment (Fig. 7D). The negative correlation indicates that in slices with plentiful microglia relative to the number of dying neurons, the microglia can clear those neurons more quickly. This idea is also supported by the data in Figure 4D, where the rate of loss of PI⁺ neurons decreased with the number of PI⁺ neurons in the preparation (Fig. 4D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Survival of AM dye⁺ neurons. A, Schematic of serial imaging experiment to measure the survival of neurons which take up the cell-permeant AM dye SBFI. Slices were imaged every 3-5 d beginning on DIV 8. The duration of survival of SBFI⁺ neurons was calculated by subtracting the experiment day on which an SBFI⁺ was first seen, from the experiment day on which it was last seen. B, Images of SBFI^– and TurboRFP⁺ neurons at three time points during an experiment in B. While TurboRFP⁺ neurons persist, SBFI⁺ neurons are largely gone by day 11 of the experiment. Experiment day 3 = DIV 11. Scale bar, 50 μm. C, SBFI⁺ neurons can survive for a substantial period of time, consistent with earlier findings indicating that microglia efferocytosis is rate-limiting after widespread neuronal injury. Neurons that became SBFI⁺ and died during an 18-d-long experiment had a τ of 4.69 d and a half-life of 3.66 d, although some were still alive after 12 d. n = 41 cells from 4 slices. D, Survival of SBFI⁺ neurons correlates with the relative levels of microglia to their phagocytic targets. In slices stained with isolectin and SBFI, both isolectin⁺ microglia and SBFI⁺ neurons were imaged over 5 d, then tracked with TrackMate. The time constant was determined from the progressive loss of the initial SBFI⁺ cells for each slice. τ was then plotted against the ratio of the number of isolectin⁺ microglia to SBFI⁺ neurons, averaged over the first 24 h of the experiment.

These data demonstrate that AM dyes are biomarkers for neurons throughout programmed cell death in hippocampal organotypic slice preparation, from the earliest stages of fluorescence quenching (Fig. 5B) to terminal effertocytosis by microglia (Fig. 6). Biomarkers based on more polar organic dyes such as PI are only positive at late stages of programmed cell death (Fig. 2), suggesting that membrane permeability must increase during programmed cell death. Although the dependence of biomarker staining on pathologic levels of membrane permeability exploits this unique feature of programmed cell death (Bouchier-Hayes et al., 2008), it also raises the possibility of false positive stains. This is because membrane permeabilization, effected by detergents and alcohols (Felix, 1982; Stadler et al., 2010), is a key step in cytochemical and immunohistochemical staining that allows the biomarkers access to the cytoplasmic and nuclear contents. To be able to evaluate potential false positive biomarker staining, we carefully characterized the membrane permeability of neurons undergoing programmed cell death.

Membrane permeability during programmed neuronal death after perinatal acute brain injury

Annexin V (AlexaFluor-594 conjugate) is a biomarker of membrane deterioration during programmed cell death. Annexin V selectively stains cells undergoing programmed cell death by binding to phosphatidylserine, a component of the inner leaflet of the cytoplasmic membrane which is found on the outer leaflet of the membranes during programmed cell death (van Engeland et al., 1998; Segawa and Nagata, 2015; Zhang et al., 2018). SBFI-AM positivity overlapped significantly with positivity for Annexin V. In slices incubated with 50 μl Annexin V per milliliter of culture media for a minimum of 150 min, 491 of 506 SBFI⁺ neurons (or 97.04%) were found to also be positive for Annexin V (Fig. 8A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Neurons in the programmed cell death pathway undergo progressive membrane deterioration. A, Evidence of membrane injury in AM dye⁺ neurons. Example image showing broad overlap between SBFI positivity and membrane injury indicated by Annexin V staining of phosphatidylserine moieties in the cytoplasmic membrane (left). Larger example images of an individual neuron that is positive for both SBFI and Annexin V (middle). Quantification of SBFI/Annexin V overlap, showing that 97% of SBFI⁺ neurons are also positive for Annexin V (right). n = 506 cells from 1 slice. Scale bar, 50 μm. B, AM dye⁺ neurons exhibit extensive Pannexin pore expression. Example images represent a group of neurons with PO-PRO-1 positivity (left), and another group in the same slice with extensive FuraRed positivity where nearly all FuraRed neurons are also PO-PRO-1⁺ (right). Scale bars, 25 μm. C, FP⁺ neurons do not exhibit Pannexin pore expression. Examples images represent a group of neurons with PO-PRO-1 positivity (left), and another group in the same slice with TurboRFP positivity that are all PO-PRO-1⁺ (right). Scale bars, 25 μm. D, Dying neurons accumulate intracellular sodium ions. In the 5 h following SBFI uptake, [Na⁺]_i increased by 8.51 ± 2.35 mm/h. n = 7 cells from 2 slices. Later in the programmed cell death process, [Na⁺]_i largely plateaued, only increasing by 0.44 ± 0.08 mm/h over 24 h. n = 46 cells from 4 slices. Finally, during terminal cell shrinkage, [Na⁺]_i rapidly decreased by 13.53 ± 2.93 mm/h in the 3 h before cell death. n = 5 cells from 5 slices. p < 0.0001 for Initial increase versus Plateau. p < 0.0001 for Plateau versus Terminal cell shrinkage. p < 0.0001 for Initial increase versus Terminal cell shrinkage. E, Blocking pannexin-1 hemichannels with 50 μm of the 10Panx for 30 min decreased [Na⁺]_i (–50.28 ± 13.49%, p = 0.0004, n = 64 neurons from 5 slices), whereas 10Panx application in microglia-depleted slices increased [Na⁺]_i (21.04 ± 8.55%, p = 0.0162, n = 74 neurons from 3 slices). F, Blockage of Na/K ATPases increases [Na⁺]_i in neurons undergoing programmed cell death. Acute application of 10 μm ouabain for 30 min increased [Na⁺]_i by 86.84% in neurons with an initial [Na⁺]_i between 0 and 15 mm (n = 84 cells, p = 0.0048), 128.19% in neurons with an initial [Na⁺]_i between 15 and 50 mm (n = 25 cells, p = 0.0017), had no effect on neurons with an initial [Na⁺]_i between 50 and 100 mm (n = 6 cells, p = 0.2692), and with all cells combined increased [Na⁺]_i by 93.32% (n = 115 cells, p < 0.0001). n = 6 slices. G, Neurons in the programmed cell death pathway have depolarized resting membrane potentials. FuraRed⁺ neurons had RMPs that were highly depolarized compared with Clm⁺ neurons (–24.78 ± 2.93 mV vs –52.78 ± 4.13 mV, n = 9 cells for each, p < 0.0001). H, Neurons in the programmed cell death pathway have compromised nuclear membranes. An example image showing that many SBFI⁺ neurons become PI⁺. Scale bar, 25 μm. I, Compromised nuclear membranes are a feature of late-stage programmed cell death. SBFI⁺ neurons that are also PI⁺ had higher [Na⁺]_i than those that are PI^– (41.93 ± 2.35 mm vs 15.21 ± 1.47 mm, n = 106 cells vs 221 cells, n = 6 slices, p < 0.0001), indicating that PI positivity does not occur until relatively late in the programmed cell death process. J, Neurons undergoing programmed cell death demonstrate a wide range of intracellular calcium concentrations (median = 12.81 μm, n = 335).

In addition to membrane deterioration, active processes may underlie the increase in membrane permeability of neurons undergoing programmed cell death. Pannexin channels are large-conductance ion channels that are inserted into the neuronal membrane after injury (Thompson et al., 2006). AM dye positivity also overlapped with positivity for PO-PRO-1, a dye that is preferentially taken up by neurons after pannexin 1 (PANX1) channels have been inserted in the cytoplasmic membrane. After overnight incubation with 28 μm of the AM dye FuraRed, and incubation with 5 μm of PO-PRO-1 for a minimum of 180 min, many PO-PRO-1⁺ cells could be seen, and neurons that were FuraRed⁺ were almost certain to also be PO-PRO-1⁺ (Fig. 8B). Conversely, even in slices with plentiful PO-PRO-1 positivity, PO-PRO-1 never overlapped with TurboRFP⁺ neurons (Fig. 8C), providing further evidence that FP⁺ healthy neurons and neurons with high membrane permeability are separate populations.

To test whether the timing of biomarker positivity reflects the degree of permeability of the cytoplasmic membrane of neurons undergoing programmed cell death, we exploited the shift in wavelengths at which AM dyes absorb light because of changes in local ionic concentrations. We used this property to test whether the timing of biomarker positivity reflects the degree of deterioration of the cytoplasmic membrane's function of separating extracellular and intracellular fluids (Macklis and Madison, 1990). We found that AM dye⁺ neurons also accumulate intracellular sodium ions, with the rate of change in the intracellular sodium concentration ([Na⁺]_i) varying widely depending on the stage of programmed cell death. Using the ratiometric properties of SBFI-AM, [Na⁺]_i was quantified in neurons over the course of 24 h serial imaging experiments. Neurons that took up SBFI-AM during the experiment exhibited an [Na⁺]_i increase of 8.51 ± 2.35 mm/h (n = 7 cells from 2 slices) over the first 5 h that they were SBFI⁺, further confirming that a large increase in membrane permeability is correlated with AM dye uptake. Later in programmed cell death [Na⁺]_i largely plateaued, only increasing by 0.44 ± 0.08 mm/h (n = 46 cells from 4 slices) over 24 h. Finally, during microglial engulfment and terminal cell shrinkage the trend reversed and [Na⁺]_i rapidly decreased by 13.53 ± 2.93 mm/h in the 3 h before cell death (n = 5 cells from 5 slices; Fig. 8D). These rates of change in [Na⁺]_i were all significantly different (p < 0.0001 for all comparisons). Thus, neurons undergoing programmed cell death experience a large early influx of Na⁺, followed by a potentially lengthy period where [Na⁺]_i is elevated but only gradually increasing, and then ultimately [Na⁺]_i drops precipitously during terminal cell shrinkage and efferocytosis. This final drop is not currently well understood, but it may be because of the near-total engulfment of the neurons by microglia at this stage (Fig. 6) cutting the neurons off from the high levels of Na⁺ in the extracellular space.

The increase in [Na⁺]_i during programmed cell death could be because of increased influx, or decreased extrusion of Na⁺. Blockade of pannexin pores with the 50 μm of the Panx-1 mimetic inhibitory peptide 10Panx for 30 min reduced [Na⁺]_i by 50.28 ± 13.49% in preparations with intact microglial populations (Fig. 8E). To test whether microglia might contribute to the role of pannexins in neuronal membrane permeability, slices were depleted of microglia by incubating them with 40 μl of liposomal clodronate (Clodrosome) suspension per milliliter of culture media as previously described (Park et al., 2015). In microglia-depleted slices, 10Panx application no longer reduced [Na⁺]_i, and indeed increased [Na⁺]_i by 21.04 ± 8.55%. These findings support the ideas that pannexins contribute to the increase in membrane permeability during neuronal programmed cell death, that the increase in permeability contributes to the increase in Nai⁺, and that microglia contribute to pannexin functioning (Yeung et al., 2020).

To address a potential role of decreased Na extrusion in the elevation of [Na⁺]_i in neurons undergoing programmed cell death, we manipulated sodium-potassium adenosine triphosphatase (Na/K ATPase) pharmacologically and by inhibition of ATP production. We found that, despite their increased membrane permeability and elevated [Na⁺]_i, neurons in the programmed cell death pathway are still producing and using ATP to power transport mechanisms and regulate ionic gradients. Acute application of 10 μm ouabain (a blocker of Na/K ATPases) for 30 min increased [Na⁺]_i by 86.84% in neurons with an initial [Na⁺]_i between 0 and 15 mm (n = 84 neurons from 6 slices, p = 0.0048), 128.19% in neurons with an initial [Na⁺]_i between 15 and 50 mm (n = 25 neurons from 6 slices, p = 0.0017), had no effect on neurons with an initial [Na⁺]_i between 50 and 100 mm (n = 6 neurons from 6 slices, p = 0.2692), and with all neurons combined increased [Na⁺]_i by 93.32% (n = 115 neurons from 6 slices, p < 0.0001) (Fig. 8F). Using whole-cell patch-clamp recordings of FuraRed⁺ and Clm⁺ neurons from the same slices, the resting membrane potential (RMP) of the FuraRed⁺ neurons in culture media (extracellular potassium concentration = 5.3 mm) was found to be greatly depolarized with respect to the RMP of Clm⁺ neurons (−24.78 ± 2.93 mV vs −52.78 ± 4.13 mV, n = 9 vs 9, p < 0.0001) (Fig. 8G). These results suggest that neurons undergoing programmed cell death are using ATP to remove excess Na⁺ from the intracellular space, but that as the membrane permeability continues to increase over time active transport is overwhelmed, resulting in membrane depolarization and increases in intracellular sodium.

The progressive increase in membrane permeability is not limited to the cytoplasmic membranes. Many SBFI⁺ neurons also demonstrated nuclear PI staining (Fig. 8H), indicating that neurons undergoing programmed cell death eventually experience a loss of nuclear membrane integrity. Indeed, PI staining would appear to be selective for neurons in late stages of programmed cell death that have already experienced a significant increase in membrane permeability, as SBFI⁺ neurons that were also PI⁺ had higher [Na⁺]_i than those that were PI^– (41.93 ± 2.35 mm vs 15.21 ± 1.47 mm, n = 106 cells vs 221 cells, n = 6 slices, p < 0.0001) (Fig. 8I).

The increase in the ionic permeability of the cytoplasmic membrane permeability in neurons was also supported by measurements of cytosolic calcium (Ca²⁺), using ratiometric determination of intracellular calcium concentrations ([Ca²⁺]_i) obtained with FuraRed-AM (Fig. 8J). As with [Na⁺]_i, neurons undergoing programmed cell death exhibited a wide range of [Ca²⁺]_i values, with a median value of 12.81 μm (n = 335 neurons), which is well above the 50 nm cytoplasmic calcium concentration of healthy hippocampal pyramidal cells (Garaschuk et al., 1997).

These findings indicate that, after acute injury, neuronal programmed cell death progresses over days to weeks. Neuronal programmed cell death is characterized by membrane deterioration and increasing membrane permeability that drives many of the biomarkers for programmed cell death (Galluzzi et al., 2009; Schmued, 2016; Zhang et al., 2018). Neuronal programmed cell death terminates with microglial efferocytosis (Fricker et al., 2018; Mike and Ferriero, 2021). As shown in Figure 4, one-time staining for a biomarker is a poor estimate for the rate of neuronal death (i.e., entry into the programmed cell death pathway); in the case of PI staining for cell death in organotypic slices, it was off by 2 orders of magnitude compared with extinction of fluorescent protein emission. However, the data also provide evidence for additional potential sources of error in the assessment of neuronal death. First, some experimental manipulations may affect microglia rather than neurons. These manipulations may affect the rate of efferocytosis as, for example, depletion of microglia by Clodrosome (Park et al., 2015), and as a consequence affect the number of visible (i.e., PI⁺) neurons undergoing programmed cell death (Fig. 1). The manipulation may also affect interactions between microglia and neurons other than efferocytosis; for example, neurons in microglia-depleted slices did not exhibit a pannexin-mediated increase in sodium permeability (Fig. 8E). Second, a manipulation that affects the permeability of the membrane of neurons undergoing programmed cell death could increase neuronal staining with biomarkers and artifactually change the size of the pool of biomarker-receptive neurons in Extended Data Figure 4-1, as schematized in Figure 1. Third, neurons that necrose or are phagocytosed before the assay for neuronal death being performed will lead to an underestimate of neuronal injury. In the next section, we consider how effects of interventions on microglia can lead to misinterpretations of the neurotoxic or neuroprotective impact of experimental interventions.

Altering microglial activity affects the number of visible dying neurons

As schematized in Extended Data Figure 4-1, the number of neurons positive for a biomarker of programmed cell death depends on the rates of neuronal entry and exit into the biomarker-receptive pool. Because exit from this pool entails microglial efferocytosis, we tested the possibility that altering the rate of microglial efferocytosis will alter the size of the biomarker-receptive pool of neurons undergoing programmed cell death. Seventeen slices (DIV 5-7 at the time of imaging) were depleted of microglia treatment with Clodrosome for 3 d before imaging (Fig. 9A). The number of PI⁺ cells in preparations depleted of microglia were counted relative to the number of PI cells in 6 control slices (Fig. 9B). In a 3D volume 590 × 590 μm in area and extending 80 μm along the z axis from the surface of the slices, Clodrosome-treated slices had significantly more PI⁺ cells than control slices (1836.53 ± 206 vs 936.17 ± 134, p = 0.0208). Similarly, in a z axis range extending from 80 to 120 μm deep from the slice surface, Clodrosome-treated slices again had more PI⁺ cells than control slices (373.35 ± 72.30 vs 34.83 ± 6.37, p = 0.0123). The depth dependence reflects the degree of traumatic injury incurred during slice preparation (Dzhala et al., 2012). The rate of microglial efferocytosis comprises the rate of exit from the biomarker receptive pool in the scheme shown in Extended Data Figure 4-1, so reducing the exit rate increases the number of neurons in the PI-receptive pool (Extended Data Fig. 4-1C vs Extended Data Fig. 4-1B). These data demonstrate that the number of visibly dying neurons in single-time point assays is strongly dependent on the rate of microglial efferocytosis, and manipulations that alter the rate of efferocytosis can be misinterpreted as neuroprotective or neurotoxic (Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 9.

Altering microglial activity affects the number of visible dying neurons. A, Chronic incubation for a minimum of 3 d with liposomal Clodrosome-depleted slices of microglia. Scale bars, 50 μm. B, Microglial depletion results in more PI⁺ neurons. Clodrosome-treated slices had more PI⁺ neurons between 0 and 80 μm deep (1836.53 ± 206 vs 936.17 ± 134, p = 0.0208) and between 80 and 120 μm deep (373.35 ± 72.30 vs 34.83 ± 6.37, p = 0.0123). n = 23 slices. C, KYNA reduces the number of visible dying neurons. Chronic application of 3 mm of KYNA for 10-12 d to block seizures resulted in a reduction in the number of PI⁺ neurons compared with control slices (–59.37 ± 3.51%, p < 0.0001). Chronic application of 0.2 mg/ml of Clodrosome resulted in an increase in the number of PI⁺ neurons compared with control slices (224.97 ± 33.24%, p < 0.0001), as did combined application of KYNA and Clodrosome (97.13 ± 25.27%, p = 0.0023). n = 53 slices. D, KYNA did not affect the total number of neurons undergoing programmed cell death, assayed by SBFI-AM staining. Chronic application of 3 mm of KYNA had no effect on the number of SBFI⁺ neurons (n = 23 slices). E, KYNA reduces the number of visible healthy neurons. Chronic application of 3 mm of KYNA for 10-12 d caused a reduction in the number of Clm⁺ neurons compared with control slices (–40.52 ± 6.36%, p < 0.0001), as did combined application of KYNA and 0.2 mg/ml Clodrosome (–22.57 ± 7.20%, p = 0.0086). Chronic application of Clodrosome by itself had no effect over the same time period. n = 53 slices. For additional data, see Extended Data Figure 9-1.

Based on PI staining, we had concluded in our prior study (Berdichevsky et al., 2012) that seizures were neurotoxic. However, it has recently been reported that neuronal activity (Grinberg et al., 2011) and seizures (Abiega et al., 2016) reduce the rate of microglial efferocytosis. To test whether the effects of seizure suppression on the number of PI⁺ neurons were because of changes in the rate of microglial efferocytosis, seizures were suppressed in control hippocampal slice cultures and in hippocampal slice cultures in which microglia had been depleted by Clodrosome. DIV 10-12 Clm slices were continuously treated with 3 mm kynurenic acid (KYNA) following the first culture media change on DIV 3, to suppress seizure activity (Berdichevsky et al., 2012). Identical Clm slices were treated with 0.2 mg/ml Clodrosome to deplete the slices of microglia, and other slices received both KYNA and Clodrosome. All slices were incubated with PI before imaging. Cell counts for a set of slices made from a given mouse pup were normalized to the counts for the untreated control slice(s) from that pup, expressed as percentages, then pooled together with the normalized values from slices from all other pups.

The number of PI⁺ neurons was significantly affected by these treatments (Fig. 9C). Seizure suppression by KYNA reduced the number of PI⁺ neurons compared with control slices (−59.37 ± 3.51%, p < 0.0001). Depletion of microglia by Clodrosome increased the number of PI⁺ neurons compared with controls (224.97 ± 33.24%, p < 0.0001), as did combined seizure suppression and microglial depletion by application of KYNA and Clodrosome (97.13 ± 25.27%, p = 0.0023) (n = 53 slices). These findings indicate that suppressing seizure activity reduces the number of neurons that are PI⁺ at any given time. This could occur either by reducing the number of neurons entering the programmed cell death pathway (i.e., a neuroprotective effect) or by accelerating the efferocytosis of dying neurons via enhanced microglial activity (a false positive neuroprotective effect; Fig. 1) (Grinberg et al., 2011; Abiega et al., 2016). The size of the increase in PI⁺ cells when both seizures and efferocytosis were blocked by KYNA (97%) was between the reduction in PI⁺ cells by seizure block alone (−59%) and the increase in PI⁺ cells by microglial depletion (225%). Thus, microglial depletion reduced the apparent neuroprotective effect of stopping seizures, supporting the idea that stopping seizures increases efferocytosis rather than reduces neuronal programmed cell death. However, KYNA reduced PI counts roughly by half in both control and microglia-depleted conditions (a 59% reduction in control conditions, and a reduction from a 225% increase to a 97% increase in Clodrosome). Thus, there is a potential microglia-independent effect of KYNA, so that we could not rule out an additional neuroprotective effect of stopping seizures. The difficulty in separating these effects is largely because of the large effect of microglial depletion on the PI counts (i.e., because of microglial phagocytosis being rate-limiting in these experiments). We therefore turned to additional biomarkers to assess these effects.

In slices ranging in age from DIV 2 to DIV 14, there was no difference in the number of neurons which took up SBFI-AM between control slices and age-matched slices in which seizures were blocked by long-term incubation in KYNA starting on DIV 1 (264.27 ± 81.04 vs 276.33 ± 83.96, n = 23 slices, p = 0.9189; Fig. 9D). Because SBFI staining is present throughout programmed cell death, this finding suggests that seizures did not alter the balance of neurons entering and exiting the programmed cell death pathway. But that interpretation does not seem to be consistent with the reduction in PI⁺ neurons when seizures were blocked with kynurenate (Berdichevsky et al., 2012) (Fig. 9C), if the reduction in PI⁺ neurons reflects enhanced exit via efferocytosis when seizures are blocked (Grinberg et al., 2011; Abiega et al., 2016). We found, however, that blocking seizures also enhanced entry into the programmed cell death pathway, evidenced by an increased rate of extinction of fluorescence of transgenically expressed fluorophores (Fig. 9E). Chronic application of 3 mm KYNA for 10-12 d resulted in a decrease in the number of Clm⁺ neurons by −40.52 ± 6.36% (p < 0.0001), and KYNA + Clodrosome resulted in a decrease of −22.57 ± 7.20% (p = 0.0086) (n = 53 slices). The application of Clodrosome without KYNA had no effect over the same time period, indicating that the increased PI⁺ count in Clodrosome-treated slices is because of reduced efferocytosis and not because of other factors, such as neurotoxicity of Clodrosome, neurotoxicity of the intracellular contents of microglia killed by Clodrosome, or a loss of pro-trophic input from microglia leading to increased neuronal death. These data indicate that prolonged seizure suppression via blockade of ionotropic glutamate receptors can adversely affect healthy neurons, but that a dearth of microglia has no such effect. The difference in survival of healthy FP⁺ neurons in KYNA was not significantly different from in KYNA + Clodrosome (Mennerick and Zorumski, 2000).

In addition to reducing the number of Clm⁺ neurons, chronic KYNA application also reduced the intensity of Clm fluorescence in surviving neurons (Extended Data Fig. 9-1A,B), as would be expected if seizure suppression were adversely affecting healthy neurons. There was also no indication that suppressing seizures with chronic TTX application for 11-15 d promoted neuronal survival compared with Control conditions when using Clm fluorescence extinction as a metric (Extended Data Fig. 9-1C).

Thus, the consistent SBFI-AM staining when seizures are blocked (Fig. 9D) reflects both the increased entry into the programmed cell death pathway (Fig. 9E) and the increased rate of exit (Fig. 9C). In contrast to our original interpretation of one-time PI staining data (Berdichevsky et al., 2012), in which we concluded that seizures were neurotoxic, we find that blocking seizures enhances entry into the programmed cell death pathway, presumably as a result of reduced neuronal activity (Ruijter et al., 1991; Heck et al., 2008; Wong et al., 2018). Blocking seizures also accelerated exit from the programmed cell death pathway by enhancing microglial efferocytosis (Grinberg et al., 2011; Abiega et al., 2016). This is evidenced by the block of the apparent neuroprotective effect of kynurenate (decreased number of PI⁺ neurons) in the slices depleted of microglia by Clodrosome (Figs. 1, 9C).

These findings indicate that the number of visibly dying neurons is dependent on the number of microglia available to consume those neurons (Fig. 6) as well as the phagocytic efficiency of the microglia that consume them (Fig. 9).

Persistent errors in neuronal death estimates

Based on Figures 5 and 7, one strategy might be to wait until organotypic slices are older (e.g., 2 weeks in vitro) before testing for neuroprotective or neurotoxic effects (Strasser and Fischer, 1995; Adamchik et al., 2000; Bonde et al., 2002; Morrison et al., 2006). This would avoid the problems of neuronal death because of brain slicing and the delayed clearance of those neurons. We performed such an experiment by imaging DIV 14 slices for GFP and PI, then exposing the slices to 0, 1, or 5 μm KA for 24 h, then continuing to image them at 24 h intervals. There was a reasonable dose-toxicity response: slices exposed to 5 μm KA had a larger number of PI⁺ cells on subsequent days of the experiment than did slices exposed to 1 μm KA (Fig. 10A). At 24 h, 5 μm KA slices had 171.2 ± 10.2% of their baseline (pre-KA) amount of PI⁺ cells, compared with 108.6 ± 9.1% in 1 μm slices (p = 0.004, n = 8 slices). At 48 h, 5 μm KA slices had 212.7 ± 42.1% of their baseline PI compared with 108.1 ± 15.0% in 1 μm slices (p = 0.0209, n = 8 slices). And at 72 h, 5 μm KA slices had 190.2 ± 50.6% of their baseline PI compared with 103.4 ± 12.6% in 1 μm slices (p = 0.0558, n = 8 slices).

• Download figure
• Open in new tab
• Download powerpoint

Figure 10.

Persistent errors in neuronal death estimates. A, Dose–response curve in older slices exposed to kainic acid. DIV 14 slices exposed to 5 μm KA had a larger number of PI⁺ cells on subsequent days of the serial imaging experiment than did slices exposed to 1 μm KA, when taken as a percentage of the baseline amount (24 h: 171.2 ± 10.2% vs 108.6 ± 9.1%, p = 0.004; 48 h: 212.7 ± 42.1% vs 108.1 ± 15.0%, p = 0.0209; 72 h: 190.2 ± 50.6% vs 103.2 ± 12.6%, p = 0.0558). There was no difference in PI staining between slices exposed to 1 μm KA and slices which only received TTX as a control (24 h: 108.6 ± 9.1% vs 107.6 ± 25.2%, p = 0.9645; 48 h: 108.1 ± 15.0% vs 117.8 ± 29.8%, p = 0.7519; 72 h: 103.4 ± 12.6% vs 102.4 ± 25.7%, p = 0.9695). n = 12 slices. B, Persistent cell death in older slices. A considerable amount of newly PI⁺ cells appeared on each day, even in 1 μm and TTX-only conditions despite the lack of increase in total PI positivity seen in those conditions in A. n = 7 slices. C, Rate of clearance of initial PI⁺ cells in older slices. The initial population of PI⁺ cells observed during baseline imaging prior to 5 μm KA application decreased to 61.0 ± 0.9% 24 h after KA. The population in 1 μm KA slices decreased to 72.4 ± 4.1% 24 h after KA, 56.0 ± 2.8% 48 h after KA, and 46.0 ± 2.7% 72 h after KA. This is again not reflected in the counts of total PI cells seen in A. n = 8 slices. D, Rate of death of initial GFP⁺ cells in older slices. The initial population of GFP⁺ cells observed during baseline imaging prior to 5 μm KA application decreased to 28.6 ± 1.3% 24 h after KA. The population in 1 μm KA slices decreased to 68.2 ± 5.7% 24 h after KA, 53.4 ± 5.1% 48 h after KA, and 40.1 ± 5.6% 72 h after KA. n = 8 slices.

While these results indicate that 5 μm KA is toxic, the relatively stable PI counts for 1 μm KA and TTX-only slices (Fig. 10A) would seem to imply a lack of additional neuronal death under these conditions. However, we found that there was a considerable amount of newly PI⁺ cells each day of the experiment, even in 1 μm KA and TTX-only slices (Fig. 10B). The 5 μm KA data are not shown here (and TTX data are limited) because of difficulties in tracking neurons over all 4 d of the experiment, particularly in slices with greater amounts of cell death.

If there is considerable additional cell death in these slices, then why was it not detected in Figure 10A? By tracking the initial population of PI⁺ cells over the course of the experiment, we can see that there is a high rate of clearance of these neurons by microglia in all slices (Fig. 10C). The initial population of PI⁺ cells observed during baseline imaging before 5 μm KA application decreased to 61.0 ± 0.9% 24 h after KA, and the population in 1 μm KA slices decreased to 72.4 ± 4.1% 24 h after KA, 56.0 ± 2.8% 48 h after KA, and 46.0 ± 2.7% 72 h after KA. This reduction in PI⁺ cells because of microglial clearance is matched by the ongoing death of healthy, GFP⁺ neurons (Fig. 10D), which in this older cohort of slices was high even in control conditions. The initial population of GFP⁺ cells observed during baseline imaging before 5 μm KA application decreased to 28.6 ± 1.3% 24 h after KA. The population in 1 μm KA slices decreased to 68.2 ± 5.7% 24 h after KA, 53.4 ± 5.1% 48 h after KA, and 40.1 ± 5.6% 72 h after KA. As with Figure 10B, 5 μm KA data are not shown (and TTX data are limited) in Figure 10C, D because of difficulties in tracking neurons over all 4 d of the experiment, particularly in slices with greater amounts of cell death. However, these data would not alter the conclusions of the experiment.

Thus, the reasonable results in Figure 10A are actually because of a combination of two serious errors: nondetection of ongoing clearance of PI⁺ neurons, and ongoing neuronal death even in control conditions (likely a consequence of subjecting older slices to daily imaging sessions). We next tested whether altering the membrane permeability of neurons already committed to programmed cell death would alter the rate of biomarker positivity, thereby producing spurious neurotoxic or neuroprotective signals.

Acute ethanol (EtOH) exposure in the developing hippocampus

EtOH and other alcohols increase the permeability of biological membranes (Komatsu and Okada, 1995; Ly and Longo, 2004) and so are used to increase access of stains to the cytoplasm of living cells (R. P. Jones and Greenfield, 1987; Vitale et al., 1993) as well as to permeabilize cell membranes after fixation for immunocytochemical (Viryasova et al., 2019) and ISH studies (Felix, 1982). EtOH consumption in the first 2 trimesters of pregnancy is also clearly linked to human fetal alcohol syndrome (May et al., 2013), which can include microcephaly and developmental disability in addition to dysmorphic facies and behavioral disorders (Wozniak et al., 2019). A well-established line of evidence for the neurotoxicity of alcohol is based on one-time assays of biomarkers of programmed cell death after alcohol exposure, but the experimental exposure was maximally neurotoxic at later stages of development than occurs in human fetal alcohol syndrome (Ikonomidou et al., 2000). These EtOH toxicity studies used the DeOlmos silver stain, a protocol that was developed to sparsely stain degenerating axons, and as such does not include detergent exposure for membrane permeabilization (DeOlmos and Ingram, 1971). We tested whether EtOH at the clinically relevant concentrations used in these studies could alter the membrane permeability of biomarkers of programmed cell death to produce a false positive neurotoxic signal.

DIV 5-7 Clm slices, as well as WT slices infected with AAV9-hSyn-EGFP, were incubated with PI for 60 min, then imaged. There were 1603.29 ± 191.42 PI⁺ neurons between 0 and 80 μm from the surface of the slices, significantly more than the 320.53 ± 68.36 between 80 and 120 μm from the surface (n = 17 slices, p < 0.0001) (Fig. 11A). These slices were then incubated with 100 mm EtOH for 24 h, then incubated with PI again and imaged again. Deep in the slices where there was less damage because of slicing (Dzhala et al., 2012), 24 h of EtOH exposure resulted in an increase in PI positivity that was larger than the change seen near the surface (89.79 ± 23.64% vs 3.11 ± 3.28%, p = 0.0022), and larger than the change seen near the surface (89.79 ± 23.64% vs 15.49 ± 9.58%, p = 0.014) or deeper (89.79 ± 23.64% vs 18.57 ± 16.75%, p = 0.0298) in Control slices (n = 9 Control slices and n = 7 EtOH slices) (Fig. 11B). These results suggest that neurons at relatively early stages of programmed cell death were more common deeper in the slices and became more permeable to PI during EtOH exposure. Consistent with this mechanism, cytoplasmic sodium increased dramatically in EtOH-exposed SBFI⁺ neurons that had had physiological sodium concentrations, but not in EtOH-exposed neurons in late stages of programmed cell death whose membranes were already permeable and whose cytoplasmic sodium was consequently already very high (Fig. 11C). Acute application of 100 mm EtOH for 60 min increased [Na⁺]_i by 266.84 ± 65.41% in neurons with an initial [Na⁺]_i between 0 and 15 mm (n = 68 cells, p < 0.0001), 59.29 ± 15.83% in neurons with an initial [Na⁺]_i between 15 and 50 mm (n = 73 cells, p = 0.0007), and had no effect on neurons with an initial [Na⁺]_i between 50 and 100 mm (n = 54 cells, p = 0.0621) (n = 10 slices) (Fig. 11C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 11.

Effects of EtOH on dying and healthy neurons. A, PI positivity is greater near the surface of slices than in deeper regions. In Clodrosome-treated slices there were 1603.29 ± 191.42 PI⁺ neurons between 0 and 80 μm from the surface of the slices, compared with 320.53 ± 68.36 between 80 and 120 μm from the surface (p < 0.0001). n = 17 slices. B, EtOH application increases PI positivity deep in the slice. Application of 100 mm EtOH for 24 h to a subset of the slices from A resulted in an increase in PI positivity deep in slices that was larger than the change seen near the surface (89.79 ± 23.64% vs 3.11 ± 3.28%, p = 0.0022), and larger than the change seen near the surface (89.79 ± 23.64% vs 15.49 ± 9.58%, p = 0.014) or deeper (89.79 ± 23.64% vs 18.57 ± 16.75%, p = 0.0298) in Control slices. n = 8 Control slices and 9 EtOH slices. C, EtOH application increases [Na⁺]_i in apoptotic neurons. Acute application of 100 mm EtOH for 60 minutes increased [Na⁺]_i by 266.84 ± 65.41% in neurons with an initial [Na⁺]_i between 0 and 15 mm (n = 68 cells, p < 0.0001), 59.29 ± 15.83% in neurons with an initial [Na⁺]_i between 15 and 50 mm (n = 73 cells, p = 0.0007), and had no effect on neurons with an initial [Na⁺]_i between 50 and 100 mm (n = 54 cells, p = 0.0621). n = 10 slices. D, EtOH application causes shrinkage in apoptotic neurons. Acute application of 100 mm EtOH for 90 minutes caused a 7.56 ± 0.56% decrease in cell area in SBFI⁺ neurons (n = 224 cells from 5 slices, p < 0.0001) but caused no change in cell area in TurboRFP⁺ neurons (n = 128 cells from 4 slices, p = 0.4174). E, Example images of Clm⁺ neurons before and after 24 h of 100 mm EtOH exposure, demonstrating little to no quenching or morphological effects. Slice imaged on DIV 7 and 8. Scale bars, 50 μm. F, EtOH does not reduce the number of visible healthy neurons. Application of 100 mm EtOH for 24 h caused no reduction in the percentage of GFP^– or Clm⁺ neurons that could be tracked using TrackMate (GFP: 88.95 ± 3.54% vs 91.44 ± 1.95%, n = 12 slices; Clm: 96.98 ± 1.09% vs 93.87 ± 1.70%, n = 6 slices). For additional data, see Extended Data Figure 11-1.

Further supporting the idea of increased membrane permeability in neurons in early stages of programmed cell death, but not healthy neurons, were the changes in somatic area in SBFI⁺ and TurboRFP⁺ neurons (Fig. 11D). Ninety minutes of 100 mm EtOH exposure caused a 7.56 ± 0.56% decrease in area in SBFI⁺ neurons (n = 224 cells from 5 slices, p < 0.0001), but caused no significant change in area in TurboRFP⁺ neurons (n = 128 cells from 4 slices, p = 0.4174).

Finally, the effect of EtOH on the permeability of neurons undergoing programmed cell death rather than healthy neurons was supported by the finding that 24 h of 100 mm EtOH had no effect on healthy neurons expressing the fluorophores GFP or Clm (Fig. 11E,F). In EtOH-treated slices, 88.95 ± 3.54% of neurons in GFP slices and 96.98 ± 1.09% of neurons in Clm slices survived the treatment and were successfully tracked using TrackMate, compared with 91.44 ± 1.95% and 93.87 ± 1.70% in Control slices (n = 12 GFP slices and 6 Clm slices). Twenty-four hours of EtOH also had no effect on the number of SBFI⁺ neurons in a slice, as the ratio of SBFI⁺ to TurboRFP⁺ neurons did not change over that time (Extended Data Fig. 11-1), even as photobleaching because of frequent imaging caused a decrease in TurboRFP fluorescence. This further supports the notion that EtOH application does not result in new neurons entering the programmed cell death pathway.

These findings indicate that acute exposure of the perinatal hippocampus to high levels of EtOH increased neuronal labeling with biomarkers of programmed cell death by enhancing the permeability of the membrane of neurons undergoing programmed cell death rather than damaging healthy neurons. These data do not indicate that EtOH is not a teratogen. Rather, these findings indicate that the effects of this experimental protocol primarily reflect a false positive signal: the effects of EtOH on the membrane permeability of neurons undergoing programmed cell death, rather than a neurotoxic effect of EtOH on healthy neurons at this developmental stage.

Prior efferocytosis can create false negatives in cell death analysis

Another source of error in single-time point analyses of neuronal death can be a failure to account for prior necrosis and efferocytosis when quantifying visible dying neurons. The total amount of neuronal death is not merely the sum of the neurons which stain positive for a cell death biomarker on a given day plus the number that have committed to programmed cell death but are not yet positive for the biomarker (Extended Data Fig. 4-1). Rather, the total amount of neuronal death includes the former categories and, in addition, the neurons that have already been ruptured by necrosis, plus the number of dying neurons that had been efferocytosed by microglia before the time of analysis.

To illustrate the effect of prior efferocytosis, we reanalyzed the data presented in Figure 4A, where slices expressing Clm were incubated once with PI for 60 min on DIV 7, then imaged over several days. There was a clear decrease in PI⁺ neurons over a 6 d period (Extended Data Fig. 4-2). When TrackMate was used to track the initial PI⁺ population over time, the number of PI⁺ neurons decreased from 546.30 ± 72.88 on DIV 7, to 63.17 ± 18.34 by DIV 10, and finally to 0.33 ± 0.33 by DIV 13 (n = 6 slices) (Extended Data Fig. 4-4A). The τ of the PI⁺ cell loss was calculated to be 1.39 d (Extended Data Fig. 4-4B), which is considerably shorter than the τ for the loss of SBFI-AM⁺ neurons (4.69) seen in Figure 7C. These data highlight two key points about the use of PI to assay cell death. First, depending on the timing of the assay relative to injury, quantification of cell death using PI can result in significant underestimation of the total amount of neuronal loss in the slice if PI is only visualized on a single day. In our example, counting PI⁺ cells on DIV 10 would only have detected a fraction of the cell death that occurred just 3 d prior. Second, the short τ for PI cell loss compared with that of SBFI cell loss indicates that neurons become capable of staining positive for PI much later in the process of programmed cell death compared with the stage at which they stain with AM dyes. Thus, using PI without a complementary stain like SBFI that stains neurons earlier in the programmed cell death pathway risks further underestimating the total amount of cell death.