Behavior of neural networks in culture suggest that sporadic and genetic forms of Alzheimer's disease may not be equivalent Authors: Fulin Ma, Himanshu Akolkar, Ryad Benosman, Karl Herrup Published: 2026 ABSTRACT AbstractINTRODUCTIONAlzheimer's disease (AD) research often assumes that familial and sporadic forms share a common pathogenesis. However, the cellular impacts of amyloid precursor protein (APP) mutations compared with age‐related overexpression of APP or the deposition of amyloid beta (Aβ) are likely different.METHODSUsing high‐density multi‐electrode arrays, we compared neural activity in cultured neurons subjected to Aβ exposure or APP overexpression (via lentiviral delivery or genetic models). Effects on neuronal firing, synaptogenesis, axonal branching, and network connectivity were assessed in both developing and mature cultures.RESULTS APP overexpression reduced individual neuron firing probability and impaired synaptogenesis and axonal branching during development. In contrast, Aβ disrupted synaptic connections in mature cultures and impaired network‐level communication without blocking early structural development.DISCUSSIONThese findings indicate that APP mis‐regulation and Aβ toxicity contribute differently to neuronal behavior. These differences raise questions about the assumptions that familial and sporadic AD are similar if not identical conditions. BACKGROUND In his oft‐cited 1976 editorial 1 (reprinted in ref.2) Robert Katzman laid out a series of cogent arguments for equating Alzheimer's disease (AD) and senile dementia. At the time, Alzheimer's was defined as a rare, early onset (pre‐senile) dementia characterized by unusual deposits of amyloid and tau. Senile dementia, by contrast, was a common but poorly understood condition of accelerated cognitive decline that began relatively late in life. By equating the two, Katzman instantly transformed AD from a rare neurological anomaly into “a major killer” with high prevalence. The subsequent demonstration that the amyloid peptide (Aβ) was the chief constituent of the plaque deposits in both conditions supported his hypothesis. Followed closely by the identification of three autosomal dominant disease genes (amyloid precursor protein [APP], ***PSEN2, and PSEN1) related to Aβ processing, decades of research ensued based on experimental mouse models of AD that equated familial genetic forms of AD (fAD) with the more common sporadic forms (sAD). Yet, the field continued to wrestle with the question of whether fAD and sAD were in fact a single disease entity. 3 , 4 Both forms are progressive, with deposits of amyloid and tau, and both have similar clinical presentations. Yet there are important differences between them. Mechanistically, fAD is due primarily to autosomal dominant genetic mutations, whereas sAD is multifactorial with heightened risks from environmental as well as genetic factors. The patterns of amyloid deposition are different, with striatal deposits featuring prominently in fAD. 5 , 6 , 7 , 8 , 9 Hypometabolism, as measured by fluorodeoxyglucose (FDG) uptake, is more dramatic in fAD. 10 The differences between sAD and fAD are still evident even when accounting for different ages at onset. 8 Naturally, much of this discussion is focused on Aβ and its parent protein, the amyloid precursor protein (APP). APP is a multifunctional type 1 transmembrane protein 11 expressed in virtually all cell types including neurons. 12 Although APP is best known as the parent protein of the Aβ peptide, an identifying feature of AD, 13 it also has physiological roles in many neuronal functions. APP is detected as early as the pre‐implantation stage, 14 and expression continues throughout development. It plays a suppressive role in neurogenesis 15 and is critical for proper neuronal migration. 16 , 17 , 18 It guides axonal growth, 19 , 20 , 21 , 22 dendritic maturation, 23 , 24 , 25 and synaptogenesis. 26 , 27 , 28 , 29 These diverse roles have largely been identified in experiments where APP levels were altered acutely. By contrast, chronic elimination of APP (in App‐null mutant mice) results in only mild phenotypes: cortical development is largely normal, and axonal connectivity is only slightly disrupted. 30 , 31 , 32 , 33 RESEARCH IN CONTEXT Systematic review: Alzheimer's disease (AD) occurs in two major forms–familial genetic forms (fAD) that are usually early onset, and sporadic forms (sAD) that typically present at later ages. Both forms have pronounced buildup of amyloid and tau, and both present with similar symptoms. Important differences are apparent. For example, advanced technologies have revealed distinct molecular signatures as well as neuronal network alterations between the two. This raises the question of whether their disease processes might differ. Interpretation: Using high‐resolution multielectrode arrays we show that neurons from a mouse model of fAD build profoundly different neural networks compared to wild‐type neurons or neurons manipulated to mimic sAD after their networks were established. We show that amyloid precursor protein (APP) mainly impaired the early development of nerve cells—reducing firing, branching, and connection building. Amyloid beta (Aβ), by contrast, disrupted communication between mature neurons without strongly affecting their early growth. These results suggest that from a neural network perspective, fAD and sAD are not the same. Future directions: Our physiological findings signal the need for caution in future research. The developmental contribution of altered APP levels should be more central in analyzing the generalizability of genomic models of AD. The distinct neural network changes seen in rodent models of familial and sporadic AD models should be validated in human neurons with the goal of identifying the specific molecular mechanisms behind their differences. Moreover, future studies involving application of exogenous Aβ aggregates should take care to ensure that they are faithfully reproducing the physiological as well as the pathological conditions of the AD brain. Systematic review: Alzheimer's disease (AD) occurs in two major forms–familial genetic forms (fAD) that are usually early onset, and sporadic forms (sAD) that typically present at later ages. Both forms have pronounced buildup of amyloid and tau, and both present with similar symptoms. Important differences are apparent. For example, advanced technologies have revealed distinct molecular signatures as well as neuronal network alterations between the two. This raises the question of whether their disease processes might differ. Systematic review: Alzheimer's disease (AD) occurs in two major forms–familial genetic forms (fAD) that are usually early onset, and sporadic forms (sAD) that typically present at later ages. Both forms have pronounced buildup of amyloid and tau, and both present with similar symptoms. Important differences are apparent. For example, advanced technologies have revealed distinct molecular signatures as well as neuronal network alterations between the two. This raises the question of whether their disease processes might differ. Interpretation: Using high‐resolution multielectrode arrays we show that neurons from a mouse model of fAD build profoundly different neural networks compared to wild‐type neurons or neurons manipulated to mimic sAD after their networks were established. We show that amyloid precursor protein (APP) mainly impaired the early development of nerve cells—reducing firing, branching, and connection building. Amyloid beta (Aβ), by contrast, disrupted communication between mature neurons without strongly affecting their early growth. These results suggest that from a neural network perspective, fAD and sAD are not the same. Interpretation: Using high‐resolution multielectrode arrays we show that neurons from a mouse model of fAD build profoundly different neural networks compared to wild‐type neurons or neurons manipulated to mimic sAD after their networks were established. We show that amyloid precursor protein (APP) mainly impaired the early development of nerve cells—reducing firing, branching, and connection building. Amyloid beta (Aβ), by contrast, disrupted communication between mature neurons without strongly affecting their early growth. These results suggest that from a neural network perspective, fAD and sAD are not the same. Future directions: Our physiological findings signal the need for caution in future research. The developmental contribution of altered APP levels should be more central in analyzing the generalizability of genomic models of AD. The distinct neural network changes seen in rodent models of familial and sporadic AD models should be validated in human neurons with the goal of identifying the specific molecular mechanisms behind their differences. Moreover, future studies involving application of exogenous Aβ aggregates should take care to ensure that they are faithfully reproducing the physiological as well as the pathological conditions of the AD brain. Future directions: Our physiological findings signal the need for caution in future research. The developmental contribution of altered APP levels should be more central in analyzing the generalizability of genomic models of AD. The distinct neural network changes seen in rodent models of familial and sporadic AD models should be validated in human neurons with the goal of identifying the specific molecular mechanisms behind their differences. Moreover, future studies involving application of exogenous Aβ aggregates should take care to ensure that they are faithfully reproducing the physiological as well as the pathological conditions of the AD brain. APP plays a homeostatic role in tuning adult neuronal activity by regulation of the axon initial segment (AIS). The AIS is a specialized region of the axon where a high density of voltage‐gated ion channels favors action potential initiation. 34 , 35 , 36 Neuronal activity shifts the position and length of the AIS in such a way as to reduce firing probability. 34 , 37 , 38 Unexpectedly, this activity‐dependent modulation of the AIS requires APP. 39 Neuronal activity increases APP and increased APP, on its own, is sufficient to shorten the AIS and cause it to shift away from the cell body, thus reducing neuronal activity. These data validate the effect of APP at the single cell level but leave unanswered several questions about how such changes might affect neuronal networks. In the current work we use chronic and acute manipulations of APP levels in dissociated neuronal cultures plated on high‐density multi‐electrode arrays (HD‐MEA). With the high spatial and temporal resolution of neuronal activity afforded by these arrays, we describe the neuronal networks as they develop in culture. We show that elevating APP levels delays network development differently depending on the timing of the manipulation, and that Aβ disrupts physiological activity in a markedly different way. The findings offer new insights into APP function and challenge our portrayal of fAD and sAD as similar. METHODS Primary culture The culture surface of the MEA chip was coated with poly‐D‐lysine (PDL, 0.5 mg/ml) in 37°C overnight. The PDL was rinsed off with water followed by an additional coating with 50 µg/µL Matrigel (Sigma L2020) for 30 min. Wild‐type E16.5 C57BL/6J embryos were harvested from gravid females. After cervical dislocation, their cerebral cortices were isolated and stored at 4°C in phosphate‐buffered saline (PBS) supplemented with 1 mM glucose. Cortices were separated into small pieces using forceps and treated with 0.25% trypsin‐EDTA (ethylenediaminetetraacetic acid) (Sigma) for 10 min at 37°C. The tissue was then washed in 10% fetal bovine serum (FBS) in Dulbecco's Modified Eagle Medium (DMEM) to inactivate the trypsin, then washed in Neurobasal medium (Invitrogen), and triturated to produce a single‐cell suspension. For cultures on glass, poly‐L‐lysine‐coated coverslips were placed in the bottom of 24‐well plates and cells added at a density of 48,000 cells/well. For MEA cultures, we used 50 µL of a 1.6 × 106 cells/mL suspension to plate 80,000 cells total in the culture chamber (Figure 1A). The medium to maintain the neurons in both conditions was Neurobasal medium supplemented with 2% B27 (Invitrogen), 1% Glutamax (Invitrogen), and 10,000 U/mL penicillin/streptomycin (Invitrogen). Cultures were maintained at 37°C in a humidified atmosphere of 5% CO2. Every 5 days, the culture medium was refreshed by replacing half of the old medium. For APP‐overexpressing cultures, the cultures were infected with Lentivirus FUGW‐APPSwe (Addgene190804) produced by Dr. Ronald Hart and used previously. 39 For Aβ cultures, Aβ oligomers were prepared as described. 40 Briefly, the 1,1,1,3,3,3‐Hexafluoro‐2‐Propanol (HFIP) pretreated Aβ1‐42 (rPeptide, A‐1163‐1) was speed‐vacuumed at 800 x g at room temperature, and then resuspended to obtain a final concentration of 100 µM. The resulting suspension was sonicated in a room water bath for 10 min then aliquoted into polypropylene vials, sealed, and stored at –80°C. The freshly prepared oligomers were added to the cultures at a final concentration of 1 µM. Medium was refreshed every 4 days after DIV14 by removing half of the medium from each culture and adding back an equal volume of complete neurobasal medium. Freshly made Aβ oligomers were added with each feeding. The experimental flow is diagrammed in Figure 1A. Characterization of electrical activity development in wild‐type cortical networks. (A) Schematic diagram of the preparation of E16.5 mouse cortical neurons for culture on the High Density‐MultiElectrode Array (HD‐MEA). (B) Overlay of immunostaining of MAP2 (white ‐ microtubule associated protein 2), traces from spontaneous recordings (red lines), and the boundaries of the 48 individual electrodes measured (gray squares) [wild‐type culture, Days in vitro (DIV26)]. (C) Examples of events detected as spikes (red triangles) from four different electrodes. (D) A heatmap of the firing rate of a wild‐type DIV18 culture during a 1‐min period. (E) Overlay of designated hotspots (red) with immunostaining of MAP2 (white) performed after the recording session. (F) Overlay of hotspots and immunostaining of Ankyrin G (white), which indicates location of the axon initial segment (AIS). (G) Successive heatmaps of the firing rate of a single MultiElectrode Array (MEA) culture made over a 2‐week period (DIV as indicated). Green circles illustrate a single group of electrodes that changed from active to inactive during the culture period. Red circles show a spot that became active and remained so at all subsequent DIVs. Yellow circles indicate a spot that became active only at late DIVs. (H) The number of active electrodes in the wild‐type culture normalized to the number of active electrodes at DIV10. (I) Mean spike amplitude (‐µV) of wild‐type cultures as a function of DIV. (J) The mean spike rate of wild‐type cultures as a function of DIV. The dashed line is the least squares line fit on the points. (K) An example of the tracking of axons and their branches based on the timing of their spiking activity (see Methods for details). (L) The number of axon branches of neurons in a wild‐type culture expressed as a function of time in culture. (M) Spikes recorded from electrodes adjacent to the reference electrode. Note the slight delay in the peak depolarization from one electrode to its neighbor. (N) Spike peak time plotted as a function of distance from the reference electrode. The slope of the line (determined using least squares fit) represents the speed of propagation. (O) The speed of spike propagation in a wild‐type culture as a function of DIV. For H–J, L, and O, n = 10 independent wild‐type cultures; error bars represent the standard error of the mean. Electrophysiology The electrical activity of the cultures was recorded using a single well complementary metal‐oxide‐semiconductor (CMOS)‐based HD‐MEA from (Maxwell Biosystems Inc, ref.41). The total array contains 26,400 tiles that function as both recording or stimulation sites. During fabrication the electrodes are coated with electrodeposited platinum black to decrease electrode impedance and improve the signal‐to‐noise characteristics. The CMOS technology allows the array to be constructed in such a way that the center‐to‐center distance of the tiles is only 17.5 µm. This enables a very high‐resolution view of the electrical activity of the culture. Although information can be retrieved from any electrode, the setup allows for simultaneous recording from only 1024 of the total 26,400 electrodes at any one time at a sampling rate of 20 kHz. Spike detection, sorting, and analysis Firing‐rate–based analysis was performed in Matlab (Mathworks, Inc.). The raw signals were filtered through a 300 Hz high‐pass second‐order Butterworth filter. Spikes were detected using the level‐crossing method where a change higher than 6 times the standard deviation (SD) of the recorded signal was considered a spike event. All analysis was performed on the firing rate based on these detected spikes. The recorded spikes were then sorted to detect individual units using the SpikingCircus 42 algorithm. All further analysis was performed using the sorted unit data. Spontaneous activity recording over the entire array. A gaussian convolution filter (sigma = 20 ms) was used to convert spike time into spike rate. The MEA electrodes are arranged in a rectangular grid of 220 columns by 120 rows. We divided the array into two halves of 110 columns each. Because the software allows a maximum of 1024 electrodes to be recorded at any one time, we sampled the entire culture as follows. The first set of electrodes we recorded included the first 4 columns of each half of the array (Columns 1–4 and 111–114). As each column contained 120 electrodes, each of the sets consisted of 960 electrodes, which we monitored for 60 s. The second set of columns (5–8 and 115–118) was then recorded, and the process continued until all 27 full sets of electrodes had been recorded. The last two columns (119–120 and 219–220) were recorded as a final set. The data from all 28 sets were collected and analyzed as one in Matlab. We calculated the total number of active electrodes (e.g., Figure 1F) as the number of electrodes that spiked at least once during the 60‐s recording. The mean spike rate of an electrode (e.g., Figure 1G) was defined using only electrodes that showed at least one spike during the recording period. We summed the total number of spikes and divided this number by 60 to obtain the spike rate (spikes per second). We calculated the mean spike amplitude as the average amplitude across all recorded spikes from the active electrodes. We also observed a tendency for cultures to demonstrate bursting behavior. Bursting was defined as a local maximum of firing rate where most of the neurons in the network fired synchronously with an amplitude higher than 1.2 times the root mean square of the firing rate activity over the entire recording. The network activity assay was used to calculate the bursting frequency defined as number of bursts per second, inter‐burst interval as the average duration between two network bursts, and number of recorded electrodes per burst as the number of electrodes spiking at least once within 100 ms around the time of the burst. To track the bursting activity more accurately, we used the activity scan to identify the 1024 most active electrodes based on the firing rates. The spontaneous activity of the selected units was recorded for 5 min. For this analysis, bursting was defined as the synchronous activity of multiple neurons, which described in more detail in the Results. The spikes from all recorded neurons were plotted as a histogram with bin size of 10 ms, and then smoothed using a gaussian kernel of width 300 ms. Axonal propagation activity The spontaneous activity of the 15 most active units of all 26,400 electrodes was recorded for 60 s. This was used to detect axons originating from the recorded 15 units. For each unit, spike‐triggered average was calculated for the rest of the electrodes to identify the course of the axons. 43 The computation was performed using MaxLive software (Maxwell Biosystems, AG, Switzerland). Once the axons were identified, the axonal length and conduction speed were calculated for each unit. 43 Axonal branching was defined as different branches of a propagating spike arising either from the soma or from the main axon. The branch was defined as a set of electrodes that form an axon and have a different spatial angle of propagation. The branching analysis and results were calculated using the MaxLive software (Maxwell Biosystems, AG, Switzerland). Stimulation activity The previously identified 15 units were stimulated using 20 bi‐phasic pulses of 200‐ms duration spaced 1 s apart. The amplitude of the pulses was from 2 mV to 240 mV, peak to peak. Following stimulation, 1000 electrodes within 300 µm of the stimulated unit were monitored. If a spike was observed in one of the monitored electrodes within 50 ms of the stimulation pulse, the unit was considered to have responded to the stimulation. The stimulation probability was then calculated as the fraction of times a spike was observed after a stimulation (n = 20). Immunocytochemistry Cells plated on coverslips were rinsed with PBS before fixation with 4% paraformaldehyde (PFA) for 15 min at room temperature. Cells plated on MEA chips were rinsed with PBS 3times before PFA fixation (25 min, room temperature). All samples were then rinsed with PBS again and blocked for 1 h at room temperature in PBS with 5% donkey serum and 0.3% Triton X‐100 (PBST). The samples were then incubated with primary antibodies (dilutions shown in Table S1) in PBST plus 5% donkey serum overnight at 4°C. The antibodies were rinsed off the following day with three PBS washes and incubated for 1 h at room temperature with secondary antibodies diluted 1:500 with PBST containing 5% donkey serum. After three additional 5 min PBS rinses, samples were incubated with DAPI (4′,6‐diamidino‐2‐phenylindole) at 1 mg/mL for 10 min at room temperature. Samples on coverslips were rinsed two more times in PBS, then mounted using Hydromount (National Diagnostics, HS‐106) before imaging. MEA chips were imaged by covering the surface with 10 µL PBS and mounting under a glass coverslip. Imaging The coverslips were imaged on a Nikon Eclipse Ti2 confocal microscope with a 60X (1.4 Numerical Aperture) oil lens and A1R (Andor, 1024 µm x 1024 µm) camera, X‐Cite 120 LED illumination system, and 4'6‐diamidino‐2‐phenylindole (DAPI, 400–418 nm), Fluorescein isothiocyanate (FITC, 450–490 nm), Tetramethylrhodamine isothiocyanate (TRITC,554–570 nm), and Cyanine‐5 (Cy5, 640–670 nm) filter sets controlled by NIS‐Elements AR software (5.21.03 64‐bit). The immunolabeling on the MEA chips was performed on a Nikon Eclipse Ni‐U inflorescent microscope with a 20X objective and equipped with DAPI (400–418 nm), FITC 450–490 nm), TRITC (554–570 nm) and Cy5 (640–670 nm) filter sets controlled by NIS‐Elements AR software [Version 5.02.00 (Build 1266), LO, 64‐bit]. Statistical analysis To analyze the effect of conditions and time on the cultures compared to the wild‐type ones, we performed two‐way analysis of variance (ANOVA) with Tukey correction for multiple tests. If a statistically significant difference was observed on ANOVA, post hoc tests were performed to test the difference on specific days. These were carried out as non‐parametric two‐tailed Mann–Whitney tests with Holm–Sidak method to adjust the p‐values. All p‐values reported here are adjusted in this way. RESULTS The electrical properties of wild‐type neuronal cultures develop on independent timelines Primary cortical neurons were isolated from C57BL/6J embryo cortices at 16.5 days of gestation (E16.5) and cultured on high‐density multi‐electrode arrays (HD‐MEA) as described in Methods (Figure 1A). By 18 days in vitro (DIV18), the neurons had become electrically active, grown processes of considerable length, and established active synapses. Using MAP2 to reveal the location of the neurons relative to the electrodes (Figure 1B), we found that a typical neuronal cell body was nearly equal in size to a single electrode and could lie in contact with up to four electrodes. Monitoring the activity of the culture, we defined a voltage shift as a “spike” when the peak potential shift exceeded 6 times the SD of the average baseline signal (Figure 1C, red triangles). We sequentially sampled all 26,400 electrodes of the MEA during a 30‐min recording session and obtained the spike rate of each electrode, which we displayed as a heat map of the array (e.g., Figure 1D). For each session, we identified the 1024 most active electrodes to use for further analysis. We found these to be scattered across the full extent of the array (Figure 1D), suggesting that clumping of cells or processes was minimal in our preparations. Superimposition of the location of the most active electrodes on a corresponding image of the immunostained culture revealed that they were located predominantly at the cell body (MAP2 immunostain—Figure 1E), with a bias toward the location of the AIS (ankyrin G immunostain—Figure 1F). Recording sessions were conducted over a period of 26 days, allowing us to track the total number of active electrodes as a function of time in culture. The images in Figure 1G are heat maps of sequential recordings from a single array over a 2 ½ week period from DIV10 through DIV26. Earlier time points are not shown because there was little electrical activity evident before DIV10. Between DIV10 and DIV14, however, the number of active electrodes increased significantly, after which there was little additional change (Figure 1G, H), a pattern seen by others. 40 , 41 , 42 Yet, although the number of active electrodes did not change, the relative activity at any one electrode could vary considerably. Some sites began with nearly no activity then gradually developed a high spiking rate (Figure 1G, yellow circles). Others oscillated between no activity in certain time windows to high activity in others, only to become quiet again at later times in culture (Figure 1G, green and orange circles). Other electrophysiological properties of the culture appeared to develop on independent time schedules. For example, unlike the near saltatory increase in the number of active electrodes between DIV10 and DIV14, the average peak spike voltage of a given element increased steadily until DIV16, after which it plateaued for the rest of the culture period (Figure 1I). The mean spike rate showed yet a third pattern, increasing nearly linearly throughout the entire 3‐ to 4‐week culture period (Figure 1J). Heat maps illustrating these properties provide a sense of how these different properties mature (Figure S1). There was also a significant increase in the number of active branches of each axon. To follow individual action potentials as they left their site of initiation and traveled across the culture substrate, we identified the 15 most active electrodes and recorded from a 9x9 matrix of 81 electrodes surrounding the 15. We then traced spike events as they moved across the MEA (Figure 1K, Movie S1) 43 These data also allowed us to count the number of active axonal branches (Figure S2). During the first 2 weeks in culture, most axons displayed an average of three to four branches, but after DIV18 the number of branches increased nearly fourfold (Figure 1L). The recording data were also used to estimate axonal conduction speed (Figure 1 M, N), which trended upward until DIV16, after which it appeared to stabilize (Figure 10). Network activity of the array begins with large scales bursts and matures to more local activity Viewed horizontally, the MEA consists of 110 rows of electrodes arrayed in 220 columns. To display network activity, we numbered the 1024 most active electrodes, starting at the top of the first column (top left in Figure 2A) and then proceeded through each of the 220 columns to the bottom right of the array. In this example, we have used a color gradient to illustrate the order in which the electrodes were sampled. These electrodes were then displayed in order along one axis, with time in seconds plotted on the other axis, resulting in a graph or raster plot such as the one shown in Figure 2B. Note how some electrodes display a near constant level of spiking. These elements appear as a vertical band in the raster plot (Figure 2B, magenta arrows); others were far less active (Figure 2B, green arrows). The network also showed evidence of coordinated bursting activity (horizontal bands, Figure 2B, orange arrows). Some bursts were quite strong and involved virtually every electrode (Figure 2B, gray arrows). Network activity development. (A) The position of all active electrodes on a typical MEA based on a 5‐min recording of a wild‐type culture done at DIV18. The active electrodes were numbered, starting at the upper left and proceeding, column by column to the lower right. A color scale was used to reflect the number assigned to each electrode. (B) Raster plot showing the activity of each of the electrodes (colored as in A) during a representative 60‐s time window of the 5‐min recording session. Each dot represents a detected spike at the electrode. Red and green arrows indicate constitutively active electrode sites. Gray arrows represent network‐wide bursts (simultaneous activity) involving a high percentage of sites. Orange arrows represent bursts involving a smaller local network. (C) Raster plots of a single wild‐type culture recorded at 4‐day intervals beginning at DIV14. (note: plots are rotated 90 degrees, relative to the plot in B). (D) Example of bursting activity on the array. In the diagram to the left, each grid square contains the activity of that electrode during a 5‐ms recording interval. Di and Dii are enlargements of the grid squares indicated by the black box taken at two different times during the recording session. Di illustrates a 5‐ms quiescent period; Dii represents the same electrodes during a burst. (E) The fraction of total number of active electrodes participating in a single burst at different DIV. (F) The number of bursts as a function of the fraction of the total number of electrodes involved in that burst. n = 10 independent cultures for E and F. Error bars represent the standard error of the mean. Using raster plots such as this we followed the development of the activity of the cultures over time. At first (DIV14), coordinated bursting activity predominated (Figure 2C—note that in this representation the axes have been rotated 90° from panel B; bursts are now vertical; active sites are horizontal). Similar patterns have also been observed by others. 41 , 42 , 44 Bursting activity matured with time in culture, such that more and more units were involved in each burst (Figure 1D–F). After DIV22, however, the number of electrodes participating in each bursting event declined sharply (Figure 2E), as did the fraction of the total electrodes that were involved in any one burst (Figure 2F). This likely reflects the establishment of local circuits that fire independently of the other parts of the network. Neurons from an fAD model have a distinct physiological pattern With these normative data as background, we asked what the consequences would be if the neurons in the culture came from a mouse carrying the human APPSwe mutation. This would mimic the situation in fAD where mutant APP or PSEN1 proteins are present during the formation of the nervous system as well as in the adult. To achieve this, we cultured neurons from R1.40 transgenic mice, a model of human AD that carries a yeast artificial chromosome transgene with the entire human APPSwe gene including introns and all cis regulatory elements within 30 kb of the coding sequence. The APPSwe transgene is expressed with the same temporal and spatial profile as the normal mouse or human APP gene. 44 , 45 Analysis of the electrical behavior of R1.40 neuronal networks revealed profound differences with wild‐type cultures. In R1.40 cultures, the number of active electrodes rose earlier than what was seen in wild‐type cultures, although with later times in culture both genotypes overlapped (Figure 3A). Other features of the activity of the cultures also revealed a precocious pattern of development. Although spike amplitude of R1.40 neurons developed to nearly the same levels as wild‐type cells, it rose earlier in the evolution of cultures (Figure 3B). Indeed, at the earliest time points (DIV5 and DIV7) the R1.40 spike amplitude was significantly greater than the values observed in untreated cultures. Similarly, the increase in the average spike rate was faster in the R1.40 cultures (Figure 3C). As early as DIV10, the R1.40 cultures had already attained a spike frequency comparable to a mature wild‐type culture. This represented an early maturation as additional time in culture did not lead to further increases. In line with the early emergence of these functional differences, there were structural differences as well. The number of axonal branches was higher in young R1.40 cultures but remained stable with time in culture. By contrast, after 3 weeks in culture, wild‐type neurons had nearly doubled their level of branching complexity while the R1.40 neurons remained unchanged (Figure 3D). The speed of action potential propagation followed a pattern similar to that seen for the spike rate (Figure 3E). Network behavior of R1.40 cultures. (A) The number of active electrodes in R1.40 cultures normalized to the values at DIV10. (B) The average spike amplitude in R1.40 cultures as a function of DIV. (C) The mean firing rate of the active electrodes in R1.40 cultures. (D) The number of active axon branches in R1.40 cultures as a function of DIV compared with wild‐type. (E) Propagation speed of action potentials in R1.40 cultures as a function of DIV. (F) Representative raster plots of R1.40 cultures at different DIVs. (G) The average percentage of active sites involved in a burst as a function of DIV. (H) Number of bursts plotted as a function of the fraction of the total number of electrodes involved in that burst. (I) Replotting of the data from H to allow the comparison of wild‐type and R1.40 cultures with respect to the number of bursts as a function of the fraction of total electrodes active at DIV 22 and DIV 26. n = 6 independent R1.40 cultures. In A–E and G, error bars represent the standard error of the mean. The behavior of the network recapitulated many of the same kinetics. At DIV14, the R1.40 cultures had already established a pattern of burst and inter‐burst activity that resembled more mature untreated cultures. Additional time in culture, however, led to little additional development (Figure 3F). This can be visualized by viewing the change in the fraction of sites contributing to a burst (Figure 3G, H). Whereas wild‐type cultures have a transient period of coordinated bursting activity, they soon lose this feature. R1.40 cultures develop earlier but seem locked in their pattern of bursting (Figure 3I). Modeling sAD by exposure to the Aβ peptide Cultures of neurons bearing a mutant APP gene offer a straightforward way to create an in vitro model of fAD. How to create a realistic model of sAD, however, is less obvious. We reasoned that we should begin by using mature cultures, since the pathological hallmarks of sAD do not begin to appear until midlife. Although in our earlier studies of the AIS 39 we had shown that exogenously applied Aβ (fibrillar or oligomeric) had no effect on the structure or location of the AIS, we used the exogenous application of Aβ oligomers to attempt to mimic the chemistry of the sAD brain in our cultures. We applied 1 µM Aβ oligomers to mature (DIV18) cultures and measured their responses. To maintain the concentration of the oligomers, we added freshly prepared Aβ aliquots every 4 days. Within hours after the first addition of Aβ, the number of active electrodes dropped by half compared to untreated, wild‐type cultures (p < 0.001 – Figure 4A). The cultures experienced a transient rebound in the next several days, but by DIV24, the continued presence of the Aβ oligomers depressed the number of active electrodes by half. The effects of the oligomers extended to the phenotypes of the individual neurons. Although they had no effect on spike amplitude (Figure 4B) the number of active branches failed to develop to the same level seen in untreated cultures (Figure 4C). In contrast, after a transient significant increase in the hour following Aβ addition (p < 0.01), the spike rate decreased and remained low (Figure 4D). The increased speed of action potential propagation seen in control cultures was completely blocked after the oligomers were added (Figure 4E) These changes were correlated with obvious changes in the activity of the overall network as seen in the raster plots (Figure 4F). We observed considerable variability from culture to culture and from day to day in the same culture. Overall, it appeared as though Aβ quieted entire regions of the array leaving other regions largely unaffected. Effects of Aβ (beta‐amyloid) treatment on mature DIV18 cultures. (A) The number of active electrodes in lentiAPP (lentiviral Amyloid Precursor Protein) cultures and Aβ oligomers. Green arrow indicates the DIV at which lentiAPP cultures and Aβ oligomers were added to the cultures. (B) Spike amplitude as a function of DIV for lentiAPP and Aβ cultures. (C) The number of active axon branches in cultures with Aβ addition compared to control and R1.40 culture as a function of DIV. (Di) Spike rate of cultures with Aβ addition compared to lentiAPP culture. (Dii) Spike rate change before and 1 h after Aβ addition to the culture. (E) The spike propagation speed of cultures with Aβ addition compared to wild‐type and R1.40 culture as a function of DIV. (F) A representative DIV22 raster plot of Aβ addition in DIV18 compared to untreated control. n = 10 independent wild‐type cultures, n = 4 Aβ cultures, and n = 8 independent lentiAPP cultures shown in B– G. Green arrow indicates the addition of virus/Aβ. Error bars represent the standard error of the mean. ** p < 0.001, paired t‐test. Altering the levels of APP in mature cultures mimics neither R1.40 nor Aβ‐treatment The data thus far make it clear that the physiological properties of neurons carrying fAD gene mutations, and the networks that they establish, differ from those found in untreated wild‐type cultures but also differ from wild‐type cultures exposed to Aβ. These results would seem to suggest that the cell biological basis of fAD and sAD are different, since recreating the chemistry of the sAD brain by adding Aβ oligomers to our cultures did not reproduce the phenotypes found in the fAD model R1.40 cultures. This led us to consider the potential contributions of the full‐length APP protein to the behaviors of the R1.40 cultures. We had shown previously that APP mRNA and protein increase with activity and that such increases alter the structure of the AIS. 39 The alterations are predicted to quiet a neuron, 34 , 46 and this prediction has been validated with calcium imaging of neurons carrying APPSwe. 39 , 47 To extend these findings using the high resolution afforded by the MEAs, we infected DIV10 MEA cultures with a lentivirus (lentiAPP) that expressed full‐length APPSwe plus an mCherry marker that allowed us to identify the infected cells (Figure 5A). On DIV18, we identified the most active neurons and applied bi‐phasic 200 µs pulses of increasing voltage spaced 1 s apart. Recordings from the surrounding electrodes were then used to determine whether the stimulus had elicited a response. For each neuron and condition, we performed 20 replicates (see Methods for additional details). In control cultures, increasing the applied voltage to uninfected cells increased the probability of firing, with a half‐maximal response seen at ≈35 mV (Figure 5B, black symbols). By contrast, lentiAPP‐infected cultures required nearly 80 mV to achieve the same half‐maximal response (Figure 5B, purple symbols). Despite these differences in firing probability, the properties of the individual units were largely unaffected. The number of active electrodes was not reduced (Figure 5D) and the amplitude of the average spike event was largely unchanged (Figure 5E). Effects of lentiAPP on mature DIV18 cultures. (A) Cultured neurons from a lentiAPP‐infected culture, 4 days after introduction of the virus. Nuclear mCherry (red) identifies an infected cell. Immunostaining for APP (green, Amyloid Precursor Protein) confirms that APP levels increase in infected (green arrow) compared to uninfected cells (blue arrow). (B) Probability of a site firing after stimulation with different voltages in wild‐type (black) and lentiAPP (purple) cultures. Orange lines indicate the voltage at which 50% firing probability is reached. (C) The number of active electrodes in control and lentiAPP cultures (normalized to DIV14). (D) Amplitude of spikes in control and lentiAPP culture. (E) Mean firing rate of lentiAPP and control cultures as a function of DIV. (F) The number of active axon branches as a function of DIV. (G) Representative raster plots recorded on DIV22 after treatment with Aβ oligomer, control, or lentiAPP on DIV18. n = 10 independent wild‐type cultures and n = 8 independent lentiAPP cultures. All error bars in B–F represent the standard error of the mean. Both Aβ and APP interfere with the development of the network and its units These data make it clear that exogenous application of Aβ, a model of sAD, has a set of impacts on neuronal networks that differs from those of endogenous overexpression of mutant APP—either “early onset” (R1.40), the situation expected in familial forms of AD, or “late onset” (lentiAPP) a mimic of sporadic AD. This offers additional support for the proposition that fAD and sAD progress by two distinct processes. Note, however, that the R1.40 and lentiAPP cultures, although similar in most properties, are not identical (e.g., the number of active wells—Figure 5C). We reasoned that the lentiAPP infections illustrated in Figure 5 altered APP levels after the cultures and their networks were largely established. By contrast, in R1.40 cultures, APP expression was elevated throughout the entire culturing process. This led us to consider the possibility that APP and its Aβ breakdown product have developmental as well as homeostatic functions. The development of activity in untreated wild‐type cultures progressed through a period of synchronous activity (bursting) and then matured to a pattern characterized by more local interactions (Figure 2C, E, F). We noted that 4 days after DIV18 lentiAPP infection, cultures appeared to regress toward a DIV14 pattern of increased bursting with less inter‐burst activity (Figure 5G). To probe this phenomenon more deeply, we applied lentiAPP to cultures on DIV10 instead of DIV18. We found that the development of the culture was profoundly blocked. The number of active electrodes increased slightly, but the dramatic increase found in the untreated cultures was virtually eliminated (Figure 6A; p < 0.001). A similar pattern was seen in the development of spike amplitude. The effect was less dramatic, but shortly after lentiAPP infection the increase in amplitude found in the wild‐type was blocked (Figure 6B). For both measures, the block was maintained for 2½ to 3 weeks following infection. The developmental block, however, was not complete. The spike rate, for example, followed a bi‐phasic response during the first week following lentiAPP infection; the rate dropped by roughly half (p < 0.001 for DIV 16 and DIV 18 compared to control). Between DIV18 and DIV20, however, the rate rapidly increased before finally plateauing slightly below the rate found in untreated cultures (Figure 6C). Effects of lentiAPP on immature DIV10 cultures. (A) The number of active electrodes after treatment with lentiAPP normalized to the values at DIV10. In this and subsequent panels, the green arrows indicate the DIV when lentiAPP was first added to the culture. (B) The average spike amplitude as a function of DIV after treatment with lentiAPP at DIV10. (C) The firing rate of the active electrodes in a culture following treatment with lentiAPP at DIV10. In A–C the green line represents the R1.40 data from Figure 3, whereas the black points are data from Figure 1 replotted for reference. (D) Representative raster plots of a single culture treated with lentiAPP on DIV10 recorded every 4 days beginning at DIV14. (E) The percentage of active sites involved in a burst as a function of DIV. (F) The number of bursts plotted as a function of the fraction of the total number of electrodes involved in that burst. (G) The number of branches in lentiAPP compared to wild‐type culture and R1.40 culture as a function of DIV. n = 5 independent cultures. In A–C and E–G, error bars represent the standard error of the mean. The block in the development was also seen in the raster plots. As early as DIV14, the activity on the array was reduced relative to untreated wild‐type cultures (Figure 6D, compare with Figure 2C) and this reduction continued through DIV18. The pattern of firing was also notably different from that seen in untreated cultures. Four days after infection, on DIV14, bursting activity was present, but its frequency trailed that seen in age‐matched untreated cultures. By DIV18 and beyond, the immature pattern persisted and did not mature with additional time in culture (Figure 6D). Note in particular the absence of the inter‐burst activity that characterized the more mature untreated control cultures (Figure 2C). Plotting the number of bursts as a function of the percentage of active electrodes in a burst offers a more quantitative picture of this lack of developmental change (Figure 6E–F). In the lentiAPP cultures the pattern of firing at DIV26 more closely resembled younger untreated cultures (Figure 2F). Consistent with the biphasic effects of lentiAPP on spike rate (Figure 6C), there was a transient increase in activity visible in the DIV22 raster plot. Despite this, by DIV26, the lentiAPP cultures returned to their immature state with a noticeable absence of continuously active recording sites (horizontal bands in the DIV26 raster plot in Figure 6D). As a final measure of the profound developmental block induced by lentiAPP infection at DIV10, we found that the axonal branching of the active neurons (Figure 6G) also failed to follow the increase seen in wild‐type cultures and instead remained similar to the reduced branching complexity of the R1.40 cultures. Based on serum and cerebrospinal fluid concentrations, the levels of Aβ in sAD do not change until well into midlife, 48 although deposits in certain brain areas appear earlier. 49 To determine whether the altered developmental profile we observed in our cultures was due to abnormal levels of Aβ peptide produced from the increased amounts of full‐length APP, we exposed cultures to Aβ oligomers at DIV14, adding fresh Aβ every 4 days to maintain its presence. The Aβ‐treated MEAs differed from both untreated and lentiAPP cultures. Although lentiAPP infection at DIV10 virtually halted the increase in the number of active sites, Aβ exposure led to only a modest decrease from controls (Figure 7A). The spike amplitude (Figure 7B) drifted lower but was largely identical to untreated cultures. The most unexpected result came from our analysis of the average spike rate. As shown in Figure 6, although lentiAPP infection at DIV10 reduced spike frequency, Aβ addition markedly increased it (Figure 7C). Curiously, the oligomers appeared to block the normal increase in conduction speed observed in untreated cultures (Figure S3). Aβ also had a strong effect on network activity. There was enormous variation in the activity pattern of the treated cultures (Figure 7D) with little evidence of bursting activity (Figure 7E–F). The drop in the number of active sites participating in any one burst was significantly reduced from controls (p < 0.05). By contrast, sites with continuous activity were more common than in lentiAPP cultures, although not as frequent as in untreated cultures. These functional changes were all the more curious because the addition of Aβ oligomers at DIV14 had only a slight effect on the development of active axonal branches (Figure 6G). To learn if there were structural correlates to these odd behaviors, we performed 6E10 immunostaining of Aβ in sister DIV18 cultures grown on poly‐lysine‐coated coverslips instead of the MEA. The staining pattern revealed small domains where synapsin‐I staining of synaptic vesicles was virtually absent. In these regions, the MAP2+ neurites and spine‐like protuberances were coated with small puncta of 6E10 immunostaining (Figure 6H). The staining suggested that there was a loss of presynaptic integrity (loss of synapsin‐I)—perhaps a physical stripping of synapses caused by aggregations of the Aβ peptide. This interpretation is consistent with the observed changes in network activity. Effects of Aβ on neuronalcultures. (A) The number of active electrodes after treatment with Aβ oligomer normalized to the values at DIV10. (B) The average spike amplitude in Aβ oligomer cultures as a function of DIV (error bars = SEM). (C) The firing rate of the active electrodes in Aβ oligomer cultures (error bars = SEM). In A–C, the purple line is a replotting of the lentiAPP data from Figure 6 and the black points are the control data from Figure 1 included for reference. (D) Representative raster plots of a single culture treated with Aβ oligomer on DIV10 recorded every 4 days beginning at DIV14. (E) The average percentage of active sites involved in a burst as a function of DIV. (F) Number of bursts plotted as a function of the fraction of the total number of electrodes involved in that burst. (G) The numbers of branches in wild‐type and Aβ cultures as a function of DIV. (H) Immunolabeling of a wild‐type neuronal culture either untreated (left) or treated with 1 µM Aβ oligomer for 24 h (right). Synapsin I (red), 6E10 (green), MAP2 (gray), and DAPI (blue). Scale bar = 20 µm. Areas in the white boxes are enlarged to illustrate how, in Aβ‐treated cultures, there are areas where synapsin‐I staining is missing and seemingly replaced by puncta of 6E10. n = 5 independent Aβ cultures. Error bars represent the standard error of the mean. As a final piece of evidence that our lentiAPP model of sAD was a useful replicate of the conditions found in Alzheimer's brain, we asked whether any of the classic histopathological features of AD were found in lentiAPPSwe‐infected neurons. We infected cultures with lentiAPP and then fixed them at either DIV14 or DIV22. Immunostaining with synapsin‐I and PSD95 to detect synapses revealed an increase in synaptic density with time in culture, but, as predicted from the electrophysiology, very little difference between infected and uninfected cells (Figure S4). In particular, the puncta along the neurites appeared largely unchanged. We also stained for PHF1, a late‐stage epitope of pathological tau (Figure 8). In control cultures we observed modest PHF1 staining in the neurites but little accumulation in the cell soma. By contrast, in APP overexpressing cells marked by the mCherry marker, somatic PHF1 staining increased steadily. By DIV22, the PHF1 signal in the soma was dramatically higher than in neighboring uninfected cells in the same culture (Figure 8, bottom row). The suggestion is that the increased levels of APP not only alter neuronal and network activity but also begin to recapitulate other aspects of AD pathology. Mouse cortical neurons from a lentiAPP infected culture at DIV10 and DIV18. Cells are fixed 4 days after infection. Red arrows showed infected cells with nuclear mCherry (red). Cells were immunolabeled with APP (gray) and PHF1 (green) followed by a DAPI (blue) labeling. White outlines in all panels show the cell bodies of cells used as examples in the detailed zoomed‐in PHF1 images in panel i, ii, iii, and iv. Scale bar = 10 µm. DISCUSSION We have exploited the possibilities afforded by CMOS technology to monitor cortical neuronal cultures with 26,400 closely spaced contiguous electrode tiles. The resulting electrical view of cultures spans considerable distances (8 x 106 µm2) at very high resolution. Despite the complexity of the electronics and their surfaces, there was no tendency for cells to clump or for neurites to track along the interfaces among the surface tiles. Furthermore, because we recorded directly in the CO2 incubator, we could easily perform longitudinal studies, tracking a single culture over several weeks, with minimal stress on the cells. We are not the first laboratory to explore the physiological effects of AD chemistry in vitro, 50 , 51 , 52 but the exceedingly high resolution of the CMOS arrays offers a far more nuanced picture than those afforded by previous studies. We found that the electrical features of the cultures developed at different rates. Neurons were largely quiet during early days, and then became electrically active between DIV10 and DIV14, consistent with previous reports. 53 , 54 , 55 , 56 , 57 Two features of this MEA data bear noting. First, the sites we record as active electrodes represent neuronal axons as well as their cell bodies. Second, not all processes that are present in the culture are active. Both features can be seen in Figure 1. Neurons are post‐mitotic in vivo and in vitro, so their numbers do not increase after they are plated. By contrast, structural features of the cultures—dendritic and axonal branches—expand continuously. Because the number of active electrodes remained constant after DIV14, the suggestion is that the constant number of neuronal cell bodies puts a ceiling on the level of activity that can be achieved. That activity of individual electrodes waxes and wanes with time in culture is consistent with this idea. Viewed as a whole, the cultures display two defining features—constitutive activity of individual sites and coordinate activity of the entire ensemble (bursting). The high resolution of the MEA thus extends the findings of earlier studies. 53 , 54 , 58 , 59 , 60 , 61 Bursting develops first, and the number of electrodes that were active during a burst increased but then dropped after DIV22. We propose that this reflects local circuits beginning to emerge and taking the place of pan‐culture activity. The results of the R1.40 cultures reveal important neurobiological features of neurons carrying fAD mutations. In nearly every domain, the presence of the APPSwe transgene altered the physiological properties of both individual units and the networks they establish, with changes apparent from the beginning of the experiment. Overall, expression of APPSwe induced a precocious development of the units but then blocked their full maturation. Placed against the clinical picture of fAD, this finding has important implications. Persons with fAD mutations and mouse AD models show no obvious neurological problems early in life. Yet the data from our cultures suggest that this “normality” is achieved by their neural networks compensating for the altered physiology of their individual units. The literature on other AD mouse models supports this idea, since the underlying physiology of the neurons, 62 and their structure, 63 are altered. The implication for humans with fAD is that the genetic factors they have inherited have altered the properties of their brains, probably from conception. Age is still required to unleash the dementia process, but the seeds of the subsequent destruction have already been planted. The R1.40 data are a near perfect in vivo genocopy of human fAD. Attempts to model sAD, however, are more challenging. The addition of Aβ aggregates to in vitro cultures as we have done is one such attempt. To the extent that it mimics the chemistry of the AD brain, our data show that its effects are distinctive. Added to mature (DIV18) cultures it reduced the spike rate similar to lentiAPP but it reduced the number of active electrodes by half compared to controls. The infection of wild‐type neuronal cultures with lentiAPP represents a second sAD model one that assumes that the increased Aβ observed with advancing age and with Alzheimer's reflects in part an increase in the levels of APP. 64 Given our earlier findings, 39 it was gratifying to find that lentiAPP increased the voltage needed to fire a neuron and thus decreased the average firing rate. However, the broader network of cells in mature cultures proved resilient to the increased APP levels, as the number of active electrodes remained largely unaffected. Thus, even while the neurons were less active, the network structures, although not as functional, remained intact. Axonal branching data combined with the early maturation of the networks suggest that APP overexpression works primarily at a single neuron level by reducing spike probability and blocking additional connections with other neurons. This leads to fewer but more inclusive network‐wide bursts. We propose that this precludes the cultures from forming smaller local networks such as those seen in untreated cultures. Introducing lentiAPP into younger cultures (DIV10) had a dramatic impact both on the behavior of the individual units (spike amplitude and spikes per second) as well as the activity of the larger network of cells (number of active electrodes and the blocked maturation of mature activity patterns). Coupled with the relative lack of impact of the same manipulation performed at DIV18, the data suggest that two different APP‐dependent processes are at work. The first is the homeostatic effect of APP on a mature neuron. Through changes in the AIS, APP quiets an active neuron and thereby functions as a feedback governor to protect it from overheating. The DIV10 data, extended by the R1.40 findings, suggest that in addition to this homeostatic role, APP has a prominent role to play in neuronal maturation. As the AIS is found in cultures as early as DIV4, it is conceivable that under a chronic increase in APP levels, the network develops more slowly (or not at all) when the activity of its individual units is blunted, but other explanations are also possible. Unlike the profound reduction in the number active electrodes found after lentiAPP at DIV10, early Aβ addition had only a minimal effect. Furthermore, although lentiAPP reduced the spike rate, Aβ significantly increased it. The response of the network to Aβ was also unlike its response to lentiAPP. Aβ suppressed bursting activity while lentiAPP appeared to block the ability of the network to stop bursting. The data thus suggest that APP and Aβ produce their effects by different mechanisms. Our interpretation of these findings is that during development, APP retards electrophysiological maturation, whereas in the adult, it dampens neuronal excitability. By contrast, at all ages, Aβ oligomers appear to break existing synaptic connections, a prediction consistent with the patches of 6E10 immunostaining that are devoid of synapsin I. A similar co‐localization of Aβ and synapses has been demonstrated recently. 65 , 66 Although mouse models and in vitro cultures are only proxies for the situation in the brains of persons with early‐ and late‐onset AD, the data we have assembled here have important implications for the understanding of AD. The lack of concordance between the actions of APP and exogenous Aβ leads to the suggestion that the clinical symptoms of AD may represent a composite response to both biological agents. Many studies have concluded that the location of amyloid plaques and the cellular structural changes related to AD are a poor predictor of the cognitive deficiencies experienced by any one person. 64 , 67 We note that analyses of the location and concentrations of full‐length APP are rarely reported, and it would appear that this information would be a valuable part of a more complete picture of AD pathogenesis. Since the editorial published by Katzman in 1976, 1 it has been widely accepted that early‐onset familial forms of AD caused by APP, PSEN1, and PSEN2 mutations represent the same disease process as the more common sporadic forms. Our results suggest that this assumption may need to be re‐examined. Infection with lentiAPP or application of oligomeric Aβ represent models of sporadic AD, whereas R1.40 neurons mimic familial/genetic forms. R1.40 neurons display cell‐autonomous alterations in their neurophysiology that contrast starkly with lentiAPP or control cultures. Higher spike amplitudes and spike rates appear precociously in R1.40 cultures, as does the bursting pattern. The cultures then appear blocked in their development and never lose the more immature coordinated bursting that disappears in controls. This suggests that genetic forms of AD may prime a person toward dementia onset through a different set of biological mechanisms than the late‐life elevations of APP and Aβ seen in sAD. This in turn implies that the experimental models that are used in most preclinical studies of AD should be interpreted with caution. Transgenic mice expressing mutant forms of human APP (with or without a second presenilin transgene) can model familial AD but may not reliably replicate the more common sporadic forms. Similarly, the addition of Aβ—either by intracerebral injection or application to neurons in culture—may capture some of the synaptotoxic features of the amyloid peptide but may not fully replicate the complex chemistry and physiology of the AD brain. The lackluster results of human clinical trials based on transgenic mouse models 68 would seem to underscore the need for caution in this area. Finally, the MEA technology is perfectly suited to the exploration of these same network phenomena in cultures of human iduced Pluripotent Stem Cell (iPSC)‐derived neurons as well as in neurons directly induced from human fibroblasts. Testing the ways in which such cultures and their networks differ from the properties we have documented with mouse neurons will be exceptionally informative. AUTHOR CONTRIBUTIONS Conceptualization: all authors. Methodology: Himanshu Akolkar and Fulin Ma. Formal Analysis: Himanshu Akolkar. Investigation: Fulin Ma and Himanshu Akolkar. Writing—Original Draft Preparation: Karl Herrup and Fulin Ma. Writing—Review & Editing: all authors. Visualization: Fulin Ma, Himanshu Akolkar, and Karl Herrup. Supervision: Karl Herrup and Ryad Benosman. Project Administration: Karl Herrup. Funding, Acquisition: Karl Herrup and Ryad Benosman. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. Author disclosures are available in the supporting information. CONSENT No human samples were used in this study. Supporting information Supporting Information Supporting Table S1. The catalogue number, RRID and dilution listed for all the primary antibodies used in this study. Supporting Figure S1. Representative heatmaps of the firing rate (top), spike amplitude (middle), and active electrodes (bottom) of a single wild‐type culture recorded every 4 days from DIV10 to 30. Supporting Figure S2. Average number of axonal branches detected in different conditions as a function of DIV. n = 10 for WT, n = 5 for Aβ, n = 8 for lentiAPP, n = 6 for R1.40. Supporting Figure S3. Conduction speed of action potential propagation in different cultures. Black: wild‐type culture (n = 10), purple: R1.40 culture (n = 6), green: lentiAPP (n = 8); red: wild‐type cultures treated with 1 µM Aβ oligomers (n = 5) at different DIVs. Supporting Figure S4. Mouse cortical neurons from a lentiAPP‐infected culture at DIV10 and DIV18. Cells are fixed 4 days after infection. Red arrows showed infected cells with nuclear mCherry (red). White arrows show uninfected cells. Cells were immunolabeled with synapsin I (green) and PSD95 (magenta). White boxes show example cells with zoomed‐in images on the right panels. Scale bar = 10 µm. Supporting Movie S1. Example of a 100 ms recording demonstrating the propagation of an action potential from its initiation site. The same site from a wild‐type culture was analyzed at DIV 14 and again at DIV 18. The propagation shows that the main axons remain the same but have developed additional branches on DIV 18 compared to DIV 14.