Research Article

In vivo modeling of human neuron dynamics and Down syndrome

See allHide authors and affiliations

Science  16 Nov 2018:
Vol. 362, Issue 6416, eaau1810
DOI: 10.1126/science.aau1810

Development of human brain neurons

The earliest stages of human brain development are very difficult to monitor, but using induced pluripotent stem cells (iPSCs) can help to elucidate the process. Real et al. transplanted neural progenitors derived from human iPSCs into the brains of adult mice. They used intravital imaging to visualize how resulting neurons grew and connected. The human cells produced neurons that integrated and developed synaptic networks with oscillatory activity. Dendritic pruning was observed and involved a process of branch retraction, not degeneration. Cells derived from individuals with Down syndrome, upon transplantation into the mouse brain, produced neurons that grew normally but showed reduced dendritic spine turnover and less network activity.

Science, this issue p. eaau1810

Structured Abstract


Scientists are building detailed maps of the cellular composition in the human brain to learn about its development. In the human cortex, the largest area of the mammalian brain, neural circuits are formed through anatomical refinement, including axon and synaptic pruning, and the emergence of complex patterns of network activity during early fetal development. Cellular analyses in the human brain are restricted to postmortem material, which cannot reveal the process of development. Model organisms are, therefore, commonly used for studies of brain physiology, development, and pathogenesis, but the results from model organisms do not always translate to humans.


Systems to model human neuron dynamics and their dysfunction in vivo are needed. While biopsy specimens and the generation of neurons from induced pluripotent stem cells (iPSCs) could provide the necessary human genetic background, two- and three-dimensional cultures lack factors that normally support neuronal development, including blood vessels, immune cells, and interaction with innervating neurons from other brain areas. On the basis of previous stem cell transplantation studies in mice, we reasoned that the physiological microenvironment of the adult mouse brain could support the growth of human cortical tissue grafts that had been generated from iPSC-derived neuronal progenitors. With human neurons implanted into the mouse brain, high-resolution, real-time in vivo monitoring of human neuron dynamics for periods of time spanning the range from subseconds to several months becomes feasible.


We found that transplanted human iPSC–derived neuronal progenitors consistently assembled into vascularized territories with complex cytoarchitecture, mimicking key features of the human fetal cortex, such as its large size and cell diversification. Single-cell-resolution intravital microscopy showed that human neuronal arbors were refined via branch-specific retraction, rather than degeneration. Human synaptic networks restructured over the course of 4 months, while maintaining balanced rates of synapse formation and elimination. Human functional neurons rapidly and consistently acquired oscillatory population activity, which persisted over the 5-month observation period. Lastly, we used cortical tissue grafts derived from the fibroblasts of two individuals with Down syndrome, caused by supernumerary chromosome 21. We found that neuronal synapses in cells derived from these individuals were overly stable and that oscillatory neural activity was reduced in these grafts, revealing in vivo cellular phenotypes not otherwise apparent.


By combining live imaging in a multistructured tissue environment in mice with a human-specific genetic background, we provide insights into the earliest stages of human axon, synaptic, and network activity development and uncover cellular phenotypes in Down syndrome. Our work provides an alternative experimental system that can be used to study other disorders affecting the developing human cortex.

Human neuron dynamics imaged in vivo.

We combined a human-specific genetic background with live imaging in cortical tissue grafts to investigate the earliest stages of human axon, synaptic, and network activity development and model Down syndrome.


Harnessing the potential of human stem cells for modeling the physiology and diseases of cortical circuitry requires monitoring cellular dynamics in vivo. We show that human induced pluripotent stem cell (iPSC)–derived cortical neurons transplanted into the adult mouse cortex consistently organized into large (up to ~100 mm3) vascularized neuron-glia territories with complex cytoarchitecture. Longitudinal imaging of >4000 grafted developing human neurons revealed that neuronal arbors refined via branch-specific retraction; human synaptic networks substantially restructured over 4 months, with balanced rates of synapse formation and elimination; and oscillatory population activity mirrored the patterns of fetal neural networks. Lastly, we found increased synaptic stability and reduced oscillations in transplants from two individuals with Down syndrome, demonstrating the potential of in vivo imaging in human tissue grafts for patient-specific modeling of cortical development, physiology, and pathogenesis.

Cellular analyses in the human brain are restricted mainly to postmortem material, which cannot provide direct observation of dynamic events, such as anatomical refinement (1) and the emergence of complex patterns of network activity. This limitation raises the question of how to model human neuron dynamics and their dysfunction in the many incurable disorders that affect the developing cortex (2).

Rodent models have been valuable for understanding the pathophysiology of complex genetic disorders, such as Down syndrome (DS) (35), which is associated with neurodevelopmental alterations and is caused by trisomy of chromosome 21 (Ts21), but certain phenotypes are better captured in the context of a human genetic background (6).

Human induced pluripotent stem cell (iPSC)–derived neurons can be used in patient-specific studies to model human cortical development (7), but in vitro two-dimensional (2D) and 3D cultures (8, 9) lack key interactions with neuroglia and vasculature (10). Therefore, systems that more closely recapitulate the complex cellular dynamics of the living brain by using patient-specific cells are urgently needed.

Building on previous transplantation work (11), we hypothesized that the existing physiological microenvironment in the adult mouse brain could support the expansion of human cortical tissue grafts from iPSC-derived neurons, thus allowing high-resolution, real-time in vivo monitoring of human neuron dynamics for extended periods of time.

In this study, we used single-cell-resolution intravital microscopy (12) in human tissue grafts to gain insights into the dynamics of pruning, synaptogenesis, and network activity during the earliest stages of cortical neuron development and demonstrated this approach by modeling human neuron structural and functional dynamics in DS. This research was approved by the U.K. Stem Cell Bank Steering Committee and the U.K. Home Office, in accordance with the U.K. Code of Practice for the Use of Human Stem Cell Lines and the U.K. Animals (Scientific Procedures) Act 1986, respectively.

Complex cytoarchitecture in human cortical tissue grafts

To study the dynamics of human axon and synaptic development and population activity in vivo, we generated cortical excitatory neurons from a control human iPSC line (13) (fig. S1) and transplanted them into the adult mouse somatosensory cortex (SCx1) for chronic multiphoton imaging (Fig. 1A). Cells were transplanted after 36 to 38 days of differentiation, a stage at which cultures contained ~50% neural progenitor cells and ~50% deep-layer cortical neurons [of which ~15% expressed T-box, brain 1 (TBR1+), and ~85% expressed COUP transcription factor–interacting protein 2 (CTIP2+)] (fig. S2, A and B). As expected, and consistent with ongoing neurogenesis after engraftment, upper-layer cortical excitatory neurons and a small proportion of astrocytes and oligodendrocytes could also be found at both 3 and 5 months posttransplantation (mpt) (fig. S2, C and D). Electron microscopy (EM) confirmed that human grafts resembled immature cortical tissue at 130 days posttransplantation (dpt) (fig. S3, A to C), with few synapses and few myelinated axons, and showed no detectable boundary with the mouse brain (fig. S3C), suggestive of structural integration (14). The grafts contained proliferating cells (fig. S3, C and D), enlarged with time (movie S1), and consisted of multiple human- and host-derived cell types (figs. S2 and S3). The cell types from the host included microglial cells, oligodendrocytes, astrocytes, and both excitatory neurons and inhibitory interneurons (fig. S3, D to F), whereas no interneurons of human origin were found (n = 3 transplants). Microglia recruitment in the graft was minimal (fig. S4). Postmortem analysis revealed that the human tissue grafts developed organizational features resembling the structural arrangement of the early fetal cortex (fig. S5) (15, 16).

Fig. 1 Single-cell-resolution in vivo imaging of human cortical tissue grafts reveals mechanisms of pruning.

(A) Schematic of experimental design (left) and two-photon in vivo imaging time line (right). NeurRef, neurite refinement; CaDyn, calcium dynamics; SynDyn, synaptic dynamics. (B) Representative two-photon overview of the cranial window over the injection site at 3 mpt. (C) Bright-field view of a cranial window (~15 mm2) at 5 mpt. Arrowheads indicate blood vessels. (D) Representative immunostaining of endothelial marker CD31 in the human graft at 5 mpt. Arrowheads indicate blood vessels. hNu, human nucleus marker. (E) Representative example of axonal bundles (arrows) along blood vessels. Dashed red lines represent a blood vessel. (F) Representative example of axonal layering in human grafts. The example shown is the same as that in movie S3. (G) Example of a human neuron migrating (*) and remodeling the leading processes (arrows) over 7 hours. (H) Representative example of extensive remodeling of a dendritic arbor in a human pyramidal neuron over 25 hours. (I) Pruning of axonal branch over 6 hours. Dashed red lines represent a blood vessel. (I′) Neurite degeneration over 22 hours. Arrows indicate axonal fragments. (J) Representative examples of axon elongation and retraction over 24 hours. The boxed area in the right panel is magnified in the inset. The arrows in the inset indicate EPBs. gc, growth cone. (K) Speed of neurite elongation and retraction at 3 mpt (n = 113 neurites from 104 cells in six animals, average 17 cells per animal). Means and SEM are indicated. Mann-Whitney U test, ***P < 0.001. (L) Proportion of neurites elongating, retracting, and stable in 24-hour intervals at 3 mpt (n = 92 neurites from 88 cells in six animals, average 15 cells per animal). Error bars indicate SEM. Bonferroni’s multiple comparisons test after one-way analysis of variance (ANOVA), F2,15 = 43.74, P < 0.0001; *P < 0.05; ****P < 0.0001. Scale bars, 500 μm (B), 100 μm (D), 50 μm [(E) and (F)], 20 μm [(G), (H), and (J)], 10 μm (I), and 2 μm (I′).

At earlier stages (<2 mpt), cortical tissue grafts contained areas with ventricular zone–like territories, with cells positive for Paired box protein 6 (PAX6), a marker of neuronal progenitors, and Nestin, a marker for radial glia, which extended processes both radially outward from the core of the rosette-like structures (fig. S5A) and arranged in parallel (fig. S5B), mimicking the organization of radial fibers in the intermediate zone of the human fetal cortex (15). Ki67-expressing proliferating cells were found in the inner apical layer, with doublecortin (DXC)-positive immature neurons toward the basal part, extending out into the rest of the graft (16) (fig. S5A). After 2 mpt, the rosettes did not persist, and although discrete cortical laminae were not clearly visible, consistent with their formation in late embryonic development (~7 months postconception) (17), immunostaining for deep- and upper-layer cortical neurons with antibodies for TBR1 and Special AT-rich sequence–binding protein 2 (SATB2), respectively, showed that these cell populations can segregate in vivo (fig. S5C). Human astrocytes were homogenously distributed in the cortical tissue grafts (fig. S5D). Lastly, human tissue grafts were vascularized, as shown in vivo and by the endothelial marker cluster of differentiation 31 (CD31) (Fig. 1, B to D), suggesting that the adult mouse brain microenvironment can support the development of a multicellular transplant.

Human axon pruning imaged in vivo

To track human neurons in vivo, we engineered them to express green fluorescent protein (GFP) via lentivirus-mediated transduction before transplantation. Human neurons were present for the duration of our experimental time course, which spanned up to 6 months, and spread away from the injection site (Fig. 1B) [on average, up to 1.2 ± 0.6 mm (mean ± SD) from the bregma in the rostral direction over the first 3 mpt (n = 4 mice)]. Consistent with the immature brain cell-cell interactions (10), human axons grew along blood vessels and as fiber bundles (Fig. 1E and movie S2), and parallel and radially oriented axonal layers could be detected below the dura mater (Fig. 1F and movie S3), similar to the ones found in the human cortex (18).

Given the widespread axonal extension outside the graft area, we asked which brain regions human neurons target 5 mpt. Main SCx1 target areas showed a higher number of human fibers than in areas known to receive fewer projections from SCx1 (fig. S6), suggesting that the direction of axon elongation is targeted. For example, the ipsilateral motor cortex, striatum, thalamus, and contralateral SCx1 received more fibers than the cerebellum and substantia nigra, and the corpus callosum had more axonal tracts than the internal capsule and cerebral peduncle (fig. S6), as expected from rodent tracing experiments (19). These data provide evidence for long-range (over centimeters) axon growth of grafted human neurons through the mouse adult brain and indicate that, although human axons are either not responsive to or can overcome the inhibitory signals present in the adult mouse brain, they may be directed by existing guidance cues or paths.

After an initial phase of growth (20), the selective pruning of axons and dendrites is thought to occur normally via retraction and degeneration during early development (2, 21). We explored the mechanisms of human neurite pruning up to 3 mpt (Fig. 1, G to L, and fig. S7, A and B). At this stage, neurons were still migrating (Fig. 1G) and developing neural processes in a highly dynamic mode (Fig. 1, G to L). We tracked the fate of 92 human neurites from 88 cells in six mice at 3 mpt (Fig. 1, G to L). Whereas most neurites (58.4% ± 5.5%) elongated in 24 hours, neurite refinement was dynamic, and interchanging retraction and elongation of individual neurites (31.0% ± 2.1%) over 24 hours were observed (Fig. 1, I to L). Developmental neurite degeneration involves cytoskeletal destruction with widespread fragmentation over a time scale of 12 to 48 hours (22), whereas retracting axons do not leave fluorescent fragments behind (23). Reducing the imaging interval from 24 hours to 8 hours showed that branch pruning (Fig. 1I) occurred mainly by retraction (91%), rather than degeneration (Fig. 1I′) (9%). Axonal en passant boutons (EPBs), one of the two types of presynaptic specialization on cortical axons (24), could be observed in branches with a growth cone elongating (Fig. 1J). Neural processes extended long distances (maximum neurite extension = 462.769 μm in 24 hours) at a speed of 10.29 ± 0.73 μm/hour (Fig. 1K), comparable to that observed in the neonatal mouse brain (23). Results were validated with tissue grafts from an independent control line (fig. S7, A and B).

Human synaptic development imaged in vivo

Next, we studied the dynamics of synaptogenesis up to 4 mpt. Hallmarks of developing synaptic networks are an increase in synaptic density over time, followed by pruning, and the acquisition of a steady state with balanced rates of synaptic gain and loss (25). However, when and how human synaptic networks acquire these properties is unclear. We first considered dendritic spine formation and elimination (Fig. 2, A to F).

Fig. 2 Developing human synaptic networks are characterized by substantial restructuring and balanced rates of gains and losses.

(A) Overview of cranial window at 136 and 138 dpt. Red arrows point to examples of cells with a stable location over a 48-hour period. (B) Detail of a representative dendrite imaged over 24 hours (white box in the top panel and red box in fig. S8A). Green, red, and white arrowheads indicate gained, lost, and stable dendritic spines, respectively. (C) Dendritic spine density over 4 to 6 days at 3 mpt (n = 8 cells, 1.40 mm of total dendritic length, from three animals) and 4 mpt (n = 6 cells, 0.93 mm of total dendritic length, from two animals). Two-way ANOVA, interaction F3,46 = 0.4357, P = 0.73. ****P < 0.0001. (D) Average fractions of dendritic spines gained and lost over 48 hours at 3 mpt (red, n = 8 cells) and 4 mpt (blue, n = 6). Two-way ANOVA, interaction F1,24 = 0.1894, P = 0.67. Sidak’s multiple comparisons test, *P < 0.05 (gains); P = 0.063 (losses). ns, not significant. (E) Dendritic spine TOR over 4 days at 3 mpt (n = 8 cells) and 4 mpt (n = 6 cells). Mann-Whitney U test, *P < 0.05. Each data point represents a cell. (F) Dendritic spine survival fraction at 3 mpt (red, n = 7 cells) and 4 mpt (blue, n = 6 cells). Two-way ANOVA, interaction F3,47 = 1.513, P = 0.22; *P < 0.05. (G) Representative example of a branched human axon at 130 dpt. The arrow indicates a growth cone. The boxed area is magnified in subsequent panels. (H) Detail of the axon shown in the boxed area in (G), imaged every 48 hours over 4 days. Green, red, and white arrowheads indicate gained, lost, and stable EPBs, respectively. (I) EPB density over 2 to 4 days at 3 mpt (n = 8 cells, 1.3 mm of total axonal length, from three animals). One-way ANOVA, F2,17 = 0.4014; P = 0.68. (J) Quantification of EPB TOR over 4 days at 3 mpt (n = 4 cells). Each data point represents an axon. (K) Quantification of EPB survival fraction at 3 mpt (n = 8 cells). (L) Average fractions of EPB gains and losses over 48 hours at 3 mpt (n = 8 cells). Wilcoxon matched-pairs signed-rank t test; ns, not significant. [(C), (D), (F), (I), (K), and (L)] Dashed lines represent individual cells, and solid lines represent means. Scale bars, 50 μm (A), 20 μm [(B), top panel], 2 μm [(B), bottom panel], 10 μm (G), and 5 μm (H).

After the initial phase of cell migration and neurite remodeling (Fig. 1, G and H), neurons stabilized, allowing us to track the same cells over time (Fig. 2A and fig. S8). Dendritic spines, the structural correlates of mammalian excitatory synapses (26), were seen as early as 20 dpt (32.8 ± 5.5 dpt for either dendritic filopodia, considered to be the precursors of dendritic spines, or spines; n = 3 mice) (27, 28). We monitored >500 dendritic segments from six mice over days. However, for most dendrites, the density of synapses was too low to quantitatively study the dynamics of dendritic spines before 3 mpt, as expected from previous human fetal cerebral cortex postmortem work (29) and the early developmental stage modeled in this study. Eight neurons had sufficient dendritic spine numbers at 3 mpt to calculate spine density and turnover during three to four consecutive sessions of 48-hour intervals (up to 6 days). The average spine density was similar to that in the human early fetal cerebral cortex (29) and constant over the imaging period (Fig. 2C) (0.043 ± 0.006 spines/μm; n = 70 spines present in the first session, 176 in total; Kruskal-Wallis test, P > 0.05). Synaptic structures were added and eliminated at equal rates, even at these early developmental stages (Fig. 2D) (Wilcoxon matched-pairs signed-rank test, P > 0.05). The turnover ratio (TOR), a function of both spine gain and loss (30), was 46.9% ± 5.3% over 4 days (Fig. 2E), indicating synaptic reorganization.

To investigate the development of synaptic remodeling over time, we repeated the same experiment after 1 month. Again, spine density was constant over time (Fig. 2C) (0.112 ± 0.024 spines/μm; n = 171 spines present in the first session, 291 in total; Kruskal-Wallis test, P > 0.05). However, the average spine density was increased at 4 mpt. The majority of dendrites had balanced rates of dendritic spine gain and loss (Fig. 2D) (paired two-tailed t test, P > 0.05), and only in one cell were we able to capture net synaptic pruning over 2 days (Fig. 2C, thick dashed line), consistent with the idea that a major phase of synaptic pruning occurs only at later developmental stages (28).

The TOR over 4 days was 27.6% ± 3.7%, which was lower than at 3 mpt (Fig. 2E). Consistently, the survival fraction, defined as the fraction of spines surviving as a function of time, was higher at 4 mpt (Fig. 2F), suggesting stabilization of dendritic spine dynamics over time.

To more thoroughly assess synaptic dynamics, we also studied presynaptic terminals along human cortical axons (Fig. 2, G to L). The density of boutons remained stable over time (Fig. 2I) (0.051 ± 0.0075 EPBs/μm; n = 69 EPBs in the first session, 145 in total), indicating that axonal boutons were also added and eliminated at equal rates (Fig. 2L). The TOR over 4 days was 45.1% ± 3.6% (Fig. 2, J and K), denoting comparable dynamics between dendritic spines and axonal boutons (at 3 mpt, Mann-Whitney U test, P = 0.34).

In summary, we were able to study early events of human cortical neuron synaptogenesis over the first 4 mpt. Despite the low synaptic density, consistent with the primordial stage modeled in this study (29), we can draw a number of conclusions about early in vivo human synaptic network development. First, transplanted human neurons initially formed synaptic structures within 4 to 12 weeks of in vivo development, similar to the human fetal cerebral cortex (29). Second, they underwent synaptic reorganization. Third, they progressively increased dendritic spine density over 1 month. Finally, human neurons balanced the rates of synaptic gain and loss over a time scale of a few days.

Functional human cortical networks imaged in vivo

Patterned neural activity is thought to be fundamental to neural circuit development in the immature brain (31, 32). Although spontaneous and sparse activity can be detected in human cortical network preparations in vitro, recapitulating patterns typical of early human cortical population activity, such as recurrent oscillatory bursts (32), remains challenging (33, 34).

We first investigated the electrophysiological properties of transplanted cells. We performed ex vivo whole-cell recordings in coronal brain slices containing the grafts (fig. S9). Current-clamp recordings were made from 18 pyramidal neurons (n = 4 mice), as identified by using differential interference contrast microscopy and expression of either GFP or tdTomato and by filling neurons with Lucifer yellow dye before post hoc anatomical inspection (fig. S9A). Patched grafted pyramidal neurons were at different stages of biophysical maturation and development, with an average resting membrane potential of −53.8 ± 1.7 mV, average capacitance of 19.4 ± 2.2 pF, and average input resistance of 1.4 ± 0.1 gigaohms. Although cells were quiescent at resting membrane potentials, depolarizing current steps evoked action potential firing in all pyramidal neurons tested (fig. S9B), with average action potential amplitudes of 91.3 ± 2.6 mV and half-widths of 2.2 ± 0.2 ms.

Immunohistochemistry showed glutamatergic and GABAergic terminals within the human graft (fig. S10, A and B). To confirm that human neurons received both excitatory and inhibitory input, pyramidal neurons were voltage clamped (−70 mV) and spontaneous miniature excitatory postsynaptic currents (mEPSCs) were observed at a frequency of 0.30 ± 0.05 Hz (5 of 18 neurons) with an amplitude of 20.1 ± 3.2 pA, which were completely blocked by the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor antagonist 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo-quinoxaline-2,3-dione (NBQX) (n = 4). Although synaptic events were observed in the remaining neurons, spontaneous frequency was insufficient to acquire enough events for statistical analysis (figs. S7, C and D, and S9C). By using a high-chloride (130 mM) internal solution and in the presence of NBQX, spontaneous miniature inhibitory postsynaptic currents were observed at a frequency of 0.24 ± 0.12 Hz (three of six neurons) with an amplitude of −73.3 ± 21.0 pA, which were fully inhibited by bicuculline (fig. S9C). Similar to mEPSCs, inhibitory synaptic events were observed in the remaining neurons, but insufficient events were acquired for detailed kinetic analysis. In summary, grafted neurons are excitable and fire action potentials. In addition, they receive both excitatory and inhibitory input, suggesting functional network connectivity.

To determine the origin of the afferent synaptic input to the functionally active neurons, we performed monosynaptic retrograde tracing by using a modified rabies virus. This virus lacks a glycoprotein needed for replication and can infect only cells expressing the avian tumor virus receptor A (TVA) (fig. S11). Human iPSC–derived cortical progenitors and neurons were transduced with a lentiviral vector containing the TVA, nuclear GFP, and glycoprotein under the control of the human synapsin promoter (fig. S11A). Five months after the transplantation, the modified mCherry expressing–rabies virus was injected in the same location, where only grafted cells expressing the TVA are susceptible to infection. Cells that are monosynaptically connected to the infected human cells also become infected and express mCherry, allowing for accurate tracing of the neural input to the cells in the human grafts (fig. S11B). We observed that whereas most of the input to the transplanted human neurons comes from other human neurons (92.5% ± 1.5%, n = 4333 cells in two brains), host neurons also innervate the human graft (7.5% ± 1.5%, n = 397 cells in two brains) (fig. S11C). The traced host neurons were located within the graft, in the cortical areas adjacent to the graft, in the contralateral cortex, and in the ipsilateral CA1 hippocampal region (fig. S11B). Although no traced neurons were found in other subcortical regions, thalamocortical terminals were present in the graft (fig. S10; see also fig. S12) (20). These results provide evidence that most synaptic input to the grafts comes from other human neurons. Furthermore, as no interneurons of human origin were found, these data, together with the demonstration that human neurons in the graft receive inhibitory input (fig. S9C, bottom), suggest that inhibition in the human grafts comes from the host.

To assess the functional development of cortical networks in vivo, we engineered neurons to express the genetically encoded calcium indicator GCaMP6s (35) before grafting and studied calcium-mediated neuronal activity in vivo (Fig. 3) (n = 8 mice). Spontaneous, sparse activity (Fig. 3, A to C) was detected as early as 2 weeks posttransplantation (wpt) and persisted up to 3 mpt (Fig. 3C). In addition, bursts of activity synchronized across the neuropil and multiple cells (31) were also detected at 1 mpt (Fig. 3C, inset) and persisted in all grafts tested up to 5 mpt (fig. S7, E to H, and movies S4 and S5).

Fig. 3 In vivo calcium imaging shows that patterned population activity emerges early and has a defined spatiotemporal order.

(A) Example of an imaged cortical region taken from a WT-1 graft at 1 mpt in the somatosensory cortex of an adult mouse. Neurons express tdTomato (red) and GCaMP6 (green). GCaMP-positive neurons are shown as a maximum-intensity projection of activity over a 4-min period of spontaneous activity. Active neurons (yellow) are shown by overlaying the images (merge). (B) Representative ΔF/F0 calcium traces (where ΔF/F0 is the ratio of the change in fluorescence to the baseline fluorescence) from five active neurons imaged in a WT-1 graft at 1 mpt. (C) Distribution of spontaneous calcium activity in WT-1 grafts at 1 to 2 mpt. Activity was measured as the integral of the average ΔF/F0 signal over the entire region of interest (ROI), normalized to the total duration of the recording in seconds (n = 88 cells, six ROIs, three mice). (Inset) Percentage of ROIs in WT-1 grafts at 1 to 2 mpt (3 of 16 ROIs, 18.8%; n = 4 mice) and 3 mpt (31 of 35 ROIs, 89.0%; n = 5 mice) that exhibit bursts. Chi-square test, *P < 0.05. (D) Montage of image frames from a typical recurrent burst in a WT-1 graft. (E) Example of burst activity over two different spatial regions (gray and black) shown on the left, taken from the bursts in (D). Scale bars, 10 μm (A) and 20 μm (D).

Many of these bursts had a defined spatiotemporal order (Fig. 3, D and E), as well as recurrent oscillatory behavior (<1 Hz between events) (Fig. 3D and fig. S7, E to I), with different incidences between 1 and 3 mpt (Fig. 3C, inset), resembling activity recorded in the developing human cortex (36, 37) and consistent with a report on transplanted human cerebral organoids (38).

Recordings of calcium signals with air-puff stimulation of the animal’s whiskers and facial skin revealed that grafted neurons in the primary somatosensory cortex can be responsive to sensory stimulation (fig. S12) (~30% of the stimulation trials in one mouse; neither of the other two animals tested showed sensory-evoked activity), indicating that thalamocortical synapses can functionally drive activity in the human graft at 6 mpt.

Imaging human neuron structural and functional dynamics in DS

So far, we have characterized the structural and functional dynamics of human cortical neurons during the earliest phases of their development in vivo (Figs. 1 to 3) and validated the main results with neurons from an independent control iPSC line [designated WT-2 (for wild-type line 2)] (fig. S7). To model the in vivo dynamics of pruning, synaptogenesis, and network activity in a complex genetic disorder, we first generated iPSC-derived progenitors and neurons from two individuals with DS (fig. S1) and then transplanted the cells into adult immunodeficient mice. During the reprogramming process of one of these lines [designated Ts21-2 (Ts21 line 2)], we identified a disomic clone that had lost one extra copy of human chromosome 21 (Hsa21) (yielding a WT-2 population) (3941). We used a microsatellite short tandem repeat (STR) assay to confirm that the parental fibroblast population was not mosaic for disomy and trisomy of chromosome 21 and that Ts21-2 and WT-2 are otherwise identical to each other and the initial fibroblasts (fig. S13). This revertant disomic line (WT-2) allowed us to highlight phenotypes caused by an extra copy of Hsa21, rather than by genetic differences between individuals, without the need for multiple control lines, which are typically required to control for genetic variations or diverse differentiation potencies observed in genetically distinct human iPSC lines (42). A genome-wide copy number single-nucleotide polymorphism assay confirmed that the two Ts21 iPSC lines (Ts21-1 and Ts21-2) had normal karyotypes, except for the extra copy of Hsa21 (fig. S14). Fluorescence in situ hybridization (FISH) on cortical tissue grafts further verified the presence of the extra copy of Hsa21 (fig. S15). The Ts21 lines generated progenitors, neurons, and proliferating cells similarly to control grafts at 5 mpt (figs. S16 and S17). Astroglia, however, were overproduced in Ts21 grafts (fig. S16), recapitulating the human pathology (43). Ts21 neurons were also present in stable locations to the end of our experimental time line, allowing for in vivo single-cell tracking (fig. S18). Chronic in vivo imaging revealed that Ts21 neurons had rates of axon growth and retraction similar to those of control neurons at 3 wpt (Fig. 4, A to D), suggesting normal early developmental axon refinement. In addition, Ts21 neurons in the graft formed morphologically mature synaptic structures, which were plastic over time (Fig. 4, E to L). To determine whether dendritic spine growth was associated with synapse formation in Ts21 neurons, we reconstructed in one transplant, with EM, a subset of the same dendrites after long-term in vivo imaging (Fig. 4, E to G). We found that newly formed dendritic spines formed synapses in 14 of 34 cases (41%) and six of them (6 of 14, 43%) within 48 hours of their first appearance. Serial EM reconstructions revealed that human dendritic spines and presynaptic terminals contained a postsynaptic density and synaptic vesicles, respectively, suggestive of complete synaptic maturation (Fig. 4G). Whole-cell recordings from coronal brain slices containing the Ts21 grafts showed normal synaptic input on the DS donor–derived neurons compared to the control (fig. S19, A to D), suggesting functional synaptic connections. Longitudinal in vivo imaging, however, showed that dendritic spines, and to a lesser extent synaptic boutons (Fig. 4, J to L), were more stable in neurons from both individuals with DS than in the control, as demonstrated by higher survival and reduced turnover (Fig. 4, H and I, and fig. S20). High density of GFP-positive neurons prevented a quantitative analysis of synaptic dynamics in the WT-2 line. To understand whether the higher dendritic spine survival rates in Ts21 lines lead to higher spine density, we quantified dendritic spine density across the four lines (Ts21-1, Ts21-2, WT-1, and WT-2). We found an increase in dendritic spine density in neurons from the Ts21-1 line compared with WT-1 (fig. S21A), although this increase did not reach significance, consistent with postmortem fetal DS brain analysis at ~5 to 8 gestational months (27). However, we found higher spine densities in Ts21-2 than in WT-2, our most reliable comparison (fig. S21A). Putting the data from the two Ts21 and WT lines together highlighted a significant spine density increase in the Ts21 cells (fig. S21B). Overall, these data raise the possibility that spine density in DS cortical neurons is higher than in the control, at least at the early developmental stages tested. No difference in EPB density was found across the four lines (fig. S21, C and D).

Fig. 4 In vivo modeling of structural and functional neuronal dynamics in tissue grafts from individuals with DS.

(A) Representative example of axon elongation in a Ts21-1 neuron over a 24-hour period. The inset corresponds to the boxed area and highlights the presence of EPBs. The red line indicates alignment between the top and bottom images. (B) Example of axonal branch retraction (arrows) in a Ts21-1 neuron over 17 hours. (C) Proportion of elongating, retracting, and stable neurites in 24-hour intervals in WT-1 (n = 96 neurites from 79 cells, seven grafted animals, average 11 cells per animal), Ts21-1 (n = 65 neurites from 60 cells, seven grafted animals, average 9 cells per animal), WT-2 (n = 65 neurites from 53 cells, four grafted animals, average 13 cells per animal), and Ts21-2 (n = 60 neurites from 51 cells, four grafted animals, average 13 cells per animal) grafts at 3 wpt. WT-2 is a revertant disomic cell line from Ts21-2. Unpaired two-tailed t test; ns, not significant. Each data point represents an animal. (D) Speed of neurite elongation and retraction in WT-1 (n = 96 neurites from 73 cells, average 10 cells per animal), Ts21-1 (n = 62 neurites from 54 cells, average 8 cells per animal), WT-2 (n = 53 neurites from 47 cells, average 12 cells per animal), and Ts21-2 (n = 54 neurites from 46 cells, average 12 cells per animal) grafts at 3 wpt. Unpaired multiple t test; ns, not significant. Each data point represents an animal. (E) Example of dendritic branches and spines on a Ts21-1 neuron, imaged at 48-hour intervals for 4 days. The boxed region in the left panel is magnified in subsequent panels. Green, red, and white arrowheads indicate gained, lost, and stable dendritic spines, respectively. (F) 3D rendering of the same dendritic region imaged in vivo in (E), obtained from EM reconstruction. Presynaptic terminals are shown in green. (G) EM images of the dendritic spines marked with 1 and 2 in (E). Arrowheads indicate the location of synapses. Asterisk, presynaptic terminal. (H) Dendritic spine survival fraction over 4 days in WT-1 (n = 10 cells from two animals), Ts21-1 (n = 9 cells from four animals), and Ts21-2 (n = 7 cells from two animals) grafts at 3 to 4 mpt. Two-way ANOVA, interaction F4,69 = 5.435, P = 0.0007; Tukey’s multiple comparisons test, ****P < 0.0001. Each data point represents a cell. (I) Quantification of dendritic spine TOR over 4 days in WT-1 (n = 10 cells from two animals), Ts21-1 (n = 9 cells from four animals), and Ts21-2 (n = 7 cells from two animals) grafts at 3 to 4 mpt. Sidak’s multiple comparisons test after one-way ANOVA, F2,23 = 3.078, **P < 0.01; ***P < 0.001. Each data point represents a cell. (J) Representative example of an axon on a Ts21-2 neuron imaged at 48-hour intervals for 4 days. The arrowheads in the insets indicate stable (white), new (green), and lost (red) EPBs. (K) EPB survival fraction over 4 days in WT-1 (n = 6 cells), Ts21-1 (n = 24 cells), and Ts21-2 (n = 10 cells) grafts from three mice each at 3 to 4 mpt. Two-way ANOVA, interaction F4,111 = 0.8211, P = 0.51; ns, not significant. Each data point represents an axon. (L) EPB TOR over 4 days in WT-1 (n = 6 cells), Ts21-1 (n = 24 cells), and Ts21-2 (n = 10 cells) grafts from three mice each at 3 to 4 mpt. Sidak’s multiple comparisons test after one-way ANOVA, F2,37 = 5.588, **P < 0.01; ns, not significant. Each data point represents an axon. (M and N) (Left) Example of imaged cortical regions taken from Ts21-1 (M) and Ts21-2 (N) grafts in the somatosensory cortices of adult mice. Neurons express tdTomato (red) and GCaMP6s (green). Active neurons (yellow) are shown by overlaying the images. (Right) Representative ΔF/F0 calcium traces from five active neurons imaged in Ts21-1 (M) and Ts21-2 (N) grafts. Note weak synchronized burst activity across different neurons compared with the traces in fig. S7E. (O) Percentage of ROIs in WT-1 (50 of 52 ROIs, 96.1%, six grafted mice), Ts21-1 (10 of 38 ROIs, 26.3%, three grafted mice), WT-2 (34 of 34 ROIs, 100%, three grafted mice), or Ts21-2 (11 of 23 ROIs, 47.8%, three grafted mice) grafts that exhibit bursts at 3 to 5 mpt. Z test, ***P < 0.001. (P) Frequency of burst events in WT-1, Ts21-1, WT-2, and Ts21-2 grafts measured at 3 to 5 mpt. Kruskal-Wallis test, **P < 0.01; ***P < 0.001. (Q) Global ROI activity in WT-1, Ts21-1, WT-2, and Ts21-2 grafts measured at 3 to 5 mpt. Kruskal-Wallis test, ***P < 0.001. Error bars in (P) and (Q) indicate SEM. Scale bars, 10 μm [(A) and (B)], 5 μm [(E), left, and (J)], 2 μm [(E), right], 0.2 μm (G), and 20 μm [(M) and (N)].

To further investigate the increased synaptic stability phenotype, we studied neural population activity, a main regulator of postnatal synaptic refinement and stabilization (26), through in vivo calcium imaging of GCaMP6-expressing Ts21 grafts (Fig. 4, M and N). We measured both burst and global activity (see methods). These measures were reduced in Ts21 grafts (Fig. 4, O to Q). Together, these data highlight in vivo synaptic stability and functional early cortical network phenotypes in DS.


We investigated the earliest stages of human axon, synaptic, and network activity development in a complex genetic disorder by combining live imaging in a multistructured tissue environment and a patient-specific genetic background.

Transplanted human neurons continued to develop and mature in vivo, in a microenvironment that retained features reminiscent of the human fetal cortex, such as large size (up to ~100 mm3 at 5 mpt) (movie S1), temporal order and duration (i.e., many months) of neurogenesis (20), vascularization, and cell diversification (human-derived cortical progenitors, neurons, oligodendrocytes, and astrocytes together with host-derived microglia and vessels), as well as complex cytoarchitecture. However, the extent to which neurons in human cortical tissue grafts, generated from either human iPSCs (present study) or embryonic stem cells (ESCs) (20, 38, 44), can mimic the maturation, complexity, and functionality of early human fetal cortical networks remains to be fully established.

Repeated imaging of single human neurons in cortical tissue grafts enabled us to gain insights on pruning, synaptic refinement, and functional neural network formation in vivo. We found that pruning occurred mainly by branch-specific retraction, rather than degeneration.

Nascent human excitatory synaptic networks already had balanced rates of synaptic gain and loss over ~1 week at the single-cell level, suggesting that immature human neurons possess intrinsic programs of synaptic turnover regulation over relatively short time scales. Human synaptogenesis and axon growth were concurrent, rather than happening at different times, confirming previous postmortem static analysis (28) and revealing conservation of this developmental growth program between species (45).

Oscillatory population activity had marked neuropil and soma synchronization, which became more prominent over 2 months, underscoring on-going modifications of cortical circuits. Results were robust across two independent control lines, providing a basis for applying this approach, which combines live imaging in a multistructured tissue environment with a patient-specific genetic background (46), to many other neurodevelopmental diseases affecting the cortex.

In this study, we modeled a complex genetic disorder and saw that whereas developmental axon refinement was normal, synapses were more stable and neural network activity was reduced in tissue grafts from two individuals with DS, suggesting a possible role for patterned activity in regulating synaptic lifetimes in the early stages of human cortical circuit development (32). These deficits were evident even after Ts21 cells were exposed to the in vivo physiological microenvironment of the mouse brain for several months, indicating cell-intrinsic deficits. By using a revertant disomic iPSC line, we showed that the population activity deficits were rescued by the loss of an extra copy of Hsa21, indicating that heightened expression of Hsa21 genes is both necessary and sufficient to disrupt oscillatory burst activity in developing cortical DS networks in vivo.

In most previous work, human ESC– or iPSC–derived neurons have been transplanted into the damaged cortex (38, 47), spinal cord (48), striatum (49, 50), or retina (51), with the aim of cell replacement (11) rather than for disease modeling (6, 52), as demonstrated in our study. Transplantation and in vivo imaging for disease modeling in mice is advantageous over that in higher species such as primates, as larger numbers of animals can be used to track cells in the grafts over long periods of time and the model provides a microenvironment containing vessels, immune cells, and innervation, not present in common in vitro preparations.

In summary, we established a new in vivo experimental model of DS to study how the chromosomal abnormality affects the earliest stages of human axon, synaptic, and functional neural network development. We expect that this single-cell-resolution intravital microscopy approach will advance the knowledge of cellular pathophysiology in this and other neurodevelopmental disorders, particularly valuable in light of the scarcity of early human fetal brain tissue material.

Supplementary Materials

Materials and Methods

Figs. S1 to S21

Table S1

References (5360)

Movies S1 to S6

References and Notes

Acknowledgments: We thank K. Alvian, R. Festenstein, T. Keck, and M. Lancaster for comments on the manuscript; S. Papadoupoulou for help with immunohistochemistry; C. Bass and A. A. Bharath (Imperial College London) for help with the calcium imaging analysis; C. Whilding for developing a customized FIJI script for image analysis; M. Tortora for help with the immunohistochemistry experiments and analysis and C. Pernaci for help with the analysis; M. Lavrov and M. Rakowska for help with synaptic dynamics analysis; E. Mustafa, A. Czerniak, K. Horan, A. Matthews, and E. Rowley for assistance with animal care and monitoring; M. Tripodi (LMB, Cambridge) for the kind gift of the modified rabies transynaptic tracer; and G. Stamp for the analysis of hematoxylin and eosin–stained material. Funding: This work was supported by the Medical Research Council (V.D.P.); the GABBA Ph.D. program (FCT fellowship PD/BD/52198/2013), the Rosetrees Trust, and ARUK (R.R.); the UK Dementia Research Institute (grant code DRIImp17/18 Q3 to S.J.B.); a Wellcome Senior Investigator award (F.J.L.); and the Alborada Trust of the ARUK Stem Cell Research Centre (M.P. and F.J.L.). Author contributions: V.D.P. conceived and planned the live imaging, characterization, and analysis of transplanted patient-derived neurons and brought F.J.L. into the project; F.J.L. independently generated the iPSC-derived neurons, conceived the in vitro aspects of the project, and contributed to the study design; R.R., A.T., and V.D.P. performed the grafting and the two-photon imaging experiments, analyzed the data, and prepared the related figures and text; M.P. performed the human iPSC–derived neuron differentiation and in vitro characterization, the copy number assay, and the lentiviral vector transductions and provided input on the design of the experiments; R.R. and S.K. performed cell marker immunohistochemistry, imaging, and analysis for the characterization of cell identity after transplantation and prepared the relevant figures with input from V.D.P.; R.R. conducted and analyzed the whole-brain rabies tracing reconstructions; M.A.S. performed the electrophysiology recordings in acute brain slices containing the human grafts and prepared the relevant figure and text; S.J.B. analyzed the in vivo calcium imaging data and prepared the relevant figures and text with input from R.R. and V.D.P.; A.M. generated and characterized the Ts21-2, WT-2, and WT-2′ lines with input from F.J.L.; A.S. analyzed the gene expression, copy number variation, and STR data and prepared the relevant figures; J.D. and S.K. characterized the graft size; J.D. characterized the axon projections from the grafts; E.V. provided input on the Hsa21 FISH experiment and analyzed the FISH data with input from V.D.P.; G.K. performed and analyzed the EM reconstructions and prepared the relevant figures; V.D.P. led the project; and V.D.P and R.R. wrote the paper with contributions from all authors. Competing interests: None. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the supplementary materials.

Stay Connected to Science


Navigate This Article