All animal care and experimental procedures were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at University of California, Santa Cruz. Thy1-YFP-H line mice (Feng et al., 2000) were purchased from Jackson Laboratory. Timed pregnant C57Bl/6 female mice were purchased from Charles River. Mice were group-housed in the UCSC animal facility, with 12 h light-dark cycle and access to food and water ad libitum. Both male and female mice were used in all experiments.

In Utero electroporation (IUE) was performed as previously described (Saito and Nakatsuji, 2001) on E13.5 or E15.5 timed pregnant C57Bl/6 mice to label L5 or L2/3 PNs, respectively. The pCAG-GFP plasmid (Addgene #11150) was purified using the NucleoBond Extra Midi EF Kit (Clontech Laboratories). The plasmid was diluted to a final concentration of 1 μg/μl with sterile phosphate buffered saline (PBS) and colorized with 0.1% Fast Green (Sigma-Aldrich) dissolved at 37°C immediately prior to use. One to two microliters DNA plasmid was injected into the lateral ventricle (LV) through a pulled glass micropipette. Five pulses (25–30 V amplitude, 50 ms duration with 950 ms intervals) were delivered, targeting the motor or barrel cortex, using a platinum plate tweezers-type electrode connected to a square-pulse electroporator (CUY21, NEPA Gene).

The mouse was transcardially perfused with 4% paraformaldehyde (PFA) in 0.1 M PBS. Following perfusion, the brain was post-fixed in 4% PFA at 4°C overnight and cryoprotected with overnight incubation in 30% sucrose. The brain was then embedded in OCT medium and cryosectioned into 25 μm thick coronal sections. For immunostaining, sections were washed in PBS for 10 min, and incubated in blocking solution (5% goat serum, 0.01% Triton in PBS) for 15 min at room temperature in a humid chamber. Sections were then quickly washed in PBS and labeled with rabbit anti-Cux1 (1:1000; Santa Cruz Biotechnology) at 4°C overnight in a humid chamber. Sections were subsequently incubated with goat anti-rabbit secondary antibody conjugated with Alexa Fluor 594 (1:500; Life Technologies) in 0.1 M PBS for 2 h at room temperature. Finally, sections were washed in PBS and mounted with Fluoromount-G mounting medium (Southern Biotech). Confocal images were taken with a Leica SP5 confocal microscope with 10×/0.3 NA, 20×/0.75 NA and 63×/1.4 NA oil-immersion objectives. All images shown in Figure 1 are representative of at least three replications. Merging of different channels into multi-color images was performed with Adobe Photoshop.

For quantification of fluorescently labeled cells co-labeled with Cux1, wide-field images of brain sections were collected on a Zeiss Axio Imager M2 microscope with a 20×/0.8 NA objective using the Axiovision software and cells were manually counted in Stereo Investigator (MicroBrightField).

Transcranial two-photon imaging and analysis of spine density and dynamics of apical dendritic tufts were performed as previously described (Zuo et al., 2005a). All images were analyzed using ImageJ. Spine density was calculated by dividing the number of spines by the length of the dendritic segment on which they reside. Only dendritic segments that lie within a single optical section are analyzed. Percentage of spines eliminated or formed was calculated as the number of spines eliminated or formed over the total spines counted in the images obtained during the first imaging session. The numbers of animals and spines analyzed under various experimental conditions are summarized in Supplementary Tables S1, S2. All data are presented as mean ± SEM. Mann-Whitney U test and Kruskal-Wallis rank sum test followed by post hoc multiple comparisons test were used for statistical analysis. p < 0.05 was considered significant.

Image processing for Figures 2A,B were performed as previously described (Xu et al., 2009). Briefly, we chose regions with sparsely labeled dendrites as examples and made maximum intensity projections of the image stack. The resulted images were then thresholded, Gaussian filtered and contrast-enhanced for presentation.

Single-pellet reaching test was performed as previously described (Xu et al., 2009; Chen et al., 2014a). Briefly, the mouse was food-restricted to maintain 90% of the ad libitum weight during the experiment. A brief shaping phase was used to familiarize the mouse with the training chamber and task requirements, as well as to determine its limb preference. Thirty pellets were used for each training session. Reach attempts were scored and the success rate was calculated as the percentage of successful reaches over total reaches per session.

Whisker trimming was performed as previously described (Lee et al., 2009). Mystacial vibrissae of both side whisker-pads were cut to skin level daily from postnatal day 0–7. Control mice were handled similarly without whisker trimming.

Whisker-dependent texture discrimination test was performed as previously described (Wu et al., 2013; Chen and Zuo, 2015). Briefly, the mouse was habituated and tested in a testing arena (38 cm × 28 cm × 23 cm). On the testing day the mouse went through three phases: encoding (5 min), rest (5 min) and testing (3 min). In the encoding phase, the mouse was presented with two identically textured columns (3 cm × 3 cm × 9 cm). One of the columns was replaced with a new column of a different texture during testing phase. The amount of time spent actively investigating the columns was recorded and analyzed using EthoVision XT 10-Noldus software. Data are presented as mean ± SEM. Student’s t-test was performed to compare the percentage of time spent investigating the columns during testing. p < 0.05 was considered significant.

To specifically label L5 or L2/3 PNs, we electroporated plasmids encoding the GFP into the mouse cortex on either embryonic day (E) 13.5 or E15.5, respectively (Figures 1A–C). By adjusting the electrode position, we selectively targeted either the barrel or the motor cortex. None of the GFP+ cells in E13.5-electroporated brains co-labeled with Cux1, a marker for upper-layer neurons (Arlotta et al., 2005; Molyneaux et al., 2007), but the majority (>94%, 407 cells from three mice) of GFP+ cells in E15.5-electroporated brains did (Figure 1D). These data demonstrate our capability to target PNs of a particular cortical region in a layer-specific manner.

As most excitatory synapses reside on spines (Gray, 1959), spine density is a good indicator of a neuron’s excitatory synaptic connectivity. To compare the spine density on apical dendrites of L2/3 vs. L5 PNs, we imaged dendritic segments in L1 of the motor cortex of both electroporated mice and YFP-H line mice with transcranial in vivo two-photon microscopy. We found that the spine density of L5 PNs was comparable between E13.5-electroporated mice (0.44 ± 0.04 spines/μm) and YFP-H mice (0.47 ± 0.01 spines/μm) at P30 (p > 0.6, Supplementary Figure S1A). However, the spine density along L2/3 PNs (0.73 ± 0.04 spines/μm) was almost twice that of L5 PNs at the same age (p < 0.001, Figures 2A,C). This difference in spine density between L2/3 and L5 PNs was also observed in the barrel cortex (p < 0.001, Supplementary Figure S1B). It is worth noting that the spine density of L5 and L2/3 PNs was comparable between barrel and motor cortices (p > 0.3, Supplementary Figure S1B). Furthermore, we found that spine density of L5 PNs decreased from postnatal day (P) 14 (early adolescent) until P120 (adulthood; Figure 2C), consistent with earlier findings in the sensory cortex (Grutzendler et al., 2002; Holtmaat et al., 2005; Zuo et al., 2005a). Given the developmental pruning of spines, the higher spine density observed on L2/3 PNs at P30 could be due to a slower or delayed spine pruning or a higher spine density to start with. To distinguish between these possibilities, we compared spine densities of L2/3 PNs and L5 PNs at three other ages (P14, P60 and P120; Figure 2C). We found no difference in spine density along L2/3 PNs among these age groups (p > 0.7), and spine densities of L2/3 PNs were significantly higher than that of L5 PNs at all ages examined (p < 0.05). In summary, apical dendrites of L2/3 PNs harbor intrinsically higher spine density than L5 PNs, but unlike L5 PNs, L2/3 PNs do not show spine pruning after P14.

Time-lapse imaging has accumulated much evidence that synaptic connections are constantly formed and eliminated in the living brain, even in adulthood (Holtmaat and Svoboda, 2009; Chen et al., 2014b). To compare baseline spine dynamics on apical dendrites of L2/3 vs. L5 PNs, we followed the same dendritic segments over a 4-day interval and compared spine changes between imaging sessions at P30 and P120 (Figures 2A,B). We found that spines on L2/3 PNs are much more dynamic than spines on L5 PNs. In the motor cortex over a 4-day interval at P30, 17.5 ± 1.5% of spines were formed on L2/3 PNs, significantly higher than that of L5 PNs (5.5 ± 0.6%, p < 0.001; Figure 2D). Similarly, 18.2 ± 1.8% of spines on L2/3 PNs were eliminated over the same period, compared to 8.8 ± 0.6% on L5 PNs (p < 0.001, Figure 2E). In addition, our results revealed that L5 PNs have significantly higher spine elimination than formation (p < 0.05, Supplementary Figure S2A), consistent with the decrease in spine density during adolescent development (Holtmaat et al., 2005). In contrast, L2/3 PNs had balanced spine formation and elimination at P30 (p > 0.6), consistent with the lack of spine pruning in the adolescent brain (Supplementary Figure S2A).

We also found that spine dynamics of both L2/3 and L5 PNs slowed down in the adult brain. In the motor cortex at P120, spine formation and elimination rates of L2/3 PNs over 4 days were 9.0 ± 0.3% and 9.7 ± 0.4%, respectively, significantly lower than those measured at P30 (p < 0.01 for both, Figures 2D,E). Nevertheless, as in adolescence, these rates were still higher than corresponding ones of L5 PNs (3.6 ± 0.4% formation, 3.7 ± 0.4% elimination, p < 0.001 for both, Figures 2D,E). Importantly, L2/3 and L5 PNs had balanced spine formation and elimination at P120 (p > 0.5 for both, Supplementary Figure S2B), suggesting that spine density reaches a constant level for both PNs in adults.

The differences in baseline structural dynamics between L2/3 and L5 PNs prompted us to ask if experience-dependent spine plasticity could also differ. To do so, we trained mice to reach for single food pellets, a forelimb-specific motor-skill learning task (Xu et al., 2009), and imaged the contralateral motor cortex over a 4-day interval at P30 and P120 to determine spine dynamics changes. Consistent with earlier work (Xu et al., 2009), we found that motor-skill learning increased spine formation and elimination of L5 PNs at both P30 and P120 (Figures 3A,B). At P30, 13.4 ± 0.9% and 14.1 ± 0.9% spines were formed and eliminated, respectively, on the apical dendrites of L5 PNs in mice undergoing daily training, significantly higher than those in control mice (p < 0.05 for both, Figure 3A). In contrast, 18.7 ± 0.3% and 20.1 ± 0.8% spines were formed and eliminated, respectively, on the apical dendrites of L2/3 PNs during motor-skill learning, not significantly different from those in control mice (p > 0.2 for both, Figure 3A). We observed a similar effect in adulthood as well. While L5 PNs responded to learning with elevated formation (7.9 ± 0.6%) and elimination (9.5 ± 0.3%; p < 0.05 for both compared to controls), L2/3 PNs failed to do so (9.9 ± 0.6% formation and 10.4 ± 0.5% elimination with training, p > 0.2 for both compared to controls, Figure 3B).

Sensory experience during early postnatal life is crucial for the proper development of neuronal morphology and sensory acuity in rodents (Hubel and Wiesel, 1964; Carvell and Simons, 1996; Shoykhet et al., 2005; Lee et al., 2009; Wimmer et al., 2010; Chen C.-C. et al., 2012; Chen et al., 2015; Erzurumlu and Gaspar, 2012; Papaioannou et al., 2013). To determine if neonatal sensory deprivation alters sensory processing later in life, we bilaterally trimmed the whiskers of pups during the first postnatal week (i.e., from P0 to P7), before active whisking starts (Landers and Philip Zeigler, 2006; Erzurumlu and Gaspar, 2012). We then waited for the whiskers to grow back to full length (p > 0.3, Supplementary Figure S3A) and assessed whisker function using the whisker-dependent textural discrimination task (Wu et al., 2013) at P30. We found that control mice spent significantly more time approaching the column with the novel texture to the column with the habituated texture (p < 0.01, Supplementary Figure S3C). In contrast, trimmed mice spent equal amount of time investigating novel and habituated texture (p > 0.4, Supplementary Figure S3D). Together, trimmed mice spent a smaller fraction of time approaching the column with a novel texture compared to control (p < 0.01, Figure 4A). It is important to note that there was no significant difference in the amount of time spent investigating the columns during encoding between control and trimmed mice (p > 0.6, Supplementary Figure S3B), suggesting no defect in exploration activity. These data suggest that early sensory experience is crucial for the development of normal whisker-dependent textural discrimination ability.

As the integrity of the barrel cortex is necessary for this texture discrimination task (Chen and Zuo, 2015), we next asked if neonatal whisker trimming alters synaptic connectivity and dynamics in the barrel cortex. To do so, we imaged apical dendrites of L2/3 and L5 PNs in the barrel cortex over a 7-day interval of both control and trimmed mice at P30. We found that the spine density of both L2/3 and L5 PNs in trimmed mice were significantly lower than that of controls (Figure 4B, 0.47 ± 0.01 vs. 0.38 ± 0.02 spines/μm for L5 PNs, p < 0.001; 0.90 ± 0.04 vs. 0.69 ± 0.05 spines/μm for L2/3 PNs, p < 0.05). Following the same dendrites over time, we also found that L5 PNs in trimmed mice had significantly higher spine formation and elimination than in control mice (formation: 11.2 ± 1.0% vs. 7.2 ± 0.6%, p < 0.05; elimination: 17.0 ± 1.5% vs. 12.3 ± 0.7%, p < 0.01; Figures 4C,D). Interestingly, L2/3 PNs responded to trimming differently from L5 PNs, with trimmed mice showing lower spine formation than control mice (7.6 ± 0.9% vs. 14.6 ± 0.3%, p < 0.01), but comparable spine elimination (17.6 ± 2.0% vs. 15.1 ± 0.6%, p > 0.7; Figures 4C,D). Together, these results suggest that neonatal sensory deprivation differentially affects the spine dynamics of PNs whose cell bodies reside in different layers, suggesting layer-specific rearrangements of excitatory connectivity.