An Activity-Dependent Determinant of Synapse Elimination in the Mammalian Brain

SUMMARY

To establish functional neural circuits in the brain, synaptic connections are refined by neural activity during development, where active connections are maintained while inactive ones are eliminated. However, the molecular signals that regulate synapse refinement remain to be elucidated. When we inactivate a subset of neurons in the mouse cingulate cortex, their callosal connections are eliminated through activity-dependent competition. Using this system, we identify JAK2 tyrosine kinase as a key regulator of inactive synapse elimination. We show: JAK2 is necessary and sufficient for the elimination of inactive connections; JAK2 is activated at inactive synapses in response to signals from other active synapses; STAT1, a substrate of JAK2, mediates inactive synapse elimination; JAK2 signaling is critical for physiological refinement of synapses during normal development; and JAK2 regulates synapse refinement in multiple brain regions. We propose that JAK2 is an activity-dependent switch that serves as a determinant of inactive synapse elimination.

eTOC blurb

Yasuda et al. identified that the tyrosine kinase JAK2 serves as an “elimination signal” of inactive synaptic connections during development. They demonstrate that JAK2 is a neural activity-dependent switch, which is turned on at inactive synapses in response to “punishment signals” from other active synapses to drive inactive synapse elimination.

Graphical Abstract

INTRODUCTION

Formation of appropriate synaptic connections is critical for the proper functioning of the brain. Early in development, neurons form a surplus of immature synapses. To establish efficient, functional neural networks, neurons selectively stabilize active synapses and eliminate less active ones. This process is known as activity-dependent synapse refinement (Katz and Shatz, 1996; Sanes and Lichtman, 1999; Yasuda et al., 2011). Defects in this fundamental process of brain wiring have been implicated in many neuropsychiatric disorders, like schizophrenia and autism (Innocenti et al., 2003; Paul et al., 2007). Thus, the identification of the molecular regulators of refinement will lead to the understanding of the fundamental mechanisms underlying appropriate brain wiring and may inform novel therapeutic strategies.

Previously, we elucidated the manner through which neural activity regulates refinement in the brain (Yasuda et al., 2011): Inactive synaptic connections are eliminated only when there are other active connections to compete with. Thus, when a subset of inputs is inactive, the inactive inputs are eliminated; contrarily, when all inputs are inactive, elimination does not occur. This suggests that active connections send a “punishment” signal to inactive ones and instruct them to leave by triggering “elimination” signals within the inactive synapses. However, the molecular identities of the “elimination” signals, which detect levels of synaptic activity and determine if a synapse should be eliminated during refinement, remain to be elucidated.

At active synapses, connections are kept and strengthened by the presence of “stabilization” signals. Several molecules, including SIRPα/CD47, C1ql1/Bai3, and Fn14, have been implicated in active synapse strengthening and maintenance (Umemori and Sanes, 2008; Toth et al., 2013; Nagappan-Chettiar et al., 2018; Kakegawa et al., 2015; Cheadle et al., 2018, 2020). In contrast, far less is known about the mechanisms driving inactive synapse elimination. Some studies have suggested that MHC/PirB regulates synapse elimination by mediating signaling within the postsynaptic neuron (Datwani et al., 2009; Lee et al., 2014). Other studies have implicated complement proteins as executors of synapse elimination, where they recruit microglia to engulf synapses tagged for elimination (Stevens et al., 2007; Schafer et al., 2012). However, the identity of the “elimination” signal, the decision-making molecule within the presynaptic neuron that detects levels of synaptic activity and determines whether the synapse should be eliminated, is unknown.

Here, we established a new in vivo system to study activity-dependent synapse refinement. Using this system, we identify JAK2, a protein tyrosine kinase implicated in cytokine signaling, as an activity-dependent determinant of synapse elimination. We show that: 1) inactive connections are not eliminated when JAK2 activity is suppressed; 2) STAT1 is required for the elimination of inactive connections; 3) JAK2 becomes activated at inactive synapses in response to signals from other active connections; 4) JAK2 activation induces synapse elimination; 5) JAK2 signaling is critical for physiological synapse refinement; and 6) JAK2 is required for synapse refinement in multiple brain regions. Together, our findings reveal a novel activity-dependent molecular determinant of synapse elimination that can be switched on at inactive synapses to drive their elimination and shape functional neural networks.

RESULTS

Activity-dependent competition drives elimination of inactive callosal axons.

We first developed an in vivo system to study activity-dependent synapse refinement (Figures 1 and S1). We focused on the refinement of callosal connections between the cerebral hemispheres. The corpus callosum is the largest axon tract in the brain, originating from cortical callosal neurons. It transfers motor, sensory, and cognitive information between the two cerebral hemispheres (Fame et al., 2011). Neural activity plays important roles in the development of callosal axons, affecting both axon refinement (Grigonis and Murphy, 1994) and layer targeting (Mizuno et al., 2007). We suppressed neural activity, in the form of synaptic transmission, in a subset of callosal neurons by introducing tetanus toxin light chain (TTLC), which inhibits neurotransmitter release by cleaving the SNARE protein, VAMP2 (Yasuda et al., 2011; Yu et al., 2004). We co-expressed EGFP to label TTLC-expressing neurons (Figure 1A). As a control, we introduced EGFP alone. We used in utero electroporation (Saito and Nakatsuji, 2001; Saito, 2006) to introduce TTLC and/or EGFP into left cingulate cortical neurons: co-electroporation efficiency was 98.7 ± 1.6% (Figures S1A and S1B). Electroporated cells were neurons and not astrocytes or microglia (Figure S1C). Of these neurons, 88.9 ± 4.6% were SATB2-positive callosal neurons (Figures S1D and S1E). We set the condition of electroporation so that ~20% of layer V callosal neurons in the cingulate cortex were electroporated (Figures S1D and S1E). Thus, when TTLC is electroporated, there can be competition between active (~80%) and inactive (~20%) callosal neurons.

Figure 1. Activity-dependent competition eliminates inactive callosal axons.

(A) In utero electroporation of EGFP into the left cingulate cortex (P10).

(B) Schematic of experiments. Electroporation of plasmids into the left cingulate cortex followed by analysis. Images were taken from the dotted boxed area.

(C and D) Coronal sections of EGFP-alone (C) and EGFP+TTLC (D) electroporated brains. The bottom panels show higher magnification images taken from the boxed areas in the top panels. Scale bars, 200 μm (top), 50 μm (bottom).

(E) Quantification of callosal axon densities in Control (gray line) and TTLC-expressing brains (red line). n = Control: P0, 10 mice; P5, 5; P10, 6; P15, 13; P30, 6. TTLC: P0, 12; P5, 9; P10, 7; P15, 16.

(F) The ratios between the axon density of TTLC-expressing axons and that of Control axons. At P15, relative to Control axons, only 21% of TTLC-expressing axons remain. n = 9 mice, P5; 7, P10; 16, P15.

(G) Coronal section of a TTLC electroporated brain (P30). Scale bar, 500 μm [dummy_in cmp]

(H) Schematic of TTX application experiments. Electroporation at E13.5, Elvax-TTX implantation at P3 on the cingulate cortex, and analysis at P15.

(I–L) Callosal axons in Control (I), TTLC without TTX (J), TTLC with TTX/Elvax (K), and TTLC with Elvax only (L) sections at P15. Scale bars, 200 μm (top), 50 μm (bottom).

(M) Quantification of axon densities, relative to Control. n = 14 mice, Control; 16, TTLC; 10, TTLC+TTX/Elvax; 12, TTLC+Elvax.

Mean ± SEM. **P < 0.01, ****P < 0.0001, n.s.: not significant, two-way ANOVA with Sidak’s test (E), ANOVA with Tukey test (F, M).

See also Figure S1

Before probing the role of activity in callosal refinement, we first examined the time course of callosal axon development. For this, we electroporated EGFP to label cortical callosal neurons in the left hemisphere at embryonic day (E) 13.5 and examined the axon density in the right hemisphere at postnatal day (P) 0 to P30 (Figures 1B and 1C). Callosal connections between the cingulate cortices are mostly homotopic, and thus, axon projections and targeting can be readily observed in the same section. Callosal axon density increased from P0 to P5 indicating a period of initial axon targeting to the contralateral cortex (see Figures S1F and S1G for calculation methods). The axon density then decreased from P5 to P15. Afterwards, axon density was maintained (Figure 1E; Gray line). These results show that callosal axons are physiologically refined between P5 and P15.

Then, to examine whether callosal axon refinement is neural activity-dependent, we co-electroporated TTLC and EGFP (Figures 1B and 1D). As in control neurons (Figures 1C and 1E, Gray line), the density of callosal axons from TTLC-electroporated neurons increased from P0 to P5 (Figure 1E; Red line), indicating that TTLC did not affect initial axon targeting (at P5, the ratio between the density of TTLC-expressing axons and that of control axons was 104 ± 10%; Figure 1F). However, the axon density from the TTLC-expressing neurons decreased significantly relative to controls from P10 to P15 (the ratios were 76 ± 7% at P10 and 21 ± 2% at P15; Figure 1F), indicating that inactive callosal axons are eliminated after they project to their target region.

Layer V callosal neurons also send axon collaterals to the ipsilateral striatum (Cowan and Wilson, 1994; Kim et al., 2015). In our system, electroporated callosal neurons also had corticostriatal projections (Figures S1H and S1I). Interestingly, we found that, while inactive callosal axons were eliminated by P15 (Figure 1D), inactive corticostriatal axons were not eliminated and remained into adulthood (Figure 1G), indicating that inactive callosal axons are selectively eliminated. This also implies that callosal axon elimination is not due to neuronal death, which is further supported by our finding that TTLC-expressing neurons were not TUNEL-positive (Figure S1J) (Le et al., 2002).

In the experiments described above, both active and inactive callosal neurons were present – so competition can occur. We next examined whether active callosal connections are necessary for inactive callosal axon elimination. For this, after TTLC electroporation, we globally suppressed the neural activity of callosal connections by applying tetrodotoxin (TTX; Figure 1H), which blocks voltage-gated sodium channels to prevent neuronal firing, and as a result, action potential-dependent synaptic transmission. TTX containing-Elvax was implanted on the left cortex at P3, and axon density was examined at P15 (Echegoyen et al., 2007; Yasuda et al., 2011). We found that TTX treatment inhibited the elimination of TTLC-expressing callosal axons (Figures 1K–1M), suggesting that active connections are necessary for the elimination of inactive connections.

While TTX application globally suppresses neural activity by silencing action potential-dependent neurotransmission, it also suppresses action potentials within TTLC-expressing neurons. To exclude the possibility that action potentials are needed in TTLC-expressing neurons for inactive axon elimination, we expressed the inhibitory DREADD (hM4Di; Armbruster et al., 2007) in a subset (~20%) of callosal neurons to suppress neuronal excitability (and subsequently, action potentials and action potential-dependent synaptic transmission) in the presence of clozapine-N-oxide (CNO; Thompson et al., 2016). At P10, there was no significant difference in the axon density between the CNO treated and untreated brains, suggesting that initial axon growth is not affected by the suppression of neuronal excitability (Figures S1K and S1L). At P15, we found that, like TTLC-expressing axons, the callosal axons from neurons expressing hM4Di (+ CNO) were significantly decreased relative to control (− CNO) axons (Figure S1M). These findings suggest that action potentials are not necessary for axon elimination, and that it is the suppression of synaptic transmission that results in axon elimination. Together, our results suggest that synaptic transmission-dependent competition regulates axon refinement and support the notion that active connections send signals to drive the elimination of inactive ones.

Identification of JAK2 as a critical regulator of inactive axon elimination

Utilizing this system, we performed a screen to identify signaling molecules in presynaptic neurons that are critical for the elimination of inactive connections. We focused on non-receptor protein tyrosine kinases (PTKs), because they regulate various aspects of neural network development, including axon growth, neurite differentiation, and axon regeneration (Maness, 1992; Liu et al., 2011). We generated kinase-dead mutants of PTKs, which serve as dominant-negative (DN) forms of the following PTKs (Figure S2): JAK2DN (Ungureanu et al., 2002), JAK1DN (Shimoda et al., 1997), FynDN (Tezuka et al., 1999), and CSKDN (not shown; Howell and Cooper, 1994), and co-expressed them with TTLC in a subset of left cortical callosal neurons. None of the mutants affected the initial targeting of callosal axons to the right hemisphere (P10; Figures 2A–2C and 2E). As described above, TTLC-expressing callosal axons were eliminated by P15 (Figures 1D–1F). However, when JAK2DN was co-expressed, TTLC-expressing axons were not eliminated (Figures 2F and 2J). Co-expression of JAK1DN, FynDN, or CSKDN did not inhibit the elimination of inactive axons (Figures 2G, 2H and 2J). These results indicate that the activity of JAK2, and not the other PTKs, is necessary for the elimination of inactive callosal axons.

Figure 2. The JAK2/STAT1 signaling pathway mediates inactive axon elimination.

(A–J) Dominant-negative (DN) forms of JAK2, JAK1 or Fyn, or wild-type SOCS3 were co-electroporated with TTLC, and callosal axons were examined at P10 (A–D) and P15 (F–I). Scale bars, 200 μm (top), 50 μm (bottom). (E and J) Quantification of axon densities at P10 (n = 6 mice, Control; 7, TTLC; 8, JAK2DN; 9, JAK1DN; 6, FynDN; 5, SOCS3) and P15 (n = 14, Control; 16, TTLC; 6, JAK2DN; 12, JAK1DN; 13 FynDN; 9, SOCS3) (Gray line: Control; Red line: TTLC).

(K) SOCS3 suppresses JAK2. JAK2 activates STATs.

(L–N) STAT1DN or STAT3DN was co-electroporated with TTLC, and callosal axons were examined at P15 (L and M). Scale bars, 200 μm (top), 50 μm (bottom). (N) Quantification of axon densities at P15. n = 5 mice, STAT1DN; 3, STAT3DN.

Mean ± SEM (normalized to Control). ****P < 0.0001, n.s.: not significant, ANOVA with Tukey test

See also Figure S2.

Overexpression of JAK2DN may not only suppress JAK2 kinase activity, but also disrupt the scaffolding function of JAK2 (Keil et al., 2014). To confirm that the inhibition of inactive axon elimination is attributed to the suppression of JAK2 kinase activity, we suppressed endogenous JAK2 activity by overexpressing a negative regulator of JAK2, SOCS3 (suppressor of cytokine signaling 3; Figure 2K). SOCS3 is expressed in developing neurons and suppresses JAK2 kinase activity by binding to JAK2 (Sasaki et al., 1999; Polizzotto et al., 2000; Croker et al., 2003). SOCS3 did not affect initial axon targeting (P10, Figures 2D and 2E); however, at P15, TTLC-expressing axons were still maintained when SOCS3 was co-expressed (Figures 2I and 2J). SOCS3 can also regulate JAK1 (Babon et al., 2012), but, since JAK1DN did not affect inactive callosal axon elimination (Figures 2G and 2J), SOCS3’s effects are likely through the suppression of JAK2 activity. Altogether, these results demonstrate that JAK2 plays a critical role in the elimination of inactive axons.

STAT1 mediates inactive axon elimination

We next asked what signals downstream of JAK2 mediate the elimination of inactive connections. In immune cells, activated JAK2 phosphorylates STATs (Signal Transducers and Activators of Transcription), which regulate transcription (Shuai and Liu, 2003) (Figure 2K). Hence, we tested whether STATs mediate inactive axon elimination. We co-expressed DN mutants of two STAT family members, STAT1 and STAT3, that are highly expressed in the developing brain (De-Fraja et al., 1998). We found that co-expression of STAT1DN (Walter et al., 1997), but not STAT3DN (Kaptein et al., 1996), suppressed TTLC-dependent elimination of callosal axons (Figures 2L–2N). These results suggest that STAT1, but not STAT3, mediates inactive axon elimination.

JAK2 is activated in inactive neurons only in the presence of other active neurons

If JAK2 activity is necessary for eliminating inactive callosal axons, JAK2 should be activated in inactive neurons during refinement. We examined this possibility using an antibody specific to active JAK2 (phospho-JAK2). Activated JAK2 is phosphorylated on tyrosine residues in the activation loop (Y1007/1008; Feng et al., 1997). We found that JAK2 was not apparently phosphorylated in EGFP-only-expressing cortical neurons at P10 (Figure 3A). In contrast, JAK2 was strongly phosphorylated in TTLC-expressing neurons (Figures 3B and 3D). Co-expression of JAK2DN or SOCS3, which prevented TTLC-induced axon elimination (Figures 2F, 2I and 2J), suppressed TTLC-induced phosphorylation of JAK2 (Figure S3). These results suggest that JAK2 is activated in inactive neurons, and that JAK2 activity correlates with callosal axon elimination.

Figure 3. JAK2 is activated in inactive neurons in response to activity from other active neurons.

(A–C) Immunostaining for active JAK2 (Phospho-JAK2) in Control (EGFP-alone; A), TTLC (B), or TTLC with TTX (C) expressing brains at P10. Images were taken from the electroporated hemisphere. Scale bar, 10 μm.

(D) Quantification of phospho-JAK2 staining intensity relative to that in Control. n (neurons, mice) = 22, 4, Control; 21, 4, TTLC; 26, 4, TTLC+TTX.

(E and F) Immunostaining for phospho-JAK2 in TTLC and hM4Di co-electroporated brains at P10 without (E) or with (F) CNO.

(G) Quantification of phospho-JAK2 staining intensity relative to that in Control (−CNO). n (neurons, mice) = 30, 3, Control (−CNO); 30, 3, hM4Di (−CNO); 30, 3, Control (+CNO); 30, 3, hM4Di (+CNO).

Mean ± SEM. ****P < 0.0001, n.s.: not significant, ANOVA with Tukey test.

See also Figure S3.

We next asked whether JAK2 is activated in response to signals from active neurons. We examined whether JAK2 is still active in TTLC-expressing neurons when TTX was applied to globally suppress the activity of all callosal neurons (Figures 3C and 3D). We found that TTX application blocked the phosphorylation of JAK2 in inactive neurons, suggesting that signals from active neurons induce the activation of JAK2 in inactive neurons. As described above, TTX suppresses action potentials in all neurons, including TTLC-expressing neurons. Thus, there is an alternative possibility that TTLC-expressing neurons require action potentials in a cell-autonomous manner to activate JAK2. To exclude this possibility, we co-expressed hM4Di and TTLC in callosal neurons to suppress the excitability (and accordingly, action potentials) of TTLC-expressing neurons without disrupting the activity of other neurons. We found that JAK2 is still activated in TTLC-expressing neurons even when the intrinsic excitability of these neurons is suppressed (Figures 3E–3G), indicating that TTLC-expressing neurons do not require action potentials to activate JAK2. These results support the notion that JAK2 serves as a sensor of signals from other active neurons during the refinement of callosal projections.

JAK2 activation is sufficient to drive axon elimination

We have shown that JAK2 activity is necessary for inactive callosal axon elimination. We next asked whether JAK2 activation is sufficient to induce the elimination of active callosal axons. For this, we introduced wild-type JAK2 (JAK2WT) or a constitutively active form of JAK2 (JAK2CA) (Figures S4A–S4D, S4G) into callosal neurons (James et al., 2005). As a comparison, we introduced JAK1WT. Overexpression of JAK2WT, JAK2CA, or JAK1WT did not apparently affect initial callosal axon targeting (P10, Figures 4A–4D, 4K). At P15, JAK1WT-expressing callosal axons were still maintained (Figures 4I and 4L). However, the majority of JAK2WT- or JAK2CA-expressing callosal axons were eliminated (Figures 4G, 4H and 4L), suggesting that JAK2 activation is enough to drive callosal axon elimination during the synapse refinement period.

Figure 4. Activation of JAK2 is sufficient to drive developmental callosal axon elimination.

(A–J) JAK2WT, JAK2CA, JAK1WT, or SOCS3DN were electroporated into callosal neurons and their axons were examined at P10 (A–E) and P15 (F–J). Scale bars, 200 μm (top), 50 μm (bottom).

(K and L) Quantification of axon densities at P10 (K) and P15 (L) (Gray line: Control; Red line: TTLC). n = P10: 8 mice, JAK2WT; 6, JAK2CA; 6, JAK1WT; 5, SOCS3DN. P15: 10, JAK2WT; 10, JAK2CA; 15, JAK1WT; 5, SOCS3DN. Mean ± SEM. ****P < 0.0001, n.s.: not significant, ANOVA with Tukey test.

See also Figure S4.

Next, to examine whether the activation of endogenous JAK2 kinase is sufficient to induce axon elimination, we introduced a SOCS3 DN mutant that lacks the kinase inhibitory region (Figure S4E). SOCS3DN still binds to JAK2 but lacks the ability to inhibit JAK2’s kinase activity; thus, it prevents endogenous SOCS3 from binding to JAK2, leading to the sustained activation of endogenous JAK2 kinase (Sasaki et al., 1999). Phospho-JAK2 immunostaining confirmed that the expression of SOCS3DN increased endogenous JAK2 activation (Figures S4F and S4G). Expression of SOCS3DN did not affect the targeting of callosal axons (P10, Figures 4E and 4K). However, at P15, expression of SOCS3DN resulted in the elimination of callosal axons (Figures 4J and 4L). Neither expression of SOCS3DN nor JAK2 induced neuronal death (Figure S1J). These results demonstrate that the activation of JAK2 is sufficient to drive callosal axon elimination in the developing brain.

JAK2 activity is required for the physiological refinement of callosal connections

We have shown that JAK2 activity is necessary in axons inactivated by TTLC to drive axon elimination, and that the activation of JAK2 is sufficient to induce axon elimination. We next asked whether JAK2 is indeed critical for physiological callosal axon refinement. Natural refinement of callosal axons occurs between P5 and P15 (Figures 1C and 1E). We first examined whether this natural callosal axon refinement is neural activity-dependent. For this, we globally suppressed neural activity with TTX from P3. TTX treatment almost completely inhibited callosal axon refinement (Figures 5A–5C), indicating that physiological callosal axon refinement is activity-dependent.

Figure 5. JAK2 is required for physiological axon refinement.

(A–C) Physiological callosal axon elimination is activity-dependent. Representative images of callosal axons at P15 without (A) or with (B) TTX. TTX was applied from P3 to P15. (C) Quantification of axon densities (relative to P5 Control without TTX). n = 9 mice.

(D–G) JAK2 is necessary for physiological axon elimination. (D) Representative images of Control and JAK2DN-expressing callosal axons at P5 and P15. Scale bars, 200 μm (top), 50 μm (bottom). (E) Quantification of axon density (relative to that of Control axons at P5). n = 14 mice, Control; 13, JAK2DN. (F) Representative images of Control (expressing the empty shRNA vector) and JAK2 shRNA-expressing (sh#1 or sh#2) callosal axons at P15. Scale bars, 200 μm (top), 50 μm (bottom). (G) Quantification of axon density (relative to Control). n = 6 images from 3 mice.

Mean ± SEM. *P < 0.05, **P < 0.01 by Mann-Whitney test (C, E), ANOVA with Tukey’s test (G).

Having established that natural axon elimination is activity-dependent, we next investigated the role of JAK2 in this process. We first suppressed JAK2 kinase activity by expressing JAK2DN in callosal neurons. Expression of JAK2DN did not affect initial axon projection (P5, Figures 5D and 5E). However, JAK2DN expression significantly suppressed the elimination of axons at P15, indicating that JAK2 activity is necessary for natural callosal axon refinement. We next knocked down JAK2 in callosal neurons by electroporating two different shRNAs against Jak2 (Zhang et al., 2013). We found that shRNA-mediated JAK2 knockdown increased axon density at P15 (after axon refinement) relative to controls (expressing the empty shRNA vector) (Figures 5F and 5G). Altogether, these results indicate that JAK2 is necessary for physiological callosal axon refinement.

JAK2 is necessary for synapse elimination

Thus far, we have examined the activity-dependent elimination of axons. Axon elimination is generally regarded as a result of activity-dependent synapse elimination (Liu et al., 2005; Yu et al., 2013). Hence, we next examined whether callosal synapses are developmentally refined, and if so, whether the refinement requires JAK2. For this, we introduced Synaptophysin-mCherry, which labels presynaptic terminals, and EGFP into left cortical neurons with or without JAK2DN (Figure 6A). We then quantified the number, size, and intensity of Synaptophysin-mCherry puncta in EGFP-labeled axons in the right hemisphere at P5 and P10 (Figures 6B–6F). In the control, the number of Synaptophysin-mCherry puncta decreased from P5 to P10 (Figures 6B and 6D), indicating that callosal synapses are innately refined between P5 and P10. The intensity of Synaptophysin-mCherry puncta increased from P5 to P10 (Figure 6F), suggesting strengthening of remaining synapses. Expression of JAK2DN significantly increased the number of Synaptophysin-mCherry puncta at P10, but not at P5, relative to control (Figures 6C and 6D). The size or intensity of Synaptophysin-mCherry puncta was not affected by JAK2DN expression (Figures 6E and 6F). These results suggest that inactivation of JAK2 prevents synapse elimination, but does not affect strengthening of the remaining synapses.

Figure 6. JAK2 regulates callosal synapse elimination and is activated at inactive synapses.

(A) Schematic of experiments. Expression plasmids were electroporated into the left cortex and imaging was done in the right.

(B and C) Confocal images from EGFP/Synaptophysin (Sphy)-mCherry (B) and JAK2DN/EGFP/Sphy-mCherry (C) electroporated brains at P5 and P10. Scale bar, 5 μm.

(D–F) Quantification of the Sphy-mCherry puncta density (relative to P5 Control; D), size (E) and intensity (F). n (images, mice) = 20, 4, P5 Control; 20, 4, P10 Control; 20, 4, P5 JAK2DN; 20, 4, P10 JAK2DN.

(G–J) Activation of JAK2 signaling results in fewer callosal synapses. (G) Schematic of experiments. AAVs were injected into the left cortex of Control (SOCS3^fl/fl) and SOCS3 KO (Emx1Cre-SOCS3^fl/fl) mice at P5. Imaging was done in layer V of the right cortex at P19 (boxed area). (H) Representative images. (I and J) Quantification of the density and size of Sphy-mCherry puncta. n (images, mice) = 10, 3, Control; 8, 3, SOCS3 KO. Scale bar, 5 μm.

(K–M) Inverse correlation between synaptic activity and JAK2 activation. Cortical cultures were subjected to the Synaptotagmin-1 (Syt-1) antibody re-uptake assay and immunostained for Phospho-JAK2 and VGLUT1. (K) Syt-1 is taken up at and labels active presynaptic terminals. (L) Representative images. (M) Intensities of Syt-1 and Phospho-JAK2 at VGLUT1-positive puncta. Reproduced in 3 independent experiments. P < 0.0001, Spearman correlation.

(N–P) Phosphorylation of JAK2 at synapses by activity-dependent competition. Cortical sections from P5 Control (N) and TTX-treated (O) brains were stained for VGLUT1 and Phospho-JAK2. n = 6 sections from 3 mice.

(Q–S) Activation of callosal (CC), but not corticostriatal (CS), axons. (Q) Schematic of experiments. AAV-EGFP was injected into the left cortex and imaging was done in the right cortex (CC) and left striatum (CS). (R) Confocal images at P6. (S) Quantification of Phospho-JAK2 puncta density on EGFP positive axons (% relative to that in CC). n = 6 sections from 3 mice. Scale bar, 5 μm.

Mean ± SEM. **P < 0.01, Mann-Whitney test.

See also Figure S5.

Developmental synapse refinement occurs between P5 and P10 in CD1 mice (synapse density decreases between P5 and P10, but not between P10 and P15; Figures 6D and S5A). Therefore, the completion of synapse refinement (by P10) precedes that of axon refinement (by P15 in CD1 mice; Figure 1E), which might suggest that axon refinement is a consequence of synapse refinement.

We next examined whether JAK2 activation can drive synapse elimination. For this, we examined callosal synapse elimination in SOCS3 conditional knockout (KO) mice. To inactivate SOCS3 (and consequently activate JAK2) in cortical neurons, we crossed the SOCS3^fl/fl mice to Emx1-Cre mice to generate SOCS3 KO mice. We introduced Synaptophysin-mCherry and EGFP into left cingulate cortical neurons (Figure 6G). We then evaluated Synaptophysin-mCherry puncta in EGFP-labeled axons in the right hemisphere after synapse refinement is complete. We found a significant decrease in the density and size of Synaptophysin-mCherry puncta in SOCS3 KO mice relative to control (Figures 6H–J), indicating that the loss of SOCS3 results in fewer and smaller callosal synapses. These results support the notion that JAK2 kinase is a critical regulator of structural synapse elimination.

JAK2 is activated at inactive synapses in response to signals from other active synapses

For JAK2 to serve as an “elimination” signal of inactive synapses, JAK2 should be more active at inactive synapses than at active synapses. To distinguish between active and inactive synapses, we used a Synaptotagmin (Syt-1) re-uptake assay. We applied fluorescently labelled-Syt-1 antibody to cultured cortical neurons. This Syt-1 antibody was taken up into presynaptic terminals as these terminals undergo neurotransmitter vesicle release and endocytosis (Figure 6K). Thus, synapses that are more active would have a higher intensity of Syt-1 fluorescence. We then immunostained the cultures for phospho-JAK2 and VGLUT1 (to label excitatory presynaptic terminals). We found that more active synapses have less JAK2 activation, and that synapses with high JAK2 activation have low levels of synaptic activity (Figures 6L and 6M). These results suggest that JAK2 is activated at less active synapses under physiological conditions.

We then examined whether JAK2 is localized at synapses in vivo, and if so, whether its activity at synapses is regulated by neural activity-dependent competition. Immunostaining of cortical sections revealed that phospho-JAK2 was localized at synapses (Figure 6N). Application of TTX to the brain diminished phospho-JAK2 staining at synapses (Figures 6O and 6P), suggesting that JAK2 is activated at synapses only when there are other active synapses in vivo.

JAK2 is activated at callosal, but not at corticostriatal projections

While inactive callosal projections are eliminated during development, inactive corticostriatal projections are not (Figure 1G). Therefore, we reasoned that as an “elimination” signal, JAK2 should be activated in callosal, but not in corticostriatal projections. To test this idea, we introduced EGFP into cortical neurons to label callosal and corticostriatal axons and evaluated the activation of JAK2 (phospho-JAK2) on EGFP-positive projections at the beginning of callosal synapse elimination (Figure 6Q). We found that JAK2 is activated in callosal axons but not in corticostriatal axons (Figures 6R and 6S). Phospho-JAK2 signals were also detected in ipsilateral corticocortical collaterals (Figure S5C), which appear to undergo JAK2-dependent synapse elimination (in SOCS3 KO mice, synapse size was decreased in ipsilateral synapses relative to controls after synapse refinement; not shown). These results suggest that projection-specific activation of JAK2 contributes to the selective elimination of the projections.

JAK2 signaling is required for functional synaptic refinement

To evaluate the role of JAK2 in functional synapse refinement, we performed electrophysiological experiments. To evaluate the functional connectivity of callosal synapses in which JAK2 activity is manipulated, we used an optogenetic approach. We electroporated Cre, as a control, into the left cingulate cortex of Ai32 mice (STOP-ChR2-EYFP mice; Madisen et al., 2010) to express ChR2 in ~ 20% of layer V pyramidal neurons (Figure 7A). To test the function of JAK2, we expressed Cre and SOCS3DN (to activate JAK2; Figures S4E–S4G). At P3–5 and P9–11, we recorded optically evoked synaptic transmission in the right hemisphere from layer V pyramidal neurons, and the percentage of cells that responded to light (innervation rate) was calculated (Figures 7B and 7C). The innervation rate decreased from P3–5 to P9–11 in the control, suggesting that callosal synapses are physiologically refined between P3 and P11. Expression of SOCS3DN did not affect initial callosal innervation (Figure 7B). However, at P9–11, JAK2 activation significantly reduced the innervation rate relative to controls, indicating that JAK2 activation drives a loss of callosal innervation (Figure 7C).

Figure 7. JAK2/STAT1 signaling regulates functional synaptic refinement.

(A–C) Optogenetic analysis of functional callosal synapse refinement. (A) Schematic of experiments. Cre or Cre+SOCS3DN were electroporated into the left cingulate cortex of Ai32 mice, and optically evoked synaptic responses were recorded in the right cortex. (B and C) The percentage of cells that responded to light (innervation rate). P = n.s., P3–5; 0.033, P9–12, Chi square test. n (cells from 4–8 mice) = 37, P3–5 Control; 76, P3–5 SOCS3DN; 143, P9–11 Control; 62, P9–11 SOCS3DN.

(D–G) The callosal fiber fraction (FF) method. (D) Image of a mouse cortex, indicating the positions of recording and stimulating electrodes. (E and F) Representative traces and quantification of single fiber (E) and maximum (F) responses at different developmental ages. (G) Quantification of the FF ratio (single fiber amplitude / maximal amplitude). n = 13 cells, P5–6; 48, P8–10; 42, P15–16.

(H–N) FF analysis with SOCS3 KO^Pre mice. (H) Schematic of experiments. AAV-EGFP (Control) or AAV-EGFP + AAV-Cre (SOCS3 KO^Pre) were injected into the left cortex of SOCS3^fl/fl mice. The electrically evoked callosal synaptic responses were recorded in the right cortex at P8–10 and P16–19. (I and L) Single fiber responses. (J and M) Maximal responses. (K and N) Quantification of the FF ratio. n (cells from 5–8 mice) = 66, P8–10 Control; 40, P8–10 SOCS3 KO^Pre; 55, P16–19 Control; 40, P16–19 SOCS3 KO^Pre.

(O–U) FF analysis with JAK2 KO^Pre mice. (O) Schematic of experiments. AAV-EGFP + AAV-Cre were injected into the left cortex of WT (Control) or JAK2^fl/fl (JAK2 KO^Pre) mice. The callosal synaptic responses were recorded in the right cortex at P10–11 and P16–17. (P and S) Single fiber responses. (Q and T) Maximal responses. (R and U) Quantification of the FF ratio. n (cells from 3 mice) = 19, P10–11 Control; 22, P10–11 JAK2 KO^Pre; 28, P16–17 Control; 22, P16–17 JAK2 KO^Pre.

(V–AA) FF analysis with STAT1 KO mice. The synaptic responses were recorded in STAT1^+/+ (Control) and STAT1^−/− (STAT1 KO) mice at P10–11 and P16. (V and Y) Single fiber responses. (W and Z) Maximal responses. (X and AA) Quantification of the FF ratio. n (cells from 3–4 mice) = 23, P10–11 Control; 23, P10–11 STAT1 KO; 29, P16 Control; 45, P16 STAT1 KO.

Mean ± SEM. ^#P = 0.0768, *P < 0.05, **P < 0.01, Kruskal-Wallis with Dunn’s test (E–G), Mann-Whitney test (H–AA).

See also Figures S6 and S7.

To physiologically assess how the number of functional callosal inputs converging onto a target neuron changes through development, we developed the callosal fiber fraction (FF) method. For this, we recorded EPSCs (excitatory postsynaptic currents) from layer V pyramidal neurons while electrically stimulating the callosal fiber bundle (Figure 7D). We recorded two types of responses: a single fiber response obtained by stimulating the callosal bundle at a minimal intensity, at which more than 50% of trials failed to induce a response (Figure S6A), and a maximum response, obtained by increasing the stimulus intensity until the response amplitude reaches a plateau. The response latency was constant across all stimulation strengths, suggesting that the recorded responses were monosynaptic (Figure S6B). The FF ratio, calculated as the ratio of the single fiber response to the maximum response (single fiber/maximum), is a measure of the contribution of a single callosal input to the total callosal input received by a cell (Hooks and Chen, 2006, 2008; Noutel et al. 2011; Narushima et al. 2016). The inverse of the FF value provides a rough estimate of the callosal input number received by a cell. In C57BL/6 mice, the single fiber response increased from P8–10 to P15–16 (Figure 7E), indicating callosal synaptic strengthening. The maximum response increased significantly between P5–6 and P8–10 and then plateaued between P8–10 and P15–16 (Figure 7F). The calculated FF ratio decreased between P5–6 and P8–10 (Figure 7G), indicating an increase in functional callosal synaptic inputs. The FF then increased between P8–10 and P15–16 (Figure 7G), suggesting that callosal inputs undergo synapse elimination during this period. The timing of synapse elimination is strain dependent: in CD1 mice, which were used for the histological and optogenetic experiments (Figures 6A–6F, 7A–7C, and S5A), the FF ratio increased between P6–7 and P10–11 (Figure S6E), suggesting that synapse elimination occurs earlier in CD1 than in C57BL/6 mice. We further confirmed this strain difference using Synaptophysin-mCherry (Figures S5A and S5B).

Having established the FF method, we next evaluated callosal synapse elimination in SOCS3 conditional KO mice. To inactivate SOCS3 in presynaptic neurons (and consequently activate JAK2), we injected AAV-Cre and AAV-EGFP into the left cortex of SOCS3^fl/fl mice (SOCS3 KO^Pre; Figures 7H and S7A). SOCS3 inactivation did not induce cell death (Figures S7B–S7C). At P10–11, there were no differences between controls and SOCS3 KO^Pre mice (Figures 7I–7K), suggesting that JAK2 signaling is not critical to callosal synapse formation. In contrast, at P16–19, the FF ratio was significantly larger in SOCS3 KO^Pre mice relative to controls (Figure 7N), indicating that the loss of SOCS3, and therefore the activation of JAK2, drives greater functional synapse elimination.

We next evaluated whether JAK2 is necessary for functional synapse elimination using JAK2 conditional KO mice (JAK2^fl/fl; Krempler et al., 2004). To knockout JAK2 from presynaptic neurons, we injected AAV-Cre and AAV-EGFP, into the left cortex of JAK2^fl/fl mice (JAK2 KO^Pre; Figure 7O). JAK2 was not critical to initial callosal synapse formation (P10–11; Figures 7P–7R). However, at P16–17, the FF ratio was significantly smaller in JAK2 KO^Pre mice relative to controls (Figure 7U), suggesting that the loss of JAK2 inhibits functional synapse elimination. We did not observe significant differences in the single fiber amplitude, which is a readout of postsynaptic strength, between control and SOCS3 KO or JAK2 KO mice, suggesting that presynaptic JAK2 signaling does not affect postsynaptic properties.

We then examined whether STAT1, which mediates inactive callosal axon elimination (Figures 2L and 2N), is necessary for functional synapse elimination. We performed the FF analysis in STAT1 KO mice (Durbin et al., 1996). STAT1 was not critical to callosal synapse formation (P10–11; Figures 7V–7X). However, at P16, the FF ratio was significantly smaller in STAT1 KO mice relative to controls (Figure 7AA), indicating that the loss of STAT1 suppresses functional synapse elimination.

Note that while statistically not significant, we see consistent trends in the amplitude of the maximal responses: after synapse refinement, SOCS3 KO^Pre mice have lower, while JAK2 KO^Pre and STAT1 KO mice have higher, maximal responses relative to control. Single fiber responses were the same between KO and control. As a result, the FF ratios were higher in SOCS3 KO^Pre and lower in JAK2 KO^Pre and STAT1 KO mice relative to control. The lack of significance in the changes in maximal responses can be attributed to the variability of the responses. Altogether, our electrophysiological experiments are consistent with the notion that JAK2–STAT1 signaling mediates functional synapse elimination.

JAK2 also regulates eye-specific segregation

Finally, we examined whether JAK2 regulates the refinement of other connections in the brain. To this end, we evaluated the role of JAK2 in eye-specific segregation in the dorsal lateral geniculate nucleus (dLGN), where initially intermingled left and right eye inputs to the dLGN segregate in an activity-dependent manner during development (Godement et al., 1984; Sretavan and Shatz, 1986). To suppress JAK2 kinase activity in retinal ganglion cells (RGCs) by the time eye-specific segregation starts (~P3; Jaubert-Miazza et al., 2005), we established a method to inject AAV in utero into embryonic eyes. We injected AAVs encoding JAK2DN and EGFP to the left eye of embryos at E13.5 (Figure 8A). EGFP expression was clearly observed at P0 and uniform throughout the retina (Figures S8A–S8C). After in utero AAV injection, mice were injected with CF594-conjugated cholera toxin subunit B (CTB 594) into the right eye at P14 and analyzed at P15 (Figure 8A).

Figure 8. JAK2 regulates eye-specific segregation.

(A–I) JAK2 inactivation suppresses eye-specific segregation. (A) Schematic diagram of RGC projections to the dLGN, labeled with AAV (EGFP; green) and CTB (CF594; pink). AAVs were injected into the left eye at E13.5. CTB was injected into the right eye at P14. dLGNs were analyzed at P15. (B and C) Projection patterns of EGFP (Control) or EGFP+JAK2DN (JAK2DN) expressing RGC axons (green) and CTB expressing axons (pink) to the right (B) and left (C) dLGN. “Overlap” shows the overlapped region between EGFP and CTB signals (15% intensity threshold). “R” shows a threshold-independent representation of the segregation of ipsi and contralateral projections to the dLGN. (D–I) Quantification of eye-specific segregation. Quantification of the variance of the R-distributions in the right (D) and left (G) dLGN. Quantification of the area encompassed by projections from the contra (E and H) and ipsilateral (F and I) eye in the right and left dLGN. JAK2DN-expressing projections (light green bars) cover larger areas than Control (dark green bars). The areas occupied by CTB-labeled axons (axons where JAK2 signaling is not altered) were not different between JAK2DN (light pink bars) and control (dark pink bars). n = 8 sections from 4 mice. Scale bar, 100 μm.

(J–N) JAK2 activation promotes eye-specific segregation. (J) Schematic diagram of RGC projections to the dLGN, labeled with CTB (CF594 = green; CF660R = pink). Chx10-Cre;SOCS3^fl/fl (SOCS3 KO^RGC) mice received intraocular injections of CTB at P14. dLGNs were analyzed at P15. (K) Projection patterns of Control and SOCS3 KO^RGC RGC axons to the dLGN. “Overlap” shows the overlapped region between CTB594 and CTB660R signals (5% intensity threshold). (L) Quantification of the variance of the R-distributions in the dLGN. (M and N) Quantification of the dLGN area encompassed by projections from the contra (M) and ipsilateral (N) eye. n = 12 sections from 6 mice. Scale bar, 100 μm.

(O) Proposed model. JAK2 senses “punishment” signals from more active synapses and is activated at less active synapses. JAK2 then activates STAT1 and drives the elimination of the less active synapses.

Mean ± SEM. *P < 0.05, **P < 0.01, Mann-Whitney test.

See also Figure S8.

Eye-specific segregation in the dLGN was quantified using an unbiased assay in which the degree of segregation is represented by the variance of the distribution of the R value, the log of the ratio of fluorescence intensity (F) of each channel (R = log₁₀(F_ipsi/F_contra)) (Torborg and Feller, 2004). High R-variance indicates more segregation, while low R-variance corresponds to increased overlap between ipsi- and contralateral inputs. We found that the suppression of JAK2 kinase activity resulted in significantly lower R-variances than control in both right and left dLGNs (Figures 8B–D and 8G). There was no significant difference in the R-variance at P3 (Figures S8D–S8F), suggesting that the differences at P15 are due to the impaired elimination of JAK2-inactivated RGC axons rather than aberrant axon targeting. Importantly, when we quantified the areas innervated by EGFP- and CTB-labeled axons in each side of the dLGN, we found an increase only in the dLGN area encompassed by JAK2DN-expressing RGC axons relative to control axons (EGFP only) (Figures 8E, 8F, 8H and 8I). The area encompassed by CTB-labeled axons (i.e., axons where JAK2 signaling is not altered) was not different between JAK2DN-expressing and control mice. These results suggest that JAK2 signaling cell-autonomously regulates RGC axon elimination, and that it does not indirectly affect naïve axons. Together, these findings support the notion that JAK2 kinase activity plays a critical role in the elimination of RGC axons during eye-specific segregation.

Finally, we asked whether JAK2 activation is sufficient to drive eye-specific segregation. For this, we inactivated SOCS3 in RGCs by crossing SOCS3^fl/fl mice to Chx10-Cre (SOCS3 KO^RGC). These mice received injections of CTB 594 into the left eye and CTB 660R into the right eye at P14 and were examined at P15 (Figure 8J). The loss of SOCS3 in RGCs resulted in significantly higher R-variances relative to that of controls (Figures 8K and 8L). RGC axons in SOCS3 KO^RGC mice also occupied a significantly smaller area of the dLGN than controls (Figures 8K, 8M and 8N), which suggest that JAK2 signaling drives increased RGC axon elimination during development. Altogether, these results indicate that JAK2 serves as an “elimination” signal at multiple connections in the brain.

DISCUSSION

Activity-dependent synapse refinement is a critical step for the development of functional circuits in the brain. Here we demonstrated that JAK2, localized at presynaptic terminals, senses “punishment” signals from other active connections and serves as a determinant of activity-dependent synapse refinement. We propose that JAK2 is a neural activity-dependent switch, which can be turned on at inactive synapses to drive synapse elimination (Figure 8O).

Synaptic transmission-dependent competition drives synapse refinement

We (Yasuda et al., 2011) and others (Penn et al., 1998; Buffelli et al., 2003; Kano and Hashimoto, 2009) have shown that neural activity-dependent competition refines synapses during development. Here we established a new in vivo system to study activity-dependent synapse refinement focusing on the cortical callosal connections, whose proper development is critical for language acquisition, social interaction, attention, visual function, and perception (Innocenti et al., 2003; Paul et al., 2007; Pietrasanta et al., 2014). We demonstrated that the expression in a subset of callosal neurons of TTLC, which inhibits synaptic transmission without affecting the cells’ ability to fire action potentials (Baines et al., 1999; Fletcher et al., 2017), drives the elimination of TTLC-expressing callosal projections (Figure 1). In this case, TTLC-expressing neurons compete with other active neurons releasing neurotransmitters, and are subsequently eliminated. This elimination was prevented when we created a non-competitive environment by inhibiting the firing of all callosal neurons (Figure 1). Together, these data suggest that signals from active synapses drive the elimination of inactive synapses.

Synaptic transmission is detected by postsynaptic neurons. Thus, we propose that competition occurs at the level of synapses where the “punishment” signal would be sent from more active to less active synapses, likely via the postsynaptic neuron, to drive elimination of inactive synapses. This is consistent with findings at the neuromuscular junction, where the focal suppression of postsynaptic neurotransmitter receptors, which also suppresses synaptic transmission in a subset of synapses, drives motor axon elimination (Balice-Gordon and Lichtman, 1994). The diffuse suppression of neurotransmitter receptors prevents motor axon elimination, suggesting that the “punishment” signal is generated by synaptic transmission. The role of the postsynaptic neuron in presynaptic synapse refinement in the brain will be addressed in future studies.

JAK2 as an activity-dependent “elimination” signal in the brain

Our results indicate that inactive synapses detect the “punishment” signals from active connections and drive the elimination of the inactive connections; thus, it stands to reason that there are “elimination” signals within inactive synapses that are activated by the “punishment” signal and subsequently drive the removal of that inactive synapse. Here, we identified JAK2 kinase as such an “elimination” signal. It appears that JAK2 signaling regulates both axon (Figures 2, 4, and 5) and synapse elimination (Figures 6 and 7). Our data show that synapse elimination occurs earlier than axon elimination. Whether JAK2 signaling directly regulates both as a synaptic and axonal “elimination” signal, or axon elimination is a consequence of synapse elimination is a future question to address.

Importantly, JAK2 signaling not only regulates callosal synapse refinement, but also eye-specific segregation of retinogeniculate connections (Figure 8). Thus, JAK2 signaling appears to be a global “elimination” signal that is critical for the proper refinement of multiple brain connections. Additionally, since JAK2 is implicated in the induction of long-term depression in the hippocampus (Nicolas et al., 2012), JAK2 might also be serving as a sensor of changes in neural activity not only during development but also during adulthood.

Mechanisms of synapse elimination

What is the signal transduction pathway by which JAK2 regulates synapse elimination? We found that STAT1, but not STAT3, regulates synapse elimination (Figures 2 and 7). Both STAT1 and STAT3 can be activated by JAK2, but they play different roles in cytokine/growth factor signaling and tumorigenesis. While STAT3 promotes cell survival and proliferation, STAT1 enhances inflammation and triggers anti-proliferative and pro-apoptotic responses (Sironi and Ouchi, 2004; Schindler et al., 2007; Avalle et al., 2012; Yu et al., 2014). The activation of pro-inflammatory genes in neurons by STAT1, including chemokines and complement factors, may be involved in the recruitment of microglia to complete synapse elimination (Di Liberto et al., 2018). In addition to STAT1, JAK2 is also known to activate local substrates (Rawlings et al., 2004). These local substrates may serve as cues for “synaptic tagging” and for the recruitment of microglia. Inactive synapses and axons may present such tags, which could then capture elimination-related proteins whose expression is driven by STAT1 signaling. Additionally, JAK2 itself, which is preferentially activated at synapses to be eliminated (Figure 6), may also translocate to the nucleus (Figure 3) to regulate gene expression (Sorenson and Stout, 1995; Lobie et al., 1996; Ram and Waxman, 1997; Zouein et al., 2011).

The JAK2 pathway has also been implicated in optic nerve regeneration (Smith et al., 2009; Sun et al., 2011). The fact that both axon elimination and regeneration require JAK2 suggests that JAK2 activation might initiate the rearrangement of cytoskeleton for dynamic changes of axons. Indeed, in smooth muscle cells, JAK2 activation causes actin cytoskeleton reorganization in arteries (Guilluy et al., 2010). Molecules involved in cytoskeletal rearrangement might be also downstream of JAK2 in synapse elimination.

Signals upstream of JAK2 that contribute to synapse elimination are still an open question. One possibility is gp130, a receptor that can activate JAK2 in immune cells in response to growth hormones, growth factors and cytokines (Argetsinger et al., 1993; Parganas et al., 1998; Narazaki et al., 1994), that is also expressed in neurons (Handel et al., 2012). We found that when a DN form of gp130 (Selander et al., 2004) was co-expressed with TTLC, the density of remaining callosal axons at P15 was significantly increased (20.7 ± 1.9% with TTLC only and 33.2 ± 4.3% with TTLC + gp130DN; n = 6 mice; P = 0.014 by Student’s t-test). These results suggest that gp130 is an upstream molecule of JAK2 in inactive synapse elimination.

The involvement of gp130 in synapse elimination suggests that ligands of gp130, such as IL-6, might be a “punishment” signal released by active synapses to target and activate gp130 on inactive ones. Indeed, IL-6 is expressed in neurons in an activity-dependent manner (Sallmann et al., 2000; Jankowsky et al., 2000; Balschun et al., 2004; Gruol, 2015). At the same time, the partial rescue of inactive synapse elimination by the gp130DN suggests that other upstream signals exist. There may exist brain-specific JAK2 signaling cascades for synapse elimination. For example, cell surface molecules implicated in synapse refinement, such as MHC/PirB (Datwani et al., 2009; Lee et al., 2014), might be using JAK2 as a signaling molecule for synapse elimination.

Implication in neuropsychiatric disease

Many forms of mental illness, including autism and schizophrenia, are associated with abnormal alterations in the circuitry in the brain. The JAK2 pathway is implicated in a range of neurological disorders, including schizophrenia and Alzheimer’s disease (Chiba et al., 2009; Costain et al., 2013; Hsu et al., 2014). Abnormal synapse refinement during development and in adults may cause such defects, and our study here may help design strategies utilizing JAK2 to treat such disorders.

STAR METHODS

Resource Availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Hisashi Umemori (hisashi.umemori@childrens.harvard.edu).

Materials availability

All materials generated for this study are available upon request from the Lead Contact without restriction.

Data and code availability

This study did not generate any unique code. Individual data are available upon request from the Lead Contact.

Experimental Model and Subject Details

Animals

CD1 mice were from Charles River, and C57BL/6, Ai32, SOCS3^fl/fl, STAT1 KO, Chx10-Cre and Emx1-Cre mice were from Jackson. JAK2^fl/fl mice were from Kay-Uwe Wagner (Krempler et al., 2004). Both male and female mice were used. Mice were housed in OptiMICE cages (4–5 adult mice or one dam with a litter per cage) with a 12-h/12-h light/dark cycle. The housing room temperature was 22 ± 1 °C. Mice were allowed ad libitum access to food and water. Experiments were performed at 0–30 days of age in animals. All animal care and experiments were performed in accordance with the institutional guidelines and approved by the Institutional Animal Care and Use Committees at Boston Children’s Hospital.

Methods Details

Plasmid constructs

The pUBC-EGFP vector was built by replacing the CAG promoter of pCAG vector (Matsuda et al., 2007) with the UBC-EGFP fragment from FUGW vector (Lois et al., 2002). pUBC-mCherry vector was built by replacing EGFP coding sequence with the mCherry coding sequence. pUBC-IRES-EGFP and pUBC-IRES-mCherry vectors were built by adding PCR-amplified IRES between the UBC promoter and coding sequences. To build the pUBC-TTLC-IRES-EGFP, PCR-amplified TTLC gene (Yasuda et al., 2011) was inserted into the pUBC-IRES-EGFP vector. The pUBC-Synaptophysin-mCherry vector was built by replacing TTLC-IRES-EGFP sequence in the pUBC-TTLC-IRES-EGFP vector with Synaptophysin-mCherry. For pUBC-JAK2WT-IRES-EGFP, PCR amplified mouse JAK2 coding sequence with primers (gcaacgcgtaccatgggaatggcctgccttaca and ggcgctagctcacgcagctatactgtcccg) was cloned into pUBC-IRES-EGFP. JAK2DN (JAK2K882E) and JAK2CA (JAK2V617F) were generated by PCR with site-directed mutagenic primers (JAK2DN: ctggcgaggtggtcgctgtggagaaactccagcacagcact and agtgctgtgctggagtttctccacagcgaccacctcgccag, JAK2CA: ttttgaattatggtgtctgtttctgtggagaggagaacatt and aatgttctcctctccacagaaacagacaccataattcaaaa). JAK1 (Shimoda et al., 1997) and FynDN (FynK295M; Tezuka et al., 1999) were previously described. JAK1DN (JAK1K907E) was generated by site-directed mutagenesis with mutation primers (caggggagcaggtagctgtcgagtccctgaagcctgagagt and actctcaggcttcagggactcgacagctacctgctcccctg). pUBC-SOCS3-IRES-EGFP was built by inserting RT-PCR amplified coding sequence of SOCS3 fragment (with gaacgcgtaccatggtcacccacagcaagttt and atcgctagcttaaagtggagcatcatactg) into pUBC-IRES-EGFP. To build pUBC-SOCS3DN-IRES-EGFP, two fragments of SOCS3 coding sequences were amplified by PCR using pUBC-SOCS3-IRES-EGFP as a template (fragment 1: gaacgcgtaccatggtcacccacagcaagttt and cttgcgcacggcgttcaccacgcgcaggctggtgtccagggggcg, fragment 2: cccctggacaccagcctgcgcgtggtgaacgccgtgcgcaag and atcgctagcttaaagtggagcatcatactg). Two amplified fragments were mixed and PCR was performed with primers used for SOCS3 coding sequence amplification. The amplified fragment was cloned into pUBC-IRES-EGFP vector. pUBC-STAT1DN-IRES-EGFP and pUBC-STAT3DN-IRES-EGFP was built by inserting RT-PCR amplified coding sequence (STAT1: with gcaacgcgtaccatgtcacagtggttcgagctt and ggcgctagcttatactgtgctcatcatact; STAT3: with gcaacgcgtaccatggctcagtggaaccagctg and ggcgctagctcacatgggggaggtagcaca) into pUBC-IRES-EGFP and then introducing mutations by site-directed mutagenesis. The primers for mutagenesis were: (STAT1DN [STAT1Y701F]: gagcttgacgaccctaagcgaactggattcatcaagactgagttgatttctgtgtct and agacacagaaatcaactcagtcttgatgaatccagttcgcttagggtcgtcaagctc; STAT3DN [STAT3Y705F]: gaagccgacccaggtagtgctgccccgttcctgaagaccaagttcatctgtgtgaca and tgtcacacagatgaacttggtcttcaggaacggggcagcactacctgggtcggcttc). pUBC-gp130DN-IRES-EGFP was built by inserting RT-PCR amplified fragment of gp130 (with gcaacgcgtaccatgtcagcaccaaggatttgg and ggctctagactagacgcccagcagggttgtcag) into pUBC-IRES-EGFP. pUBC-hM4D(Gi)-mCherry was built by subcloning hM4D(Gi)-mCherry from pAAV-hSyn-DIO-hM4D(Gi)-mCherry (Addgene # 44362) into pUBC vector. pUBC-JAK2DNp2aEGFP was built by replacing the IRES sequence in pUBC-JAK2DN-IRES-EGFP with the p2a sequence. pAAV-JAK2DNp2aEGFP and pAAV-Synaptophysin-mCherry were built by replacing hChR2(H134R)-EYFP in pAAV-hSyn-hChR2(H134R)-EYFP (Addgene # 26973) with JAK2DNp2aEGFP and Synaptophysin-mCherry, respectively. Short-hairpin RNA (shRNA) to JAK2 was expressed by HuSH shRNA constructs in pRFP-C-RS shRNA vector (Origene Technologies, Rockville, MD, USA). We built four constructs, two by BLOCK-iT™ RNAi Designer (Thermo Fisher Scientific) and two from published data. These shRNA vectors were transfected to COS7 cells to examine the efficiency of JAK2 knockdown by Western blot using ImageJ Gel Analysis program. After that we used two constructs for in utero electroporation experiments (JAK2sh#1, GAAAGGATCTGGTACCCACCCAATCATGT, BLOCK-iT™ RNAi Designer and JAK2sh#2, AGCAAGCAAACCAGGAATGCTCAAATGAA, Zhang et al., 2013). All DNA sequences were verified by DNA sequencing. The knockdown efficiencies of these shRNAs as assessed by Western blotting were 90.5% (JAK2sh#1) and 98.8% (JAK2sh#2).

In utero electroporation

E13.5 pregnant female mice were anesthetized by isoflurane inhalation. The abdominal skin and wall were cut and the uterus containing embryos was pulled out, and 2 μl of DNA [up to 3 μg/μl; the amounts of DNA electroporated were: EGFP 0.5 μg, TTLC 0.5 μg, JAK2 and JAK1 1.5 μg, Fyn and CSK 1 μg, SOCS3 0.5 μg (the amount of DNA was determined based on the molecular weight), dissolved into mammalian ringer solution] was injected with a glass micropipette into the left lateral ventricle of embryos. The head of the DNA-injected embryo was placed between tweezer electrodes and five electric pulses (a voltage of 40–50 V was applied so that an 80 mA current flows, 50 ms duration, 950 ms intervals) were delivered using CUY21 electroporator (NepaGene). After electroporation of embryos, the uterus was returned into the abdominal cavity. The cavity was filled with warm sterile saline, and the abdominal wall and skin were closed with suture (Saito and Nakatsuji, 2001; Saito, 2006).

Elvax (ethylene vinyl acetate copolymer resins)-TTX implantation

P3 mice were anesthetized, their scalp was incised, and a rectangular window was made on the left side of the skull by cutting 3 sides (2.5 × 1.5 mm). The cut skull piece was flipped to expose the cortex. A 25 G needle, which was bent 0.5 mm from the tip to a 90° angle, was used to make a 0.5 mm depth cut to the cortex (1.5 mm from the midline) and TTX-Elvax or Elvax only (0.5 × 1 × 0.02 mm) (Yasuda et al., 2011) was inserted into the cut. The wound was closed and sutured. The animals were placed on a heated pad for recovery and returned to their cage. Implantation of Elvax-only did not affect TTLC-dependent axon refinement (Figures 1L and 1M).

Brain fixation

Mice were euthanized and perfused transcardially with PBS followed by 4% paraformaldehyde/PBS. Brains were removed and post-fixed with 4% paraformaldehyde/PBS for 16 hours. After cryoprotection in 30% sucrose/PBS, brains were frozen in OCT embedding compound.

Electroporation efficiency calculation (see Figures S1B–S1E)

To evaluate the efficiency of electroporation, the EGFP-expressing plasmid was in utero electroporated into the cingulate cortex at E13.5. At P5, 20 μm-thick coronal sections containing the cingulate cortex were prepared. Sections were screened by observing with an Olympus epifluorescence microscope (BX63): sections that had off-target EGFP expressions (e.g., electroporated neurons in the sensory cortex) or with a low EGFP signal intensity (sections that required over 500 ms of exposure time to acquire images with the microscope) were excluded from further processing. Sections were then immunostained for SATB2 to label callosal projecting neurons. The stained sections were imaged with a confocal microscope (Zeiss LSM700; 40X objective, 0.5X zoom). Images were taken from layer V of the electroporated cingulate cortex at ~500 μm from the dorsal surface and 150 μm from the midline (Figure S1D). Images were taken from 5 sections per mouse (spanning at least 100 μm of the cingulate cortex) from 3 mice. The electroporation efficiency was calculated as the percentage of EGFP-positive cells among all SATB2-positive cells (Figure S1E).

To evaluate the co-electroporation efficiency, a DNA mixture solution consisting of equal amounts of the EGFP-expressing plasmid and the mCherry-expressing plasmid was in utero electroporated into the cingulate cortex at E13.5. At P10, 20 μm-thick coronal sections containing the cingulate cortex were prepared. Layer V of the cingulate cortex was imaged with confocal microscopy (Zeiss LSM700; 25X objective, 0.5X zoom). Single focal plane images were taken randomly from the cingulate cortex (Figure S1B). “Snapshot” images were taken from 4 different sections (spanning at least 80 μm of the cingulate cortex) per mouse. The numbers of EGFP- and mCherry-positive cells were manually counted. The co-electroporation efficiency was calculated as the percentage of EGFP and mCherry double positive neurons per total electroporated neurons (which include EGFP and mCherry-double positive, EGFP-only, and mCherry-only neurons).

Axon density analysis (see Figures S1F and S1G)

For the analysis of callosal axon density in the brains expressing EGFP in callosal neurons, 50 μm-thick frozen sections were cut and mounted on glass slides. Images including both cortical hemispheres were taken with a 4X lens on an Olympus BX63 epi-fluorescence microscope using cellSens Dimensions software to obtain unsaturated images. Focus was adjusted to obtain the highest EGFP signal intensity in the electroporated hemisphere. Images were then analyzed with ImageJ software. To quantify the callosal axon density, the total EGFP fluorescence intensity of a 200-μm length of callosal axon bundle in the right hemisphere (adjacent to the midline; the red box in Figure S1F) was divided by the total value of the EGFP intensity in the cingulate cortex of the electroporated hemisphere (the white-lined area in Figure S1F). For more details, please see Figure S1F and S1G.

DREADD experiments

A plasmid driving expression of hM4Di under the UBC promoter was co-electroporated with an EGFP-expressing plasmid or EGFP- and TTLC-expressing plasmids by in utero electroporation at E13.5. From P3 to P15, electroporated pups were treated with CNO twice a day to suppress the activity of hM4Di expressing neurons. Brains were then collected from the CNO treated and untreated control mice at P15.

Immunohistochemistry and quantification

Immunohistochemistry was performed as described (Yasuda et al., 2011). 20 μm coronal sections were cut on a cryostat. Sections were blocked in blocking buffer (2% BSA, 2% goat serum and 0.1% Triton X-100 in PBS) for 1 hour, followed by incubation with primary antibodies in blocking buffer at 4°C for 16 hours. Sections were washed 3 times in PBS, and secondary antibodies in blocking buffer were applied for 2 hours at RT. Antibodies and dilutions used were: Phospho-JAK2 (Millipore, 1:200), VGLUT1 (Millipore, 1:5000), NeuN (Millipore, 1:100), GFAP (Synaptic Systems, 1:500), IbaI (Wako, 1:500), and SATB2 (Santa Cruz, 1:200). Stained sections were observed with a confocal microscope (Olympus FV1000 and Zeiss LSM700).

For phospho-JAK2 staining, mice were euthanized and perfused transcardially with ice-cold PBS containing 1 mM sodium orthovanadate followed by ice-cold 4% paraformaldehyde/PBS containing 1 mM sodium orthovanadate. Post-fixation, cryoprotection, and immunohistochemistry were performed as described above except with 1 mM sodium orthovanadate in all solutions. For phospho-JAK2 quantification (Figure 3), images of single focal planes were taken with sequential scanning for EGFP and Alexa 568 (Figures 3A–3D) or Alexa 647 (Figures 3E–3G) (secondary for phospho-JAK2). Note that the secondary antibodies used for phospho-JAK2 were labeled with Alexa 568 in Figures 3A–3D and Alexa 647 in Figures 3E–3G.

Signal intensity of phospho-JAK2 in EGFP-expressing neurons was quantified by using ImageJ software. For co-localization analysis of phospho-JAK2 with VGLUT1 or EGFP (Figure 6), image stacks consisting of 10 optical sections (separated by 0.52 μm) were made by ImageJ, and the number of phospho-JAK2 puncta co-localized with VGLUT1 or EGFP was counted. TUNEL (terminal deoxynucleotidyl transferase-mediated deoxyuridine triphosphate nick-end labeling) staining was carried out using the Roche In Situ Cell Death Kit (TMR red).

Western blotting

Protein lysates were prepared with lysis buffer (50 mM Tris, pH 8.0, 150 mM sodium chloride, 1% TritonX-100, 5 mM EDTA), and the protein concentration was determined with the Micro BCA Protein Assay kit (Thermo Fisher Scientific). Western blotting was done as described previously (Nagappan-Chettiar et al., 2018) with primary antibodies for JAK2 (Cell Signaling Technology, 1:1,000), JAK1 (Santa Cruz, 1:500), Fyn (Novus Biologicals, 1:1,000), Phospho-JAK2 (Millipore, 1:500), SOCS3 (1:500, Novus Biologicals) and α-tubulin (1: 1,000, Thermo Fisher Scientific). Bands were detected using the ECL detection system (Bio-Rad), and images were scanned with ImageQuant (GE Healthcare).

Synaptophysin assay

For the quantification of callosal synapses, Synaptophysin-mCherry and EGFP were expressed in cortical neurons via in utero electroporation (Figures 6A–F) or AAV injection (Figures 6G–J) to label presynaptic terminals and axons, respectively. 20–50 μm-thick coronal sections were cut, and images were taken from layer V of the right hemisphere with a Zeiss confocal microscope (63X objective, 2X zoom). Image stacks consisting of 10 optical sections (0.52 μm apart) were made by ImageJ, and the density of Synaptophysin-mCherry puncta (per EGFP-positive axon length), size and intensity of Synaptophysin-mCherry puncta were quantified.

Primary cortical neuronal cultures

Cortices were dissected from P0 pups. Cortical neurons were dissociated with 0.5% trypsin, plated on poly-D-lysine-coated coverslips (60,000 cells/coverslip; diameter 12 mm), and grown in culture medium (B27 plus (Gibco), 2 mM L-glutamine, 100 units/ml penicillin-streptomycin in Neurobasal plus medium (Gibco)).

Synaptotagmin re-uptake assay

At DIV13, cultured neurons were treated with Oyster-550-labeled Synaptotagmin-1 antibody (Synaptic Systems; 1:200) for 10 minutes at 37 °C. Neurons were fixed with 4% paraformaldehyde for 10 min at 37 °C and blocked in 2% BSA, 2% normal goat serum and 0.1% Triton X-100 for 1 hour at room temperature. Neurons were then incubated with Alexa-488-labeled phospho-JAK2 (1:100) and VGLUT1 (1:5000) antibodies overnight at 4 °C. Neurons were washed with PBS, incubated with the secondary antibody for 1 hour at room temperature, washed with PBS, and mounted in glycerol with n-propyl gallate (Sigma). Neurons were imaged with a confocal microscope (Zeiss LSM700; 63X objective, 2.5X zoom). Image stacks consisting of three optical sections (0.5 μm apart) were made by ImageJ and the intensity of Syt-1 and phospho-JAK2 at VGLUT1-positive puncta were quantified.

Stereotaxic viral injection

P0.5–P5 pups were anesthetized by hypothermia. 800 nl of viral solution was stereotactically injected (100 nl/min) into the left cortex using the following coordinates: 1.2 mm anterior from lambda, 0.4 mm lateral from midline, 0.4 mm ventral from the skull surface.

Slice preparation for optogenetics and electrophysiology

Mice were decapitated and the forebrain was quickly removed. Coronal sections (300 μm for optogenetics or 350 μm for fiber fraction analysis) of the cingulate cortex were then cut using a VT1200S (Leica, Germany) vibratome. Sections were cut in ice cold solution containing: 2.8 mM KCl, 1 mM MgCl₂, 2 mM MgSO₄, 1.25 mM NaH₂PO₄, 1 mM CaCl₂, 206 mM sucrose, 10 mM glucose, and 26 mM NaCO₃. Slices were then recovered in artificial cerebral spinal fluid (aCSF) at room temperature for at least one hour. aCSF contained 118 mM NaCl, 2.5 mM KCl, 1.3 mM MgCl₂, 1.2 mM NaH₂PO₄, 2.5 mM CaCl₂, 10 mM glucose, and 26 mM NaCO₃. All solutions were bubbled continuously with 95% O₂/5% CO₂.

Optogenetic stimulation and recording

Pyramidal cells in layer V of cingulate cortex in the right hemisphere (opposite to the electroporated side) were targeted for patch clamping. Pipettes (4–6 MΩ) filled with 100 mM cesium gluconate, 0.2 mM EGTA, 5 mM MgCl₂, 2 mM MgATP, 0.3 mM LiGTP, and 40 mM HEPES (pH 7.2). Data were collected with a Multiclamp 700B amplifier (Molecular Devices, San Jose, CA) using Clampex 10.7 (Molecular Devices), and digitized using a Digidata 1440 (Molecular Devices). Cells were held at −65 mV. Responses were evoked by a 5 ms light pulse (470 nm) delivered every 10 sec for a total of 12–14 sweeps. Analysis was based upon whether there was a time-locked consistent response evoked by light stimulation in all sweeps or not. If mice were not electroporated with the Cre-expressing plasmid, no light response could be evoked (n = 50 cells from 4 mice). In additional experiments, we found that these responses were not blocked by TTX or picrotoxin, but were blocked by the application of CNQX with APV (data not shown), suggesting that we were recording mono-synaptic callosal responses.

Fiber fraction analysis

Pyramidal cells in layer V of cingulate cortex, opposite to the virally infected side, were targeted for patch clamping. Pipettes (4–6 MΩ) filled with internal solution (35 mM CsF, 100 mM CsCl, 10 mM EGTA, and 10 mM HEPES, pH adjusted to 7.3 with CsOH) supplemented with the voltage-gated calcium channel blocker, D600 (methoxyverapamil hydrochloride; 0.1 mM; Sigma). A pair of glass electrodes, filled with 1 M NaCl and 25 mM HEPES, were placed dorsoventrally flanking the callosal tract at the midline of cortical hemispheres (Figures 7D, 7H and 7O) in a location that evoked the maximal callosal current and allowed the isolated activation of callosal inputs only (Kumar and Huguenard, 2001). Data were collected with a Multiclamp 700B amplifier (Molecular Devices, San Jose, CA) using Clampex 10.7 (Molecular Devices), and digitized using a Digidata 1440 (Molecular Devices). Cells were held at −70 mV and the callosal bundle was stimulated at an inter-trial interval of 20 seconds.

Single fiber callosal response measurement

The single fiber callosal response was determined using the “failures” method (Kumar and Huguenard, 2001). The callosal bundle was stimulated five times at a fixed stimulus intensity (between 30–80 μA). If the stimulus intensity failed to elicit a response (defined as a synaptic response of at least 5 pA with a fixed latency of 5–8 ms from stimulus onset) in more than 50% of trials (response detected in at least one but no more than two of the five trials), the cell was stimulated 5–10 more times at the same stimulus intensity (Figure S6A). All detected synaptic responses were averaged and recorded as the single fiber response. If no responses were detected, the stimulus intensity was raised by 5 μA and the “failures” method was repeated. The stimulus intensity was decreased by 5 μA if a response was evoked in more than 50% of trials and the “failures” method was repeated.

Maximal callosal response measurement

The callosal bundle was stimulated at 100 μA and the stimulus strength was gradually increased with 10–50 μA increments until the maximum callosal response was evoked. The maximum evoked response was defined as the biggest response evoked where at least three more increases in stimulus intensity no longer evoked a larger callosal response.

Fiber fraction calculation

The fiber fraction ratio is calculated as the average single fiber response / maximal response. The fiber fraction indicates the contribution of a single callosal fiber to the total callosal input received by a single neuron. The inverse of the fiber fraction is an estimate of the number of callosal inputs received by a neuron (Hooks and Chen, 2006, 2008; Noutel et al. 2011; Narushima et al. 2016).

In utero intraocular viral injection

Pregnant mice with embryos at E13.5 were anesthetized by isoflurane inhalation. The abdominal skin and wall were cut and the uterus containing embryos was pulled out. Embryos received intraocular injection of 0.5 μl of AAV virus (1 × 10¹² gc/ml) (with 0.01% Fast Green) through the uterine wall.

Anterograde labeling of retinogeniculate projections

Pups were anaesthetized and received intraocular injection of 1 μl of cholera toxin subunit B (CTB, 1 mg/ml in PBS, Biotium) into the eye to label RGC axons at P2 or P14. 24 hours after the CTB injection (P3 or P15), mice were perfused transcardially with PBS followed by 4% paraformaldehyde/PBS. Brains and eyes were removed and post-fixed in 4% paraformaldehyde/PBS overnight and subjected for analysis.

Eye-specific segregation

Fluorescence images of dLGNs for EGFP and CTB (CF594) were acquired with an Olympus BX63 epi-fluorescence microscope using a 10X objective. Fluorescence images of dLGNs for CTB (CF594) and CTB (CF660R) were acquired with a Zeiss confocal microscope (Zeiss LSM700; 20X objective, 0.5X zoom).

Variance of the R-distribution analysis was conducted based upon Torborg and Feller, 2004. The background fluorescence intensity calculated from the area adjacent to the dLGN was subtracted from each image. Image intensities were then normalized to a 0–255 scale using ImageJ. For each pixel in the image, the logarithm of the intensity ratio [R = log₁₀(F_ipsi/F_contra)], where F_ipsi is the fluorescence intensity of ipsilateral projections and F_contra is the fluorescence intensity of the contralateral projections, was calculated. More ipsi-dominant pixels would give large positive R’s and more contra-dominant pixels would give large negative R’s. The variance of R-distributions in each image was calculated and compared.

To quantify the areas encompassed by labeled retinal projections in the dLGN, images were converted into binary black and white images using ImageJ with a fixed threshold value for all images. The percentage of the area occupied by ipsilateral or contralateral projections in the total dLGN area was calculated using ImageJ.

Quantification and Statistical Analysis

All data are expressed as mean ± SEM. Statistical analyses were performed using GraphPad Prism. Sample size, P values, and statistical tests used are provided in the figure legends. All statistical comparisons were conducted on data originating from 3 or more biologically independent experimental replicates. Sample sizes were similar to those reported in previous publications (Schafer et al., 2012; Suarez et al., 2014; Cheadle et al., 2018). Data acquisition and quantification for all electrophysiological experiments were performed in a blinded fashion. Steps in the experiments were randomized, to minimize the effects of confounding variables, including how mice were chosen for experiments, order of treatments, etc. Imaging was done in same fashions among conditions. For quantification, cells and fields from brain sections were chosen randomly from the region of interest.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-phospho-JAK2 (Tyr1007/Tyr1008)	Millipore	Cat#04–1098; RRID: AB_10561932
Guinea pig anti-Vesicular Glutamate Transporter 1 (VGLUT1)	Millipore	Cat#AB5905; RRID: AB_2301751
Mouse anti-JAK2	Millipore	Cat#04–001; RRID: AB_1587214
Goat anti-JAK1	Novus Biologicals	Cat#AF602; RRID: AB_2128512
Mouse anti-Fyn	R and D Systems	Cat#MAB3574; RRID: AB_2232192
Mouse anti-α-Tubulin	Sigma	Cat#T 6074; RRID: AB_477582
Mouse anti-NeuN	Millipore	Cat#MAB377; RRID: AB_2298772
Rabbit anti-GFAP	Synaptic Systems	Cat#173 002; RRID: AB_887720
Mouse anti-SATB2	Santa Cruz Biotechnology	Cat# sc-81376; RRID: AB_1129287
Mouse anti-SOCS3	Novus Biologicals	Cat#MAB5696; RRID:AB_2193299
Rabbit anti-Synaptotagmin 1, Oyster® 550-labeled	Synaptic Systems	Cat#105 103C3; RRID: AB_887829
Rabbit anti-IbaI	Wako	Cat#019–19741; RRID:AB_2314666
Bacterial and Virus Strains
AAV-DJ CAG-EGFP	Boston Children’s Hospital Viral Core	N/A
AAV-DJ hSyn-JAK2-P2A-EGFP	This paper	N/A
AAV-DJ hSyn-Synaptophysin-mCherry	This paper	N/A
AAV-8 CAG-Cre	Boston Children’s Hospital Viral Core	N/A
Chemicals, Peptides, and Recombinant Proteins
Cholera Toxin Subunit B, CF594 Dye Conjugates	Biotium	Cat#00072
Cholera Toxin Subunit B, CF660R Dye Conjugates	Biotium	Cat#00078
Tetrodotoxin	Tocris	Cat#1078
Clozapine N-oxide	ApexBio A3317	Cat#A3317
B27 plus	Thermo Fisher Scientific	Cat#A3582801
Neurobasal plus	Thermo Fisher Scientific	Cat#A3582901
Critical Commercial Assays
In Situ Cell Death Detection Kit, TMR red	Sigma-Aldrich	Cat#12156792910
Experimental Models: Organisms/Strains
Mouse: CD1	Charles River	Cat. No. 22
Mouse: C57BL/6J	Jackson Laboratories	Stock No: 000664; RRID: IMSR_JAX:000664
Mouse: Ai32	Jackson Laboratories	Stock No: 012569; RRID: IMSR_JAX:012569
Mouse: B6;129S4-Socs3tm1Ayos/J	Jackson Laboratories	Stock No: 010944; RRID: IMSR_JAX:010944
Mouse: JAK2flox/flox	Kay-Uwe Wagner	Krempler at al., 2004
Mouse: B6.129S(Cg)-Stat1tmDlv/J	Jackson Laboratories	Stock No: 012606; RRID: IMSR_JAX:012606
Mouse: Tg(Chx10-EGFP/cre,-ALPP)2Clc/J	Jackson Laboratories	Stock No: 005105; RRID: IMSR_JAX:005105
Mouse: B6;129S2-Emx1tm1(cre)Krj/J	Jackson Laboratories	Stock No: 005628; RRID: IMSR_JAX:005628
Oligonucleotides
Primers used for cloning and mutagenesis	See Table S1	N/A
Recombinant DNA
Expression plasmids generated	See Table S2	N/A
Software and Algorithms
FIJI/ImageJ	Rueden et al., 2017	https://fiji.sc/
Adobe Photoshop	Adobe	https://www.adobe.com/products/photoshop.html
cellSens	Olympus	https://www.olympus-lifescience.com/en/software/cellsens/powerful-analysis-tools/
GraphPad Prism 6	GraphPad	https://www.graphpad.com RRID:SCR_002798
Clampfit 10.7	Axon Instruments	http://mdc.custhelp.com/app/answers/detail/a_id/18779/~/axon%E2%84%A2pclamp%E2%84%A2–10-electrophysiology-data-acquisition-%26-analysis-software-download

Highlights.

1. The tyrosine kinase JAK2 acts as an “elimination signal” for synapse refinement

2. JAK2 is activated at inactive synapses in response to signals from active synapses

3. JAK2–STAT1 signaling drives the elimination of inactive synaptic connections

4. JAK2 regulates synapse refinement at multiple connections in the brain.