Neural circuit changes in neurological disorders: Evidence from in vivo two-photon imaging

Neural circuits, such as synaptic plasticity and neural activity, are critical components of healthy brain function. The consequent dynamic remodeling of neural circuits is an ongoing procedure affecting neuronal activities. Disruption of this essential process results in diseases. Advanced microscopic applications such as two-photon laser scanning microscopy have recently been applied to understand neural circuit changes during disease since it can visualize fine structural and functional cellular activation in living animals. In this review, we have summarized the latest work assessing the dynamic rewiring of postsynaptic dendritic spines and modulation of calcium transients in neurons of the intact living brain, focusing on their potential roles in neurological disorders (e.g. Alzheimer ’ s disease, stroke, and epilepsy). Understanding the fine changes that occurred in the brain during disease is crucial for future clinical intervention developments.


Introduction
Brain is the dominant organ in human beings, regulating activities required to sustain a functional life. There are nearly one hundred billion neurons in the human brain (Azevedo et al., 2009;Herculano-Houzel, 2009), coupled up through trillions of synapses in the central nervous system, and clarification of the structure and function of neurons are critical to understand how brain works. The functional operations of the brain are dependent upon the exchange and transmission of information between neuronal synapses, and the subsequent neural activities are utilized to provide excitatory, inhibitory, or modulatory signals. The paths of information transmissions between neurons are referred as neural circuits, and they form complex networks within a brain region (Luo, 2021); each altered in the paths would be a sensitive indicator of the neural circuits (Hunnicutt and Krzywinski, 2016).
The neuronal synapse is the intersection where information is disseminated between neurons, and the dynamic rewiring of synapses could be defined as neural circuits (Yuste and Bonhoeffer, 2001). The spine is the post-synapse of most excitatory synapses and receives synaptic inputs (Hering and Sheng, 2001). The arrangement and constant rearrangement of spine function and structure are thought to be essential for the regulatory function of the central nervous system (CNS) in both physiological and pathological conditions (Buonomano and Merzenich, 1998;Katz and Shatz, 1996). Calcium is the one essential messenger located intracellularly, where its versatility and involvement are in almost all physiological and pathological cellular processes within the organism (Berridge et al., 2000). Accordingly, dynamic changes in calcium flux in neurons could reflect the functional variation of local cortical circuits (Greenberg et al., 2008;Ohki et al., 2005). Both the activity and structure of neurons play essential roles in the normal function execution of the nervous system. As such, single-time-point in vitro observation of synaptic structure or calcium activity is limited in revealing changes over time. Therefore, long-term dynamic observation in the intact brain would be more approximate to revealing the degenerative status.
The traditional in vivo neural circuit imaging devices include computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET). In Alzheimer's disease (AD), PET images have demonstrated that the neuronal activity was decreased in brain areas like the posterior cingulate cortex and precuneus, while MRI scans have shown brain atrophy resulting in communication impairment in several brain regions (Zott et al., 2018). Indeed, partial neuronal activity was increased inside the hippocampus with higher resolution laser scanning microscopy, while only a decrease of integral neuron activity was observed with PET (Busche et al., 2008;Drzezga et al., 2011). In general, each of these devices can detect the intact brain function or structure at a scale of centimeters, even though PET could reach the resolution of milli-microliters , thus representing the circuits within a region rather than at a dynamic single neuron scale, restricted resolution of these large-scale circuit imaging limits our understanding of cellular circuitry associated with pathological damage at the earliest stage of the disease, making it challenging to expound on the disease mechanism.
The two-photon effect in live-cell in vitro was first reported in 1976 (Berns, 1976), leading to the development of the two-photon microscope in 1990 (Denk et al., 1990) and the subsequent commercial application of the technology by Bio-Rad in 1997 (Robertson et al., 1997). In conjunction with various new fluorochromes and the continuous improvement of two-photon imaging technology, the two-photon laser scanning microscope (TPLSM) has become an indispensable tool in biomedical research. For example, with TPLSM, the scattering tissue such as living brain slices or intact brain could be used for small structure calcium imaging, or functional imaging (Nikolenko et al., 2007;Stosiek et al., 2003).
Two-photon molecular excitation manifested as one fluorescin absorb two photons nearly at the same time to achieve excitation, the excitation wavelength here is approximately twice compared to traditional one-photon excitation wavelength. Therefore, TPLSM has advantages for its higher 3D resolution, reduced signal-to-noise ratio (RSD), and reduced photodamage (Denk and Svoboda, 1997;Helmchen and Denk, 2005). Especially, it allows for superior imaging of live tissues such as the brain (Wilt et al., 2009), and has been used for long-term, high-resolution imaging of the neocortex or hippocampus with the fluorescent protein expressed or labeled calcium in neurons through a chronic cranial window (Feng et al., 2000;Holtmaat et al., 2009;Li et al., 2017;Yang et al., 2010) (Fig. 1).
TPLSM presents as an excellent tool to determine local circuitry on a single neuron over time, it has been applied in studies on neurological disorders. This review aims to focus on these applications to understand the pathogenesis of neurodegenerative diseases such as AD, Parkinson's disease (PD), Huntington's disease (HD), as well as nonneurodegenerative neurological diseases such as stroke and epilepsy. Additional remarks, the criteria to distinguish neurodegenerative disease is that they are expressing specific pathological in specific subset neurons, the cause is unclear and they are incurable, which may not significant in non-neurodegenerative neurological disease (Przedborski et al., 2003). We have also reviewed the calcium signaling changes in neurons and structural changes of synapse obtained by TPLSM in intact animals modeling acute or chronic neurological disorders. A comprehensive understanding on changes of fine neural circuits in disease may facilitate reveal the network mechanism of these neurological diseases.

Alzheimer's disease
AD is the most common neurodegenerative disease without a cure. Functional MRI has shown that the functional memory network involved in regions such as the medial temporal lobe system altered in clinical Fig. 1. Schematic of neural circuits changes in neurological diseases with in vivo two-photon imaging. (a) Experimental arrangement for TPLSM in vivo imaging. The head of the live mouse was fixed under TPLSM, with the partly skull thinned to approximately 20 µm (below) or replaced with a cover glass (up). (b) Time-lapse imaging of dendritic spines located mainly on the apical of pyramidal neurons. Spines present at the first imaging were shown in the middle, then the spines would be stable or eliminated, also some new spine formed at the second imaging during physiological. But when during pathological such as PD, the rate of eliminated and formed spines were both increased. Modulated from ref Guo et al. (Guo et al., 2015). (c) Imaging of calcium transient. The neurons are hyperactive, inactivated or normally active in proportion during physiological, but when in AD with amyloid plaque exist near neurons, the ratio of hyperactive or silent neurons were both increased. Modulated from ref Busche et al. (Busche et al., 2008). AD, and the subjects at-risk for AD (Sperling et al., 2010). The synaptic loss was also reported to occur in the inferior temporal gyrus and hippocampus in individuals with early AD or mild cognitive impairment (Scheff et al., 2006(Scheff et al., , 2011. Comprehending the functional and structural changes of neuronal circuits during the long-term pathological changes may significantly benefit our understanding of AD pathogenesis. Dendritic spines serve as the basic units for learning and memory, and the loss of dendritic spines is significantly correlated with cognitive impairment in AD (Selkoe, 2002;Terry et al., 1991). How spines are impaired has been investigated using long-term in vivo imaging. Using fluorescence dyes (e.g. thioflavin S, Pittsburgh Compound B (PIB), or Methoxy X04) to label the senile plaques (Skoch et al., 2005), it was observed that the somatosensory cortex spine density of Tg2576 mice was reduced at 18-24 months of age with the amyloid deposits compared to the control, but was the same at a younger age (8-to 10-month-old) before amyloid deposits present (Spires-Jones et al., 2007). In APP/PS1 transgenic mice, the amyloid deposits could be visible in the cerebral cortex from 2 months of age (Radde et al., 2006). Spine elimination was increased in the vicinity of plaques both in young (3-to 4-month-old) and old (18-to 19-month-old) APP/PS1 mice, and the spine formation was similar regardless of its relative location to plaques (Bittner et al., 2012). Collectively these results indicate dendritic spinal abnormality may be an early event in disease progression (Table 1).
Dendrites and axons are the neuronal branches, presynapse axon button and postsynapse dendritic spine located on axon and dendrite respectively. Dendrites or axons passing through or being very close by the amyloid deposits displayed neurite breakages, spine loss, shaft atrophy, and axon varicosities in AD mice models (Bittner et al., 2012;Tsai et al., 2004;Zhao et al., 2017). Progressive cytoskeletal derangements in neurites have been observed after days of plaque deposition (Meyer--Luehmann et al., 2008). Moreover, in 7-month-old PS1-deltaE9 or 3-month-old APP/PS1 mice, both of which are at the initial stage of the amyloid pathology, axonal dystrophy proximate to Aβ plaques could be observed in vivo even after 200 days of continuous two-photon imaging (Blazquez-Llorca et al., 2017).
The neuronal circuits in the hippocampus that is primarily associated with memory have also been found compromised in an AD mouse model. The dendritic spine density of Oriens Lacunosum-Molecular (O-LM) interneurons in the hippocampus was increased with age from 4months to 11-months in wild-type mice, but unchanged in APP/PS1 mice, indicating a decreased axonal survival rate of O-LM interneurons.
Additionally, synaptic rewiring in O-LM interneurons after fear learning was impaired in the APP/PS1 mice (Schmid et al., 2016). On a technical note, the dendritic spines in the hippocampus could not be detected with TPLSM directly due to light scattering from the neocortex over one millimeter above the hippocampus; hence the neocortex needs to be removed when imaging the hippocampus, a process that may affect the brain activity within 40 d post-surgery (Gu et al., 2014).
Neuronal activity altered is in conjunction with the synaptic rewiring, both of which are associated with the onset of AD (DeKosky and Scheff, 1990). The numbers of hyperactive neurons and silent neurons in the frontal cortex and hippocampal CA1 were remarkably increased, and the hyperactive neurons were mainly observed around plaques in 6-to 8-month-old Aβ-depositing APP/PS1 mice (Busche et al., 2012(Busche et al., , 2008. Besides, the neurons in CA1 were hyperactivated but not silent before the formation of plaques (in 1-to 2-month-old mice) (Busche et al., 2012). In the hippocampal CA1 of transgenic 6-to 9-month-old AD rats (McGill-R-Thy1-APP), the numbers of both hyperactive neurons and silent neurons were significantly increased (Sosulina et al., 2021). In addition to the hyperactivity of neurons near amyloid plaques, the neural activity pattern during sensorimotor circuits was also affected. Neuronal tuning for orientation in the visual cortex progressively declined as Aβ-load increased gradually with aging (Grienberger et al., 2012), and the standard action of neurons during visuomotor integration was not evident (Liebscher et al., 2016). Notably, these observations detected at the single-neuron level were achievable by using in vivo two-photon calcium imaging, but were not evident using traditional imaging methods such as PET.
The modulation of neuronal activity and synapse structure in AD were likely affected by factors including molecular and protein regulation (Fig. 2, Table 1, Table 2). Amyloid precursor protein (APP) is cleaved by β-secretase and γ-secretase to generate Aβ (Zheng and Koo, 2011). Inhibition of γ-secretase reduced the dendritic spine density in wild-type mice using in vivo two-photon imaging, but that was not observed in APP deficient mice (Bittner et al., 2009). The enzymatic unit of γ-secretase, presenilin1, altered spine density in vivo alone (Jung et al., 2011). Additionally, intraneuronal APP overexpression reduces spine density and decreases spine formation, while extracellular Aβ exhibits similar effects on spine plasticity at the place adjacent to the plaque (Zou et al., 2015).
Within in vivo two-photon calcium imaging, the neural hyperactive in the CA1 of 1-to 2-month-old APP/PS1 mouse without plaque deposit is rescued by γ-secretase inhibitor LY-411575, which targets and reduces -Abbreviations: Elim = elimination, Form = formation, Refs = references, HP = hippocampus, WT = wild type. "-" indicates the value is the same with the control, "↑" indicates the value is increased compared to the control, "↓" indicates the value is increased compared to the control. The characteristic of the blank cell in the table were not mentioned in the references.
soluble Aβ (Busche et al., 2012). After the wild-type C57 somatosensory cortex is topically applicated with soluble oligomeric Aβ extracted from Tg2576 mice, the neural resting calcium increased within 1 h and the spine density decreased in 24 h (Arbel-Ornath et al., 2017). On the contrary, immunotherapy with monoclonal anti-Aβ 3D6 (the murine equivalent of bapineuzumab) antibody reduced Aβ deposition, but increased the percentage of hyperactive neurons in the cortex of 12-to 14-month-old plaque-deposit PDAPP mice (Busche et al., 2015), consistent with the failure of bapineuzumab in Phase III trials of AD patients (Salloway et al., 2014;Vandenberghe et al., 2016). The calcium activity obtained from in vivo imaging therefore may provide some insight into the failure of clinical Phase 3 trials with Aβ-targeted immunotherapy (Panza et al., 2019). Tau protein is also featured pathologically in AD, and it mediates Aβ toxicity (Lei et al., 2012;Li et al., 2015b). Detected by in vivo two-photon calcium imaging, the silent neurons are increased in the cortex of Tg4510 (overproducing human tau and formed neurofibrillary tangles) mice and Tg21221 (overproducing non-aggregating human tau) mice, and either of which crossing with APP/PS1 mice could rescue the neural hyperactivity in APP/PS1 mice without affecting the proportion of the silent neurons in Tg4510 or Tg21221 mouse (Busche et al., 2019). Moreover, neuronal silencing can be rescued in tau transgenic mice by tau suppression with doxycycline, but not in Aβ/tau co-expressing transgenic mice (Busche et al., 2019), suggesting that soluble tau rather than neurofibrillary tangles may be involved in neuronal activity suppression while Aβ promotes neuronal activity. Another research using the P301S transgenic mouse (overexpressing a disease-related mutant tau) with an exogenous injection of tau fibrils also revealed that the neural activity in the cortex was strongly reduced independent of the presence of neurofibrillary tangles (Marinkovic et al., 2019). In addition, the neuron activity of the visual cortex in response to visual stimuli was intact in awake rTg4510 mice with neurofibrillary tangles (Kuchibhotla et al., 2014). These studies collectively suggest that tau, rather than its aggregation, plays a key role in AD-associated neural circuits impairment (Fig. 2, Table 2). APOE ε4 allele is the most significant genetic risk factor for AD (Lambert et al., 2013), while the APOE ε2 allele is related to a lower Alzheimer's dementia odds ratios (Reiman et al., 2020). A Postmortem study of AD patients has found a worsen intensity of dendrite growth with more severe neuronal degeneration in patients with the APOE ε4 allele (Arendt et al., 1997). Consistently, the neural activity responding to visual stimulation is damaged in awake 8-to 10-month-old APP/PS1 mice, and can be rescued in age-matched APP/PS1/APOE null mice (Hudry et al., 2019). In addition, the protection of APOE deficiency in APP/PS1 mice vanishes with aging (Hudry et al., 2019). Therefore, APOE may differentially modulate neural circuits impairment depending on the pathology ( Table 2). Considering that the APOE ε4 allele and APOE ε2 allele has opposite role in human AD (Serrano-Pozo et al., 2021), human APOE alleles knockin mice instead of APOE knockout mice would be a superior model to research the role of APOE alleles in dendritic spines and neural activity.
Reduced blood circulation and overactive astrocytes were also

Fig. 2.
Elements involved in AD related neural circuit change. In physiological conditions, dendritic spines will arise a normal elimination and formation rate between the first and second imaging. But the synaptic structural plasticity would be affected by several AD-related factors. When presenilin1 is overexpressed, the spine density will increase (Jung et al., 2011), while the γ-secretase is inhibited, the dendritic spine density will decrease (Bittner et al., 2009). What's more, if the distance from the nearest amyloid plaque is less than 50 µm, the spine elimination will increase, thereby the dendritic spine density will decrease (Bittner et al., 2012;Schmid et al., 2016;Spires-Jones et al., 2007;Zou et al., 2015). As to neural activity, neurons within 60 µm from the nearest amyloid plaque or Aβ oligomers show an increase in the proportion of hyperactivated neurons (Busche et al., 2012(Busche et al., , 2008(Busche et al., , 2015(Busche et al., , 2019Sosulina et al., 2021), and the overexpression of tau protein would increase the proportion of silent neurons (Busche et al., 2019;Marinkovic et al., 2019).
implicated in AD (Iadecola, 2010;Kuchibhotla et al., 2009;Takano et al., 2007). In 6 month-old APP/PS1 mice, cerebral microcirculatory vascular density, and red blood cell numbers were altered but remain recoverable through voluntary exercise . Using in vivo two-photon imaging, blood protein fibrinogen deposit was shown to induce dendritic spine elimination through microglia activation independent of amyloid plaques (Merlini et al., 2019). Moreover, the disruption of experience-dependent remodeling of dendritic spines in young 4-to 5-month-old APP/PS1 (APPswe/PS1deltaE9) mice (mice with amyloid deposits but without cognitive deficits) could be rescued by anti-inflammatory treatment with pioglitazone or the interleukin-1 receptor antagonist (Zou et al., 2016). The micro neural circuit in the intact brain may accurately present changes during AD, which can be detected using in vivo two-photon imaging.

Parkinson's disease
PD is a complex disorder that physically presents with motor irregularities such as tremors, bradykinesia, and rigidity. Cognitive impairment is also commonly presented in PD cases, accompanied or presented later with motor disability, resulting in dementia eventually (Aarsland et al., 2009;Hely et al., 2008;Williams-Gray et al., 2007). While the cause of both dysfunctions is yet to be investigated, understanding the function and structure of the neuronal circuit throughout PD progression may provide additional insights.
Traditionally, pathological changes of PD were thought to occur mainly in the midbrain (substantia nigra and striatum), and researches did discover the loss of dendritic spine in substantia nigra neurons and neostriatum medium spiny neurons of human postmortem autopsy (Patt et al., 1991;Zaja-Milatovic et al., 2005). Although it is challenging to survey the neural circuits within these regions in the intact animal brain, conventional TPLSM could determine that PD was implicated in the olfactory bulb and cortex in addition to substantia nigra (Berendse et al., 2001;Haehner et al., 2011), which could be considered as alternative targets for PD therapy. Indeed, in vivo two-photon imaging has revealed that both the elimination and formation of dendritic spines were increased in the motor cortex of the 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) treated mouse, a toxin model of PD. The rearrangement of the elimination and formation of the dendritic spine was shown to be modulated by the D1 and D2 receptors, respectively (Guo et al., 2015). D1 and D2 receptors were expressed in direct and indirect medium spiny neurons respectively in the striatum, where the deficiency of dopamine is the hallmark of PD. In vivo calcium imaging has shown that the increasing and decreasing dopamine differentially modulates the direct and indirect spiny projection neurons (Maltese et al., 2021), consistent with the results on the dendritic spine dynamics. In addition, mood disturbance such as anxiety is commonly seen in PD, and calcium imaging in awake Parkinson-like PARK7 knockout mice has found that the neurons in the frontal association cortex were hyperactive, which may closely link with anxiety .
The pathogenesis of PD contains the abnormal accumulation and propagation of the synaptic protein α-synuclein. Dementia with Lewy body (DLB) and multiple system atrophy, two other known diseases of gait disorder, also exhibit α-synuclein pathology (Spillantini et al., 1997;Wakabayashi et al., 1998). The accumulation of α-synuclein in synapses within the neocortex could be observed in real-time with in vivo two-photon imaging, and the reduction of α-synuclein synaptic accumulation can ameliorate parkinsonism symptoms (Wrasidlo et al., 2016). Impaired olfaction is another incipient symptom of Parkinson's disease. Researches have shown that neuronal survival and dendritic No Improved (compared to APP/PS1) (Hudry et al., 2019) Abbreviations: Refs = references, HP = hippocampus, WT = wild type, Tau-PFFs = Tau preformed fibrils. "-" indicates the value is the same with the control, "↑" indicates the value is increased compared to the control, "↑↑" indicates the value is increased compared to the value of "↑" group in the same reference. The characteristic of the blank cell in the table were not mentioned in the references.
spine density in the olfactory bulb of α-synuclein overexpressed A30P α-synuclein transgenic mice were both decreased with in vivo two-photon imaging (Neuner et al., 2014a;Neuner et al., 2014b). Neural activity detected with in vivo two-photon imaging is also altered in the barrel cortex of α-synuclein overexpressed transgenic mice, where it is augmented with both single or train-repetitive electrical stimulation on the contralateral whisker pad (Reznichenko et al., 2012).

Huntington's disease
Huntington's disease (HD) is another inherited neurodegenerative disorder with the symptom of abnormal involuntary movements, cognitive deficits, and psychiatric presentation, caused by the mutation of the huntingtin (HTT) gene. The mutation leads to an expanded polyglutamine tract existing in HTT protein, results in cellular dysfunction and cell death within the striatum and cortex. The course of HD is usually divided into a premanifest stage and a manifest stage, based on the disease progression. Cognitive decline often emerges in the premanifest stage and then continues to progress throughout the progression of HD (Ross et al., 2014). Studies on neural circuit mechanisms in the striatum and cortex in the early stages of the disease may help predict the clinical and pathological progression of HD with recent developments with TPLSM.
Mutant HTT is gradually aggregated from around three weeks of age in the transgenic HD R6/2 mouse model (Li et al., 2005). Chronic in vivo dendritic spine structure imaging in R6/2 mice beginning from 1-month old reveals altered spine stability and a resultant progressive loss of persistent spines in the somatosensory cortex over 40 days, encompassing the visualization from pre-symptomatic stage to an advanced phase (Murmu et al., 2013). Additionally, sensory deprivation through whisker trimming in the symptomatic phase promotes the loss of persistent spines in R6/2 mice (Murmu et al., 2015). Two-photon calcium imaging indicates that the neuronal activity was increased in the motor cortex of awake behaving pre-symptomatic stage R6/2 mice (Burgold et al., 2019). The reoccurrence rate of these active neurons was higher in R6/2 mice after three weeks, and observation was possibly caused by the loss of inhibitory synapses (Burgold et al., 2019). Heterozygous Q175 mice is a late-onset transgenic model of HD with motor deficits at an age older than 8-month, and cognitive deficits at around 1 year (Menalled et al., 2012). During motion and non-motion states, the frequency and amplitude of calcium transient within different periods were all decreased in symptomatic R6/2 mice. However, the amplitude decreased, and the frequency increased in 2-3 months (pre-symptomatic) heterozygous Q175 mice. While the amplitude decreased, the frequency increased during motion states, and both of them decreased during non-motion states in 12-18 months (symptomatic) heterozygous Q175 mice (Donzis et al., 2020). These results suggest that the cortical network deficits in HD models are dependent on genotypes and symptoms.
Metformin is the first-line drug for type 2 diabetes mellitus, and it was shown to ameliorate motor dysfunction in HD mice and patients (Sanchis et al., 2019;Shi et al., 2019). Consistently, research on neural circuits with in vivo calcium imaging has indicated that enhanced synchronicity and hyperactivity of neurons in the visual cortex of the premanifest Hdh150 transgenic HD mice were rescued by metformin treatment, as well as the behavioral abnormalities in the premanifest stage . Since Hdh150 mice exhibit significant motor abnormalities at 17 months old (Heng et al., 2007), the study (5 months old animals) here indicates early changes in neural circuits may serve as a predictor for the disease, and the time at the early changes occurred may be considered as an early intervented target.

Stroke
Stroke is the second cause of death and a critical cause of disability worldwide (Katan and Luft, 2018). The mechanisms of cell death involved in ischemic stroke have been extensively investigated, however, the neuroprotective drugs for stroke remain limited (Tuo et al., 2021a). Considering that cognitive impairment post-stroke occurs frequently, understanding the neural circuit dynamics in stroke may provide additional insights into the disease (Cuartero et al., 2019;Tuo et al., 2021b). In fact, the utilization of TPLSM, to image the neural circuits before, during, and after stroke in vivo, could provide relevant clinical information to predict stroke outcomes (Table 3).
Photothrombosis is a minimally invasive and highly reproducible procedure to induce focal ischemia. It employs two-photon laser microscopy that causes lesions in superficial vessels, resulting in selective blood flow interruption in the areas exposed to light (Watson et al., 1985). Murphy and colleagues have implemented in vivo two-photon imaging and molecular manipulating in a photothrombotic ischemia model to induce an infarction in the somatosensory cortex of YFP (yellow fluorescent protein)-H line transgenic mice, and found that the dendritic damage close to the infarction site was located on the apical tuft of Layer 5 pyramidal neurons (Enright et al., 2007). While the dendrites proximal to the infarction site were shortened, dendrites distal to the infarction site were progressively grown-up (Brown et al., 2010). The spine density near infarction was lower than the control group after a 1-week recovery from the stroke, indicating impaired dynamic rewiring of spines. However, the spine formation was elevated in the surgical group and lasted for 6 weeks post-surgery, while the spine density was recovered to the level of the control group over time (Brown et al., 2007), indicating that the neural circuits lost in stroke could be rebuild, and target neural circuits rewiring would be a potential therapy to stroke.
In the transient 20-minutes bilateral common carotid artery ligation (BCAL) model of stroke, rapid spine beading recovers at 1-hour postreperfusion, while spine formation increases until 30 days postreperfusion (Zhu et al., 2017b). In the unilateral distal middle cerebral artery occlusion (dMCAO) model of ischemia stroke combining chronic imaging of dendritic spines up to 3 months post-injury, it was also found that spine formation was related to hemodynamics near the infarction site (Mostany et al., 2010). However, there was no observed dendritic growth or an increase in branching in the dMCAO model (Mostany and Portera-Cailliau, 2011), inconsistent with an early in vitro study that the rat sensorimotor cortex lesion caused neuronal dendritic branches growth and subsequent decrease (Jones and Schallert, 1992).
The recovery rate of spines is critically involved in the degree of ischemia and perfusion. Rose Bengal could induce severe ischemia through photoinduced thrombus formation, while endothelin-1 (ET-1) only induces tiny ischemia caused by a moderate drop in blood flow. The spine damage would occur rapidly and unrecoverable in the Rose Bengal-induced severe ischemia but not in the ET-1-induced tiny ischemia . In contrast, in the BCAL model, blebbed dendrite spines could recover up to 80% at 3 h post-reperfusion after moderate 1-hour ischemia, or up to 30% after 3 h ischemia; However, limited recovery can be found with 6 h ischemia (Zhu et al., 2017a). Moreover, after severe MCAO ischemia of 60 min in mice with a subsequent 3 h reperfusion, dendritic spines recovered only in restrict penumbral regions with adequate blood flow supply .
With the combination of dendritic spine imaging and intrinsic optical signal imaging, it was found that the typical spine structure could be maintained at a distance of 80 µm from the flowing blood vessels after ischemia stroke; by contrast, it is only 13 µm under normal conditions . Moreover, the brain activation induced by peripheral sensory stimulation was only present 400 µm from the infarction boundary . In addition to the extent of ischemia, thermal changes also affect the rewiring of spines. Hypothermia could not prevent the dendritic damage observed after ischemia; however, it did promote the recovery of spines during reperfusion (Tran et al., 2012).
If the neural circuits can be rewired in the ipsilateral cortex near ---  Abbreviations: Elim = elimination, Form = formation, Refs = references, RB = rose bengal, BCAL = bilateral common carotid ligation, dMCAO = distal middle cerebral artery occlusion, MCAO = middle cerebral artery occlusion, ET-1 = endothelin-1. The units in column of ischemia duration and reperfusion are present as: h = hour, w = week, min = minute, d = day, M = month. "-" indicates the value is the same with the control, "↑" indicates the value is increased compared to the control, "↑↑" "↑↑↑" "↑↑↑↑" indicate the value is increased compared to the value of "↑" group in the same reference. The characteristic of the blank cell in the table was not mentioned in the references.
infarction, it seems logical to discover if a compensatory neural circuit response in the contralateral cortex exists. Consistent with the hypothesis, the spine formation increased with reduced spine elimination in the contralateral somatosensory cortex after a 1-week recovery from photothrombosis in the ipsilateral somatosensory but not visual cortex (Takatsuru et al., 2009). However, the same results were not observed in the dMCAO model . These inconsistent results may be related to the difference of chronic imaging window (a thinned-skull window v.s. an open-skull window), mouse line, stroke model, and imaging intervals. The neuronal activity pattern is also affected in stroke. The hindlimb stimulus-induced neuronal activity in hindlimb somatosensory was reduced after penetrating vascular occlusion, revealed by in vivo twophoton calcium imaging (Shih et al., 2013). Hindlimb somatosensory can selectively respond to the contralateral forelimb after ischemia in forelimb somatosensory with photothrombosis (Winship and Murphy, 2008), consistent with the functional compensatory mechanism after brain damage. In stroke, the rapid dendritic beading in the penumbra near infarction is promoted by spreading depolarizations (SDs) (Risher et al., 2010), which usually occur after brain damage and can arise from the core of the microinfarct in the cortex (Shih et al., 2013). Local amide anesthetic or the sodium channel blocker dibucaine can delay the SDs and inhibit dendritic damage (Risher et al., 2011). In single neurons, the mitochondrial depolarization can impact the dendritic beading in the BCAL model (Liu and Murphy, 2009), and the neuronal mitochondrial fragmentation may be a sensitive indicator for the dendritic damage (Kislin et al., 2017). The depolarization depending on NMDA receptors activation that occurs after ischemia is responsible for rapidly reversible spine beading. Blocking the NMDA receptors could reduce the spine damage , and the NMDA receptor antagonists MK-801 or memantine could reduce the infarction volume in the tiny ischemia model (Shih et al., 2013). The activation of NMDA receptors located outside of the synapse is mediated by astrocyte calcium waves, which can exacerbate brain damage after ischemia. The inhibition of the IP3R2-mediated astrocyte calcium signaling pathway can lessen the damage evidenced by electrophysiology and calcium imaging in vivo (Dong et al., 2013;Li et al., 2015a).
Glial cells are known to provide support for neurons in the brain (Allen and Barres, 2009;Araque and Navarrete, 2010), and there is evidence that glia is critical for the recovery from damage to neural circuits in stroke. As stroke is more prevalent in the elderly population, an observation that the fraction of neural activity in the penumbra after MCAO was not changed in 18-to 24-months-old mice but reduced in 3to 4-months-old mice, may result from the different activity patterns of astrocytes in the penumbra (Fordsmann et al., 2019). Indeed, intracellular calcium elevations in both neurons and astrocytes in the penumbra after stroke is accompanied by peri-infarct depolarizations (PIDs), and the calcium elevation has been shown to affect the degree of damage negatively (Rakers et al., 2017). Moreover, the transient receptor vanilloid 4 (TRPV4) channel in astrocytes may also contribute to the calcium elevation in neurons during PIDs (Rakers et al., 2017). On the other hand, the microglia contract the synapse every few minutes in healthy conditions; however, after ischemia stroke, the contraction lasts for 1 h, which may alternatively support the reversible spine beading (Wake et al., 2009). The SDs could promote dendritic beading after ischemia (Risher et al., 2010), and a lack of SDs could decrease neuronal activity which leads to excitotoxic injury. Therefore microglia may suppress the generation of SDs and relieve brain damage after MCAO (Szalay et al., 2016).
There are also physical strategies to ameliorate the damage of stroke through rewiring the impaired neural circuits. Physical training is one of the strategies that contribute to the improvement of neural circuits (Allegra Mascaro et al., 2019). Another one is thalamocortical circuit stimulation, as stroke in the somatosensory cortex can inhibit the thalamocortical circuits, activating the thalamocortical circuit with optogenetics could potentially relieve brain damage after stroke (Tennant et al., 2017).
The observations from animal models of stroke are consistent with clinical findings, particularly with that the progression of the disease without post-stroke treatment can be detrimental to neuronal circuitry. Nonetheless, advances in TPLSM with modified cranial window mentioned in summary and outlook section can yield clues to neural circuit recovery during drug treatment during stroke, which may deliver potential therapeutic strategies.

Epilepsy
Epilepsy is a chronic neurological disease concomitant of unprovoked and recurrent seizures. It is known that brain disorders trigger seizures, which may cause brain injury via rewiring of the brain circuitry (Bronen, 2000). Seizure and epilepsy usually represent symptoms of an underlying disorder, and as such, epilepsy patients are known to suffer from memory impairment and psychiatric disorders (Elger et al., 2004). Long-term changes in synaptic morphology in epilepsy may therefore serve as the basis of the cognitive deficits. Specimens from epilepsy patients and animal models indicate profound dendritic changes, including an increase or decrease in dendritic density, size alterations, and abnormal morphology (Swann et al., 2000;Wong, 2005). However, a precise understanding of the mechanism is not clear. Nonetheless, abnormal hypersynchronous neuronal firing seems to be operated within a circuitry network (Luo et al., 2014;Smith and Schevon, 2016), altering the normal neuronal circuitry patterns. With TPLSM, the dynamic changes of neural circuits, dendritic spines, and subsequent neural activity in the intact brain during different epilepsy stages can be observed.
Using in vivo two-photon imaging of hippocampal CA1 dendritic spines, spines were found to be stable for the first several hours postseizure, but then begin to disappear after the epileptiform activity induced by pilocarpine in GFP-O line mice (a mouse that expressing green fluorescence protein in a subset of pyramidal neurons) (Mizrahi et al., 2004). However, the observed lifespan for dendritic spines in CA1 was only several hours limited by surgery and the resolution of TPLSM. Therefore, dendritic spines in the hippocampus were detected at high-resolution and long-term with stable cranial window surgery (Gu et al., 2014), and the stability of spines for several hours was found to be associated with the anesthetics. For example, dendritic spines remained stable for several hours in the neocortex after 4-Aminopyridine (4-AP) induced electrographic seizures in anesthetized and unanesthetized mice. However, when coming up to 4-AP induced clinical electrographic seizures without anesthesia, the spines swell and lose rapidly (Rensing et al., 2005). In the kainate model of epilepsy, dendritic spines exhibited reversible dendritic beading and lost more spines with higher-level seizure activity, however, calcineurin inhibitor FK506 could inhibit cofilin activation and ameliorate dendritic spine injury in the cortex (Zeng et al., 2007), and in this kainite induced severe seizures model, microglia helped beading dendrites recovering through microglial process pouches instead of damaging the dendrites through phagocytosis (Eyo et al., 2021). These researches implicated that spine dynamic was closely related to the epilepsy status and the probable mechanisms were actin depolymerization, calcineurin signal, and microglia activation (Eyo et al., 2021;Zeng et al., 2007).
Dendritic spines undergo beading or are lost during seizures, but how neural activity appears hypersynchronous and leads to the spread of seizures? The cellular and circuitry patterns depend on GABAergic interneurons, modulate the neuronal activity in cortical seizures induced by 4-AP (Wenzel et al., 2017). During the pre-ictal period, seizures first recruit neuron activation at an initiation site, and when a threshold is reached, seizures are propagated to a different focal region. In this region, the electrographic seizure can be detected, concomitantly with the increase in firing over parvalbumin (PV) containing interneurons (Wenzel et al., 2019). The inhibition of PV interneurons contributes to the ictal and propagates of seizures via such a mechanism that has been observed in rats using two-photon calcium imaging (Liou et al., 2018).
Interneurons are also crucial in the appearance of recurrent seizures (Drexel et al., 2017;Zhu et al., 2018). The interictal spikes are characterized as synchronous bursts occurring between seizures in CA1 neurons of the pilocarpine-induced temporal lobe epilepsy, the interictal spikes in the CA1 neurons are primarily occurred in GABAergic neurons and may contribute to the inhibition of downstream neuronal activity (Muldoon et al., 2015). A selective GABAa receptor-positive modulator, imidizodiazepine (KRM-II-81), can decrease the spontaneous discharges in the hippocampus of mice with kainite-induced seizures. It has been shown to inhibit the neuronal hyperactivity in the cortex of the post-traumatic epilepsy model in vivo (Witkin et al., 2020). Desynchronization of PV positive GABAergic interneurons in response to temperature elevation can also contribute to the ictal seizures, and PV, somatostatin (SST), or vasointestinal polypeptide (VIP) positive GABAergic interneurons are recruited separately during seizures, as seen in Darvet syndrome, a neurodevelopmental disorder manifesting as severe epilepsy (Somarowthu et al., 2021;Tran et al., 2020).

Multiple sclerosis
Multiple sclerosis (MS) is an autoimmune disease in which the immune system causes demyelination within the central nervous system, lead to neuronal damage in spinal cord and many brain areas (Lassmann et al., 2007), and whether MS is a kind of neurodegenerative disease still controversial (Przedborski et al., 2003;Trapp and Nave, 2008). Cognitive impairment and sensory symptoms include numbness and paresthesia are both reported as an prevalent symptom of MS (Rao, 1995). Experimental autoimmune encephalomyelitis (EAE) is a classic animal model for MS, and which is inducted through immunization of myelin oligodendrocyte glycoprotein peptide (MOG 35-55 ) (Ransohoff, 2012).
Reconstruction of single cortical projection neurons in autopsies from MS patients has shown that the density of dendritic spine decreased independent of cortical demyelination and axon loss (Jurgens et al., 2016), consistent with many in vitro slice studies (Centonze et al., 2009;Potter et al., 2016). Research with in vivo two-photon imaging has shown that not only the elimination but also the formation of dendritic spine and axon boutons in the somatosensory were both elevated at the 7 days post MOG injection, which is earlier than the usually 10 days when MOG-induced EAE clinical scores were increased, and earlier than T-cell infiltration and microglial activation (Yang et al., 2013). Homozygous transgenic mice of myelin basic protein specific expressed T cell receptors has a high incidence of EAE (Lafaille et al., 1994), Huang et al. used this transgenic mice and found the similar increasing of dendritic spine elimination and formation in somatosensory cortex before the onset of EAE, and this dendritic spine turnover also appeared in motor cortex (Huang et al., 2021). What's more, the early instability of dendritic spine dynamic and axon boutons in somatosensory cortex of MOG-induced EAE would reversed through inhibition of the elevated peripheral TNF-α (Yang et al., 2013), the abnormal spine plasticity in motor cortex of the homozygous transgenic mice would alleviative when treated with anti-IFN-γ antibodies (Huang et al., 2021). The early abnormal spine plasticity main influenced by elevated cytokines in the cortex.
Using fluorescence resonance energy transfer (FRET)-TPLSM, calcium activity was found to be elevated before interactions between CD4 + Th17 immune cells and neurons in the brain stem in EAE mice, and it is hypothesized that the calcium oscillations pattern could be an indicator of subsequent neuronal damage during inflammatory infiltration (Siffrin et al., , 2010. Moreover, the neuronal damage caused by elevated calcium oscillations could partially be reversed by NMDA receptor blocker MK-801, fingolimod, or dimethyl fumarate (Luchtman et al., 2016;Siffrin et al., 2010). The in vivo axonal damage caused by calcium hyperactivity in the axon was associated with CD4 + Th17 immune cells but not Ovalbumin-transgenic (OT-1) CD8 + T cells. CD8 + T cells presented in MS may, therefore, have other not yet clearly defined roles in the pathogenesis of MS (Reuter et al., 2015). In this study, the neuronal calcium was labeled with genetically encoded sensors based on FRET, which is more suitable for monitoring the dynamic of low but long-lasting pathological calcium increase than ordinary calcium indicators (Mank et al., 2008;Naoto et al., 2012;Siffrin et al., 2015).
A cuprizone diet with daily rapamycin injection can lead to demyelination in mice, with a reversal to remyelination when the diet is withdrawn, and can be considered as a model for MS. In vivo neuronal activity in the CA1 and the dentate gyrus of the hippocampus during this baseline-demyelination-remyelination cycle has revealed that the neuronal firing rates are substantially reduced during demyelination and then partially recovered during remyelination (Das et al., 2019). These hyperactive neurons also appeared in the frontal cortex and visual cortex of the EAE mouse in the remission phase, but not in the relapse phase (Ellwardt et al., 2018). Moreover, an anti-TNFα antibody injection reversed the hyperactivity and alleviate the anxiety-related behavior disorder that is often observed at a later stage of the disease (Ellwardt et al., 2018). These in vivo observations highlight the correlation between immune attack and subsequent neurodegeneration.

Other neurological disease
Neuropathic pain is often caused by progressive nerve injury, and chronic pain of which without apparent reason is associated with nearly the health status of 10% of Americans (van Hecke et al., 2014). In the partial sciatic nerve ligation (PSL) model of neuropathic pain, the synaptic plasticity in the somatosensory cortex rearranges during the development phase and stabilizes in the maintenance phase, detected by long-term in vivo two-photon imaging (Kim et al., 2011). In Freund's adjuvant-related chronic inflammatory pain, the response intensity and amplitude of inhibitory interneurons in the somatosensory cortex monitored with in vivo calcium imaging have been enhanced and the modulation of the inhibitory interneural circuit could ameliorate the pain thresholds (Eto et al., 2012). Neural activity in the contralateral or ipsilateral dorsal anterior cingulate cortex of awake, head-restrained mice was higher after the sensory stimulation in unilateral one limb with spared sciatic nerve injury when compared to the no-sciatic nerve injury group (Zhao et al., 2018a). These observations indicate that the neural activity in different cortical regions and the development phase of it during neuropathic pain could present as avenues to investigate therapeutic interventions for pain therapy.

Summary and outlook
TPLSM is currently the preferred choice of light microscopy for in vivo functional and structural research on the subcellular and multi-cell levels. As reviewed, the observed pathological changes and related biological variations in neurological diseases obtained through in vivo two-photon imaging on precise neural circuits have provided critical insights into the mechanisms of diseases. This, in turn, has created enormous potential for unveiling the pathogenesis of neurological disorders and identifying novel therapeutic strategies. What's more, the method to measure the dynamics of neural circuits in vivo can also be used to study neurodevelopmental disorders, psychiatric disorders, learning and memory deficits, and so on.
Notwithstanding the promising development, there are some obstacles limited the extensive application of this in vivo TPLSM in neurological diseases. The first one is the image depth in the live interior of the intact brain. The usually image depth of TPLSM is about 0.7 mm (Gu et al., 2014;Nemoto, 2008), approaching 1 mm by combining with a laser regenerative amplifier (Theer et al., 2003). With the advent of three-photon microscopy can enable visualization of neuronal calcium transients in the stratum pyramidal layer of the CA1 hippocampus at approximately 1.4 mm below the pia in anesthetized mice (Ouzounov et al., 2017), and 1.1 mm below the pia in freely moving rats (Klioutchnikov et al., 2020). Limited by the depth of penetration in live tissue of TPLSM, studies of the neural circuits on neurological disease in vivo that we have summarized here are mainly focused on the cortex, or hippocampus after the surgical removal of the cortex. Considering that the involved brain regions in neurological diseases include not only cortex and hippocampus, but also deep internal area inside the brain, such as the primarily effected substantia nigra in PD, which is nearly 4 mm below the pia in mice, the deeper of the image depth in the further would facilitate the study of neural circuits in live interior of the intact animal brain, even in the human brain.
The second obstacle is the cranial window for imaging. In the in vivo TPLSM, the made of chronic cranial window is vital and highly dependent on the experience of operator, the success rate varied from 30% to 80% (Holtmaat et al., 2009), which may limit the application of TPLSM in researching on neural circuits of neurological disease. The optical clearing is a potential tool for improve imaging quality by reducing scattered light caused by the skull over the cortex. With this technique, the skull does not require to be as thin as 20 µm or replace with a cover glass when used for synaptic imaging in vivo (Zhao et al., 2018b), which may reduce the difficulty of cranial surgery and promote the applications of TPLSM in neural circuits researching.
In addition, the two classic chronic cranial windows were not appropriate to drug application, cellular manipulations, or electrophysiological recording at the same time of neural circuits recording. Therefore, there are many variants based on open skull window for drug application, cellular manipulations or recording. A micro-optical fluidic devices equipped to open skull window could help to deliver drug or chemical into the brain (Takehara et al., 2014), this contribute to analyze how the chemical or drug influence in the neural circuits during the progress of neurological disease such as fragile X syndrome (Nagaoka et al., 2016). Another modified open skull window is that drill a circular hole into the cover glass (Roome and Kuhn, 2014), use a removal and replacement glass (Goldey et al., 2014), use silicone-based polydimethylsiloxane (PDMS) (Heo et al., 2016) or polyethylene-oxide-coated amorphous perfluoro (1-butenyl vinyl ether) polymer (Takahashi et al., 2020) for drug application, cellular manipulations, or electrophysiological recording during neural circuits recording. The PDMS materials has been used to observe the dynamic change of neuronal activity in different cortex area before and after local ischemic stroke model with ET-1 injection (Chornyy et al., 2021). All of the variants of cranial windows are hopeful in drug screening and mechanism researching at the level of neural circuits in neurological disorders.
Chronic cranial window could detect multiple brain area at the same time were also important, a polished and reinforced thinned skull window allowed for a larger imaging field of view than classic thinned-skull window (Drew et al., 2010;Yang et al., 2010), which is appropriate to applicated in multi area neural circuit imaging.
The third limitation is fluorescence labeling. While the near-infrared reflectance microscopy may be useful to observe myelinated axons in the label-free live brain (Allegra Mascaro et al., 2015), the resolution is limited in other structures without myelin, rendering them invisible. It would be critical to developing labels associated with neural circuits that are also visible for TPLSM. To date, there has been much efficient and specific labeling of targeted populations or substructures of neurons in animals in vivo. At present, human pyramidal neurons obtained from epileptic patients have been successful labeled with fluorescence protein using viral transfection, and the spine dynamics could be detected as long as 24 h in vitro (Schwarz et al., 2019), these fluorescence labeling provide a foundation for future in vivo human neuron labeling. Moreover, an appropriate label can help to figure out the structure and function of the brain, in addition to neural circuits. For example, a calcium indicator of neural activity utilizing a genetically encoded fluorescence sensor has recently been used to detect the intracellular Clconcentration, and pH in an intact mouse brain in vivo (Sulis Sato et al., 2017).

Conclusions
In conclusion, we have presented here evidence on mechanisms of neural circuit dysfunctions in neurological disorders, using TPLSM. The continued advances in TPLSM will further aid in the quest to understand the dynamic modeling of neural circuits and lead to discoveries that will provide hope to develop effective pharmaceutical interventions for brain disorders.

Funding
This work was supported by the National Natural Science Foundation of China (81722016, 31800899, 91632115, 32000724), the China Postdoctoral Science Foundation (2018M633375), the Sichuan Science and Technology Program (2019YFS0212).

CRediT authorship contribution statement
PL and HX conceived of the article, HX and FT wrote the first draft, with some advice provided by YJG and RXX. All authors read the draft and critically revised it and approved the decision to publish, PL is the guarantor and accepts full responsibility for the article, controlled the decision to publish.

Declaration of Competing Interest
The authors have declared that no competing interest exists.