Alessandro Gozzi | Istituto Italiano di Tecnologia / Italian Institute of Technology (original) (raw)

Papers by Alessandro Gozzi

The mechanisms underlying the signal changes observed with pharmacological magnetic resonance ima... more The mechanisms underlying the signal changes observed with pharmacological magnetic resonance imaging (phMRI) remain to be fully elucidated. In this study, we obtained microdialysis samples in situ at 5-min intervals during phMRI experiments using a blood pool contrast agent to correlate relative cerebral blood volume (rCBV) changes with changes in dopamine and cocaine concentrations following acute cocaine challenge (0.5 mg/kg iv) in the rat over a duration of 30 min. Three brain areas were investigated: the dorsal striatum (n = 8), the medial prefrontal cortex (mPFC; n = 5), and the primary motor cortex (n = 8). In the striatum and mPFC groups, cocaine and dopamine temporal profiles were tightly correlated, peaking during the first 5-min period postinjection, then rapidly decreasing. However, the local rCBV changes were uncorrelated and exhibited broader temporal profiles than those of cocaine and dopamine, attaining maximal response 5 – 10 min later. This demonstrates that direct vasoactivity of dopamine is not the dominant component of the hemodynamic response in these regions. In the motor cortex group, microdialysis revealed no local change in dopamine in any of the animals, despite large local cocaine increase and strong rCBV response, indicating that the central hemodynamic response following acute iv cocaine challenge is not driven directly by local dopamine changes in the motor cortex. The combination of phMRI and in situ microdialysis promises to be of great value in elucidating the relationship between the phMRI response to psychoactive drugs and underlying neurochemical changes.

Translational Psychiatry, 2015

Increasing evidence points to a role for dysfunctional glutamate N-methyl-D-aspartate receptor (N... more Increasing evidence points to a role for dysfunctional glutamate N-methyl-D-aspartate receptor (NMDAR) neurotransmission in schizophrenia. D-aspartate is an atypical amino acid that activates NMDARs through binding to the glutamate site on GluN2 subunits. D-aspartate is present in high amounts in the embryonic brain of mammals and rapidly decreases after birth, due to the activity of the enzyme D-aspartate oxidase (DDO). The agonistic activity exerted by D-aspartate on NMDARs and its neurodevelopmental occurrence make this D-amino acid a potential mediator for some of the NMDAR-related alterations observed in schizophrenia. Consistently, substantial reductions of D-aspartate and NMDA were recently observed in the postmortem prefrontal cortex of schizophrenic patients. Here we show that DDO mRNA expression is increased in prefrontal samples of schizophrenic patients, thus suggesting a plausible molecular event responsible for the D-aspartate imbalance previously described.

Journal of Neuroscience Methods, Feb 15, 2010

Multiple experimental neuroscience techniques rely on the use of general anaesthesia to minimize ... more Multiple experimental neuroscience techniques rely on the use of general anaesthesia to minimize the discomfort associated to animal restraint and to achieve a more effective control of relevant physiological parameters. In order to minimise potential interference on brain neuronal activity, such studies are typically conducted at low anaesthetic doses. This practice is often coupled to peripheral infiltration of local anaesthetics to provide supplementary analgesia and prevent sub-threshold activation of pain pathways that may confound central measurements of brain function. However, little is known of the effect of peripheral anaesthesia on central measurements of brain activity in small laboratory animal species. In order to begin to address this question, we measured total and free brain exposure of five different local anaesthetics following subcutaneous infiltration of analgesic doses in a surgical protocol widely used in rodent neuroimaging and electrophysiology studies. Notably, all the anaesthetics exhibited detectable total and free brain concentrations at all the time points examined. Lidocaine and mepivacaine showed the highest free brain exposures (>525 ng/g), followed by bupivacaine and ropivacaine (>70 ng/g). The ester-type local anaesthetic tetracaine produced the lowest free brain exposure (<8.6 ng/g). Our data suggest that peripheral administration of local anaesthetics in small laboratory animals could result in pharmacologically active brain exposures that might influence and confound central measurements of brain function. The use of the ester-type anaesthetic tetracaine produced considerably lower brain exposure, and may represent a viable experimental option when local anaesthesia is required.

We have applied pharmacological magnetic resonance imaging (phMRI) methods to map the functional ... more We have applied pharmacological magnetic resonance imaging (phMRI) methods to map the functional response to nicotine in drugnaïve rats. Nicotine (0.35 mg/kg intravenous (i.v.)) increased relative cerebral blood volume (rCBV) in cortical (including medial prefrontal, cingulate orbitofrontal, insular) and subcortical (including amygdala and dorsomedial hippocampus) structures. The pharmacological specificity of the effect was demonstrated by acute pretreatment with the nicotinic acetylcholine receptor (nAChR) ionchannel-blocking agent mecamylamine, which suppressed the rCBV response to nicotine. Control experiments with norepinephrine, a potent non-brain-penetrant vasopressor, at a dose that mimics the cardiovascular response induced by nicotine were performed to assess the potential confounding effects of peripheral blood pressure changes induced by nicotine. In an attempt to highlight the relative contribution of different nAChR subtypes to the observed activation pattern of nicotine, we also investigated the central phMRI response to an acute challenge with (R)-N-(1-azabicyclo[2.2.2]oct-3-yl)(5-(2-pyridyl)thiophene-2-carboxamide) (cpdA, at 5, 10, 20, and 30 mg/kg i.v.) and 5-iodo-A-85380 (5IA, 5 mg/kg i.v.). CpdA is a selective agonist at homomeric a7 nAChRs, while 5IA features high in vivo affinity for the a4b2* and other less-abundant b2-containing nicotinic receptors. CpdA did not produce significant rCBV changes at any of the doses tested, whereas 5IA induced a pattern of activation very similar to that induced by nicotine. The lack of phMRI response to cpdA together with the high spatial overlap between the activation profile of nicotine and 5IA, suggest that the acute functional response to nicotine in drug-naïve rats is mediated by b2-containing nAChR isoforms, presumably belonging to the a4b2* subtype.

The mechanisms underlying the signal changes observed with pharmacological magnetic resonance ima... more The mechanisms underlying the signal changes observed with pharmacological magnetic resonance imaging (phMRI) remain to be fully elucidated. In this study, we obtained microdialysis samples in situ at 5-min intervals during phMRI experiments using a blood pool contrast agent to correlate relative cerebral blood volume (rCBV) changes with changes in dopamine and cocaine concentrations following acute cocaine challenge (0.5 mg/kg iv) in the rat over a duration of 30 min. Three brain areas were investigated: the dorsal striatum (n = 8), the medial prefrontal cortex (mPFC; n = 5), and the primary motor cortex (n = 8). In the striatum and mPFC groups, cocaine and dopamine temporal profiles were tightly correlated, peaking during the first 5-min period postinjection, then rapidly decreasing. However, the local rCBV changes were uncorrelated and exhibited broader temporal profiles than those of cocaine and dopamine, attaining maximal response 5 -10 min later. This demonstrates that direct vasoactivity of dopamine is not the dominant component of the hemodynamic response in these regions. In the motor cortex group, microdialysis revealed no local change in dopamine in any of the animals, despite large local cocaine increase and strong rCBV response, indicating that the central hemodynamic response following acute iv cocaine challenge is not driven directly by local dopamine changes in the motor cortex. The combination of phMRI and in situ microdialysis promises to be of great value in elucidating the relationship between the phMRI response to psychoactive drugs and underlying neurochemical changes. D 2004 Elsevier Inc. All rights reserved.

In this contribution we focused on the application of graphics processing units (GPUs)accelerated... more In this contribution we focused on the application of graphics processing units (GPUs)accelerated computing in reconstruction of diffusion weighted nuclear magnetic resonance (DW-NMR) images by using non-Gaussian diffusion models, such as the diffusional kurtosis imaging (DKI) 1 and the stretched exponential model imaging (STREMI) 2 , which allow to increase the sensitivity and specificity of the DW-NMR mapsin detecting several pathological conditions 3,4 . However, the post-processing of DW-NMR images based on these models currently requires too long times for any application in real-time diagnostics. Typically, for the elaboration of these images, 10 6 -10 7 voxels have to be managed. For each voxel the algorithm calculates at least 3 parameters by non-linear functions optimization. This is computational demanding and takes some hours on recent multi-core processors(i.e. CPU Intel Xeon E5 and E7) to obtain a brain map. The aim of this work is to implement non-Gaussian diffusion imaging processing on the massively parallel architecture of GPUs and optimize different aspects to enable online imaging. METHODS. Code implementation: a modern GPU device can have a number of "multiprocessors" (MP), each of which executes in parallel with the others. Using the Nvidia compute unified device architecture (CUDA), multiple thread blocks (and thus multiple fittings) can execute concurrently with many parallel threads on one multiprocessor. Here we implemented an efficient and robust fitting algorithm, based on a highly parallelized Levenberg-Marquardt (LM) method on GPU. The LM algorithm is based on an iterative numerical optimization procedure that minimizes the sum of squared model residuals. The function we used to perform LM fittings on GPU device is the GPU-LMFit function 5 . GPU-LMFit uses a scalable parallel LM algorithm optimized for using the Nvidia CUDA platform. Here we used two versions of the GPU-LMFit function: a single-precision (S) and a double-precision (D) version. The code kernel calls GPU-LMFit to perform the LM algorithm on each CUDA block, which is mapped to a single voxel. Because the processing of different voxels is totally independent, the CUDA blocks do not need to synchronize, and the kernel launches as many blocks as voxels contained in a particular slice to speed up performance. The code was optimized to be fully integrated within Matlab (The Mathworks, Natick, MA, USA) scripts. Two different multi-core central processing unit (CPU) configurations and three different Nvidia GPUs were used for the analysis and the cross-comparison of CPU and GPU performance (see Table1). In particular, lsqcurvefit function with Parallel Computing Toolbox was used to test multicore CPU performance. DW-NMR data acquisition: an in vivo and an ex vivo healthy mouse brain, fixed in paraformaldehyde and stored in PBS, was scanned at 7.0 T (BRUKER Biospec). An imaging version of PGSTE sequence was performed with TE/TR = 25.77/4000 ms, Δ/δ= 40/2 ms, NA = 14; 16 axial slices with STH= 0.75mm, FOV=6cm, matrix 128x128 with in plane resolution of 470μm 2 were acquired with10 b-values ranging from 100 to 8000s/mm 2 along 30 no-coplanar directions plus 5 b=0s/mm 2 . DW-NMR data analysis: parametric maps of kurtosis metrics, i.e. Kapp-maps, were obtained by fitting on a voxel-by-voxel basis the following relationto the DW image signal intensities (for b≤3000s/mm 2 ): S(b)/S(0)=exp(-bDapp+1/6b 2 Dapp 2 Kapp), where Dapp and Kapp are the apparent diffusion coefficient and kurtosis, estimated in the direction parallel to the orientation of diffusion sensitizing gradients, respectively. STREMI parametric maps, i.e. γ-maps, were obtained by similar procedure, using the following relation: S(b)/S(0)=exp[-(Dappb) γ ], where γ is the stretching parameter, being between 0 and 1.Finally, the non-Gaussian diffusion parametric maps Mγ and MK were computed by averaging across the 30 directions in the corresponding γ and Kapp-maps, respectively. The analyzed set of DW-NMR images was of ~ 100-200 Mb. The total number of fittings to be performed was in the range of(0.5-5)x10 6 . RESULTS & DISCUSSION. Non-Gaussian diffusion parametric maps Mγ and MK, obtained by using lsqcurvefit on multiple CPU threads and GPU-LMFit on GPU, are displayed in Figure1. The specific performances of each CPU configuration and GPU employed are reported in Table1, and the cross-comparison between lsqcurvefit and GPU-LMFit results is reported in it is evident that the GPU approach for STREMI is in perfect agreement with conventional CPU one. On the contrary, for DKI, GPU-LMFit slightly overestimates MK values with respect to lsqcurvefit. However,it is important to note that MK-maps obtained with GPU-LMFit show a better contrast-to-noise ratio than the lsqcurvefit ones. Finally, from Table1 it is possible to appreciate the relative speed-up obtainable using GPU-LMFit, which for medium level GPUs is ~100x but for high level ones can reach a factor ~1000x. This means that the GPU implementation reported here allowsto reduce the time for massive image processing from some hours to some seconds (see . CONCLUSION. The proposed implementation of LM algorithm on GPU makes it excellent for extensive GPU-based applications such as massive MRI processing. Our results show that the GPU application proposed here can further improve the efficiency of the conventional LM model fittings, finally enabling automated parametric non-Gaussian DW-NMR analysesin real-time. REFERENCES: 1 Jensen JH, et al.

We have recently shown that the mouse brain contains distributed resting-state fMRI (rsfMRI) netw... more We have recently shown that the mouse brain contains distributed resting-state fMRI (rsfMRI) networks including a putative “default-mode” like network (DMN). To map the direction of information flow within this network, we have used Granger Causality (GC), a directed measure of functional connectivity, to describe the dominant direction of causation among four key nodes of the mouse DMN. We provide evidence of directional information transfer between nodes of this network that replicate patterns recently described for the human DMN, thus corroborating the presence of an evolutionary conserved DMN precursor in the rodent brain.

contributed equally to this work.

Pharmacological MRI (phMRI) methods have been widely applied to assess the central hemodynamic re... more Pharmacological MRI (phMRI) methods have been widely applied to assess the central hemodynamic response to pharmacological intervention as a surrogate for changes in the underlying neuronal activity. However, many psychoactive drugs can also affect cardiovascular parameters, including arterial blood pressure (BP). Abrupt changes in BP or the anesthetic agents used in preclinical phMRI may impair cerebral blood flow (CBF) autoregulation mechanisms, potentially introducing confounds in the phMRI response. Moreover, relative cerebral blood volume (rCBV), often measured in small-animal phMRI studies, may be sensitive to BP changes even in the presence of intact autoregulation. We applied laser Doppler flowmetry and MRI to measure changes in CBF and microvascular CBV induced by increasing doses of intravenous norepinephrine (NE) challenge in the halothane-anesthetized rat. NE is a potent vasopressor that does not cross the blood-brain barrier and mimics the rapid BP changes typically observed with acute drug challenges. We found that CBF autoregulation was maintained over a BP range of 60-120 mmHg. Under these conditions, no significant central rCBV responses were observed, suggesting that microvascular rCBV changes in response to abrupt changes in perfusion pressure are negligible within the autoregulatory range. Larger BP responses were accompanied by significant changes in both CBV and CBF that might confound the interpretation of phMRI results. D

Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention, 2014

There is a growing conviction that the understanding of the brain function can come through a dee... more There is a growing conviction that the understanding of the brain function can come through a deeper knowledge of the network connectivity between different brain areas. Resting state Functional Magnetic Resonance Imaging (rs-fMRI) is becoming one of the most important imaging modality widely used to understand network functionality. However, due to the variability at subject scale, mapping common networks across individuals is by now a real challenge. In this work we present a novel approach to group-wise community detection, i.e. identification of functional coherent sub-graphs across multiple subjects. This approach is based on a joint diagonalization of two or more graph Laplacians, aiming at finding a common eigenspace across individuals, over which clustering in fewer dimension can then be applied. This allows to identify common sub-networks across different graphs. We applied our method to rs-fMRI dataset of mouse brain finding most important sub-networks recently described i...

Frontiers in Neuroinformatics, 2015

† These authors have contributed equally to this work.

Purpose. The departure from purely mono-exponential decay of the Pulsed Field Gradient signal S(b... more Purpose. The departure from purely mono-exponential decay of the Pulsed Field Gradient signal S(b)=S(0)exp(-bD) as a function of the b-value observed in biological tissues, prompted the search for alternative models to characterize non-Gaussian dynamics of water diffusion. Several approaches have been proposed in the last years 1 . In this study we investigated the so-called anomalous diffusion (AD) stretched exponential γ-imaging model 2,3 which is based on fitting the stretched function S(b)=S(0)exp(−(bD) γ ) to PFG data obtained by varying the gradient strength g. Although γ is not a tensor, it is possible to use a "first order approximation approach in which γ is projected along the DTI main directions of diffusion 3 . As a consequence, invariant indices, such as the mean γ (Mγ) and the γ

While the fMRI response to excitatory somatosensory and pharmacological stimuli is well character... more While the fMRI response to excitatory somatosensory and pharmacological stimuli is well characterised, negative responses are more difficult to interpret and have been the subject of much debate in the recent literature . Here, we have employed a multimodal approach to investigate the individual components of the haemodynamic response to pharmacological challenge with Tiagabine, a GABAergic drug with anticonvulsant properties [2], in the anaesthetised rat. Tiagabine inhibits reuptake of GABA, the principal inhibitory neurotransmitter in the brain, thus inducing widespread cortical deactivation [3]. We have used MR methods, Laser Doppler Flowmetry, and fluorescence quenching methods to measure the changes induced by Tiagabine in cerebral blood volume, blood flow and tissue oxygen tension, respectively. Oxygen levels in the brain parenchyma are sensitive to the balance between oxygen supply and consumption, and provide important information to correlate haemodynamic changes with underlying changes in brain metabolism.

2013 International Workshop on Pattern Recognition in Neuroimaging

We present an unsupervised approach based on the Dominant Sets framework to automatically segment... more We present an unsupervised approach based on the Dominant Sets framework to automatically segment the white matter fibers into bundles. This framework, rooted in the Game Theory, allows for the automatic determination of the number of clusters from the data itself, without any prior assumption. The clustered bundles are a key information for the generation of unbiased structural connectivity atlases. We have thoroughly validated our algorithm both quantitatively and qualitatively. Indeed, we used biologically plausible synthetic datasets to numerically validate the performance in terms of Precision, Recall and other measures employed in the literature. We also evaluated the algorithm on a real Diffusion Tensor Imaging tractography of a whole mouse brain obtaining promising results. In fact, some of the most prominent brain structures determined by the algorithm correspond to white matter expected anatomy.

D-aspartate (D-Asp) is an atypical amino acid, which is especially abundant in the developing mam... more D-aspartate (D-Asp) is an atypical amino acid, which is especially abundant in the developing mammalian brain, and can bind to and activate N-methyl-D-Aspartate receptors (NMDARs). In line with its pharmacological features, we find that mice chronically treated with D-Asp show enhanced NMDAR-mediated miniature excitatory postsynaptic currents and basal cerebral blood volume in fronto-hippocampal areas. In addition, we show that both chronic administration of D-Asp and deletion of the gene coding for the catabolic enzyme D-aspartate oxidase (DDO) trigger plastic modifications of neuronal cytoarchitecture in the prefrontal cortex and CA1 subfield of the hippocampus and promote a cytochalasin D-sensitive form of synaptic plasticity in adult mouse brains. To translate these findings in humans and consistent with the experiments using Ddo gene targeting in animals, we performed a hierarchical stepwise translational genetic approach. Specifically, we investigated the association of variation in the gene coding for DDO with complex human prefrontal phenotypes. We demonstrate that genetic variation predicting reduced expression of DDO in postmortem human prefrontal cortex is mapped on greater prefrontal gray matter and activity during working memory as measured with MRI. In conclusion our results identify novel NMDAR-dependent effects of D-Asp on plasticity and physiology in rodents, which also map to prefrontal phenotypes in humans.

Pharmacological MRI (phMRI) experiments utilise fMRI time series methods to map the central effec... more Pharmacological MRI (phMRI) experiments utilise fMRI time series methods to map the central effect of pharmaceutical compounds. The typical univariate maps may, however, integrate the effects of several different neurotransmitter systems or underlying mechanisms. The results may thus be spatially and/or mechanistically nonspecific. Intersubject correlation analysis based on the phMRI response amplitude can more directly identify patterns of functional connectivity underlying the central effects of an acutely administered compound. In this article, we extend this approach to experiments where the effects of one compound in modulating the response to another are of interest. Specifically, we show a modulation of the correlation structure of a probe compound (d-amphetamine) by pretreatment with the selective dopamine D 3 receptor antagonist SB277011A in the rat. The strongest modifications in the correlation patterns occurred in connection with the ventral tegmental area, the source of mesolimbic dopamine projections and a key substrate in the reward system. D

Current Topics in Behavioral Neurosciences, 2011

Expert Opinion on Drug Discovery, 2012

Translational Psychiatry, 2015

Journal of Neuroscience Methods, Feb 15, 2010

The mechanisms underlying the signal changes observed with pharmacological magnetic resonance ima... more The mechanisms underlying the signal changes observed with pharmacological magnetic resonance imaging (phMRI) remain to be fully elucidated. In this study, we obtained microdialysis samples in situ at 5-min intervals during phMRI experiments using a blood pool contrast agent to correlate relative cerebral blood volume (rCBV) changes with changes in dopamine and cocaine concentrations following acute cocaine challenge (0.5 mg/kg iv) in the rat over a duration of 30 min. Three brain areas were investigated: the dorsal striatum (n = 8), the medial prefrontal cortex (mPFC; n = 5), and the primary motor cortex (n = 8). In the striatum and mPFC groups, cocaine and dopamine temporal profiles were tightly correlated, peaking during the first 5-min period postinjection, then rapidly decreasing. However, the local rCBV changes were uncorrelated and exhibited broader temporal profiles than those of cocaine and dopamine, attaining maximal response 5 -10 min later. This demonstrates that direct vasoactivity of dopamine is not the dominant component of the hemodynamic response in these regions. In the motor cortex group, microdialysis revealed no local change in dopamine in any of the animals, despite large local cocaine increase and strong rCBV response, indicating that the central hemodynamic response following acute iv cocaine challenge is not driven directly by local dopamine changes in the motor cortex. The combination of phMRI and in situ microdialysis promises to be of great value in elucidating the relationship between the phMRI response to psychoactive drugs and underlying neurochemical changes. D 2004 Elsevier Inc. All rights reserved.

contributed equally to this work.

Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention, 2014

Frontiers in Neuroinformatics, 2015

† These authors have contributed equally to this work.

2013 International Workshop on Pattern Recognition in Neuroimaging

Current Topics in Behavioral Neurosciences, 2011

Expert Opinion on Drug Discovery, 2012