Exploring the Role of Gut Microbiota in Major Depressive Disorder and in Treatment Resistance to Antidepressants

Major depressive disorder (MDD) is a common severe psychiatric illness, exhibiting sub-optimal response to existing pharmacological treatments. Although its etiopathogenesis is still not completely understood, recent findings suggest that an altered composition of the gut microbiota might play a role. Here we aimed to explore potential differences in the composition of the gut microbiota between patients with MDD and healthy controls (HC) and to identify possible signatures of treatment response by analyzing two groups of MDD patients characterized as treatment-resistant (TR) or responders (R) to antidepressants. Stool samples were collected from 34 MDD patients (8 TR, 19 R and 7 untreated) and 20 HC. Microbiota was characterized using the 16S metagenomic approach. A penalized logistic regression analysis algorithm was applied to identify bacterial populations that best discriminate the diagnostic groups. Statistically significant differences were identified for the families of Paenibacillaceae and Flavobacteriaceaea, for the genus Fenollaria, and the species Flintibacter butyricus, Christensenella timonensis, and Eisenbergiella massiliensis among others. The phyla Proteobacteria, Tenericutes and the family Peptostreptococcaceae were more abundant in TR, whereas the phylum Actinobacteria was enriched in R patients. Moreover, a number of bacteria only characterized the microbiota of TR patients, and many others were only detected in R. Our results confirm that dysbiosis is a hallmark of MDD and suggest that microbiota of TR patients significantly differs from responders to antidepressants. This finding further supports the relevance of an altered composition of the gut microbiota in the etiopathogenesis of MDD, suggesting a role in response to antidepressants.

symptoms severity [19]. Of interest they focused on response to antidepressants, but not on treatment resistance, finding that fecal bacterial α-diversity was increased in non-responsive but not in responsive MDD patients when compared to healthy controls [19]. Finally, Valles-Colomer et al. found that MDD corresponded to a higher prevalence of Bacteroides enterotype 2 [20]. However, little is known on the relationship between treatment resistance to antidepressants and microbiota variation. Studies have so far focused either on different phenotypes (clinical response to antidepressants) [19], or have been hampered by a limited sample size (n = 7 in [20]). Thus, it appears timely to investigate if and how composition of the gut microbiota might influence and correlate with treatment resistance in MDD.
In this scenario, our study had a two-fold aim. First, we explored differences in microbiota composition in patients affected by MDD compared to healthy controls (HC) (H 0 : microbiota MDD = microbiota HC ). Second, we tested whether a specific microbiota composition was associated with the presence of treatment resistance (TR) to antidepressants (H 0 : microbiota TR = microbiota R ).

Participants
Thirty-four patients affected by MDD were recruited at the community mental health center of the Section of Psychiatry of the Department of Medical Science and Public Health, University of Cagliari and University Hospital Agency of Cagliari and the Unit of Clinical Pharmacology, University Hospital Agency, Cagliari, Italy. In addition, twenty healthy control (HC) subjects were recruited through word of mouth among hospital staff, their families, and university students. Assessment procedures have been detailed in Manchia et al. [21]. Briefly, patients were included in the study if: (1) they had a diagnosis of MDD according to Diagnostic and statistical manual of mental disorders DSM-IV-TR (DSM IV-TR) [22] criteria; (2) were able to express a consent to participate formulated by signing the consent form; (3) were of age between 18 and 70 years old; (4) were in euthymic phases. Patients with (1) presence of acute infections; (2) presence of chronic autoimmune inflammatory conditions (e.g., rheumatoid arthritis, thyroiditis); (3) presence of eating disorders; (4) presence of post-traumatic stress disorder; (5) presence of current substance use disorders; (6) presence of neurological disorders; (7) presence of past traumatic brain injury; (8) presence of severe co-morbidities that may influence microbiota variation were excluded. The following exclusion criteria were applied to patients with MDD and HC: (1) use of antibiotics in the three months preceding the sampling procedure, (2) chronic use of probiotics. At the time of recruitment, patients were assessed by trained mental-health professionals (psychiatry residents or senior clinical staff). We collected detailed clinical information through direct interview as well as with a systematic assessment of existing medical records. Treatment resistance for MDD was defined according to the criteria of Souery et al. [23] consisting of the presence of a poor response to at least two adequate trials of different classes of antidepressants. The assessment of TR was based on a retrospective assessment of longitudinally collected information of the clinical course as well as on evaluation of treatment response patterns. All socio-demographic and clinical information, as well as dietary, lifestyle and smoking habits, were collected at the moment of recruitment when the stool specimen was also collected. Given the bidirectional relationship between gut microbiota and drug treatment (i.e., gut microbiota could modulate treatment response and, conversely, pharmacological treatments could impact on gut microbiota), and the need to identify the specific signatures associated with TR, we analyzed three subgroups of MDD patients (treatment responsive, TR and untreated at the moment of sampling). The study protocol was approved by the Ethics Committee of the University Hospital Agency of Cagliari (PG/2018/11693) on 5 September 2018. The study was conducted in accordance with the principles of good clinical practice, with the Declaration of Helsinki and in compliance with the national legislation. Written, informed consent was obtained from all participants.

Sample Collection and DNA Extraction
Collection tubes with a DNA stabilization buffer (Canvax Biotech, Cordoba, Spain) were filled with fresh stool by each participant. A total of 250 µL of each sample was used to perform DNA microbial extraction using the QIAamp DNA Stool Mini Kit (Qiagen, Milan, Italy) according to the manufacturer's protocol. After assessing DNA concentration and purity, samples were stored at −80 • until processing.

Next-Generation Sequencing of Bacterial 16S rRNA Gene
The V3-V4 region of 16S rRNA gene was amplified using specific primers selected from Klindworth et al. [24] with Illumina adapter sequences, followed by index PCR according to the Illumina MiSeq 16S Metagenomic Sequencing Library preparation protocol, as described elsewhere [25]. Libraries were purified, quantified on a Qubit 3.0 Fluorometer (Thermo Scientific, Milan, Italy), pooled and paired-end sequenced (2 × 300 cycles) on an Illumina MiSeq (San Diego, CA, USA) platform.

Statistical Analysis
Clinical characteristics of patients with MDD and HC were reported as median along with interquartile range (i.e., first-third quartiles) and observed frequencies (and percentages) for continuous and categorical variables, respectively. For each continuous variable, the assumption of normality distribution was checked by means of quantile-quantile (Q-Q) plots and Shapiro-Wilks test. In the presence of non-normal distributions, comparisons between groups were performed by Mann-Whitney U test (or Kruskal-Wallis test as appropriate) and χ 2 test (or Fisher exact test, as appropriate) for continuous and categorical variables, respectively. Stacked bar charts were used to show the gut microbiota composition (i.e., mean relative abundance %) at phylum, family, genus and species levels between MDD and HC. To identify pathways of bacterial populations that best discriminated groups (i.e., MDD versus HC or comparisons among MDD subgroups according to presence/absence of TR to antidepressants), we applied the penalized logistic regression analysis (PELORA) algorithm [26]. This algorithm is mainly used to find predictive gene signatures from microarray data by using supervised grouping techniques. To this purpose, the relative abundance (%) of each bacterium was first logistic transformed (i.e., by calculating the natural logarithm of the ratio between the relative abundance proportion and its complimentary) and then standardized (computing a Z-score) by subtracting its mean and dividing by its standard deviation (SD). Both mean and SD were computed in the sample which included all the subjects involved in the comparison. When the relative abundance was exactly 0%, the logistic transformation could not be performed for that value and, to overcome this issue, such percentage was replaced by 0.001% for the computation of Z-score only. Once a pattern was identified, its centroid was computed by the mean of the Z-scores of the involved bacteria. To calculate centroids, Z-scores of some bacteria could be sign-flipped (reversed) in order to put their values in the same direction suggested by the centroid. The PELORA algorithm was also set to accommodate clinical variables: when a new predictor is added to the model, this can either be a group centroid or a clinical variable, depending on which yields better predictive value [26]. In detail, when comparing patients with MDD versus HC, penalized logistic models which included centroid as predictor were adjusted for the effect of age at the sample collection, gender and body mass index (BMI), whereas, when comparing subgroups of patients with MDD, models were adjusted for the effect of age at MDD onset, illness duration, gender, BMI, treatment duration and the presence of concomitant drugs. Moreover, when comparing patients with MDD versus HC, covariates related to lifestyle (i.e., diet, smoke and drink habits or presence of physical activity) were not considered because they were intrinsically related to the HC profile. In accordance with the analysis protocol, two different free parameters were set in the PELORA algorithm: the number of centroids and the penalty parameter (λ). The number of centroids was set to 1, because we were mainly interested to detect only one informative pathway for each scenario, whereas a number of different combinations of λ = 0, 1/32, 1/16, 1/8, 1/4, 1/2, 1 were evaluated, performing 200 bootstrap resampling of data and recording the overall misclassification rate. For each specific scenario, the penalty parameter that achieved the lowest median misclassification rate (across the bootstrap samples) was chosen. Comparisons between Z-scores were performed using the two-sample t-test. Heatmaps of normalized Z-scores (from 0 to 1) of relative abundance of bacterial populations identified by the PELORA algorithm along with the corresponding centroid and boxplots of centroid Z-scores were produced. Two-sided p < 0.05 was set as the statistically significance threshold. All statistical analyses and plots were performed by the computing environment R (version 3.6, R Core Team (2020). R: A language and environment for statistical computing, Vienna, Austria), [27] (packages: supclust, ggplot2, gridExtra).

Sample Characteristics
Since it is known that gut bacteria may influence drug metabolism, possibly affecting response to treatments and, conversely, that pharmacological treatments might alter gut microbiota composition, the study participants affected by MDD were classified in three subgroups, according to whether they had treatment-resistance to antidepressants or not or were untreated at the time of enrollment. Clinical and demographic characteristics of these three subgroups of MDD patients as well as of HC are summarized in Table 1. The four groups were homogeneous for all the examined characteristics except for gender distribution (p = 0.036), physical activity (p = 0.032), history of suicide attempt (p = 0.014) and intake of any concomitant drugs other than antipsychotics (p = 0.011). Two untreated MDD patients had a history of treatment with antipsychotics and/or mood stabilizers, but treatments were stopped a minimum of six months before the sampling procedure.

Comparison of Gut Microbiota Composition between Patients with MDD and HC
In order to assess whether a different gut microbiota discriminates MDD patients and HC, its composition in the two cohorts of subjects was analyzed by next generation sequencing. A total of 8,135,346 quality-filtered read pairs were obtained from 54 study participants (34 MDD patients and 20 HC), with an average of 150,654 read pairs per sample. Gut bacterial communities at the phylum, family, genus and species level detected in MDD and HC subjects are represented in Figure 1.
As expected, Firmicutes and Bacteroidetes were the most abundant phyla, accounting for about 90% of all bacteria in both groups. Based on the relative abundances generated by taxonomic analyses, the PELORA algorithm was used first to identify patterns of bacterial populations that best discriminate all patients with MDD from HC. The results are listed in Table 2 and graphically represented by the heatmap in Figure 2, showing slight variations at the phylum level, but more pronounced differences deeper in the taxonomic scale, with Paenibacillaceae detected in MDD but absent in HC and vice versa for the family of Flavobacteriaceaea, the genus Fenollaria, the species Flintibacter butyricus, Christensenella timonensis, Eisenbergiella massiliensis, Pseudoflavonifractor capillosus, Fenollaria timonenis, Robinsoniella sp. MCWD5 and Clostridum citroniae.
In Figure 3 the distribution of centroid Z-scores computed by PELORA in MDD patients and HC shows that the identified bacterial patterns were able to discriminate between the two groups, with a higher discriminatory power at family, genus and species levels.

Comparison of Gut Microbiota Composition between Patients with MDD and HC
In order to assess whether a different gut microbiota discriminates MDD patients and HC, its composition in the two cohorts of subjects was analyzed by next generation sequencing. A total of 8,135,346 quality-filtered read pairs were obtained from 54 study participants (34 MDD patients and 20 HC), with an average of 150,654 read pairs per sample. Gut bacterial communities at the phylum, family, genus and species level detected in MDD and HC subjects are represented in Figure 1.    In Figure 3 the distribution of centroid Z-scores computed by PELORA in MDD patients and HC shows that the identified bacterial patterns were able to discriminate between the two groups, with a higher discriminatory power at family, genus and species levels.

Comparison of Gut Microbiota Composition between Treated and Untreated Patients with MDD
As a second step, we investigated the eventual difference in gut microbiota in treated and untreated patients with MDD. A comparison was performed from which a conspicuous pattern of differentially represented taxa emerged, as reported in Table 3 and graphically represented by the heatmap in Figure 4. Proteobacteria, Propionibacteriaceae, Peptococcaceae, Murimonas, Murimonas intestini, Parabacteroides sp J1502 were increased in treated subjects; Candidatus Saccharibacteria, Lentisphaerae, Euryarchaeota, Acidaminococcaceae, Micrococcaceae, Fusibacteriaceae, Victivallaceae, Figure 3. Boxplots of centroid Z-scores, computed by the penalized logistic regression analysis algorithm, which discriminated all patients with major depressive disorder (MDD) from healthy controls (HC).

Comparison of Gut Microbiota Composition between Treated and Untreated Patients with MDD
As a second step, we investigated the eventual difference in gut microbiota in treated and untreated patients with MDD. A comparison was performed from which a conspicuous pattern of differentially represented taxa emerged, as reported in Table 3 and graphically represented by the heatmap in Figure 4. Proteobacteria, Propionibacteriaceae, Peptococcaceae, Murimonas, Murimonas intestini, Parabacteroides sp J1502 were increased in treated subjects; Candidatus Saccharibacteria, Lentisphaerae, Euryarchaeota, Acidaminococcaceae, Micrococcaceae, Fusibacteriaceae, Victivallaceae, Eggerthellaceae, Methanobacteriaceae, Sanguibacteroides, Phascolarctobacterium, Anaeromassilibacillus, Streptomyces, Raoultibacter, Denitrobacterium, Prevotella sporal clone IK062, Ruminococcus torques, Sanguibacteroides justesenii, Flintibacter butyricus, Roseburia intestinalis and Dialister sp S7D were instead more represented in untreated patients; Elusimirobia, Dakarella and Desulfovibrio fairfieldensis were solely detected in the treated group.  Abbreviations: IQR: interquartile range (i.e., first-third quartiles); SD: standard deviation; Absent: all values are 0%. • Standardized Z-score: the relative abundance of each bacterium was first logistic transformed, and then the Z-score was calculated subtracting its mean and dividing by its standard deviation (SD). Both the mean and SD were computed in the sample which included all MDD treated at sample collection and untreated MDD. Centroid is computed by the mean of Z-scores; * to calculate the centroid, the sign of the specific bacterial's Z-score was reversed; # p-values from two-sample t-test on Z-scores; § p-values from Mann-Whitney U test, calculated in the presence of no variance in one of the two groups.

Comparison of Gut Microbiota Composition between MDD Patients with and Without TR
We then evaluated the gut microbiota profile of patients classified as responders (R) and (TR). The subsequent analysis included only antidepressant-treated patients characterized for the presence/absence of TR at the time of the enrollment. As reported in Table 4, Proteobacteria, Tenericutes and Peptostreptococcaceae were more abundant in TR patients, whereas Actinobacteria were enriched in responders. Moreover, a number of bacteria characterized exclusively the microbiota of TR MDD patients (Thaumarchaeota, Yersinia and its species Yersinia pseudotuberculosis, Peptococcus, Fenollaria timonensis, Blautia sp. canine oral taxon 337, Papillibacter cinnamivorans), and many other were only detected in non-resistant patients (Candidatus Saccharibacteria, Planctomycetes, Bacillus, Candidatus Soleaferrea, Intestinibacillus, Porphyromonas, Robinsoniella sp. MCWD5 and Massilioclostridium coli).  Abbreviations: IQR: interquartile range (i.e., first-third quartiles); SD: standard deviation; Absent: all values are 0%. • Standardized Z-score: the relative abundance of each bacterium was first logistic transformed, and then the Z-score was calculated subtracting its mean and dividing by its standard deviation (SD). Both the mean and SD were computed in the sample which included all patients with and w/o TR. Centroid is computed by the mean of Z-scores; * to calculate the centroid, the sign of the specific bacterial's Z-score was reversed; # p-values from two-sample t-test on Z-scores; § p-values from Mann-Whitney U test, calculated in the presence of no variance in one of the two groups.

Comparison of Gut Microbiota Composition between MDD Patients with and without TR and HC
As listed in Table 5, the presence of Elusimicrobia, Flavobacteriaceae, Fenollaria and Robinsoniella sp. MCWD5 was found exclusively in responder MDD patients, the presence of Nitrospirae and Peptostreptococcaceae only in HC and the enrichment of Proteobacteria in HC with respect to responders emerging as the bacterial pattern best discriminating between these two groups.
On the other hand, as reported in Table 6, the exclusive detection of Flavobacteriaceae, Hungatella, Yersinia, Citrobacter, Fenollaria and Fenollaria timonensis in patients with treatment-resistant MDD and the exclusive detection of Candidatus Saccharibacteria and Massilioclostridium coli in HC were the bacterial patterns distinguishing these groups. • Standardized Z-score: the relative abundance of each bacterium was first logistic transformed, and then the Z-score was calculated subtracting its mean and dividing by its standard deviation (SD). Both the mean and SD were computed in the sample which included all R patients and HC. Centroid is computed by the mean of Z-scores; * to calculate the centroid, the sign of the specific bacterium's Z-score was reversed; # p-values from two-sample t-test on Z-scores; § p-values from Mann-Whitney U test, calculated in the presence of no variance in one of the two groups. Abbreviations: IQR: interquartile range (i.e., first-third quartiles); SD: standard deviation; Absent: all values are 0%. • Standardized Z-score: the relative abundance of each bacterium was first logistic transformed, and then the Z-score was calculated subtracting its mean and dividing by its standard deviation (SD). Both the mean and SD were computed in the sample which included all TR patients and HC. Centroid is computed by the mean of Z-scores; * to calculate the centroid, the sign of the specific bacterium's Z-score was reversed; # p-values from two-sample t-test on Z-scores; § p-values from Mann-Whitney U test, calculated in the presence of no variance in one of the two groups.

Discussion
The identification of specific changes in microbiota in relation to illness status and the presence of TR to antidepressants should be interpreted in the context of several limitations. First, due to the cross-sectional design, our study was not able to establish causality (i.e., whether microbiota variation resulted from or preceded the onset of MDD). Second, the relatively small sample size could have impacted on the statistical power leading to a decreased sensitivity and specificity of our findings. Thus, we shall consider our study as hypothesis generator, capable of clarifying the strength of the association between microbiota variation and MDD and/or TR. In any instance, it should be noted that most of the available evidence in the literature has been gathered in samples of comparable size. Third, although the definition of TR was based on an accurate assessment of longitudinally collected clinical information, its assessment was mainly retrospective. However, given that TR was based on the analysis of sensible and validated psychometrics measures such as the Hamilton Depression Rating Scale [28], collected at each follow-up, it is reasonable to assume that this might have been relevant in terms of underrepresentation of TR cases, rather than of the misidentification of responsive patients as TR. Finally, although information on diet was accurately collected at the moment of stool specimen sampling, it is not possible to establish reliably how consistent were dietary habits among our patients.
However, even in light of these limitations, our study was able to identify the following relevant findings: (a) a statistically significant overrepresentation of Paenibacillaceae in MDD compared to HC, while the opposite was found for the family of Flavobacteriaceaea, the genus Fenollaria, the species Flintibacter butyricus, Christensenella timonensis, Eisenbergiella massiliensis, Pseudoflavonifractor capillosus, Fenollaria timonenis, Robinsoniella sp. MCWD5 and Clostridum citroniae; (b) a substantial pattern of bacterial taxa was differentially represented in the comparison between untreated and treated MDD patients; (c) a number of bacteria were identifiable in the microbiota of TR MDD patients (Thaumarchaeota, Yersinia and its species Yersinia pseudotuberculosis, Peptococcus, Fenollaria timonensis, Blautia sp. canine oral taxon 337, Papillibacter cinnamivorans) but not in responsive patients; (d) compared to HC, Flavobacteriaceae, Hungatella, Yersinia, Citrobacter, Fenollaria and Fenollaria timonensis were identified exclusively in TR MDD patients, while Elusimicrobia, Flavobacteriaceae, Fenollaria and Robinsoniella sp. MCWD5 were found exclusively in responsive MDD patients. These findings should be discussed in the context of the existing literature. There is increasing evidence of a link between gut microbiota imbalance and neurological disorders, including mood disturbances such as depression [29,30]. Supporting this link are the observations that MDD patients often exhibit metabolic and gastrointestinal symptoms [29,31] and that some antidepressants exhibit antimicrobial properties, while some antibiotics produce antidepressant effects [32].
According to the monoamine deficiency hypothesis, the main pathogenetic mechanism underlying MDD would be a shortage of monoaminergic neurotransmitters serotonin, norepinephrine or dopamine [33,34]. Consistently, the vast majority of marketed antidepressants impact the monoamine neurotransmission, with serotonin being the main target [34]. Serotonin derives from the essential amino acid tryptophan, which is introduced with diet. Once in the gut, about 90% of tryptophan enters the kynurenine pathway for the production of nicotinamide adenine dinucleotide, about 4-6% is metabolized by gut microbiota to produce indole and its derivatives, and only 3% is made available for serotonin synthesis [35][36][37]. Bacteria not only degrade tryptophan to produce indole compounds but also regulate the activity of rate-limiting enzymes in the kynurenine pathway [35][36][37], evidence that explains why gut microbiota might affect tryptophan availability for serotonin production in the brain [38].
Previous observational studies have examined the gut microbiota in patients suffering from MDD [19,[39][40][41][42][43]. The novelty of our study lies in two main aspects: (1) the accurate characterization of TR cases and their numerosity leading to what is, to date, the largest sample size for microbiota analysis; and (2) the application of the PELORA algorithm to find out the most discriminative bacterial patterns at different taxonomic levels in our study population. First, we compared gut microbiota of all MDD patients with that of HC. Interestingly, we observed that all the microorganisms identified by the algorithm at the genus and species level, which were only detected in patients but not in HC, belonged to the class of Clostridia. Since among other gut bacteria, Clostridia are able to degrade tryptophan to tryptamine in the indole pathway [44], we speculate that they could divert tryptophan from serotonin production in the brain of MDD patients. Secondly, we distinguished MDD patients according to whether they were or not under pharmacological treatment at the moment of sample collection, and we observed a conspicuous number of bacteria discriminating between these two groups, thus supporting previous findings that antidepressant drugs may induce gut microbiota variations [32,45]. Some of these drug-induced changes could be potentially disadvantageous, such as the decrease in butyrate producing bacteria like Ruminococcus torques, Flintibacter butyricus and Roseburia intestinalis or the drop in Phascolarctobacterium, previously found to be correlated with positive mood [46]. Thirdly, we focused on patients with MDD who had developed TR during their illness course. Again, the relationship between antidepressants (and resistance to them) and microbiota appears to be bidirectional. If, on the one hand, antidepressant medications affect microbiota composition [32,45], on the other hand data suggest that the gut microbiota phenotype may modulate their efficacy [19,47,48]. Of note, we found that Proteobacteria were considerably increased in TR patients, in agreement with previous finding by Jiang et al., who observed this phylum enriched in active, non-responsive MDD patients compared with healthy subjects [19]. The phylum of Proteobacteria includes intestinal pathogenic microorganisms belonging to Enterobacteriaceae, such as Yersinia and Citrobacter, which we only found in TR patients but not in responders nor in HC. Intriguingly, previous studies reported that gastrointestinal infections caused by pathogenic bacteria, including Citrobacter, elicit anxiety-like behaviors [49][50][51], thereby supporting the hypothesis that these microorganisms could contribute to the failure of antidepressant trials. Striking is also the case of the genus bacterial genus Bacillus, which was detected in HC (data not shown), in all MDD responders, but in none of the TR MDD patients (Table 4). Bacillus species are able to produce norepinephrine and dopamine [52], which may contribute to restore dysregulated levels of these neurotransmitters in the brain of MDD subjects.
Overall, the results of our study confirm that gut dysbiosis is a feature of MDD patients and that the microbiota of TR subjects significantly differs from responders to antidepressants, suggesting a role of gut microbiota in the etiopathogenesis of the disease as well as in its response to therapies. Elucidating the role of microbiota in MDD pathogenesis might pave the way to a possible use of gut bacterial profiles as disease biomarkers. In addition, the identification of microorganisms discriminating treatment-resistant from responsive patients could lead to set up specific interventions of microbiota manipulation in order to improve the clinical efficacy of the current antidepressant therapies. Funding: This research was partly funded with the "Ricerca Corrente RC2018-2020" funding granted by the Italian Ministry of Health to VP and by the grant "Fondo di Finanziamento per le Attività Base di Ricerca (FFABR)", call 2017, assigned to AS by Ministero dell'Istruzione, Ministero dell'Università e della Ricerca (MIUR), Italy.