Gut Microbiota Markers and Dietary Habits Associated with Extreme Longevity in Healthy Sardinian Centenarians

This study was aimed at characterizing the gut microbiota (GM) and its functional profile in two groups of Sardinian subjects with a long healthy life expectancy, overall named Long-Lived Subjects (LLS) [17 centenarians (CENT) and 29 nonagenarians (NON)] by comparing them to 46 healthy younger controls (CTLs). In addition, the contribution of genetics and environmental factors to the GM phenotype was assessed by comparing a subgroup of seven centenarian parents (CPAR) with a paired cohort of centenarians’ offspring (COFF). The analysis was performed through Next Generation Sequencing (NGS) of the V3 and V4 hypervariable region of the 16S rRNA gene on the MiSeq Illumina platform. The Verrucomicrobia phylum was identified as the main biomarker in CENT, together with its members Verrucomicrobiaceae, Akkermansia and Akkermansia muciniphila. In NON, the strongest associations concern Actinobacteria phylum, Bifidobacteriaceae and Bifidobacterium, while in CTLs were related to the Bacteroidetes phylum, Bacteroidaceae, Bacteroides and Bacteroides spp. Intestinal microbiota of CPAR and COFF did not differ significantly from each other. Significant correlations between bacterial taxa and clinical and lifestyle data, especially with Mediterranean diet adherence, were observed. We observed a harmonically balanced intestinal community structure in which the increase in taxa associated with intestinal health would limit and counteract the action of potentially pathogenic bacterial species in centenarians. The GM of long-lived individuals showed an intrinsic ability to adapt to changing environmental conditions, as confirmed by functional analysis. The GM analysis of centenarians’ offspring suggest that genetics and environmental factors act synergistically as a multifactorial cause in the modulation of GM towards a phenotype similar to that of centenarians, although these findings need to be confirmed by larger study cohorts and by prospective studies.


Introduction
The aging process is influenced and determined by complex interactions between genetic and environmental factors which, together with the stochastic process, can contribute to the attainment of longevity [1,2]. Evidence that relatives of people with a long lifespan are more likely to live longer and have a reduced risk of developing significant age-related diseases makes aging a potential therapeutic target [3][4][5].

Sampling
Stool samples from each subject were independently collected. The collection was carried out at home or in the host structures by the staff, using standard safety procedures. Transport was carried out by the staff and delivery was made to the laboratory within 3 h. Fresh samples were stored at −80 • C until further processing.

Total DNA Extraction from Fecal Sample and Quantification of Bacterial DNA
Genomic DNA was isolated from the fecal sample utilizing the QIAamp Fast DNA Stool Mini Kit following the manufacturer's instructions (Qiagen, Hilden, Germany). The concentration of the fecal bacterial DNA of each patient was quantified through realtime PCR (qPCR) on the genomic DNA samples, performed using degenerate primers encompassing the V3 and V4 hypervariable region of the bacterial 16S rRNA gene, as previously described [32].

16S Libraries Preparation and Sequencing
The protocol of library preparation and sequencing has been described in detail elsewhere [29]. 16S barcoded amplicon libraries were generated using primers targeting the V3-V4 hypervariable region of the bacterial 16S rRNA gene and the Nextera XT index kit (Illumina, inc., San Diego, CA, USA), and their size and quality were verified using Agilent DNA 1000 Analysis kit (Agilent Technologies, Santa Clara, CA, USA) on the Agilent 2100 Bioanalyzer system (Model G2939B, Agilent Technologies, Santa Clara, CA, USA). Genomic libraries were quantified with a Qubit 3.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA), normalized to a concentration equal to 4 nM, then pooled. The pooled library, and the adapter-ligated library PhiX v3 used as a control, were denatured and diluted to equal concentration (8 pM) and subsequently combined to obtain a PhiX concentration equal to 5% of the total volume. Combined 16S library and PhiX control were further denatured and sequenced on the MiSeq platform using MiSeq v3 Reagent Kit (Illumina).
Alpha diversity was assessed with the script alpha rarefaction.py in QIIME in order to obtain the Shannon index. Alpha diversity and Firmicutes/Bacteroidetes ratio were analyzed using the Kruskal-Wallis test followed by Bonferroni correction for multiple comparisons. Beta diversity was generated in R-vegan, using the Bray-Curtis distance. The Non-Metric Multidimensional Scaling (NMDS) based on the Bray-Curtis distance matrix was conducted in R software v.3.5.2 (ggplot2 package). The statistical significance of beta diversity among the groups was determined with Permutational Multivariate Analysis of Variance (PERMANOVA) (R-vegan, function adonis). The overall p-value obtained from multiple comparisons was confirmed through a pairwise PERMANOVA test performed in R (RVAdeMemoire package). The analysis at taxonomic levels was performed in SPSS software v. 28.0.1.0 (IBM, New York, NY, USA) using the Kruskal-Wallis test. Pairwise comparison was performed only for significant taxa, followed by Bonferroni correction for multiple comparisons. Only bacteria present in at least 25% of the samples and with a relative abundance of ≥0.1% in cases and/or controls were considered. The Linear Discriminant Analysis Effect Size (LEfSe) was additionally performed on statistically significant bacterial taxa obtained by the Kruskal-Wallis test and confirmed after Bonferroni adjustment. The LEfSe algorithm was performed on the Galaxy computational tool v. 1 [53] algorithm was performed on Galaxy software v.1.0. (https://galaxy.morganlangille. com/, accessed on 30 August 2021) to infer metagenome composition in the samples by analyzing OTUs generated by QIIME pipeline. Bacterial metabolic pathways were predicted and classified by Kyoto Encyclopedia of Genes and Genomes (KEGG) [54].
Statistical differences were analyzed for all metabolism pathways present in at least 25% of the samples and with a minimum abundance of 0.1% using Statistical Analysis of Metagenomic Profiles (STAMP) software [55]. The statistical significance was tested using Welch's test, with a Storey False Discovery Rate correction (FDR) correction. Overall, p ≤ 0.05 was considered statistically significant.

Clinical and Lifestyle Data of Subjects
Clinical characteristics of the study cohorts are shown in Tables 1 and 2. Overall, the study cohorts were rather homogeneous with each other regarding demographic, anthropometric and lifestyle data. Notably, the analysis of clinical data showed no statistically significant differences between CENT and NON in terms of gender, BMI, comorbidities and lifestyle factors, except for MMSE, ADL and PASE scores, while LLS groups diverged significantly from CTLs in some clinical factors, such as BMI, number of medications per day and comorbidities. Furthermore, the number of smokers was significantly higher in CTLs than in the NON group (p = 0.027) and NON presented a lower adherence to MD compared to CTLs, though the statistical value was not highly significant (p = 0.036). The CPAR and COFF cohorts were not significantly different in terms of gender, BMI and most lifestyle factors, except bowel function, number of medications, MMSE, ADL, PASE and MNA scores.

Alpha and Beta Diversity Analysis
The Kruskal-Wallis test showed statistically significant differences in the Shannon index across different study cohorts (p = 0.037), confirmed by pairwise testing only for the long-lived group comparison that showed an alpha diversity in CENT significantly lower than that in NON (CENT = 2.39 ± 0.31, NON = 2.46 ± 0.29, p = 0.020). Alpha diversity was higher in both LLS cohorts compared to controls, albeit not significantly (see Figure 1A and Supplementary Table S1 online). No statistically significant differences in the Shannon index between CPAR and COFF subgroups were observed (p = 0.398; see Supplementary Table S1 online). The Kruskal-Wallis test showed statistically significant differences in the Shannon index across different study cohorts (p = 0.037), confirmed by pairwise testing only for the long-lived group comparison that showed an alpha diversity in CENT significantly lower than that in NON (CENT = 2.39 ± 0.31, NON = 2.46 ± 0.29, p = 0.020). Alpha diversity was higher in both LLS cohorts compared to controls, albeit not significantly (see Figure 1A and Supplementary Table S1 online). No statistically significant differences in the Shannon index between CPAR and COFF subgroups were observed (p = 0.398; see Supplementary Table S1 online).
The Non-Metric Multidimensional Scaling (NMDS) based on the Bray-Curtis distance matrix showed a marked separation between the GM communities of LLS and CLTs (see Figure 1B and Supplementary Table S2 online), confirmed by PERMANOVA analysis, which indicated a significant difference in beta diversity between cohorts (sum of squares = 1.498, mean of squares = 0.749, F = 6.074, R = 0.1201, p = 0.001). Significant segregation persisted only in the comparison between CENT and CTLs (p = 0.006) and between NON and CTLs (p = 0.003) following the pairwise PERMANOVA test.
No statistically significant differences in beta diversity between CPAR and COFF subgroups were obtained (sum of squares = 0.08, mean of squares = 0.08, F = 0.706, R = 0.056, p = 1) (Supplementary Table S3     The Non-Metric Multidimensional Scaling (NMDS) based on the Bray-Curtis distance matrix showed a marked separation between the GM communities of LLS and CLTs (see Figure 1B and Supplementary Table S2 online), confirmed by PERMANOVA analysis, which indicated a significant difference in beta diversity between cohorts (sum of squares = 1.498, mean of squares = 0.749, F = 6.074, R = 0.1201, p = 0.001). Significant segregation persisted only in the comparison between CENT and CTLs (p = 0.006) and between NON and CTLs (p = 0.003) following the pairwise PERMANOVA test.

Compositional Analysis of the Gut Microbiota
Illumina MiSeq generated a mean value of 108,775 (+/− 16,901 SD) reads per patient. The Firmicutes/Bacteroidetes ratio was significantly higher in LLS compared to CTLs (see Table 3), while no statistical significance persisted when CPAR and COFF were compared (p = 0.499). 3.83 ± 4.28 The statistical significance was calculated by the non-parametric Mann-Whitney test in SPSS software v.28.0.1.0. Pairwise comparison was performed only for significant taxa, followed by Bonferroni correction for multiple comparisons. The Firmicutes/Bacteroidetes ratio was significantly higher in CENT and NON compared to CTLs (p = 0.015 and p = 0.022 respectively). No statistical significance persisted in the comparison between CPAR and COFF (p = 0.499). Bold values denote statistical significance (p ≤ 0.05). CENT = centenarian subjects, NON = nonagenarian subjects, CTLs = healthy younger controls, CPAR = centenarian parents, COFF = centenarians' offspring.
The Kruskal-Wallis test on GM composition between CENT, NON and CTLs showed 105 statistically significant results (see Table 4). Pairwise analysis showed 29 common significant differences in the two classes of LLS compared to CTLs, 24 significant divergences were found only from the comparison between CENT and CTLs, 41 only from the comparison between NON and CTLs, while the GM of CENT and NON differed significantly in 8 bacterial taxa (see Supplementary Tables S4-S7 online).    The Kruskal-Wallis test on GM composition between CPAR and COFF showed five statistically significant results, which did not maintain statistical significance after Bonferroni correction (see Table 5). Relative abundance differences of bacterial taxa between CPAR and COFF were performed by Kruskal-Wallis test on R software v.3.5.2 followed by Bonferroni correction for multiple comparisons. Bonferroni p equal to or less than 0.05 was considered statistically significant. CPAR = centenarian parents, COFF = centenarians' offspring, ↓ = significantly reduced in CPAR.
LEfSe plots of taxonomic biomarkers were generated on the Galaxy computational tool v.1.0. (https://huttenhower.sph.harvard.edu/galaxy/) accessed on 30 August 2021. Results were ranked by their Linear Discriminant Analysis (LDA) score. Blue bacterial taxa were more abundant in NON, green bacterial taxa were more abundant in CTLs, red bacterial taxa were more abundant in CENT. CENT = centenarian subjects, NON = nonagenarian subjects, CTLs = healthy younger controls.

Spearman Correlation between Gut Microbiota Alterations and Dietary, Lifestyle and Clinical Variables in CENT and Non
Taxa significantly associated with longevity were correlated with dietary, lifestyle and clinical parameters in both CENT and NON. Most taxa associated with CENT were correlated with MDS score and bowel function (see Figure 3A and Supplementary Table S8 online). As for the former, seven bacterial taxa were positively correlated, while six were negatively correlated. The taxa related to MDS score mainly concerned the Firmicutes phylum, such as Lactobacillus taiwanensis, Clostridiaceae and its members Clostridium and Dorea, Peptoniphilus, Thermicanaceae and Thermicanus, all of which were positively correlated, and Catenibacterium, Veillonella and Dialister invisus, which were all negatively correlated. Furthermore, Bacteroides rodentium and Parabacteroides merdae, which belong to Bacteroidetes phylum, and Eggerthella, which belongs to Actinobacteria phylum, were negatively and positively correlated with MDS, respectively. As for bowel function, eight bacterial taxa were positively correlated, and one was negatively correlated. The positively correlated taxa mostly concerned the Firmicutes phylum (Thermicanaceae, Thermicanus, E. inopinata), but also included Desulfohalobiaceae, Desulfonauticus, D. autotrophicus (Proteobacteria) and Synergistetes phylum with its Synergistaceae family. The taxa negatively correlated to bowel function belonged to the Firmicutes phylum (L. taiwanensis). Some bacterial taxa have frequently shown a correlation with several clinical variables. The Thermicanaceae family and related Thermicanus genus were also positively related to levels of current and former alcohol consumption, whereas Desulfovibrio was negatively correlated with MMSE and MNA scores and to former alcohol consumption; D. piger negatively correlated with ADL, MNA and PASE scores.
In NON, there was a greater number of correlations than in CENT (see Figure 3B and Supplementary Table S9 online), mainly related to the number of medications, MDS and PASE score. Taxa belonging to Actinobacteria phylum, such as Bifidobacterium, Bifidobacterium spp., Blautia wexlerae (Firmicutes) and Bacteroidaceae (Bacteroidetes phylum) were negatively correlated to the number of medications, while the Bacteroidetes phylum (and related Bacteroides, Bacteroides spp., Sphingobacterium shayense) and Alcaligenaceae and Sutterella (Proteobacteria phylum) showed a positive correlation. Members belonging to the Actinobacteria phylum were all negatively correlated to MDS scores (Bifidobacteriaceae, Bifidobacterium, Bifidobacterium spp., Streptomycetaceae), as were those belonging to Bacteroidetes (Odoribacteraceae, Bacteroides spp., Pedobacter kwangyangensis and Parabacteroides spp.) and to Firmicutes (Alkaliphilus, Clostridium frigoris, Lactobacillus ultunensis, Peptoniphilus). Several significant associations were found for other clinical variables, such as ADL and PASE scores. Taxa belonging to Proteobacteria phylum (Serratia, S. entomophila, Escherichia, E. albertii) and Eggerthella (Actinobacteria) were inversely related to ADL scores. In relation to PASE scores, Firmicutes members (Thermicanaceae, Thermicanus, Blautia wexlerae), on one hand, and Bacteroidetes and Proteobacteria members, on the other, showed a positive and negative correlation, respectively.

Functional Metagenome Prediction Analysis
A comparative prediction analysis of the functional metagenome was performed using PICRUSt. A total of three different significantly metabolic pathways were identified by comparing CENT and NON (see Figure 4A). In particular, the pathway related to the biosynthesis of secondary metabolites (tropane, piperidine and pyridine alkaloid) were most expressed in CENT, while the pathways related to lipid metabolism (Ether lipid metabolism) and amino acid metabolism (D-Arginine and D-ornithine metabolism) were enriched in NON. The comparative functional metagenome prediction between LLS groups and CTLs showed a common significative decrease in glycan degradation in both LLS groups and a significant increase in secretion systems and in signal transduction (twocomponent system) in the same subjects ( Figure 4B,C). In CENT the bacterial secretion system and the pathway of replication, recombination and reparation of proteins were also most expressed compared to CTLs; on the other hand, the metabolism of pyrimidine, amino and nucleotide sugar was reduced. A comparative prediction analysis of the functional metagenome was performed using PICRUSt. A total of three different significantly metabolic pathways were identified by comparing CENT and NON (see Figure 4A). In particular, the pathway related to the biosynthesis of secondary metabolites (tropane, piperidine and pyridine alkaloid) were most expressed in CENT, while the pathways related to lipid metabolism (Ether lipid metabolism) and amino acid metabolism (D-Arginine and D-ornithine metabolism) were enriched in NON. The comparative functional metagenome prediction between LLS groups and CTLs showed a common significative decrease in glycan degradation in both LLS groups and a significant increase in secretion systems and in signal transduction (twocomponent system) in the same subjects ( Figures 4B and C). In CENT the bacterial secretion system and the pathway of replication, recombination and reparation of proteins were also most expressed compared to CTLs; on the other hand, the metabolism of pyrimidine, amino and nucleotide sugar was reduced.
Comparing the functional metagenome prediction profile of NON and CTLs has shown that in the former, a significant decrease in starch and sucrose metabolism was observed, whereas in the latter, a reduction in transporters and ABC transporter pathways was found. No statistically significant differences in the functional metagenome comparing CPAR and COFF subgroups were observed.  Comparing the functional metagenome prediction profile of NON and CTLs has shown that in the former, a significant decrease in starch and sucrose metabolism was observed, whereas in the latter, a reduction in transporters and ABC transporter pathways was found. No statistically significant differences in the functional metagenome comparing CPAR and COFF subgroups were observed.

Discussion
The present study aimed at characterizing human GM and its functional profile in two groups of Sardinian subjects with long, healthy lifespans (17 CENT and 29 NON) by comparing them to 46 younger CTLs. In addition, the contribution of genetics and environmental factors to the GM phenotype was assessed by comparing a subgroup of centenarian parents (CPAR) with a paired cohort of centenarians' offspring (COFF). The analysis was performed through NGS of the V3 and V4 hypervariable regions of the 16S bacterial rRNA gene on the MiSeq Illumina platform.
The alpha diversity in CENT and NON was higher than in CTLs, although no significant difference in the Shannon index was observed. These data confirm previous studies' results, as the literature agrees on a greater alpha diversity being associated with aging among elderly and extremely elderly adults; however, findings are not always statistically significant [56,57].
Beta diversity analysis showed a significant dissimilarity between both groups of long-lived subjects (CENT and NON) compared to CTLs, as previously reported [19,43,58,59].
Our microbial diversity findings represent a strong indicator of GM implications in advanced aging, in agreement with previous hypotheses. In fact, greater alpha diversity in both CENT and NON cohorts compared to healthy controls reflects a rich and complex microbial ecosystem, indicative of an adaptable intestinal microbiota, capable of adapting to multiple environmental perturbations. Due to this peculiarity, high intestinal microbial diversity has been defined as an indicator of longevity [60].
We also observed a statistically significant reduction in the alpha diversity of CENT compared to NON. It should be pointed out that lower alpha diversity has been associated with poor cognitive function [61], in line with a significantly lower MMSE score in the CENT cohort compared to NON. Furthermore, although the CENT cohort represents a healthy population, during aging, and especially in advanced aging, a physiological reduction in gastrointestinal function and host immune response has been observed linked to the development of chronic low-grade inflammation [6]. In this regard, it has been observed that reduced alpha diversity is related to metabolic and inflammatory diseases [62,63]. Therefore, a reduction in alpha diversity in CENT compared to NON is not surprising.
At the taxonomic level, we observed a significant increase in the Firmicutes/Bacteroidetes ratio in both CENT and NON groups compared to CTLs, in contrast to previous studies on centenarian subjects with the same [43] or a different geographic origin [36] and also to studies on the elderly [9,64]. Notably, among our CENT, NON and CTLs cohorts, the relative abundance of Firmicutes was approximately the same and did not change significantly (40.54, 47.72 and 43.15, respectively; p = 0.387), while that of Bacteroidetes almost doubled in CTLs compared to CENT and NON groups (43.59, 24.74 and 25.28, respectively; p CENTvsCTLs = 0.015 p NONvsCTLs = 0.022). In other words, the higher Firmicutes/Bacteroidetes ratio in our long-living subjects compared to controls reflected a significant reduction in the relative abundance of the Bacteroidetes phylum, rather than an increase in that of Firmicutes.
Multilevel taxonomic analysis showed greater divergences between LLS and CTLs, while the GM composition between CENT and NON did not diverge considerably. Specifically, more statistically significant differences (n = 41 taxa) were found between NON and CTLs. Twenty-four taxa significantly diverged when comparing CENT to CTLs, while twenty-nine common taxa in both CENT and NON were significantly altered compared to CTLs. Lastly, only eight bacterial taxa were significantly altered when comparing CENT to NON. This demonstrates that the GMs of the older cohorts are more similar to each other than to the cohort of younger subjects.
LEfSe showed that the Verrucomicrobia phylum was identified as the main biomarker in CENT, together with its members Verrucomicrobiaceae, Akkermansia and Akkermansia muciniphila, as often reported in previous studies of centenarians [57,65,66], while one study observed an opposite trend [64]. In addition, Akkermansia has been reported to increase with aging in several studies [6]. Akkermansia muciniphila is a mucin degrading bacterium that resides in the human intestinal mucous layer and is able of promoting intestinal integrity due to its capacity for restoring mucous thickness and thus stimulating the mucous turnover rate [67,68]. It is considered a significant biomarker of intestinal homeostasis, as its depletion has been associated with many diseases such as inflammatory bowel diseases and metabolic disorders [69]. Several studies confirm its protective effects. A. muciniphila has been reported to increase anti-tumor responses during anti-programmed cell death protein 1 (PD-1) immunotherapy [70], improve metabolic status and clinical outcomes after a dietary intervention in overweight/obese adults [71] and have protective effects in diet-induced obesity [72,73]. A. muciniphila supplementation in patients with overweight/obesity has reduced inflammation marker levels and improved several metabolic parameters [74], while in animal models of diabetes and obesity, it has restored the integrity of the epithelial mucosa, improved glucose tolerance and improved metabolic parameters, such as endotoxemia and inflammation [75].
Other strongly associated taxa in CENT were the Synergistaceae family, which belong to the Synergistetes phylum, Eggerthella, Collinsella intestinalis and Bifidobacterium bifidum (Actinobacteria), Methanobrevibacter and Methanobrevibacter smithii (Euryarchaeota phylum), as well as Rikenellaceae and Prevotella within the Bacteroidetes phylum. These associations are consistent with previous studies in which an increase in the abundance of all these taxa, with the exception of Prevotella, was observed in centenarians compared to younger subjects [19,43,58,66].
To date, the metabolic role of most of the taxa mentioned above has not been characterized. However, the beneficial effect of some of them has been described. For instance, Bifidobacterium bifidum constitutes one of the most dominant taxa of human intestinal microbiota in healthy breast-fed infants [76] and it has been individuated as one of the most abundant Bifidobacteria species in Italian centenarians [77].
Bacteroides intestinalis is able to degrade complex arabinoxylans from dietary fibre with the consequent release of the beneficial ferulic acid metabolite. It has been demonstrated that cultured Bacteroides intestinalis in the presence of complex insoluble arabinoxylans enhances the Th1-type immune response in mice and exerts anti-inflammatory activity in dendritic cells under inflammatory conditions [78].
Phascolarctobacterium is an acetate/propionate-producer, whose increase, observed after treatment with berberine and metformin in high-fat diet-induced obesity in rats, has been hypothesized to contribute to the beneficial effects of these two drugs [79]. Phascolarctobacterium faecium exerted beneficial effects on the host in rat models with nonalcoholic fatty liver disease [80], and has been associated with the supplementation of cruciferous vegetables in a controlled fruit and vegetable-free diet [81].
Regarding the Euryarchaeota phylum, the Methanobacteriaceae family and its members, Methanobrevibacter and Methanobrevibacter smithii, were strongly associated with CENT. These data are in agreement with previous studies that reported a high abundance of Methanobrevibacter smithii and of Methanobrevibacter genus in the centenarian gut microbiota of Sardinian and Chinese subjects, respectively [43,82]. Methanobrevibacter smithii represents the most dominant methanogen in the human gut due to its ability to reduce CO 2 by using H 2 (or formate) [83]. Several studies have investigated the possible link between the presence of methanogens and some human diseases, such as colorectal cancers (CRC), inflammatory bowel disease (IBD), irritable bowel syndrome (IBS), obesity and constipation, although contradictory findings make them difficult to interpret [83]. However, the mutualistic activity of M. smithii and B. thetaiotaomicron, inoculated into a germ-free mice model, has shown to promote an increase in caloric intake from diets, in lipogenesis and in host fat [84]. Noteworthy is the relationship between methanogens and aging, based on the excreting methane breath test, which has been consistently observed [83]. This association has not yet been defined, but different hypotheses have been formulated. Among them, the possibility that methanogens are selected during aging for their insensitivity to most of the antibiotics used throughout a lifespan, their slower transit time observed in aging, which possibly contributes to their over-representation, and favorable environmental exposure to methanogens in extreme long-living subjects during their life compared to current adults (different dietary habits, exposure to livestock) [83].
In NON, the strongest associations observed concerned the Actinobacteria phylum and related members Bifidobacteriaceae, Bifidobacterium angulatum, Bifidobacterium asteroides, Bifidobacterium catenulatum, Bifidobacterium choerinum, Bifidobacterium indicum and Bifidobacterium kashiwanohense. It should be pointed out that the abundance of these taxa was elevated in NON compared with CTLs and underwent a slight decrease in extreme longevity (CENT) compared with NON, albeit still higher than in CTLs. Interestingly, several studies have demonstrated that Bifidobacteria and its species diversity are decreased in elderlies [85]. Consistent with our findings, higher proportions have been identified in successful aging compared to younger elderlies or younger adult [14,43,86,87]. A reduction in Bifidobacteria has been associated with impaired adhesion to the intestinal mucosa, but it remains to be clarified whether the cause is attributable to changes in the mucus structure in the microbiota of elderly subjects [85]. Furthermore, Bifidobacteria depletion has been correlated with enhanced susceptibility to infections and impaired intestinal activity [88].
A strong association in NON was also found for Lactobacillaceae, Lactobacillus and Lactobacillus spp., which all belong to the Firmicutes phylum.
The probiotic effect of Lactobacillus and Bifidobacterium taxa has been well documented [89]. It has been demonstrated that Lactobacillus spp. or Bifidobacterium spp. probiotic supplementation attenuates oxidative stress and inflammation and improves physiological parameters such as gut barrier function, learning and memory ability in aged mice [90]. In addition, improved immunity in elderly humans and aged mice has been observed [91].
The taxa Streptococcaceae, Streptococcus, Streptococcus spp., Veillonellaceae, Veillonella, Veillonella dispar, Veillonella atypica, Thermicanaceae, Thermicanus, Bacillales_Family X_Incertae Sedis (Firmicutes phylum), Desulfovibrionaceae, Desulfovibrio, Desulfovibrio piger (Proteobacteria) and Bacteroides fragilis (Bacteroidetes) were also strongly associated with NON. Veillonella species are known to use lactic acid as a source of carbon and energy and are believed to ferment the lactic acid produced by Streptococcus, derived from the fermenta-tion of carbohydrates [92], and positively correlates with the abundance of Streptococcus in irritable bowel syndrome (IBS) [93,94]. Furthermore, this bacterial genus has the ability to ferment organic acids through the production of propionic and acetic acids, carbon dioxide and hydrogen. The propionic acid produced by Veillonella potentially presents greater risks than benefits regarding the neurotoxic character linked to its accumulation [95,96].
As for Desulfovibrio bacteria, it has been demonstrated, using a Stress-Induced Premature Senescence Model of Bmi-1 Deficiency, that mice in whom this bacterial genus penetrated the epithelium underwent an induced TNF-α secretion by macrophages, which caused impairment of TNF-α-dependent intestinal barrier permeability and aging. Furthermore, Desulfovibrio, one of the predominant sulphate-reducing bacterial generates residing in the human gut, is capable of leading to the formation of hydrogen sulfide, which is toxic for intestinal epithelial cells and exerts a pro-inflammatory effect. In fact, its abundance has been correlated with IBD [97,98] and obesity [29].
The association with the Bacteroidetes phylum was in line with previous studies [19,77,99] and disagrees with the initial hypothesis concerning the increase in the abundance of Bacteroidetes in old age, and with the reduction of the Firmicutes/Bacteroidetes ratio in older adults [6,64]. It should be borne in mind that the harmonic balance between Firmicutes and Bacteroidetes phyla in the human microbiota can be indicative of good health, but it is subject to the influence of lifestyle factors. The significant reduction in Bacteroidetes in both CENT and NON subjects compared to CTLs could be explained by the significantly higher BMI in these cohorts than in controls, given that Bacteroidetes are known to positively correlate with a reduction in body fat [29]. Furthermore, the relative abundance of Bacteroidetes has been shown to be substantially accentuated as a consequent exercise intervention in an early obesity and NAFLD model and in controls, compared with corresponding untrained group [100]. It should be pointed out that our LLS cohorts, mainly CENT, had a low PASE score, which indicates a sedentary lifestyle, consistent with a reduction in the Bacteroidetes phylum in these subjects compared to controls.
Taxa significantly associated with longevity were correlated with dietary, lifestyle and clinical variables in both CENT and NON. Most taxa associated with CENT were correlated with DMS score and bowel function. As for the former, most of the bacteria belonged to the Firmicutes phylum. Specifically, taxa belonging to the Clostridiaceae (Clostridium), Lachnospiraceae (Dorea), Peptostreptococcaceae (Peptoniphilus) and Thermicanaceae (Thermicanus) families positively correlated with DMS scores. The Clostridiaceae family has been associated with increased dietary fibers in rodent models [101], responding to dietary carbohydrates. Interestingly, fiber represents one of the nutrients with a beneficial impact evaluated for the attribution of adherence score to the Mediterranean diet [50]. Clostridium species can ferment carbohydrates, proteins, organic acids, and other organics, and produce acetic acid, propionic acid, butyric acid (SCFAs) and some solvents, such as acetone and butanol. SCFAs and most of the metabolites they produce, such as bile acids (BAs), proteins and other metabolic substances, bring many benefits to gut health [102]. As for Dorea, its ability to produce SCFAs from vegetables has been reported [103]. In CENT, Catenibacterium correlated negatively with DMS scores. This bacterial genus belongs to the Erysipelotrichidae family (Firmicutes), which has been associated with high fat diets in humans and in rodent models [104,105] and with inflammation-related intestinal disease and metabolic disorders [106], although a subsequent study observed an increase related to their diets [107] in a group of Egyptian adolescents compared to US Children.
In regard to bowel function, eight bacterial taxa were positively correlated, and one was negatively correlated. The positively correlated taxa mostly belonged to the Firmicutes phylum (Thermicanaceae, Thermicanus, E. inopinata), but also included Desulfohalobiaceae, Desulfonauticus, D. autotrophicus (Proteobacteria) and the Synergistetes phylum, with its Synergistaceae family. The taxa negatively correlated with bowel function belonged to the Firmicutes phylum (L. taiwanensis). Further studies are needed to clarify the significance of these correlations with the intestinal function of the study subjects, given that, to our knowledge, the literature does not describe the physiological implications of these taxa on human health. Diet may mediate some of these correlations, as Thermicanaceae and Thermicanus are positively associated with both DM and bowel function.
In CENT, Desulfovibrio negatively correlated with MMSE and MNA scores and to former alcohol consumption. D. piger negatively correlated with ADL, MNA and PASE scores. This data is not surprising, given the pro-inflammatory implications of these taxa (discussed above).
In NON there was a greater number of correlations with DMS, to the number of medications and to PASE score. Members belonging to the Actinobacteria phylum were all negatively correlated to DMS score (Bifidobacteriaceae, Bifidobacterium, Bifidobacterium spp., Streptomycetaceae), as well as those belonging to Bacteroidetes (Bacteroides clarus) and Firmicutes (Lactobacillus ultunensis), whereas others belonging to Bacteroidetes (Odoribacteraceae, Bacteroides dorei, Pedobacter kwangyangensis and Parabacteroides spp.), Firmicutes (Alkaliphilus, Clostridium frigoris, Lactobacillus ultunensis, Peptoniphilus) and Proteobacteria (Bilophila and Bilophila wadsworthia) were positively correlated. Taxa belonging to the Actinobacteria phylum, such as Bifidobacterium, Bifidobacterium spp., Blautia wexlerae (Firmicutes) and Bacteroidaceae (Bacteroidetes phylum) were negatively correlated with the number of medications, while the Bacteroidetes phylum (and related Bacteroides, Bacteroides spp., Sphingobacterium shayense) and Alcaligenaceae and Sutterella (Proteobacteria phylum) showed positive correlations. Several significant associations have also been found for other clinical variables, such as ADL score and PASE scores. Taxa belonging to the Proteobacteria phylum (Serratia, S. entomophila, Escherichia, E. albertii) and Eggerthella (Actinobacteria) were inversely related to ADL scores; however, in relation to PASE score, Firmicutes members (Thermicanaceae, Thermicanus, Blautia wexlerae) on one hand, and Bacteroidetes and Proteobacteria members on the other, showed positive and negative correlations, respectively.
We performed a comparative prediction analysis of the functional metagenome using PICRUSt.
It was previously observed that older adults and long-lived subjects have reduced pathways related to carbohydrate metabolism and amino acid synthesis [6]. It should be pointed out that, with aging, dietary habits change due to a reduction in appetite, loss of teeth, decrease in gustatory perception and decreased efficiency of the digestive system, which results in a reduction in the absorption of essential nutrients [6]. In this regard, we observed a depletion in glycan metabolism in both CENT and NON, and a reduction of starch and sucrose metabolism (carbohydrate degradation related pathway) in NON. This result agrees with a previous study carried out on a group of subjects recruited in the same territory (Sardinia, Italy), in which a reduction of pathways related to carbohydrate degradation was observed compared to elderly and younger subjects [43]. In our study cohort, this finding is of particular interest in the light of the contextual reduction of the Bacteroidetes phylum in both CENT and NON. In fact, Bacteroidetes encode more carbohydrate-degrading enzymes than Firmicutes (more representative than Bacteroidetes in our long-lived cohorts) and possess a lesser number of ABC carbohydrate transporters. Genes encoding ABC transporters specific for glycans are often located adjacent to those encoding glycoside hydrolases (with which they are co-expressed) in Firmicutes but not in Bacteroidetes; this might be a glycan acquisition strategy that Firmicutes have evolved [108]. Furthermore, ABC transporters are involved in the transport of a variety of substrates, including nutrients, toxins, antibiotics and xenobiotics [109]. NON showed a significant increase in ABC transporter expression, which may be related to more frequent use of medications in this cohort.
Another noteworthy finding of our research is the significant increase in the twocomponent system pathway in both CENT and NON compared to CTLs, which points out the greater adaptability of the long-lived intestinal microbial ecosystem compared to that of younger subjects. Two-component signal transduction systems represent a means of communication through which bacteria perceive and respond to their environment, including stress conditions, nutrient availability, quorum signals, chemokines, pH and other factors [110]. It is a strategy developed by bacteria to adapt their cellular physiology to changes in the environment. The importance of such a sophisticated signaling mechanism justifies their prevalence throughout the bacterial kingdom [111].
Assuming that the microbial phenotypic patterns observed in our cohort of centenarians were peculiar and/or predisposing to the state of longevity, we hypothesized that they were not significantly divergent in terms of abundance from those of the paired centenarian's offspring cohort, considering both genetic and environmental effects as predisposing factors for the state of longevity. As expected, CPAR and COFF differed significantly only in five bacterial taxa, which lost statistical significance following Bonferroni's post-hoc correction. In CPAR, we have observed a reduction in taxa belonging to the Firmicutes phylum, such as Faecalibacterium, Faecalibacterium prausnitzii (Clostridiaceae) and Roseburia faecis (Lachnospiraceae), as well as a reduction in Bacteroides denticanum and Bacteroides plebeius, which belong to the Bacteroidetes phylum. Furthermore, no statistically significant differences in alpha and beta diversity, nor in metabolic function, between CPAR and COFF were observed. These findings were not surprising and seemed to confirm our hypothesis. The genetic make-up and environmental factors, such as diet, geographical environment, type of residence, modality of childbirth or type of breastfeeding, act synergistically as a multifactorial cause in the modulation of GM. In fact, it should be noted that all subjects of our COFF cohort were born through natural childbirth and were nursed with their mother's milk and all, except one, lived in their own home and maintained an identical diet to that of their parents until adulthood. Moreover, their diet remained similar until before the sampling of the study.

Conclusions
In conclusion, long-lived subjects were more similar to each other than to younger controls and the greatest divergences, in terms of microbial composition, emerged from the comparison between nonagenarians and controls. Nonagenarians showed an increase in both anti and pro-inflammatory bacterial taxa compared to younger subjects. This is not entirely surprising, given that these are subjects in an advanced aging phase, thus subjects in whom the likelihood of successful extreme aging, as we see in centenaries, is unknown. In this context, a prospective analysis of the nonagenarian population might be useful in order to understand what intestinal microbial pattern would predispose a subject to reach the age of a hundred.
Our population of centenarians diverged less from younger subjects in term of bacterial taxa compared to NON. Overall, the main biomarkers associated with centenarians belonged to the Verrucomicrobia phylum, including the Akkermansia muciniphila species, considered to be a significant biomarker of gut homeostasis for its ability to promote intestinal integrity; at the same time, there is a significant increase in taxa with an antiinflammatory phenotype, biomarkers of a state of health. This intestinal microbial ecosystem could guarantee intestinal health, which would then translate into the health of the whole organism.
The results deriving from the GM analysis of centenarians' offspring suggest that genetics and environmental factors act synergistically as a multifactorial cause in the modulation of GM towards a phenotype similar to that of centenarians, although these findings need to be confirmed by larger study cohorts and by prospective studies in order to clarify whether such microbial phenotypic patterns are predisposing factors in longevity.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/ 10.3390/nu14122436/s1. Table S1: Alpha diversity analysis between CENT, NON and CTLs and between CPAR and COFF; Table S2: GM beta diversity analysis between CENT, NON and CTLs;  Table S3: GM beta diversity analysis between CPAR and COFF; Table S4: 29 Statistically significant common differences in the relative abundance of bacterial taxa in the two classes of LLS compared to CTLs; Table S5: 24 Statistically significant divergences in the relative abundance of bacterial taxa observed only from the comparison between CENT and CTLs; Table S6: 41 Statistically significant divergences in the relative abundance of bacterial taxa observed only from the comparison between NON and CTLs; Table S7: 8 Statistically significant divergences in the relative abundance of bacterial taxa observed from the comparison between CENT and NON; Table S8: Spearman's correlation between significant taxa and dietary, lifestyle and clinical parameters in CENT; Table S9: Spearman's correlation between significant taxa and dietary, lifestyle and clinical parameters in NON. The anthropometric and lifestyle factor assessments can also be found in Supplementary Materials, in the sections "Anthropometric evaluation" and "Lifestyle factor assessment".
Funding: This research received no external funding.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committees of the University of Cagliari, Italy (PROT. NP/2020/4418, 28/10/2020).

Informed Consent Statement:
Informed consent was obtained from all subjects involved in the study.