A Data-Mining Approach to Identify NF-kB-Responsive microRNAs in Tissues Involved in Inflammatory Processes: Potential Relevance in Age-Related Diseases

The nuclear factor NF-kB is the master transcription factor in the inflammatory process by modulating the expression of pro-inflammatory genes. However, an additional level of complexity is the ability to promote the transcriptional activation of post-transcriptional modulators of gene expression as non-coding RNA (i.e., miRNAs). While NF-kB’s role in inflammation-associated gene expression has been extensively investigated, the interplay between NF-kB and genes coding for miRNAs still deserves investigation. To identify miRNAs with potential NF-kB binding sites in their transcription start site, we predicted miRNA promoters by an in silico analysis using the PROmiRNA software, which allowed us to score the genomic region’s propensity to be miRNA cis-regulatory elements. A list of 722 human miRNAs was generated, of which 399 were expressed in at least one tissue involved in the inflammatory processes. The selection of “high-confidence” hairpins in miRbase identified 68 mature miRNAs, most of them previously identified as inflammamiRs. The identification of targeted pathways/diseases highlighted their involvement in the most common age-related diseases. Overall, our results reinforce the hypothesis that persistent activation of NF-kB could unbalance the transcription of specific inflammamiRNAs. The identification of such miRNAs could be of diagnostic/prognostic/therapeutic relevance for the most common inflammatory-related and age-related diseases.


Introduction
The nuclear factor (NF)-kB is a transcription factor (TF) activated by an evolutionarily conserved inflammatory signaling, induced by a wide range of external and internal danger signals [1][2][3]. The complex modulation of this signaling can be envisaged considering the different activation strategies, well known as "canonical" and "non-canonical" NF-kB activation signaling (reviewed in [4,5]). A fine-tuning activation of NF-kB promotes the expression of pro-inflammatory genes and participates in the regulation of survival, activation, and differentiation of innate immune cells and T cells [6]. On the contrary, a persistent activation of NF-kB signaling was described in conditions of cellular senescence
We analyzed primarily genome-wide PROmiRNA predictions, as well as TF-binding site predictions as reported in [23], to identify miRNAs with potential NF-kB binding sites in their promoter sequences. PROmiRNA is a miRNA promoter recognition method, based on a semi-supervised statistical model trained on multi-tissue deepCAGE FANTOM4 libraries and other sequence features. It is tailored to score the potential of CAGE-enriched genomic regions to be promoters of either intergenic or intragenic miRNAs, thereby modulating miRNA expression in a tissue-specific manner [23]. To identify the TFs that regulate specific miRNAs, for each predicted miRNA transcription start site (TSS), we retrieved the 1 kb We analyzed primarily genome-wide PROmiRNA predictions, as well as TF-binding site predictions as reported in [23], to identify miRNAs with potential NF-kB binding sites in their promoter sequences. PROmiRNA is a miRNA promoter recognition method, based on a semi-supervised statistical model trained on multi-tissue deepCAGE FANTOM4 libraries and other sequence features. It is tailored to score the potential of CAGE-enriched genomic regions to be promoters of either intergenic or intragenic miRNAs, thereby modulating miRNA expression in a tissue-specific manner [23]. To identify the TFs that regulate specific miRNAs, for each predicted miRNA transcription start site (TSS), we retrieved the 1 kb centered on it and used the TRAP approach [29] to compute the affinity of TF binding sites for all predicted miRNA promoters using TF models stored in the JASPAR database [30].
NF-kB appears among the first 10 TFs with the highest affinity for the 1000 bp-long region surrounding the predicted TSSs for 722 miRNA hairpin precursors (Table S1).
Since tissues show specific miRNA expression patterns, we aimed to highlight the list of putative NF-kB-responsive miRNAs expressed in tissues strictly involved in the modulation of the inflammatory processes, including inflammaging. To achieve this goal, we focused our subsequent research on those miRNAs transcribed in human tissues such as "T cells", "T cells 2", "monocytic-cells", "immune system cells", "bone marrow", "blood", and "liver". Only the libraries relative to healthy tissues have been taken into consideration. This approach retrieved 399 miRNA hairpin precursors showing "expression at the promoter level" in at least one of these tissues (Table S2). In general, this is a good indication that the mature forms of these miRNAs are expressed in a specific tissue. However, each step from DNA-RNA transcription to mature miRNA expression can be modulated, thereby modifying or blocking the final expression. Moreover, FANTOM4 libraries are characterized by a certain level of "transcriptional noise", so we should expect false positives in mature miRNA predictions [23]. Therefore, among these putative NF-kB responsive miRNAs, we selected the "high confidence" hairpins in miRbase [27], retrieving 73 pre-miRNAs (Table S3). A growing body of evidence suggests that mature sequences derived from both arms of the hairpin might be biologically functional and even that the dominant mature sequence can be processed from opposite arms [31,32]. Following the approach of selecting only the "high confidence" miRNA hairpins and filtering the dataset for "Human Expression dataset" [28], 68 "high confidence" expressed miRNAs were identified. This pool of miRNAs, reported in Table  1, constitutes our final set of putative NF-kB responsive miRNAs expressed in healthy tissues linked to inflammatory processes. NF-kB appears among the first 10 TFs with the highest affinity for the 1000 bp-long region surrounding the predicted TSSs for 722 miRNA hairpin precursors (Table S1).
Since tissues show specific miRNA expression patterns, we aimed to highlight the list of putative NF-kB-responsive miRNAs expressed in tissues strictly involved in the modulation of the inflammatory processes, including inflammaging. To achieve this goal, we focused our subsequent research on those miRNAs transcribed in human tissues such as "T cells", "T cells 2", "monocytic-cells", "immune system cells", "bone marrow", "blood", and "liver". Only the libraries relative to healthy tissues have been taken into consideration. This approach retrieved 399 miRNA hairpin precursors showing "expression at the promoter level" in at least one of these tissues (Table S2). In general, this is a good indication that the mature forms of these miRNAs are expressed in a specific tissue. However, each step from DNA-RNA transcription to mature miRNA expression can be modulated, thereby modifying or blocking the final expression. Moreover, FANTOM4 libraries are characterized by a certain level of "transcriptional noise", so we should expect false positives in mature miRNA predictions [23]. Therefore, among these putative NF-kB responsive miRNAs, we selected the "high confidence" hairpins in miRbase [27], retrieving 73 pre-miRNAs (Table S3). A growing body of evidence suggests that mature sequences derived from both arms of the hairpin might be biologically functional and even that the dominant mature sequence can be processed from opposite arms [31,32]. Following the approach of selecting only the "high confidence" miRNA hairpins and filtering the dataset for "Human Expression dataset" [28], 68 "high confidence" expressed miRNAs were identified. This pool of miRNAs, reported in Table 1, constitutes our final set of putative NF-kB responsive miRNAs expressed in healthy tissues linked to inflammatory processes.

Genomic Features of Putative NF-kB Responsive miRNAs
According to their genomic location, it is possible to distinguish two classes of miRNAs: "intergenic miRNAs" are those located in intergenic regions of the genome, whereas "intragenic miRNAs" are those embedded in introns or exons of annotated genes [23]. Among the latter, "intronic miRNAs" are those located inside the introns of other genes and can either be co-transcribed with their host gene [33] or have an independent promoter [34][35][36], whereas intergenic miRNAs can derive from a primary miRNA transcript (pri-miRNAs) located in independent gene units [23,37]. Parallelly, it is possible to distinguish different categories of miRNA promoters: "intergenic promoters" are promoters assigned to intergenic miRNAs; "intragenic promoters" are promoters assigned to intragenic miRNAs and include both "host gene promoters" and "intronic promoters"; finally, "hybrid promoters" are those promoters that fall into intergenic regions upstream of intragenic miRNAs and could not be assigned unambiguously to the miRNA [23]. Table 1. The sixty-eight putative NF-kB responsive miRNAs expressed in healthy human tissues linked to inflammatory processes. The following attributes are reported: name of the mature miRNAs which derives from the pre-miRNAs previously identified, type of predicted TSS ("intergenic", "host gene", "intronic", or "hybrid"), names of the healthy "Human Expression dataset" libraries in which the miRNAs are expressed (i.e., "liver" and "immune system"), the chromosome where the miRNA precursor is located, the age of the miRNAs corresponding to the predicted TSSs. Note: miRNAs highlighted in bold are those processed starting from two or more pre-miRNA hairpins, each one transcribed starting from two different promoter types. * In PROmiRNA, NF-kB is among the top 10 TFs with the highest affinity for the 1000 bp-long region surrounding the predicted TSSs. ** Mature miRNAs have been selected based on the "Human Expression dataset" (microrna.org, accessed on 10 January 2023). This selection allows to review mature miRNA expression patterns across the tissues of interest.
Regarding the phylogenesis of the 68 putative NF-kB responsive miRNAs, we showed that 22 miRNAs are conserved up to the vertebrate lineage (v), 38 miRNAs are conserved up to the mammal lineage (m), miR-194 and miR-19b up to the mammal and vertebrate lineage, and, finally, only 6 miRNAs are conserved in the primate lineage (p).

Characterization of the Interplay Linking NF-kB, miRNAs, and Their Host Genes
To better characterize miRNAs that share the promoters of the host gene and to determine whether those host genes are also known to be regulated by NF-kB, multiple assessments were conducted. Firstly, we retrieved available information regarding the host genes and their intragenic miRNAs, as reported in Table 2, whereas expression correlation plots between miRNAs and their host gene are shown in Figure S1.
No experimental evidence was found regarding the host gene of hsa-mir-374a, hsamir-545, or hsa-mir-15a. All the others are intronic miRNAs of genes involved in various biological processes ranging from DNA replication to differentiation: • NFYC (Nuclear transcription factor Y subunit gamma) is a component of the sequencespecific heterotrimeric TF (NF-Y) which specifically recognizes a 5 -CCAAT-3 box motif found in the promoters of its target genes. NF-Y can function as both an activator and a repressor, depending on its interacting cofactors [ complex (MCM complex) which is the replicative helicase essential for "once per cell cycle" DNA replication initiation and elongation in eukaryotic cells. It is the core component of CDC45-MCM-GINS (CMG) helicase, the molecular machine that unwinds template DNA during replication, and around which the replisome is built [43][44][45][46][47][48]; • NR6A1 (Nuclear receptor subfamily 6 group A member 1) is an orphan nuclear receptor that binds to a response element containing the sequence 5 -TCAAGGTCA-3 . By similarity, it may be involved in the regulation of gene expression in germ cell development during gametogenesis. It is involved in regulating embryonic stem cell differentiation, reproduction, and neuronal differentiation [49]; • TENM4 (Teneurin-4) is involved in neural development, regulating the establishment of proper connectivity within the nervous system. It plays a role in the establishment of the anterior-posterior axis during gastrulation. Moreover, it regulates the differentiation and cellular process formation of oligodendrocytes and myelination of small-diameter axons in the central nervous system (CNS) [50]; • COPZ1 (Coatomer subunit zeta-1) is a cytosolic protein complex involved in intracellular trafficking, endosome maturation, lipid homeostasis, and autophagy [51,52]. It is associated with iron metabolism through the regulation of transferrin [53,54]; • DDIT3 (DNA damage-inducible transcript 3 protein) is a multifunctional TF in endoplasmic reticulum stress response. It plays an essential role in the response to a wide variety of cell stresses and induces cell cycle arrest and apoptosis [55][56][57]; • WWP2 (NEDD4-like E3 ubiquitin-protein ligase WWP2) plays an important role in protein ubiquitination and inhibits activation-induced T cell death by catalyzing EGR2 ubiquitination [58]. In human embryonic stem cells, WWP2 promotes the degradation of TF OCT4, which not only plays an essential role in maintaining the pluripotent and self-renewing state of embryonic stem cells but also acts as a cell fate determinant through a gene dosage effect [55]; • HOXB3 (Homeobox protein Hox-B3) is a sequence-specific TF that is part of a developmental regulatory system that provides cells with specific positional identities on the anterior-posterior axis. Therefore, it may regulate gene expression, morphogenesis, and differentiation [59]; • SREBF1 (Sterol regulatory element-binding protein 1) is a precursor of the TF form (Processed sterol regulatory element-binding protein 1), which is embedded in the endoplasmic reticulum membrane [60]. Its processed form is a key TF that regulates the expression of genes involved in cholesterol biosynthesis and lipid homeostasis [60][61][62]; • PANK2 (Pantothenate kinase 2) is the mitochondrial isoform that catalyzes the phosphorylation of pantothenate to generate 4 -phosphopantothenate in the first and ratedetermining step of coenzyme A (CoA) synthesis [63][64][65][66]. It is required for angiogenic activity of the umbilical vein of endothelial cells (HUVEC) [67].
Experimentally validated interactions shared among the three groups of molecules, namely (i) the 21 NF-kB responsive miRNAs sharing the host gene promoter, (ii) their host genes, and (iii) the three TF members (NFKB1, REL, and RELA) are depicted in Figure 2. Important nodes can be identified on the basis of their node centrality measures, such as degree and betweenness. The degree of a node is the total number of connections to other nodes. High-degree nodes are considered important "hubs" in a network [70,71]. The betweenness measures the number of shortest paths going through a node, taking into consideration the global network structure. Nodes with higher betweenness are important "bottlenecks" in a network [70,71]. Nodes identified by NFKB1, REL, miR-16- Figure 2. TF-miRNA co-regulatory network from experimentally validated data. In this visualization, a tripartite layout has been chosen. This provides an easy abstraction of relations between different types of molecular entities in complex networks composed of several types of nodes, such as miRNAs, genes, and TFs [68,69].  Important nodes can be identified on the basis of their node centrality measures, such as degree and betweenness. The degree of a node is the total number of connections to other nodes. High-degree nodes are considered important "hubs" in a network [70,71]. The betweenness measures the number of shortest paths going through a node, taking into consideration the global network structure. Nodes with higher betweenness are important "bottlenecks" in a network [70,71]. Nodes identified by NFKB1, REL, miR-16-5p, miR-103a-3p, and NR6A1 have high degree centrality values, whereas RELA, miR-10a-5p, and miR-30e-5p represent nodes that occur between two dense clusters and have a high betweenness centrality even if their degree centrality values are not high.
Therefore, we performed an explorative evaluation of known and potential proteinprotein interactions among REL, RELA, NFKB1, and miRNA-host genes ( Figure 3) by querying the STRING Database [72][73][74]. . Protein-protein interaction network. Network nodes represent proteins: splice isoforms or post-translational modifications are collapsed, i.e., each node represents all the proteins produced by a single, protein-coding gene locus. Edges represent protein-protein associations and are meant to be specific and meaningful, i.e., proteins jointly contribute to a shared function; this does not necessarily mean they are physically binding to each other [72][73][74]. The greater the number of edges shared between two nodes, the greater the confidence of the interaction score. The line color indicates the type of interaction evidence.
Finally, the significantly differentially expressed host genes in ARDs have been identified (Table 3). Worth a mention is the downregulation of DDIT3, SMC4, and TENM4 in replicative senescence of human fibroblasts; the upregulation of SMC4 and MCM7 after vitamin C treatment; the upregulation of HOXB3 and TENM4 in Alzheimer's disease; and the deregulation of DDIT3 and SMC4 in COVID-19 disease.  . Protein-protein interaction network. Network nodes represent proteins: splice isoforms or post-translational modifications are collapsed, i.e., each node represents all the proteins produced by a single, protein-coding gene locus. Edges represent protein-protein associations and are meant to be specific and meaningful, i.e., proteins jointly contribute to a shared function; this does not necessarily mean they are physically binding to each other [72][73][74]. The greater the number of edges shared between two nodes, the greater the confidence of the interaction score. The line color indicates the type of interaction evidence.
The STRING network shows that almost all host gene proteins have some degree of interaction. Experimental and biochemical data confirm the functional association of NFKB1, REL, and RELA. On the other hand, the higher confidence interaction values suggest a functional link between DDIT3, NFYC, MCM7, and SREBF1, as well as between IARS2, SMC4, and WWP2. Of note, experimental evidence in Figure 2 indicated that NFBK1, REL, RELA, DDIT3, NFYC, MCM7, SREBF1, and SMC4 are all targets of miR-16-5p, but miR-103a-3p, in turn, regulates IARS2, MCM7, and WWP2.
Finally, the significantly differentially expressed host genes in ARDs have been identified (Table 3). Worth a mention is the downregulation of DDIT3, SMC4, and TENM4 in replicative senescence of human fibroblasts; the upregulation of SMC4 and MCM7 after vitamin C treatment; the upregulation of HOXB3 and TENM4 in Alzheimer's disease; and the deregulation of DDIT3 and SMC4 in COVID-19 disease.

Pathways Targeted by the 68 Putative NF-kB Responsive miRNAs
By performing an Ingenuity Pathway Analysis (IPA) Target Filter Analysis, we identified mRNAs targeted by the putative NF-kB responsive miRNAs. A total of 18,095 mRNAs were retrieved, of which 9613 were experimentally observed or highly predicted. The significance was reported as p-value in Table S4. The let-7a-5p was the miRNAs with the highest associated number of mRNA targets (2014 targets).
Then we performed a network analysis focusing on putative NF-kB responsive miR-NAs targeting mRNAs coding for molecules belonging to the NF-kB pathways (Figure 4).   Interestingly, the NF-kB responsive miRNAs do not directly target genes coding for the NF-kB different subunits, but most of them are able to target genes coding for molecules belonging to NF-kB activation pathways, such as TLR and MYD88. This result is very interesting, considering that the modulation of NF-kB biological activity is related to its activation, rather than to the modulation of NF-kB subunits expression.
Further, to discover the main diseases and functions associated with the selected miRNAs dataset, we performed an IPA Core Analysis ( Figure 5). The diseases and functions are shown by bar chart, sorted by their −log p-value (p-value from Fisher's Exact test).   Cancers, immunological diseases, neurological diseases, and metabolic diseases, all well-recognized as inflammatory-based diseases, are among the diseases associated with the highest probability with NF-kB responsive miRNAs. Focusing on metabolic diseases, the most affected diseases are the non-insulin dependent diabetes mellitus (−log p-value 11.955), Alzheimer disease (−log p-value 9.532), and diabetes mellitus (−log p-value 7.680).
To better explain the association of identified NF-kB putative responsive miRNAs with these human diseases, we depicted miRNAs-diseases relationship in Figure 6. Figure 6A depicts NF-KB putative responsive miRNAs associated with metabolic diseases, whereas Figure 6B-D, show the association between identified NF-kB responsive miRNAs and cardiovascular diseases, neurological diseases, and cancer, respectively.
In addition, we have chosen as a reference all available data on the miRNAs relevant to aging, inflammation, and immunity that can be referred as inflammamiRs [81]. A detailed comparison table has been provided in Table S5. Figure 7A shows the "word cloud" with the 68 "high confidence" expressed miRNAs. The more features a specific miRNA holds (such as: the number of promoter types, the number of miRNA precursors, if it is expressed in more than one tissue, and, finally, if it is known to target NF-kB), the bigger and bolder it appears in the figure. Figure 7B depicts a Venn diagram modified from [81], displaying the miRNAs related to inflammation, immunity, and aging based on their circulating shuttles.
To better explain the association of identified NF-kB putative responsive miRNAs with these human diseases, we depicted miRNAs-diseases relationship in Figure 6. Figure  6 panel A depicts NF-KB putative responsive miRNAs associated with metabolic diseases, whereas Figure 6, panels B, C, and D, show the association between identified NF-kB responsive miRNAs and cardiovascular diseases, neurological diseases, and cancer, respectively.
In addition, we have chosen as a reference all available data on the miRNAs relevant to aging, inflammation, and immunity that can be referred as inflammamiRs [81]. A detailed comparison table has been provided in Table S5. Figure 7A shows the "word cloud" with the 68 "high confidence" expressed miRNAs. The more features a specific miRNA holds (such as: the number of promoter types, the number of miRNA precursors, if it is expressed in more than one tissue, and, finally, if it is known to target NF-kB), the bigger and bolder it appears in the figure. Figure 7B depicts a Venn diagram modified In the inner circles are grouped exosome-associated miRNAs, while, in the outer circles, the circulating miRNAs associated with Ago-2, HDL, or other microparticles are grouped. In this version, it is important to note that bold characters indicate miRNAs overlapping among the two groups. Most of the 68 high-confidence NF-kB responsive miRNAs (reported in panel A) were previously identified as circulating miRNAs associated with aging, immunological functions, and inflammation, i.e., inflammaging [81]. Only three miRNAs, such as miR-154, miR-377, and miR-885-5p, were not retrieved in previous analysis [81]. However, based on recent literature, all of them are related to NF-kB/inflammation pathways [82][83][84]. All of the 68 NF-kB responsive miRNAs are therefore included in the Venn diagram reported in panel B, highlighting that these miRNAs identified as tissues expressed miRNAs are also detectable in blood, and most of them were identified inside extracellular vesicles, i.e., exosomes (miRNAs depicted in inner circles Figure 7B).

mRNAs Targeted by the 68 Putative NF-kB Responsive miRNAs Belonging to Pathways Involved in Aging Process and/or Age-Related Diseases
By further analyzing the IPA Target Filter Analysis results, we finally identified the mR-NAs, either experimentally validated or highly predicted, to be targeted by the 68 putative NF-kB responsive miRNAs, belonging to pathways related to aging or to the most common ARDs. Among the 9613 mRNAs predicted to be targeted by such NF-kB responsive miR-NAs, 189 mRNAs targeted by 46 out of 68 miRNAs were associated to "cellular senescence pathway" (Table S6). In addition, out of the 9613, 8599 mRNAs were related to diseases reported in Figure 6, such as metabolic diseases, cardiovascular diseases, neurological diseases, and cancer. All these conditions share an inflammatory etiopathogenesis and are prototypical ARD. from [81], displaying the miRNAs related to inflammation, immunity, and aging based on their circulating shuttles. In the inner circles are grouped exosome-associated miRNAs, while, in the outer circles, the circulating miRNAs associated with Ago-2, HDL, or other microparticles are grouped. In this version, it is important to note that bold characters indicate miRNAs overlapping among the two groups. Most of the 68 high-confidence NF-kB responsive miRNAs (reported in panel A) were previously identified as circulating miRNAs associated with aging, immunological functions, and inflammation, i.e., inflammaging [81]. Only three miRNAs, such as miR-154, miR-377, and miR-885-5p, were not retrieved in previous analysis [81]. However, based on recent literature, all of them are related to NF-kB/inflammation pathways [82][83][84]. All of the 68 NF-kB responsive miRNAs are therefore included in the Venn diagram reported in panel B, highlighting that these miRNAs identified as tissues expressed miRNAs are also detectable in blood, and most of them were identified inside extracellular vesicles, i.e., exosomes (miRNAs depicted in inner circles Figure 7, panel B).

mRNAs Targeted by the 68 Putative NF-kB Responsive miRNAs Belonging to Pathways Involved in Aging Process and/or Age-Related Diseases
By further analyzing the IPA Target Filter Analysis results, we finally identified the mRNAs, either experimentally validated or highly predicted, to be targeted by the 68 putative NF-kB responsive miRNAs, belonging to pathways related to aging or to the The "word cloud" has been used to highlight the values of a miRNA based on its characteristics (such as: the number of promoter types, the number of miRNA precursors, if it is expressed in more than one tissue, and, finally, if it is known to target NF-kB). The more features a specific miRNA holds, the bigger and bolder it appears in the "word cloud". This word cloud has been drawn using Wordaizer version 6.0 APP Helmond (www.apphelmond.com, accessed on 10 January 2023). (B) Venn diagram showing the NF-kB and inflammamiRs research in context. Modified version of the Venn diagram from [81]. The Venn diagram displays the 68 NF-kB responsive miRNAs related to inflammation, immunity, and aging based on their circulating shuttles. In bold, the 21 experimentally validated miRNAs; in red, the 47 not yet experimentally validated miRNAs. In the inner circles are grouped exosome-associated miRNAs, while in the outer circles are the circulating miRNAs associated with Ago-2, HDL, or other microparticles.

Discussion
NF-kB is an ubiquitously and evolutionarily conserved TF activated by a plethora of external and internal proinflammatory stimuli [85][86][87]. The crucial role as a mediator of the inflammatory responses, together with the finding that the activation or inhibition of NF-kB can induce or reverse, respectively, the main features of aged organisms, has brought NF-kB under consideration as a key TF that drives the biological aging process [88]. In this framework, the identification of genes modulated by NF-kB can be considered a cutting-edge issue [89][90][91].
NF-kB-responsive genes were extensively investigated, whereas NF-kB-responsive genes for non-coding RNAs were only recently highlighted.
Here, we demonstrated that, by applying a data-mining approach, it is possible to select the most reliable NF-kB responsive miRNAs. Most notably, the availability of data on TFs binding sites on human miRNAs sequences constituted a starting point and the foundation for studying all human miRNAs with potential NF-kB binding sites in their promoter regions.
Some years ago, a general hypothesis was advanced that the aging process and the development of the most common ARDs could be fostered by a low-grade, chronic, systemic inflammatory process named "inflammaging" [92]. Inflammaging, which is principally sustained by the activation of the innate immune cells, is paralleled by the increased burden of senescent cells acquiring a senescence-associated secretory phenotype (SASP), which turns senescent cells into proinflammatory cells [86,[93][94][95][96]. In immune cells and tissues obtained from patients affected by the most common ARDs, NF-kB is commonly constitutively activated [97]. Of note, NF-kB activation should be an inducible, but transient, event in physiological conditions. However, despite the presence of multiple checks and balances that control NF-kB activation, in cellular and organismal aging, as well as in many ARDs, NF-kB activation becomes persistent [98,99].
In this study, using PROmiRNA software and a data-mining approach, we provide a list of 73 putative "high confidence" pre-miRNAs sequences corresponding to 68 NF-kB responsive mature miRNAs sequences.
Likewise, we highlighted the presence of distinct types of promoters that can regulate NF-kB responsive miRNAs.
A total of 33 miRNAs of the 68 high confidence expressed miRNAs identified have an "intronic" promoter, and 5 of these have both an "intronic" and "host gene" promoter, whereas only one miRNA (miR-194) shares both "intergenic" and "host-gene" promoters. Alternative promoters are a common mechanism to create diversity in the transcriptional regulation of miRNA [38].
It has been demonstrated that "intronic" promoters convey an additional degree of freedom over intragenic miRNA transcriptional regulation by virtue of some peculiar characteristics, thus allowing the modulation of miRNA expression levels in a tissue-and condition-specific manner [23]. Besides the other features, in this context, it is important to stress that: 1.
"Intronic" promoters can explain cases of poor correlation between host gene and miRNA expression, functioning as a real alternative promoter [23]. As shown in Figure S1, the expression levels of NF-kB-miRNAs modulated by both "host gene" and "intronic" promoters (i.e., miR-16, miR-103, miR-186, and miR-33b) or by both "host gene" and "intergenic promoters" (i.e., miR-194) are not correlated with the expression levels of their host gene, whereas most of the miRNAs that share the host gene promoters are characterized by directly (e.g., miR-15b) or inversely (e.g., miR-30c, miR-616, and miR-93) correlated transcription levels. 2.
"Intronic" promoters are expressed in a tissue-specific manner, but "host gene" promoters are considered primarily for housekeeping gene regulation [23]. Housekeeping genes are required for the maintenance of essential functions of any cell type, so they are expected to be constitutively expressed in all cells and at any development stage [100]. Among the NF-kB-miRNA host genes, COPZ1, NFYC, and ZRANB2 have been cataloged as housekeeping genes (Table 2). 3.
"Intronic" promoters are mainly triggered by tissue-specific master regulator TFs, instead of TFs of "host gene" promoters, which broadly overlap with those of protein coding genes and can be considered mainly for housekeeping ("intergenic" promoters are regulated by a combination of intronic-specific and host-gene specific TFs). This suggests a different evolutionary mechanism [23]. In this study, the expression levels of the three housekeeping host genes (COPZ1, ZRANB2, and NFYC) and their miR-NAs (miR-148b-3p, miR-186-5p, and, lastly, miR-30c-5p and miR-30e-5p, respectively) are mainly inversely correlated or not showing clear correlation trends ( Figure S1). 4.
Moreover, those intragenic miRNAs that share the promoters of the host gene interact with their own host genes (miR-16-2::MSC4; miR-106b::MCM7, miR-181b-2::NR6A1, miR-708::TENM4, miR-148b::COPZ1, and miR-10a::HOXB3), but also with the other functionally related host genes, creating a complex regulatory mechanism ( Figure 2). NFKB1, REL, miR-16-5p, miR-103a-3p, and NR6A1 are the most important hub nodes in the network, whereas miR-10a-5p connects the hub nodes identified by NFKB1, NR6A1, and HOXB3, and miR-30e-5p connects REL, NR6A1, and ZRAMB2 hubs. Interestingly, in the network, it is possible to identify a clear TF-miRNA feed-forward loop involving DDIT3, miR-16-5p, and NFYC. In a TF-miRNA feed-forward loop, TF and miRNA co-regulate the target genes: in a "coherent" feed-forward loop, the TF and miRNA have the same effects on their common targets, whereas, in an "incoherent" feed-forward loop, the TF and miRNA carry out opposing effects, which precisely fine-tune gene expressions to minimize noise and maintain stability [68,101]. TF-miRNA feed-forward loops have a specific function in noise buffering effects, which can minimize the response of stochastic signaling noise to maintain steady-state target levels [102,103]. Disruption of feed-forward loops could lead to serious dysregulations at the origin of diseases and cancers, e.g., interference in the NF-kB/miR-19/CYLD loop can induce T cell leukemogenesis [103,104]. Therefore, investigating the regulatory motifs among DDIT3, 16-5p, and NFYC could provide valuable insights to dissect the molecular mechanisms underlying biological processes and diseases triggered by NF-kB constitutive activation.
Protein-protein interaction analysis of protein-coding host genes revealed that most of them could be functionally related (Figure 3). Beyond the well-known functional association of NFKB1, REL (cREL), and RELA, some data have highlighted the association with endoplasmic reticulum stress, providing opportunities to fine-tune cellular stress responses [105]. In the framework of atherosclerosis, multiple links between NF-kB and ER stress were suggested. A disturbed flow can cause endoplasmic reticulum stress, leading to SREBF1 activation with nuclear localization and to DDIT3 expression triggered by endoplasmic reticulum stress response elements [106]. NFYC is a subunit of a trimeric complex (NFY) known to interact with several TFs to enable the synergistic activation of specific classes of promoters. NFY directly controls the expression of TF genes such as P53 (DNA-damage), XBP1, CHOP/DDIT3 (endoplasmic reticulum stress), and HSF1 (heat shock) [107,108]. Of note, experimental data have shown the upregulation of both SMC4 and MCM7 in mesenchymal stem cells after vitamin C treatment; the downregulation of DDIT3, SMC4, and TENM4 in replicative senescence of human fibroblasts; the upregulation of HOXB3 and TENM4 in Alzheimer's disease; and, finally, the deregulation of DDIT3 and SMC4 in COVID-19 disease (Table 3).
In this scenario, targeting NF-kB signaling is becoming a promising strategy for drug development and ARD treatments [91,109].
Almost all of the 68 miRNAs that we identified in our current analysis were previously associated with inflammaging processes and with the most common ARDs, such as metabolic diseases, cardiovascular diseases, neurodegenerative diseases, and cancers [110,111].
Out of the 9613 mRNAs targeted by the 68 NF-kB responsive miRNAs, 8599 mR-NAs were related to such diseases. Of note, 189 mRNAs were associated with "cellular senescence pathway", which is recognized as the main culprit of the aging process.
Notably, the identified NF-kB responsive miRNAs are not able to directly modulate gene expression of NF-kB subunits but are able to target molecules belonging to NF-kB activation pathways (canonical and non-canonical pathway). Interestingly, among the NF-kB-responsive miRNAs genes identified with our approach, the most relevant examples of mRNAs that can target molecules belonging to the NF-kB canonical and non-canonical pathways, or related molecules, are miR-146a and miR-155. In fact, miR-146a and miR-155, control NF-kB activity during inflammation by a combinatory action without directly targeting NF-KB subunits [112]. miR-155 is rapidly upregulated by NF-kB during the early phase inflammatory response through a positive feedback loop necessary for signal amplification. miR-146a is rather gradually upregulated by NF-kB and forms a negative feedback loop attenuating NF-kB activity in the late phase of inflammation. The combined action of these two positive (NF-kB::miR-155) and negative (NF-kB::miR-146a) NF-kB-miRNA regulatory loops provides optimal NF-kB activity during inflammatory stimuli, and eventually lead to the resolution of the inflammatory response in physiological condition.
Another example is miR-16 that targets the IKKα/β complex of the NF-kB canonical pathway polarizing macrophages toward an M2 phenotype [113].These results are in line with the known modulation of NF-kB biological activity, based on the activation and not on the expression of its subunits [5].
Interestingly, all the 68 NF-kB responsive miRNAs are detectable in blood, and most of them were identified inside extracellular vesicles, i.e., exosomes. Exosomes are currently considered to be a crucial intercellular cross-talk mechanism, acting at both the paracrine and systemic levels [114]. This result highlights the complexity of the feed-back loops between NF-kB activation in specific tissues, the expression of NF-kB responsive miRNAs, and their release in the bloodstream as a systemic intercellular communication mechanism. A further level of complexity can be envisaged considering that NF-kB is known to indirectly regulate miRNA expression through the modulation of other TFs. NF-kB can modulate AP-1 TF [115], which, in turn, is able to modulate different miRNAs genes, i.e., miR-21 [116].
Of note, among the 68 miRNAs, 21 were already experimentally identified as NF-kB responsive, reinforcing the reliability of our results. Our data also highlight the potential value of the 47 NF-kB putative responsive miRNAs (listed in Figure 7B) that are yet to be experimentally validated.
Overall, our results are of interest in the framework of the research on the biomarkers/drugs of aging and inflammation related diseases. If NF-kB responsive miRNAs are hyper-transcribed in tissues involved in the modulation of inflammatory responses, the hypothesis that circulating miRNAs could be useful tools to track the trajectories of healthy or un-healthy aging is reinforced [117][118][119][120][121] and possible therapeutic strategies based on the inhibition of those miRNAs could be further tested.

Limitation of the Study
The data-mining process frequently encompasses a further phase involving the extraction of implicit relational patterns through traditional statistics or machine learning, but the particularity of the research question and the type of data available have been a hindrance to this kind of analysis.

Data Mining Process
In the field of Knowledge Discovery in Databases (KDD), a data-mining approach is used to extract meaningful information and to develop significant relationships among variables stored in large data sets [122]. In this study, we have mined and integrated data from multiple databases to select NF-kB responsive miRNAs, and the process has been tailored based on the research question. Four main steps can be distinguished:

•
PROmiRNA provides an interesting approach for miRNA promoter annotation based on a semi-supervised statistical model trained on deepCAGE data and sequence features [23]. It was used to identify all human miRNAs potentially modulated by NF-kB, i.e., "NF-kappaB", "NFKB1", "REL", and "RELA". • FANTOM4 libraries, generated by the FANTOM4 project [26], collects a wide range of genome-scale data from several tissues. The analysis of FANTOM4 libraries retrieved those miRNAs showing "expression at the promoter level" in different human tissues.

Data Extraction and Integration
This phase includes downloading, extracting, filtering, and combining the data from the databases previously identified. The integration of multiple datasets has been possible through the following steps.

Data Cleaning and Transformation
Because the data originates from multiple sources, the integration often involves converting data formats, cleaning, removal of incorrect data, generating new variables, resolving redundancy, and checking against miRNA nomenclature consistency, both between miRNAs names originating in different miRBase versions and between the names of pri-miRNAs and the mature forms. This issue has been manually curated by comparing miRNA names in miRBase database version 21.

Assessment of the Results
This is the final stage of a KDD process, involving the translation of aggregated data into comprehensible knowledge. The validity and reliability of the data were tested by comparing the results obtained in the data-mining process with those already published in the literature.
The whole data-mining process is illustrated in the data flow diagram in Figure 1. Data obtained at each intermediate step are provided in Supplementary Tables S1-S3. The final miRNA-pool is reported in Table 1.
STRING database Version 11.5 (https://string-db.org/, accessed on 10 January 2023) was used to discover known and potential interactions among REL, RELA, NFKB1, and miRNA-host gene proteins (Figure 3). STRING is a database of predicted and known protein-protein interactions. The interactions include direct (physical) and indirect (functional) associations; these stem from knowledge transfer between organisms, from interactions aggregated from other (primary) databases, and from computational prediction [72][73][74]. The network was created by setting a minimum required interaction score of 0.15.
The RNA-seq datasets in Aging Atlas (https://ngdc.cncb.ac.cn/aging/index, accessed on 10 January 2023) were examined to explore age-related changes in host gene expression [134]. Table 3 shows differentially expressed host genes in strictly age-related conditions; only those genes showing |log2FC| > 1 and q-value < 0.005 (or p-value < 0.005 if q-value was not provided) have been reported. Data relative to particular experimental conditions (e.g., gene knockdown) have not been reported. All websites and online tools were accessed in the period between January and February 2023.

Ingenuity Pathway Analysis
Bioinformatic evaluations (networks and disease analyses) were performed by the Ingenuity Pathway Analysis software (Qiagen, Hilden, Germany). The putative NF-kB responsive miRNAs identified through the data-mining process were analyzed to explore the experimentally observed or high predicted mRNA targets via the microRNA Target Filter Analysis.
Furthermore, an IPA Core Analysis was performed to define the associated diseases and functions. Direct and indirect relationships from the Ingenuity Knowledge Base (gene only) datasets were considered. We filtered only molecules and/or relationships experimentally observed in any tissue from human, rat, or mouse. Across the observations, 51 miRNAs were ready to be analyzed (Table S4) [135]. All the networks, diseases, and biological functions were assessed using IPA software (Qiagen, Hilden, Germany).

Conclusions
Here, we demonstrated that a well-settled data-mining approach may disclose the most reliable miRNAs having a key role in the regulation of specific pathways of interest. Deciphering the crosstalk between miRNAs and NF-kB is one of the major topics to be investigated to understand the complex derailment of several metabolic pathways in normal and pathological aging. Future studies are needed to confirm that the identification of such miRNAs is of diagnostic/prognostic/therapeutic relevance for the most common inflammatory-and age-related conditions.