The Roles of Histone Deacetylases and Their Inhibitors in Cancer Therapy

Genetic mutations and abnormal gene regulation are key mechanisms underlying tumorigenesis. Nucleosomes, which consist of DNA wrapped around histone cores, represent the basic units of chromatin. The fifth amino group (Nε) of histone lysine residues is a common site for post-translational modifications (PTMs), and of these, acetylation is the second most common. Histone acetylation is modulated by histone acetyltransferases (HATs) and histone deacetylases (HDACs), and is involved in the regulation of gene expression. Over the past two decades, numerous studies characterizing HDACs and HDAC inhibitors (HDACi) have provided novel and exciting insights concerning their underlying biological mechanisms and potential anti-cancer treatments. In this review, we detail the diverse structures of HDACs and their underlying biological functions, including transcriptional regulation, metabolism, angiogenesis, DNA damage response, cell cycle, apoptosis, protein degradation, immunity and other several physiological processes. We also highlight potential avenues to use HDACi as novel, precision cancer treatments.


INTRODUCTION
In the nuclei of eukaryotic cells, the entire genome of an organism is condensed into chromatin. The nucleosome is the basic unit of chromatin: it contains 147 DNA base pairs coiled around a core histone octamer, which includes histones H2A, H2B, H3 and H4 (Luger et al., 1997;Tessarz and Kouzarides, 2014). The additional linker histone H1 interacts with chromatin outside of the core octamer to regulate higher order chromatin structure (Fyodorov et al., 2018). There are two major higher order structures: heterochromatin refers to condensed chromatin, and euchromatin refers to loosely packed chromatin that is more accessible to transcriptional regulators and RNA polymerase complexes (Allfrey et al., 1964). Thus, alteration and regulation of chromatin structure impacts gene expression by making certain genes more or less available for transcription.
The epigenome is comprised of modifications to chromatin, including DNA methylation and histone modifications. For example, DNA accessibility is regulated by nucleosome sliding or post-translational modifications (PTMs), which include phosphorylation, methylation and acetylation. These covalent modifications control the structure and function of chromatin through a number of regulators. These regulators can be broadly divided into "readers" (enzymes that bind to modifications and facilitate epigenetic activities), "writers"(enzymes that establish DNA methylation or histone modifications), and "erasers"(enzymes that remove these markers) (Taverna et al., 2007;Kutateladze, 2011;Musselman et al., 2012;Huang et al., 2014;Xu et al., 2017). As an example, acetylation occurs at the fifth NH2 (N ε ) of histone lysine residues (Strahl and Allis, 2000;Jenuwein and Allis, 2001), and is read by the bromodomain-containing protein (BRD), written by histone acetyltransferases (HATs), and erased by histone deacetylases (HDACs) (Dawson and Kouzarides, 2012;Filippakopoulos and Knapp, 2014;Zaware and Zhou, 2019).
The acetylation of lysine residues (Kac) on histone tails generates positive charges, which neutralize negatively charged DNA and the unwinding of tightly coiled heterochromatin (Shogren-Knaak et al., 2006). Histone acetylation can increase the inner pore space of chromatin from 20 nm to 60−100 nm, altering spatial distance and accessibility during interphase (Gorisch et al., 2005); it also ensures sufficient space for local transcriptional events, including initiation and elongation (Wang et al., 2009). Acetylation is of particular importance because the interaction between histones and chromatin is generally very stable, and interruption of this interaction requires a high concentration of NaCl or acetate (Von Holt et al., 1989;Shechter et al., 2007). Notably, acetylation is often a necessary precursor to other modifications, such as phosphorylation, methylation and ubiquitylation (Yang and Gregoire, 2007;Yang and Seto, 2008a).
Acetylation is controlled by two antagonistic enzyme families: HATs and HDACs. HDACs are expressed by various tumors, and are involved in vital chromosomal translocation-mediated oncogenic protein fusion and carcinogenic events (Falkenberg and Johnstone, 2014;West and Johnstone, 2014). These enzymes were first revealed to remove acetyl groups from histones by Vincent Allfrey (Inoue and Fujimoto, 1969). The first HDAC that was discovered, HDAC1, was originally isolated by utilizing a microbe-derived cyclic tetrapeptide, Trapoxin, which inhibits histone deacetylation and induces cell-cycle arrest (Taunton et al., 1996). Sequence homology-dependent HDACs were subsequently identified, and shown to be involved in major biological functions such as transcription, metastasis, autophagy, cell cycle, DNA damage repair, angiogenesis, stress responses and senescence (Yang and Seto, 2008b;Li and Zhu, 2014).
Histone deacetylase inhibitors (HDACi) might be able to reverse the activation of tumor suppressor genes (TSG), and in this way inhibit the viability and malignant proliferation of tumor cells (Glozak and Seto, 2007). The efficacy of HDACi treatment has been demonstrated in numerous clinical studies. This review discusses HDACs and their inhibitors in the context of potential cancer treatments.

Class II HDACs
All Class IIa HDACs include an extended N-terminal domain that contains conserved serine (Ser) residues and other motifs for localization and function (Yang and Gregoire, 2005). Based on these Ser residues, several kinases such as
HDAC10 is a polyamine deacetylase that preferentially catalyzes N 8 -acetylspermidine hydrolysis to generate acetate Shinsky and Christianson, 2018). It contains a leucine-rich domain, a deacetylase domain and an inactivity domain (Guardiola and Yao, 2002;Kao et al., 2002). Similar to FIGURE 1 | Domain structure of human HDACs. The fundamental structure of all deacetylases. Total number of amino acid residues and molecular weights in each HDAC were shown on the right of each protein.

Class III HDACs
A total of seven NAD + -dependent class III HDACs, or SIRTs, have been identified in the cytoplasm, nucleus and mitochondria (Yao et al., 2014;Chalkiadaki and Guarente, 2015;Zhao and Zhou, 2019). SIRT1, which is distributed in the cytoplasm, mitochondria and the nucleus, has two CRM1-mediated NES and two NLS (Tanno et al., 2007): it undergoes conformational shifts in response to adenosine triphosphate (ATP), which impedes its ability to interact with substrates in the C-terminal domain (Kang et al., 2017). SIRT2 is even more widely distributed than SIRT1, being found in the plasma membrane and cytoskeletonassociated organelles in addition to the cytoplasm, nucleus and mitochondria: it contains a CRM1-dependent NES and a putative leucine-rich NES (Wilson et al., 2006). By contrast, SIRT3, SIRT4, and SIRT5 are primarily found in the mitochondria. SIRT3 has the capability to shuttle from the nucleus to the Frontiers in Cell and Developmental Biology | www.frontiersin.org mitochondria via its mitochondrial localization sequence, which is also responsible for its mitochondrial deacetylation activity (Onyango et al., 2002;Schwer et al., 2002;Lombard et al., 2007;Bao et al., 2010). Of note, these mitochondrial regulators transfer to the nucleus in response to DNA damage induced by etoposide treatment or ultraviolet (UV) irradiation . SIRT6 is widely distributed, being found in the nuclear plasma, the heterochromatin, the nucleolus, as well as the cytoplasm (Michishita et al., 2005;Ardestani and Liang, 2012;Jedrusik-Bode et al., 2013). SIRT6 has a slower catalytic rate than other active SIRTs on substrates because SIRT6 lacks the conserved, highly flexible NAD + -binding loop, and instead contains a stable single helix (Pan et al., 2011). Finally, SIRT7, predominantly locates in the nucleolus but also exists in the cytoplasm (Nahalkova, 2015;Zhang et al., 2016) and is involved in mitochondrial function (Ryu et al., 2014;Mohrin et al., 2015). Two sequences in the N-terminal and C-terminal regions of SIRT7 permit nuclear and nucleolar localization, respectively (Kiran et al., 2013).

Class IV HDACs
HDAC11 is the exclusive member of the class IV HDACs. Recent studies have indicated that HDAC11 might predominantly be more involved in the fatty acylation of proteins compared to its weak deacetylation (Kutil et al., 2018;Cao et al., 2019).
Here, we recapitulate the detailed distributions of all 18 HDACs in Table 2.

BIOLOGICAL FUNCTIONS OF HDACs
Histone deacetylases are expressed in different tumors: class I and II HDACs are considered to be general oncoproteins that interact with substrates and regulate gene expression to promote tumorigenesis and cancer development either individually or alongside with co-repressors (Falkenberg and Johnstone, 2014;West and Johnstone, 2014). Paradoxically, SIRTs can serve as both oncoproteins and tumor suppressors (Kugel et al., 2016;Costa-Machado et al., 2018;Funato et al., 2018). We list the demonstrated knockout (KO) or knockdown models of 18 HDACs ( Table 2). Because of the diverse biological function of HDACs, it is not surprising that HDACi regimens influence many cellular processes, including those that contribute to cancer progression.

Transcription Modulators
Transcription factors (TFs) can either directly target DNA or undergo various PTMs to alter gene expression. In this manner, HDACs negatively modulate transcription through forming a complex with TFs or by directly regulating TF transcription (Grunstein, 1997). For instance, the v-myc avian myelocytomatosis viral oncogene homolog (Myc) is a wellcharacterized proto-oncogene that promotes tumorigenesis by directly recruiting and interacting with HDACs to regulate gene expression (Liu et al., 2007;Zhang et al., 2012a). Meanwhile, Myc acetylation is also modulated by HDACs either directly or indirectly. For example, SIRT2 stabilizes N-Myc and c-Myc proteins by deacetylating and repressing neuronal precursor cellexpressed developmentally downregulated 4 (NEDD4), which mediates Myc ubiquitination and degradation . Consequently, the SIRT2-specific inhibitor thiomyristoyl (TM) promotes Myc ubiquitination and degradation (Jing et al., 2016). HDACi suberoylanilide hydroxamic acid (SAHA) and entinostat (also called MS-275) induce Myc acetylation at K323, downregulating Myc and accompanying with tumor necrosis factor (TNF)-related apoptosis-inducing ligand (TRAIL) activation (Nebbioso et al., 2017). Therapeutic regimens that target Myc suppression using HDACi combined with DNA demethylation reagents seems to have a notable effect on nonsmall cell lung cancer (NSCLC) through activating immune system (Topper et al., 2017).
In addition to TFs, HDACs also modulate the activity of super enhancers (SEs) (Gryder et al., 2019). Enhancer RNAs (eRNAs) are short, non-coding RNA molecules that alter the transcription of target genes in cooperation with promoters (Melo et al., 2013;Hsieh et al., 2014;Danko et al., 2015;Mao et al., 2019). Trichostatin A (TSA) and SAHA reduce eRNA synthesis by inhibiting HSP90 (Greer et al., 2015). MEF2D and HDAC4/9 form a corepressor to recognize intergenic regions. HDAC4/9 depleted cells show increased H3K27ac level around the gene transcriptional start sites where show the features of active enhancers within corresponded topologically associated domains (TAD) (Di Giorgio et al., 2020). Class I-specific HDACi 4SC-202 globally increases both of H3K27ac and H3K4me3 levels around the TSS of genes, but notably decreases occupancy at proximal regions of TSS of genes such as SMAD family member 6 (SMAD6) and E2F transcription factor 8 (E2F8), which are associated with enhancer deactivation (Mishra et al., 2017). Panobinostat and   Potente et al., 2007;Dioum et al., 2009;Lim et al., 2010 Impairs nicotinamide mononucleotide (NMN)-induced amelioration of liver fibrosis and NMN-dependent telomere integrity in premature aging mice Amano et al., 2019 Causes methionine restriction-induced lethality in mouse ESC Tang S. et al., 2017 Impairs myeloid-derived suppressor cells (MDSC) differentiation by disturbing glycolytic pathway Liu et al., 2014 Impairs various DNA repair Cohen et al., 2004;Yuan et al., 2007;Ming et al., 2010;Meng et al., 2020 Inhibits autophagy in MEFs Lee et al., 2008 Causes differentiation defects of mice ESC Tang Frontiers in Cell and Developmental Biology | www.frontiersin.org Suppresses lymph node metastases in breast cancer Leslie et al., 2019 romidepsin alter the acetylation status of H3K27 by disrupting the SE topology in paired box 8 (PAX8) . Largazole (a cyclic peptides similar to depsipeptide) preferentially disturbs SE-driven transcripts that are frequently associated with oncogenic activities .

Transcriptional Activation
Although HDACs generally function as gene silencers, they can also activate transcription Wang et al., 2002). Besides the regulation of enhancers, a potential mechanism underlying this role includes the modulation of RNA polymerase II (RNAP2) by HDACs. HDACs participate in the crosstalk between RNAP2 C-terminal domain acetylation and phosphorylation (Wang et al., 2009;Blank et al., 2017;Ali et al., 2019). SIRT6 recruits and mono-ADP-ribosylates switch/sucrose non-fermenting (SWI/SNF) related, matrix associated, actin dependent regulator of chromatin subfamily c member 2 (SMARCC2/BAF170) to form active chromatin at the enhancer of heme oxygenase-1, which subsequently recruits RNAP2 (Rezazadeh et al., 2019). SIRT6 can also bind p53 to effectively recruit RNAP2 to local promoters . SIRT6 deficiency mediates the activation of cyclin-dependent kinase 9 (CDK9) that can phosphorylate negative elongation factor (NELF) and mediate NELF release from RNAP2, facilitating the enrichment of TFs and RNAP2-related elongation factors to promote elongation of specific gene sets (Etchegaray et al., 2019). Consistently, TSA and SAHA disturb RNAP2-mediated transcriptional elongation by promoting the association between RNAP2 and NELF (Greer et al., 2015). Moreover, high doses of largazole can cause RNAP2-mediated transcriptional pausing and cell death .
The BRD family proteins are readers of Kac (Dhalluin et al., 1999;Fujisawa and Filippakopoulos, 2017). Class I HDACi 4SC-202, mocetinostat and entinostat induce increase of hundreds of gene expression, which are mostly enriched upon BRD4-and MYC-targeted TSS-proximal regions. p21 is activated by 4SC-202 to inhibit cell proliferation (Mishra et al., 2017). Similarly, JQ1 cooperates with SAHA to inhibit the growth of pancreatic ductal adenocarcinoma (PDAC) by upregulating p57 (CDKN1C) that usually blocked by Myc (Mazur et al., 2015). Besides, HDACs also interact with protein arginine methyltransferases (PRMTs) to regulate gene transcription . These findings all highlight the competition among the "readers", "writers" and "erasers" at acetylated histones and nonhistones, and may provide additional, novel and combination epigenetic approaches for cancer therapy in the future (Figure 2).

Metabolism
Various kinases and metabolic pathways form a complex network with epigenetic co-repressors to dynamically regulate metabolic flux and enzyme activity; aberrations in these processes can result in tumorigenesis and cancer progression. Metabolism can affect protein acetylation by altering the concentration of NAD + and acetyl-CoA. In turn, HDACs also mediate metabolic reprogramming in cancer cells (Verdin and Ott, 2015).
Cancer cells are often characterized by their strong glycolytic activity, with aerobic glycolytic activity being preferred for tumor energy metabolism (Weinhouse, 1956). Increased glycolysis is associated with the abnormal regulation of glycolytic enzymes and other glucose metabolism pathways. Class II HDACs induce trans-repression of gluconeogenic enzymes from the cytoplasm to the nucleus in an HDAC3-dependent manner, and mediate the deacetylation and activation of the forkhead box class O (FoxO) family in the nucleus (Mihaylova et al., 2011). SIRT2 deacetylated isocitrate dehydrogenase 1 (IDH1) at K224 and promotes IDH1 enzymatic activity. The hypoacetylated IDH1 converts isocitrate into α-ketoglutarate in the tricarboxylic acid (TCA) cycle to inhibit liver metastases of colorectal cancer (CRC) . Pyruvate kinase (PKM2) promotes tumorigenesis by regulating oncogene expression and proliferation pathway activation in HDAC3-dependent way (Yang W. et al., 2011;Yang et al., 2012). SIRT6 can directly interact with and deacetylate PKM2, resulting in its nuclear export via exportin 4 and suppression of PKM2-related oncogenic functions (Bhardwaj and Das, 2016). Conversely, SIRT3 and SIRT6 also act as tumor suppressors, restricting aerobic glycolysis in cancer cells through destabilization of hypoxia inducible factor 1 subunit alpha (HIF-1α) and inhibition of glycolytic kinases, respectively (Finley et al., 2011;Sebastian et al., 2012). p53 directly binds and activates SIRT6 to regulate gluconeogenesis by mediating the nuclear exclusion and deacetylation of FoxO1 .
Amino acids are also involved in tumorigenesis. SIRT3 depletion suppresses glutamate dehydrogenase (GDH/GLUD), which impairs glutamine flux to the TCA cycle and causes reduction of acetyl-CoA pools . SIRT4 is a lipoamidase that diminishes the activity of the pyruvate dehydrogenase complex (PDH) by hydrolyzing the lipoamide cofactor dihydrolipoyllysine acetyltransferase (DLAT) (Mathias et al., 2014). Furthermore, SIRT4 represses GDH activity through its ADP-ribosyltransferase function. SIRT4 deficiency activates GDH, stimulating amino acid-mediated insulin secretion in insulinoma cells (Haigis et al., 2006). SIRT4 also mediates other PTMs, including methylglutarylation, hydroxymethylglutarylation and 3-methylglutaconylation, and intermediates of these PTMs contribute to leucine oxidation. Indeed, SIRT4-KO induces leucine disordered metabolism and leads to glucose intolerance and insulin resistance (Anderson et al., 2017). Meanwhile, elevated SIRT5 expression in breast cancer mediates glutaminase desuccinylation and protects glutaminase from ubiquitin-mediated degradation; this effect has been associated with a poor prognosis in breast cancers (Greene et al., 2019). SIRT3 and SIRT5 also mediate desuccinylation and deacetylation of SHMT2, respectively, suggesting that suppression of serine catabolism might represent a novel strategy to restrain tumor growth (Wei et al., 2018;Yang et al., 2018).

Redox and Oxidative Stress
Histone deacetylase inhibitors treatment is often accompanied by oxidative stress related DNA damage that is primarily caused by the generation of reactive oxygen species (ROS) (Xu et al., 2006). In mammalian cells, two redox systems respond to oxidative stress: the thioredoxin (Trx) system and the glutathione-glutaredoxin (Grx) system. In response to nitric oxide (NO), HDAC2 is S-nitrosylated at Cys 262 and Cys 274, which induces chromatin remodeling to promote gene expression (Nott et al., 2008). A pair of redox-sensitive cysteine residues (Cys-667/Cys-669) in HDAC4 are involved in oxidative stress via the formation of intramolecular disulfide bonds (Ago et al., 2008). Compared with normal cells, tumor cells are enriched with the antioxidant Trx reductase (TrxR), which might represent a novel therapeutic target (Lu and Holmgren, 2014;West and Johnstone, 2014). Depsipeptide causes robust DNA damage and apoptosis by inducing ROS generation, primarily through the suppression of TrxR . HDAC5 represses mitochondrial ROS generation, and depletion of HDAC5 provokes nuclear factor, erythroid 2 like 2 (NRF2)-associated transcription . The DNA and RNA binding protein Y-box binding protein 1 (YB-1) binds to NRF2 in response to oxidative stress. Entinostat induces YB-1 acetylation and blocks its binding to NRF2, reducing NRF2 synthesis and increasing ROS levels in sarcoma cells (El-Naggar et al., 2019).

DNA Damage Response
The DDR is a vitally important regulatory mechanism that protects genomic DNA from damage induced by various stimuli (Jackson and Bartek, 2009). Different levels of DNA damage are inevitably caused by UV radiation and DNA adducts, which are produced by ROS and reactive nitrogen species (RNS), as well as exposure to chemical agents (Barker et al., 2015;Roos et al., 2016;Pouget et al., 2018;. DNA double-strand breaks (DSBs) are the most severe form of DNA damage and are repaired via one of two pathways: homologous recombination (HR) or non-homologous endjoining (NHEJ) (Scully et al., 2019). Deacetylation of H3K56 and H4K16 by HDAC1/2 are involved in mediating dynamic chromatin regulation in response to NHEJ (Miller et al., 2010). Although H4K16 acetylation attenuates binding of p53 binding protein 1 (53BP1) to H4K20me2, euchromatic histone lysine methyltransferase 1 (EHMT1, also called GLP or KMT1D)catalyzed H4K16 monomethylation could significantly enhance this binding . Meanwhile, SIRT1 redistributes to DSB foci to promote HR during oxidative stress (Oberdoerffer et al., 2008). NuRD complex subunit, chromodomain-helicase-DNA-binding protein (CHD4), was recently discovered to be recruited by SIRT6 to replace heterochromatin 1 (HP1) at H3K9me3 to ultimately promote chromatin relaxation through HR (Hou et al., 2020). SIRT6 also mono-ADP-ribosylates and displaces KDM2A (also called JmjC domain-containing histone demethylase 1A, JHDM1A) from chromatin, which leads to HP1α-dependent H3K9me3 deposition at DSBs and transient transcriptional repression, accompanying with the recruitment of NHEJ factors (Rezazadeh et al., 2020).
The DDR is controlled by three related kinases: ataxia telangiectasia mutated (ATM), ataxia telangiectasia mutated and Rad3 related (ATR), and DNA-dependent protein kinase catalytic subunits (DNA-PKcs) (Blackford and Jackson, 2017). Once a DSB occurs, the MRE11-RAD50-NBS1 (MRN) complex and the Ku family are rapidly recruited to DSB sites. As the sensor, ATM is recruited to DSB sites by the MRN complex (Blackford and Jackson, 2017). The interplay between HDAC1 and ATM increases chromatin condensation to prevent radio-sensitivity in response to ionizing radiation (Kim et al., 1999). SIRT1 binds to deleted in breast cancer 1 (DBC1), which induces p53 activation Zhao et al., 2008). ATM also mediates DBC1 phosphorylation at threonine (Thr) 454 which contributes to DBC1-SIRT1 interactions during DNA damage (Yuan J. et al., 2012). Moreover, SIRT1 deacetylates and maintains the hypoacetylation of Nijmegen breakage syndrome protein 1 (NBS1), which is necessary for the ionizing radiationinduced phosphorylation of NBS1 and subsequent MRN complex recruitment to DSB sites (Yuan et al., 2007). SIRT7 directly binds and deacetylates ATM, which is prerequisite for ATM dephosphorylation and inactivation in the final stage of DNA repair . Panobinostat-induced downregulation of meiotic recombination 11 homolog (MRE11) enhances radiosensitization of bladder cancer cells by promoting MRE11 ubiquitination that relies on the upregulated E3 inhibitor of apoptosis protein 2 (cIAP2) (Nicholson et al., 2017).
ATR and its downstream kinase CHK1 are also involved in the response to DNA replication stress. SAHA slows down replication forks by restricting the ATR pathway (Conti et al., 2010). Following the conserved mechanism in yeast, VPA disrupts the formation of single-strand-DNA-RFA nucleofilaments and the activation of the Mec1 (ATR in human) and Rad53 (CHK2 in human) by suppressing the recruitment of replication factor A protein 1 [RFA1, replication protein A (RPA) in human] and DNA damage checkpoint protein Ddc2/LCD1 [ATR interacting protein (ATRIP) in human] to DNA damage sites (Robert et al., 2011). Entinostat represses checkpoint signaling during replication stress. Mechanically, HDAC1/2 suppress the cell cycle kinases WEE1 and CDK1 and induce the dephosphorylation of ATM and CHK2 by suppressing the expression of PP2A subunit. Entinostat also induces the incorrect incorporation of NTPs and metabolites during the induction of checkpoint kinase inactivation, which can result in mitosis catastrophe (Goder et al., 2018). Therefore, CHK inhibitors might be designed to prevent this event from occurring. Indeed, CHK1 inhibitor treatment combined with HDACi induces cell death via extensive mitotic disruption in a range of solid tumors .
DNA-PKcs is another sensor that is recruited to DSBs by Ku-bound DSB ends (Blackford and Jackson, 2017). Under conditions of fasting-induced oxidative stress, SIRTs act as protective factors in the DDR. Specifically, SIRT1 deacetylates Ku70, resulting in Ku70-Bcl-2 associated protein X (BAX) disassociation and the transport of BAX away from the mitochondria, leading to stress-induced resistance to apoptosis (Cohen et al., 2004). SIRT3 also physically interacts with and deacetylates Ku70 to impede BAX translocation to the mitochondria (Sundaresan et al., 2008).
SIRTs are highly important for DNA damage repair and genome stability Ng and Huen, 2020). Recent data have shown that SIRT6 is a novel sensor for initiating the DDR (Onn et al., 2020). Deacetylated SIRT6 at K33 by SIRT1 results in SIRT6 polymerization and deposition at γH2AX foci. Moreover, a SIRT6 K33R hypoacetylation mimic can rescue DNA repair defects in SIRT1-deficient cancer cells . SIRT7 is also associated with NHEJ, as a Sirt7 deficiency impairs the recruitment of 53BP1 to DSB sites and inhibits NHEJ efficiency (Vazquez et al., 2016). SIRT7 also deacetylates ATM to mediate ATM inactivation in the final stage of DNA damage repair . SIRT7 acts as a deglutarylase to regulate H4K91 glutarylation (H4K91glu). This process is closely associated with chromatin remodeling in response to DNA damage . Treatment with 5-Fluorouracil (5-FU) induces SIRT7 degradation in the Tat-binding protein 1 (TBP1)-mediated proteasome-dependent pathway, increasing cell radiosensitivity in combination therapy .
With the exception of DSBs, cancer cells can overcome DNA damage-induced cytotoxicity through BER, nucleotide excision repair (NER) and mismatch repair. Uracil-DNA N-glycosylase isoform 2 (UNG2) has a role in BER and can be deacetylated at K78 by HDAC, boosting disassociation from its E3 ubiquitinlike containing PHD and ring finger domain 1 (UHRF1) when stimulated by ROS. HDACi combined with genotoxic agents results in UNG2 degradation, resulting in a robust cell death effect . SIRT1 interacts with xeroderma pigmentosum group A (XPA) in NER by directly deacetylating XPA or mediating XPA binding to ATR to prevent UV irradiation (Fan and Luo, 2010;Jarrett et al., 2018). HDAC10 is mainly involved in DNA mismatch repair by deacetylating mutS homolog 2 (MSH2) at K73 (Radhakrishnan et al., 2015).
Besides, a number of other histone modifications are also actively involved in these DNA repair pathways, but are beyond the scope of this review (Cao et al., 2016;. Suffice to say that the multiple sites of H3 and H4 acetylation are not absolutely related to checkpoint activation because the conversion of lysine to other amino acids can still activate checkpoints (Robert et al., 2011).

Cell Cycle
Cell cycle dysregulation is a central hallmark of oncogenesis; as such, cell cycle regulators are considered promising targets for cancer treatment. HDACs are often involved in cell cycle checkpoints. HDAC3-mediated deacetylation of cyclin A affects the progression of the S phase and G2/M transitions (Bhaskara et al., 2008(Bhaskara et al., , 2010. HDAC10 depletion induces G2-M transition arrest through the regulation of cyclin A2. Mechanically, HDAC10 depletion induces the downregulation of high mobility group AT hook 2 (HMGA2), which leads to enrichment of E4F transcription factor 1 (a cyclin A2 repressor) at the cyclin A2 promoter and G2-M arrest . Regarding combination therapies, the CDK9 inhibitor dinaciclib and panobinostat together induce apoptosis over the short-term in MLL-AF9-driven acute myeloid leukemia (AML) (Baker et al., 2016). The emerging hybrid inhibitor Roxyl-zhc-84, which concordantly inhibits HDACs and CDKs, induces G1phase arrest and apoptosis in ovarian and breast cancer cells (Huang et al., 2018c).

Apoptosis
Apoptosis is a physiologically programmed cell death pathway that is essential for the maintenance of organismal homeostasis. Apoptosis is controlled by the B-cell lymphoma 2 (Bcl-2) family of proteins, which includes both pro-survival and pro-apoptotic proteins that control cell fate .
Regarding the intrinsic apoptotic pathway, the Bcl-2 interacting mediator of cell death (Bim), a Bcl-2 homology 3 (BH3)-only proapoptotic protein, is upregulated by depsipeptide via FoxO1 acetylation (Yang et al., 2009). Panobinostat elevates Sry-box transcription factor 7 (SOX7) expression and suppresses lung cancer cell proliferation. Mechanically, SOX7 triggers apoptosis by preventing Bim from proteasome-mediated degradation . The N-terminal truncated form of p63, Np63, belongs to the p53 family, but acts as an oncoprotein. In squamous cell carcinoma, HDAC1 and HDAC2 form a complex with Np63 to suppress the proapoptotic gene expression such as p53 upregulated modulator of apoptosis (PUMA) (Ramsey et al., 2011).
In summary, HDACi promote apoptosis via the intrinsic mitochondrial pathway, decreasing the expression of key antiapoptotic factors (eg. Mcl-1, Bcl-2, and Bcl-xL), and/or increasing the expression of pro-apoptotic proteins (eg. BAX, Bim, Noxa and PUMA). HDACi also facilitate the activation of extrinsic apoptotic pathways, such as TRAIL (Nebbioso et al., 2017), driving mitochondrial outer membrane polarization (MOMP) and ultimately caspase-mediated cell death.

Degradation System
The modulation of protein degradation is of critical importance for cell function. Protein degradation occurs via two major pathways: the ubiquitin-dependent proteasome pathway and autophagy system.

Autophagy
Autophagy is a degradation process whereby autophagosomes engulf and recycle nutrient sources in response to energetic demands and organelle turnover (Mizushima et al., 2008;Mizushima, 2018). Autophagy can be effectively promoted by HDACs. For example, depletion of HDAC10 perturbs autophagy flux through increased LC3-II/I, and the accumulation of p62 and acidic vesicular organelles. HDAC10 inhibition results in increased sensitivity to cytotoxic reagents (Oehme et al., 2013). Sirt1 also forms complexes with autophagy related protein 5 (ATG5), ATG7 and ATG8 to promote autophagy, with organelles in Sirt1 −/− mice being markedly damaged . The SIRT1 and -2 inhibitor tenovin-6 activates p53 and seems to be a specific regulator of mitochondrial acetylation (Lain et al., 2008;Scholz et al., 2015). Tenovin-6 suppresses Ewing sarcoma cells by regulating the NOTCH signaling pathway (Ban et al., 2014), and perturbs autophagic flux in chronic lymphocytic leukemia (CLL) cells and pediatric soft tissue sarcoma cells . However, HDACs also interrupt autophagy, and various HDACi induce cancer cell death by promoting autophagy. In HDAC10-KO HeLa cells, chaperonemediated autophagy (CMA) instead of macroautophagy is activated by the accumulation of lysosome-associated protein type 2A (LAMP2A)-positive lysosomes and the degradation of CMA substrate glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (Obayashi et al., 2020). In response to serum starvation or oxidative stress, SIRT2 inhibition induces acetylated FoxO1 to locate in the cytoplasm, accelerating autophagy through interaction with ATG7 (Zhao et al., 2010). Under nutrient-rich conditions, FoxK1/2 bind to HDAC complex and restricts autophagic flux through the transcriptional repression of autophagy gene (Bowman et al., 2014). Under condition of nutrient deprivation, inhibition of the AKT serine/threonine kinase pathway facilitates nuclear import of FoxO3, which competitively replaces FoxK to bind the autophagy-associated gene promoters and upregulation of autophagy (Brunet et al., 1999;Bowman et al., 2014). Moreover, VPA activates autophagy by blocking HDAC1-mediated regulation of AKT pathway . The nutrient-sensor mammalian target of rapamycin (mTOR) negatively modulates downstream Unc-51like autophagy activating kinase 1 (ULK1) that is involved in the non-transcriptional autophagic pathway. SAHA induces mTOR suppression, which ultimately activates autophagy by upregulating ULK1 (Gammoh et al., 2012). Of note, SAHAinduced autophagy seems to serve as a pro-survival mechanism to ameliorate SAHA-induced apoptosis by downregulating apoptotic factors (Gammoh et al., 2012). VPA also induces Sae2 [C-terminal-binding protein interacting protein (CtIP) in human] degradation in an autophagy related manner to impair HR-mediated DNA repair (Robert et al., 2011). As such, it seems that autophagy performs a dual role in DNA damage repair, depending on the cell states or DNA damage degree (Guo and Zhao, 2020).

Proteasome-Dependent Degradation
In addition to autophagy, proteasome-dependent degradation is also critical for cell function. HDACs target various E3s to affect basal cellular function. For example, panobinostat upregulates the E3 cIAP2 that causes the ubiquitination and proteasomal degradation of MRE11, elevating cellular sensitivity to chemoradiation (Nicholson et al., 2017). HDAC6 also modulates aggresome formation and the clearance of polyubiquitinated and misfolded proteins (Kawaguchi et al., 2003). In terms of therapeutic development, suppressing the aggresome pathway results in the accumulation of misfolded proteins, causing autophagy-associated DNA damage and apoptosis of cancer cells (Rodriguez-Gonzalez et al., 2008). Proteasome inhibitors (PIs), such as the United States Food and Drug Administration (FDA)-approved bortezomib (BTZ), have similar roles in preventing the degradation of polyubiquitinmisfolded proteins, which increases the production of ROS and disturbs DNA repair in tumor cells (Perez-Galan et al., 2006). However, long-term treatment with BTZ leads to drug-resistance in most patients. Low concentrations of HDACi combined with BTZ can downregulate anti-apoptotic proteins and upregulate pro-apoptotic proteins, thus accelerating cell death (Dai et al., 2008;. Combining the HDAC6-selective inhibitor tubacin with BTZ induces significant anti-tumor activity triggering c-Jun NH2-terminal kinase (JNK)-caspase signaling and endoplasmic reticulum (ER) stress (Hideshima et al., 2005;Nawrocki et al., 2006). Another HDAC6 inhibitor, WT161, promotes the accumulation of acetylated tubulin and overcomes BTZ resistance to promote multiple myeloma (MM) cell death (Hideshima et al., 2016). RTS-V5, a dual inhibitor that targets HDAC6 and the 20S subunit of the proteasome, also possesses potent and selective anti-tumor activity in leukemia and MM cell lines (Bhatia et al., 2018).

Epithelial-Mesenchymal Transition, Cancer Stem Cells, and Senescence
Epithelial-mesenchymal transition (EMT) is characterized by the loss of the tight intercellular connections normally found in epithelial cells that then undergo cytoskeleton rearrangement and adopt the mesenchymal cell phenotype, which is associated with migration. Notably, cancer cell migration and invasion are promoted by a series of EMT-associated factors (such as SNAIL, ZEB, SLUG and TWIST) . These factors induce EMT-related stem cell properties and promote tumorigenesis via PTMs (Mani et al., 2008;Tam and Weinberg, 2013;Ye et al., 2015;Dai et al., 2020). S-nitrosylation of HDAC2 is regulated by endothelial nitric oxide synthase (eNOS) that is a crucial enzyme for NO synthesis, allowing ZEB1 re-activation (Cencioni et al., 2018). During hypoxia, HDAC3 is essential for the activation of mesenchymal gene expression by the interaction with WD repeat domain 5 (WDR5) (Wu et al., 2011). The transforming growth factorβ (TGF-β)-SMAD signaling pathway is the most important EMT stimulation pathway. HDAC6 also has an essential role in EMT by activating SMAD3 (Shan et al., 2008). SMAD3 and -4 induce SIRT7 transcriptional repression by forming a complex with HDAC8. HDAC8 inhibition significantly suppresses TGFβ signaling via SMAD-SIRT7 axis, and as a consequence, attenuates lung metastases of breast cancer . Class I HDACi 4SC-202 notably attenuates TGF-β-induced EMT (Mishra et al., 2017). By contrast, HDAC10 exhibits a potential TSG role by downregulating Sry-box transcription factor 9 (SOX9) in KRAS-driven lung adenocarcinoma. Furthermore, HDAC10 deficiency results in TGF-β pathway activation, leading to the induction of SOX9 and KRAS-expressing stem-like tumor growth . Cancer-associated fibroblasts (CAFs) secrete extracellular matrix (ECM) that assists tumor progression and invasion. Scriptaid, a selective inhibitor of HDAC1, −3, and −8, represses TGF-β-mediated CAF by inhibiting ECM secretion and cell invasion (Kim D. J. et al., 2018). Of note, E-cadherin inhibits EMT, thus reduced E-cadherin expression indicates that "stemness" is increasing in cancer cells. HDAC and the 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMGR) dual inhibitor JMF3086 restores E-cadherin expression and attenuates vimentin expression and stemness in NSCLC, which recovers sensitivity to gefitinib which is an epidermal growth factor receptor (EFGR) tyrosine kinase inhibitor (TKI) (Weng et al., 2019).
Cancer stem cells are hard to eradicate and prone to drug-resistance. HDAC3 interacts with p53 and forms complexes with tumor antigens melanoma antigen family A2 (MAGE-A2), establishing the resistance of melanoma cells to chemotherapeutic agents (Monte et al., 2006). In refractory and recurrent leukemia, HDAC8-selective inhibitor significantly restores acetylation and p53 activity, inducing apoptosis of AML cells but not of normal hematopoietic stem cells . In addition, SIRT1 inhibition increases the efficiency of BCR-ABL TKI imatinib mesylate to eliminate quiescent leukemia stem cells by reactivating p53 .
Sirtuins are also closely involved in aging-related oncogene expression. Both SIRT6 and SIRT7 modulate long interspersed elements-1 (LINE-1, L1) expression and retrotransposition. SIRT6 mono-ADP-ribosylates the Krüppel-associated box domain-associated protein 1 (KAP1/TRIM28) and facilitates the KAP1 interaction with HP1α, resulting in the packaging of L1 elements into heterochromatin. SIRT7 directly binds L1 elements and promotes L1 sequences association with the nuclear lamina protein (Lamin A/C) by deacetylating H3K18 (Van Meter et al., 2014;Vazquez et al., 2019). These repressive functions of L1 highlight the protective roles of SIRTs on genome stability through preventing retrotransposition events. SIRT6 and SIRT7 deficiency result in aberrant heterochromatin and L1 activation especially in age-related diseases. Nucleoside reversetranscriptase inhibitors can reverse a SIRT6 and SIRT7 deficiency by upregulating different immune signalings, such as the type I IFN pathway and cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway, respectively Bi et al., 2020;Figure 4).

HDAC INHIBITORS IN CANCER THERAPY
In the 1970s, sodium butyrate was discovered to transform red leukemia cells into normal cells, and to resynthesize hemoglobin. This process was accompanied by strong histone hyperacetylation, and resulted in the discovery of the first HDACi (Ginsburg et al., 1973;Riggs et al., 1977;Vidali et al., 1978). In Tsuji et al. (1976) isolated the first natural HDACi, TSA, which was derived from Streptomyces hygroscopicus. Following the discovery of TSA, trapoxin was isolated from fungi and also found to act as an HDACi (Itazaki et al., 1990). A number of natural inhibitors have since been extracted from fungi, marine life, and plants that contain sulfur, polyphenol, flavonoid, terpenoid, selenium, and other organic molecules (Newkirk et al., 2009;Lascano et al., 2018;Singh et al., 2018). At present, HDACi are mainly divided into four categories following the FDA approval: (i) hydroxamic acids or hydroxamates, such as SAHA, panobinostat and belinostat; (ii) cyclic peptides, including depsipeptide; (iii) benzamides, such as chidamide; and (iv) shortchain fatty acids, including VPA (Li and Zhu, 2014;Seto and Yoshida, 2014;Li and Seto, 2016; Table 3).

Efficiency of HDACi
From preclinical studies to clinical trials, HDACi have demonstrated powerful therapeutic effects in various cancers. HDACi can significantly attenuate tumor burden by limiting tumor growth and restraining aberrantly proliferated vessels (Guerriero et al., 2017). HDACi can also induce DNA damage, cell cycle arrest, apoptosis and autophagy to promote cancer cell death mentioned above. Some novel SIRT inhibitors, such as MC2494, MHY2245, MHY2256, tenovin-6, and YC8-02, also perform diverse anti-tumor activities through mediating FIGURE 4 | Overview of HDAC-involved biological functions and therapeutic targets. An overview of HDAC-involved biological functions including transcription, metabolism, oxidative stress, redox, protein degradation, cell cycle, DNA damage repair, apoptosis, angiogenesis, EMT, immunity, and stemness. There diverse functions could establish single or synergistic therapeutic targets. apoptosis or autophagy (Carafa et al., 2018(Carafa et al., , 2019De et al., 2018;Tae et al., 2018Tae et al., , 2020Igase et al., 2020).
Activation of the immune response by HDACi could also be an effective innate method to prevent cancer relapse when administered in a regimen with immunotherapeutic (Guerriero et al., 2017;. The class IIa HDACi TMP195 efficiently improves the durability of tumor reduction in breast cancer by strengthening the phagocytic role of macrophages that are involved in the IFNγ axis; it also activates the adaptive antitumor immune response. Upon immune checkpoint blockade, TMP195 combined with an anti-programmed cell death-1 (PD-1) regimen could significantly reduce the tumor volume and induce a durable response in breast cancer (Guerriero et al., 2017). Myeloid-derived suppressor cells (MDSCs) suppress T-cell functions and promote tumor metastasis via the formation of an immunosuppressive tumor microenvironment (Azzaoui et al.,  2016; Betsch et al., 2018). Entinostat exhibits remarkable curative effects when combined with PD-1 and cytotoxic T-lymphocyteassociated antigen-4 (CTLA-4) blockade; it has been shown to significantly reduce the number of MDSCs (Kim et al., 2014). 5azacytidine combined with entinostat can also suppress MDSCs by downregulating C-C chemokine receptor type 2 (CCR2) and C-X-C chemokine receptor type 2 (CXCR2) expressions that ultimately, stimulates MDSC differentiation into a macrophagelike phenotype (Lu Z. et al., 2020). In human epidermal growthfactor receptor 2-positive (HER2 + ) breast cancer, panobinostat combined with trastuzumab (anti-HER2) stimulates the release of CXCR3-reactive chemokines and enhances the recruitment of tumor-associated natural killer (NK) cells to achieve eradication of tumors (Medon et al., 2017). The first FDA-approved HDAC inhibitor, SAHA, belongs to the hydroxamic acid class, approved to treat patients with cutaneous T cell lymphoma (CTCL). In many clinical trials, SAHA has proven effective against advanced and refractory tumors, alone or in combination with other inhibitors. Subsequently, the cyclic peptide romidepsin was approved by the FDA in 2009 to treat CTCL. Panobinostat and belinostat were both approved by the FDA in 2014 to treat peripheral T cell lymphoma (PTCL), with belinostat gaining additional approval from the European Medicines Agency (EMA). Both panobinostat and belinostat are classified as hydroxamic acids. Meanwhile, chidamide has become the first benzamide HDACi to be approved by the China Food and Drug Administration(CFDA) in 2015, for the treatment of relapsed and refractory PTCL 1,2,3,4 . Most SIRT inhibitors still remain in the preclinical stages. So far, only nicotinamide (vitamin B3) has been used to treat cancer in clinical trials (e.g., NCT02416739 and NCT00033436). Nicotinamide has shown a potential role in inhibiting nonmelanoma skin cancers that are principally generated by UV . Compared to other HDACi, nicotinamide exhibits the most catalytic sites: it is predominantly sensitive to acetylation sites in nuclear proteins that are involved in diverse biological processes (Scholz et al., 2015).

Application of Selective Inhibitors and Combination Therapy
With the development of HDACi, numerous clinical trials are ongoing or completed currently for cancer therapy. Many HDACi have already been approved for hematological malignancies and have also been developed and tested in clinical trials, such as the dual HDAC and phosphoinositide-3 kinase (PI3K) inhibitor CUDC-907 (also called fimepinostat), which has been reported to inhibit Myc transcriptional expression and reduce Myc-mediated proliferation of multiple cancer cell lines (Pei et al., 2016;Kotian et al., 2017;Pal et al., 2018;Guo et al., 2019). The safety, tolerability and efficacy of CUDC-907 has been assessed in phase I/II trials (Younes et al., 2016). CUDC-101 is another multipletarget inhibitor that blocks HDACs, the epidermal growth factor receptor (EGFR) and HER2 in head and neck squamous cell cancer (Galloway et al., 2015). Combination drugs can inhibit tumorigenesis from different aspects. A clinical phase I trial of SAHA combined with the autophagy inhibitor MLN9708 shows potential for this regimen in advanced p53-mutant malignancies (NCT02042989). Hydroxychloroquine (HCQ) is a common autophagy-targeting reagent that has been used in clinical research (Mahalingam et al., 2014). Several phase I/II clinical studies (e.g., NCT02316340 and NCT01023737) have thus tested the safety, tolerability and pharmacological efficacy of SAHA combined with HCQ in solid tumors. A phase I/II trial targeting BTZ-resistant MM cancer has found that a synergistic regimen of using ricolinostat and dexamethasone could be safe and well-tolerated in affected patients (Vogl et al., 2017). A recent phase III clinical trial found that a combination of chidamide and exemestane has therapeutic potential for postmenopausal patients with advanced, hormone receptor-positive (HR + ), HER2 − breast cancer and who have failed to respond to endocrine therapy . Combining the SIRT1 inhibitor Ex527 with the WEE1 inhibitor MK-1775 produces efficient effects in lung cancers by impairing HR repair and mitotic catastrophe-associated apoptosis .
The overall survival (OS) and progression-free survival (PFS) are common and quite important indicators in the clinical trials. A phase I/II clinical study targets patients with metastatic clear cell renal cell carcinoma (ccRCC) by treatment with entinostat and high-dose interleukin-2 (IL2) that downregulates forkhead box P3 (Foxp3) expression and function of regulatory T cells (Treg). The median PFS reaches 13.8 months, and the median OS is 65.3 months (Pili et al., 2017) [NCT01038778]. A phase III trial enrolled 768 patients with relapsed MM exhibits that panobinostat shows the median OS of panobinostat (40.3 months) versus that of placebo (35.8 months) based on the existing treatments of bortezomib and dexamethasone. And patients who had received previous regimens such as immunomodulatory drug and bortezomib, median OS was only 25.5 months when received panobinostat, bortezomib, and dexamethasone versus that was merely 19.5 months who received placebo (San-Miguel et al., 2016) [NCT01023308]. These HDACi significantly exhibit potential median OS and PFS for patients with advanced tumors.

Limitations of HDACi
However, in a randomized, double-blind and placebo-controlled phase III trial, vorinostat did not improve OS and could not be recommended as a therapy as a second-line or third-line drug for patients with advanced malignant pleural mesothelioma (Krug et al., 2015). Moreover, a phase III study that recruited 370 patients with Sézary syndrome or relapsed or refractory mycosis fungoides in CTCL from different countries of the world showed that the overall response rate to vorinostat was less efficient than that to mogamulizumab, a novel monoclonal antibody against CCR4 that significantly prolongs PFS . The overall response to vorinostat in this study was significantly lower than reported in a previous study (Olsen et al., 2007). Besides, in a phase II trial, mocetinostat did not reverse chemoresistance in patients with previous gemcitabine-resistant leiomyosarcoma and could not significantly prolong the median PFS of patients (Choy et al., 2019).
Sirtuins constitute a relatively unique class of HDACs; as such, SIRT modulators are attracting great interest in the research community (Yang et al., 2017;Dai et al., 2018;Wang Y. et al., 2019). A novel SIRT7 inhibitor has been identified to inhibit tumor growth by blockade of the direct interaction of SIRT7 and p53 (Vakhrusheva et al., 2008;Kim J. H. et al., 2019). However, there are also several reports demonstrating an indirect interaction between SIRT7 and p53 (Barber et al., 2012;Lu Y. F. et al., 2020;Yu et al., 2020). Thus, it is urgent to develop specific SIRT inhibitors for cancer therapy according to reasonable mechanisms. It is worth noting that SIRT activators and sirtuin-activating compounds (STACs) have been developed and studied in clinical trials to investigate their anti-aging, anti-inflammatory and metabolic regulatory effects (Howitz et al., 2003;Hubbard et al., 2013;Sinclair and Guarente, 2014;Bonkowski and Sinclair, 2016;Carafa et al., 2016;Huang et al., 2018b). Nevertheless, the SIRT1 activator SRT2183 has also been shown to inhibit growth and to promote cell death by causing ER stress in glioma cells (Ye et al., 2019).
Besides, most of clinical trials of HDACi have reported many adverse effects, including bleeding caused by different grades of thrombocytopenia, susceptibility to infection caused by neutropenia, anemia caused by hemoglobin reduction, arrhythmia, myocardial hypertrophy, neurotoxicity, and gastrointestinal toxicity such as nausea, vomiting, fatigue, diarrhea as well as electrolyte disturbance such as hypophosphatemia and hyponatremia. The most adverse effect reported is cell death caused by continuous cytotoxicity (Medina et al., 1997;Zhu et al., 2001b), as these agents also kill immortalized and normal cells from different tissues (Lee et al., 1996;Burgess et al., 2004).

DISCUSSION AND PERSPECTIVES
To date, only five HDACi have been approved by health authorities globally. Numerous clinical trials are currently evaluating the safety, application, and therapeutic benefits of HDAC inhibition when treating cancers, neurological disorders and other human diseases. Thus far, it is clear that FDA-approved HDACi have both beneficial and adverse effects. These effects might be accompanied by genomic instability, abnormal gene transcription, interference with chaperone protein function and free radical generation, which currently limit the therapeutic potential of HDACi.
Analyses of HDAC structural conformations and molecular mechanisms are necessary to improve treatment development. Complicating the application of HDACi is that the modification of the substrates by HDAC has spatio-temporal and tissue specificity. Consequently, dose-dependent and time-dependent treatments have different effects on gene expression regulation, various protein PTMs and chromatin remodeling. For now, combining different drugs to inhibit pathways such as tumor proliferation or angiogenesis, or to stimulate apoptosis requires more consideration.
Isolating the anti-cancer effects of HDACi and then synthesizing molecules with highly specific targets could be a promising avenue for cancer treatment in the future. Further investigation into combination treatments involving oncoprotein inhibitors and specific HDACi is also warranted. Overall, it seems that combination therapies have the advantage of reducing drug toxicity and lowering dose demand. SIRT protective factors should also be considered. Pending additional work to clarify HDACi that target specific HDACs or can be combined with other treatments, such as DNA methylation inhibitors or autophagy inhibitors, could be of great benefit to patients with cancers that have failed to respond to conventional treatments.

AUTHOR CONTRIBUTIONS
GL wrote the primary manuscript and YT revised the manuscript. W-GZ conceived and designed the manuscript. All authors contributed to the article and approved the submitted version.