Abstract

   Poly(ADP-ribose) polymerase 1 (PARP1) is an abundant nuclear protein
   involved in DNA repair, chromatin structure, and transcription.
   However, the regulation of its different functions remains poorly
   understood. Here, we report the role of PARP1 acetylation status in
   modulating its DNA repair and transactivation functions. We demonstrate
   that histone deacetylase 5 (HDAC5) determines PARP1 acetylation at
   Lys498 and Lys521 sites. HDAC5-mediated deacetylation at Lys498 site
   regulates PARP1 DNA damage response and facilitates efficient
   recruitment of DNA repair factors at damaged sites, thereby promoting
   cell survival. Additionally, HDAC5-mediated deacetylation at Lys521
   site promotes PARP1 coactivator function, resulting in induction of
   proliferative and metabolic genes in an activating transcription factor
   4-dependent manner. Thus, PARP1 induces metabolic adaptation to spur
   malignant phenotype. Our studies in mouse tumor models suggest that
   pharmacological inhibition of PARP1 enzymatic activity does not block
   tumor progression robustly as transactivation function remains
   unperturbed. These findings provide key mechanistic insights into PARP1
   regulation and expand its role in tumor development.
     __________________________________________________________________

   HDAC5-mediated regulation of acetylation status determines PARP1
   oncogenic functions.

INTRODUCTION

   Poly(ADP-ribose) polymerase 1 (PARP1) is a cellular stress responder
   and among the first signaling proteins being recruited to both
   single-stranded breaks and double-stranded breaks ([26]1). The
   catalytic function of PARP1 is activated in response to genotoxic
   stress by attaching a negatively charged polymer termed
   poly(ADP-ribose) (PAR) to itself and other protein targets ([27]2).
   Poly-ADP-ribosylation (PARylation) facilitates the recruitment of DNA
   repair proteins, including XRCC1, MREII, RAD51, and histone-modifying
   enzymes to DNA lesions ([28]3). In addition to its role as a DNA damage
   responder, PARP1 acts as a potent modulator of gene transcription which
   may not necessarily require its enzymatic activity ([29]4). The
   transcriptional regulatory functions of PARP1 encompass enhancer
   binding, association with insulators, modulation of chromatin
   structure, and direct regulation of transcription factors.
   PARP1-mediated transcriptional regulation plays a key role in human
   malignancies, including colorectal, breast, and gastric cancers
   ([30]5). However, the dynamic regulation of its DNA repair and
   transactivation functions is poorly understood.

   Histone deacetylase 5 (HDAC5) is a histone deacetylase belonging to the
   highly conserved class IIa family of Zn^2+-dependent histone
   deacetylases. HDAC5 is predominantly regulated by posttranslational
   modifications, which determine its subcellular localization.
   Ca^2+/calmodulin-dependent protein kinase has been reported to
   phosphorylate HDAC5, which promotes its binding with 14-3-3, resulting
   in cytoplasmic retention ([31]6). On the other hand, cAMP-dependent
   protein kinase A phosphorylates HDAC5 to disrupt its interaction with
   14-3-3, which promotes nuclear accumulation of HDAC5 ([32]7). At the
   posttranscriptional level, a few microRNAs have also been reported to
   regulate HDAC5 expression. miR-2861 has been observed to repress HDAC5
   expression to promote osteoblast differentiation ([33]8). HDAC5 not
   only catalyzes the deacetylation of nuclear histones but also
   deacetylates or forms complexes with other proteins to modulate
   oncogenic signaling pathways ([34]9). Consequently, HDAC5 plays a
   prominent role in tumor cell proliferation ([35]10), metastasis
   ([36]11), and metabolism ([37]12). Aberrant expression of HDAC5 has
   also been reported in diverse cancers, including breast, lung, and
   colorectal cancer ([38]13). Hence, exploring its substrate repertoire
   can enhance our understanding of its role in altered gene expression
   during tumorigenesis.

   In this study, we sought to investigate the role of PARP1 acetylation
   status in modulating its DNA repair and transactivation functions.
   Using a biochemical approach, we established HDAC5 as the bona fide
   deacetylase that determines PARP1 acetylation. HDAC5-mediated
   deacetylation at the Lys^498 site is critical for PARP1-dependent DNA
   damage response by promoting PARylation, thereby preventing chromatin
   trapping and triggering the recruitment of repair factors. Moreover,
   HDAC5 also deacetylates PARP1 at Lys^521, which modulates its
   transactivation function, thereby inducing metabolic reprogramming
   under genotoxic stress conditions. Thus, our study sheds light on the
   intricate regulatory mechanisms that determine the impact of PARP1
   acetylation status on DNA damage response and metabolic adaptation.

RESULTS

PARP1 interacts with HDAC5

   To understand the molecular axis and determine the precise role of
   HDAC5 under genotoxic stress conditions, we analyzed the acetylome upon
   HDAC5 depletion ([39]Fig. 1A). Our analysis revealed increased
   acetylation of various proteins, prominent among which was PARP1. To
   determine whether PARP1 is a bona fide HDAC5 substrate, we first
   performed coimmunoprecipitation experiments. Our data suggest that
   HDAC5 coimmunoprecipitated with PARP1 ([40]Fig. 1B, left). Concurring
   results were obtained upon reverse coimmunoprecipitation ([41]Fig. 1B,
   right). Direct interaction between PARP1 and HDAC5 was also confirmed
   by performing glutathione S-transferase (GST) pull-down experiments
   using bacterially expressed and purified proteins ([42]Fig. 1C and fig.
   S1A). We next examined the interaction between PARP1 and HDAC5 under
   genotoxic stress conditions including 5-fluorouracil (5FU) and
   cisplatin treatment. We observed that during the early phase of
   genotoxic stress (6 and 12 hours) when HDAC5 is nuclear, HDAC5
   coimmunoprecipitated with PARP1 ([43]Fig. 1, D and E, and fig. S1B).
   However, upon prolonged genotoxic stress (24 hours) when HDAC5
   undergoes nuclear export ([44]14), this interaction was lost. Reverse
   coimmunoprecipitation confirmed these findings ([45]Fig. 1F and fig.
   S1C). Similar observations were also made in other cell types including
   SW480 and normal colon epithelial cells CCD 841 CoN (fig. S1, D to G).
   We further examined the interaction between PARP1 and HDAC5 by
   performing proximity ligation assay (PLA). The PLA foci were notably
   reduced upon prolonged genotoxic stress as compared to the early phase
   of genotoxic stress ([46]Fig. 1, G and H). Moreover, upon HDAC5
   depletion, the foci were diminished, while no decline was observed over
   the time course of genotoxic stress in the presence of a constitutively
   nuclear HDAC5 mutant (HDAC5^S259/498A).

Fig. 1. PARP1 is an HDAC5-interacting protein.

   [47]Fig. 1.
   [48]Open in a new tab

   (A) Schematic workflow (left): HT29 cells were stably transfected with
   control (scrambled) or HDAC5 shRNA. Volcano plot of up-regulated
   acetylated peptides upon HDAC5 depletion (right). (B) HT29 cells were
   infected with indicated adenoviruses. Twenty-four hours after
   infection, immunoprecipitation (IP) followed by Western blotting was
   performed. HA, hemagglutinin. (C) Glutathione S-transferase (GST)
   pull-down assay was performed. Western blotting was then performed for
   the indicated proteins. (D) HT29 cells were treated with 5-fluorouracil
   (5FU; 10 μM) for indicated time points. Western blotting was then
   performed from nuclear and cytoplasmic extracts. (E) HT29 cells were
   treated with 5FU (10 μM) for indicated time points, and the nuclear
   extracts were subjected to immunoprecipitation followed by
   immunoblotting. (F) HT29 cells were treated with 5FU (10 μM) for
   indicated time points, and the nuclear extracts were subjected to
   immunoprecipitation. Western blotting was then performed. (G and H)
   HT29 control [Control; scrambled HDAC5 short hairpin RNA (shRNA)],
   HDAC5 knockdown (HDAC5kd; HDAC5 shRNA), and HDAC5 knockdown cells
   expressing HDAC5 mutant (HDAC5kd/HDAC5^S259/498A) were subjected to
   proximity ligation assay (PLA) [(G), left]. Representative images are
   shown [(G), right]. Scale bars, 10 μm. The number of PLA foci per cell
   was quantified and plotted (H). Statistical analyses were done using
   two-way analysis of variance (ANOVA) (Bonferroni’s post hoc test). The
   data are representative of three independent experiments. Error bars
   are means ± SD. ***P < 0.001. (I) HT29 cells were transfected with an
   empty vector (EV) or the indicated FLAG-tagged PARP1 constructs.
   Twenty-four hours after transfection, nuclear extracts were subjected
   to immunoprecipitation. Western blotting was then performed (left).
   Schematic representation of the PARP1 full-length and deletion mutants
   (right). (J) HT29 cells were transfected as indicated. Twenty-four
   hours after transfection, nuclear extracts were subjected to
   immunoprecipitation. Western blotting was then performed (left).
   Schematic representation of the HDAC5 full-length and deletion mutants
   (right). (A), (G), (I), and (J) created with [49]BioRender.com.

   To map the domain of PARP1 to which HDAC5 binds, we performed
   immunoprecipitation experiments using FLAG constructs expressing
   different domains of PARP1. The PARP1-FLAG segment containing 332 to
   518 amino acids specifically bound to the HDAC5 protein ([50]Fig. 1I).
   Using Flag-tagged constructs, we also mapped the domain of HDAC5, which
   binds to PARP1. We observed that the HDAC5-Flag segment containing
   amino acids 672 to 1122 bound to PARP1 ([51]Fig. 1J). Together, these
   results indicate that the C-terminal deacetylase domain of HDAC5 binds
   to the automodification domain (BRCT) of PARP1.

HDAC5 deacetylates PARP1 at Lys^498 and Lys^521

   PARP1 has been reported to be acetylated by p300 which determines its
   transactivation function ([52]15). To explore the functional
   consequence of PARP1-HDAC5 interaction, we examined PARP1 acetylation
   status upon genotoxic stress. Our results suggest that HDAC5
   deacetylates PARP1 during the early phase of genotoxic stress ([53]Fig.
   2A). However, upon prolonged genotoxic stress when HDAC5 undergoes
   nuclear export, PARP1 was acetylated. Furthermore, upon HDAC5
   depletion, PARP1 was constitutively acetylated over the time course of
   genotoxic stress. The PARP1 acetylation observed was p300-mediated as
   it was abolished upon p300 depletion ([54]Fig. 2B).

Fig. 2. PARP1 is deacetylated at Lys^498 and Lys^521 sites by HDAC5.

   [55]Fig. 2.
   [56]Open in a new tab

   (A) HT29 control (Control; scrambled HDAC5 shRNA) and HDAC5 knockdown
   (HDAC5kd; HDAC5 shRNA) cells were treated with 5FU (10 μM) for
   indicated time points. Western blotting was then performed for the
   indicated proteins from nuclear extracts. (B) HT29 cells were stably
   transfected (pooled zeomycin-resistant population) with control
   (Control; Scrambled p300 shRNA) or p300 knockdown (p300kd; p300 shRNA).
   These cells were treated with 5FU (10 μM) for indicated time points.
   Western blotting was then performed for the indicated proteins from
   nuclear extracts. (C) HT29 PARP1 and HDAC5 double knockdown
   (HDAC5kd/PARP1kd; HDAC5 and PARP1 shRNA) cells were transfected with
   FLAG-tagged wild-type PARP1, PARP1^K498R, PARP1^K505R, PARP1^K508R,
   PARP1^K521R, or PARP1^K532R constructs as indicated. Twenty-four hours
   after transfections nuclear extracts were prepared and subjected to
   immunoprecipitation followed by Western blotting for the indicated
   proteins. (D) AcK498-PARP1 and AcK521-PARP1 peptides were incubated
   either alone (Control) or in the presence of recombinant HDAC5. Mass
   spectrometry was then performed. The relative positions of acetylated
   Lys^498 (1068.60 Da) and Lys^521 (1389.79 Da), while deacetylated
   Lys^498 (1026.60 Da) and Lys^521 (1347.79 Da) PARP1 peptides are
   indicated. m/z, mass/charge ratio. (E) Results of HDAC5 deacetylation
   reactions using acetylated PARP1 peptides. (F) Schematic representation
   of the PARP1 domains. The lysine residues in the automodification
   domain are sites of HDAC5-mediated deacetylation. Created with
   [57]BioRender.com. (G) The amino acid sequence encompassing human PARP1
   K498 (red) and K521 (blue) residue. (H) HT29 control (Control;
   scrambled HDAC5 shRNA) and HDAC5 knockdown (HDAC5kd; HDAC5 shRNA) cells
   were treated with 5FU (10 μM) for indicated time points. Western
   blotting was then performed for the indicated proteins. (I) HT29 HDAC5
   knockdown (HDAC5kd; HDAC5 shRNA) cells were stably transfected with
   Flag-tagged HDAC5 mutant constructs. These cells were treated with 5FU
   (10 μM) for indicated time points. Western blotting was then performed
   for the indicated proteins.

   To identify the site of PARP1 deacetylation, we mutated all five lysine
   residues of the BRCT domain reported to be acetylated by p300. We
   observed that the mutation of K498 and K521 residues results in the
   down-regulation of acetylation in the absence of HDAC5 ([58]Fig. 2C).
   We performed an in vitro deacetylation assay using PARP1 peptides
   containing different acetylated lysine residues. MS analysis revealed
   that HDAC5 exhibits specific deacetylase activity toward the PARP1
   peptides containing acetylated K498 and K521 residues ([59]Fig. 2, D
   and E), which lie in the automodification domain of PARP1 and are well
   conserved ([60]Fig. 2, F and G). We next examined the effect of HDAC5
   on the PARP1 K498 and K521 acetylation status under genotoxic stress
   conditions. Acetylation at PARP1 K498 and K521 sites was observed only
   at extended periods of genotoxic stress concomitant to the decline in
   nuclear HDAC5 levels ([61]Fig. 2H). On the other hand, in the absence
   of HDAC5, PARP1 K498 and K521 residues were constitutively acetylated
   over the time course of genotoxic stress. Similar results were obtained
   in other cell types (fig. S2, A to C). In addition, in the presence of
   deacetylase-dead HDAC5 mutant (HDAC5^H893F) PARP1 K498 and K521
   acetylation remained unaltered upon genotoxic stress, which was
   abrogated in the presence of a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A) ([62]Fig. 2I). Previous studies suggest that
   homologous recombination repair (HRR) proteins determine PARP1’s
   functions ([63]1). Since the cell lines used in our study do not harbor
   a mutation in HRR genes ([64]16, [65]17), we further examined the
   effect of HRR factors BRCA1 and BRCA2 on HDAC5-mediated regulation of
   PARP1 acetylation status. Our results suggest that HDAC5 deacetylates
   PARP1 independent of the presence or absence of BRCA1 and BRCA2 (fig.
   S2, D and E). Together, these results suggest that HDAC5 deacetylates
   PARP1 at the conserved lysine 498 and lysine 521 residues. However,
   upon prolonged genotoxic stress when HDAC5 undergoes nuclear export,
   PARP1 is acetylated at these sites.

HDAC5-mediated deacetylation at Lys^498 determines PARP1 DNA damage response

   We next investigated the effect of PARP1 deacetylation on DNA damage
   response. We observed that upon prolonged genotoxic stress (24 hours),
   levels of phosphorylated histone H2AX (γH2AX), a DNA damage marker,
   were elevated concomitant to increased acetylated PARP1 levels due to a
   decline in nuclear HDAC5 levels. Likewise, γH2AX levels were also
   up-regulated upon HDAC5 depletion ([66]Fig. 3A). However, in the
   presence of a constitutively nuclear HDAC5 mutant (HDAC5^S259/498A),
   γH2AX levels declined upon prolonged genotoxic stress together with
   down-regulation of acetylated PARP1 levels, but not in the presence of
   deacetylase-dead HDAC5 mutant (HDAC5^H893F). These results suggest that
   acetylation status determines PARP1-mediated DNA damage response.

Fig. 3. Deacetylation of PARP1 at Lys^498 by HDAC5 regulates DNA damage
response.

   [67]Fig. 3.
   [68]Open in a new tab

   (A) Indicated cells were treated with 5FU (10 μM). Nuclear extracts
   were subjected to Western blotting was then performed for the indicated
   proteins. (B) Indicated cells were treated with 5FU (10 μM) for
   indicated time points. Nuclear extracts were prepared. Western blotting
   was then performed for the indicated proteins. (C) Indicated cells were
   treated with 5FU (10 μM) for 6 hours. Nuclear extracts were then
   prepared and subjected to Western blotting for the indicated proteins.
   (D) Indicated cells were treated with 5FU (10 μM) for 6 hours. The
   cells were harvested and Western blotting was performed for the
   indicated proteins from the nuclear and chromatin fractions. (E and F)
   Indicated cells were treated with 5FU (10 μM) for indicated time
   points. The cells were subjected to PLA. Representative images are
   shown (E). Scale bars, 10 μm. The number of PLA foci per cell was
   quantified and plotted as a nested scatter dot plot (F). Statistical
   analyses were done using two-way ANOVA (Bonferroni’s post hoc test).
   The data are representative of three independent experiments. Error
   bars are means ± SD. ***P < 0.001. (G and H) Indicated cells were
   treated with 5FU (10 μM). The cells were cells subjected to alkaline
   comet assay (G). Scale bars, 50 μm. The tail moment of cells is shown
   in the nested scatter dot plot (H). The data are representative of
   three independent experiments. (I) Indicated cells were treated with
   5FU (10 μM). 5-Bromo-2-deoxyuridine (BrdU) incorporation followed by
   fluorescence-activated cell sorting analysis was performed. FITC,
   fluorescein isothiocyanate.

   Previous studies suggest that genotoxic stress results in DNA
   damage-induced PARP1 auto-PARylation which is determined by its
   acetylation status ([69]18). We observed that upon prolonged stress,
   abrogation of HDAC5-mediated deacetylation inhibits PARP1 PARylation.
   Upon HDAC5 depletion or in the presence of deacetylase-dead HDAC5
   mutant (HDAC5^H893F), PARylation was repressed throughout the course of
   genotoxic stress. On the other hand, in the presence of a
   constitutively nuclear HDAC5 mutant (HDAC5^S259/498A), PARP1 PARylation
   was observed throughout the course of genotoxic stress ([70]Fig. 3B).
   We further examined the role of each of the sites of deacetylation in
   determining PARP1 PARylation upon genotoxic stress. We observed that
   PARylation was down-regulated in the case of acetyl-lysine mimic PARP1
   mutant (PARP1^K498Q), but not in the case of acetyl-lysine mimic PARP1
   mutant (PARP1^K521Q) and non-acetylable PARP1 mutants (PARP1^K498R and
   PARP1^K521R) ([71]Fig. 3C). Upon HDAC5 depletion, PARylation was
   diminished in the case of PARP1^K498Q, PARP1^K521Q, and PARP1^K521R
   mutants but not in the case of PARP1^K498R mutant. These results
   suggest that Lys^498 deacetylation plays a key role in determining
   PARP1 PARylation. PARylation has been reported to prevent cytotoxic
   PARP1 trapping onto the chromatin and facilitate DNA repair ([72]19).
   Our data suggest that upon HDAC5 depletion or in the presence of
   deacetylase-dead HDAC5 mutant (HDAC5^H893F), there was increased PARP1
   trapping but not in the presence of a constitutively nuclear HDAC5
   mutant (HDAC5^S259/498A) ([73]Fig. 3D). Furthermore, increased
   chromatin binding was observed in the case of acetyl-lysine mimic PARP1
   mutant (PARP1^K498Q) but not in the case of a non-acetylable PARP1
   mutant (PARP1^K498R). Thus, we concluded that K498 acetylation status
   plays a key role in determining PARP1 chromatin trapping.

   We next examined the effect of acetylation status on PARP1-mediated
   recruitment of repair factors including MRE11 at sites of DNA damage
   ([74]3). MRE11 participates in both HR and NHEJ repair pathways
   ([75]20). It plays a critical role in the cellular decision of which
   mechanism will be used to repair damaged DNA. We observed a substantial
   decline in MRE11-PAR foci upon HDAC5 depletion or in the presence of
   HDAC5 mutant (HDAC5^H893F) from early periods of genotoxic stress
   ([76]Fig. 3, E and F, and fig. S3, A to F). However, robust MRE11
   recruitment was observed in the presence of HDAC5 mutant
   (HDAC5^S259/498A) or PARP1 mutant (PARP1^K498R). On the other hand, a
   notable reduction in foci was observed in the presence of PARP1 mutant
   (PARP1^K498Q). Thus, our data demonstrate that HDAC5-mediated
   deacetylation at Lys^498 promotes PARP1 PARylation which facilitates
   recruitment of repair factors. We further assessed the effect of PARP1
   acetylation status on DNA repair. We observed increased levels of DNA
   strand breaks upon HDAC5 depletion or in the presence of HDAC5 mutant
   (HDAC5^H893F) from early periods of genotoxic stress ([77]Fig. 3, G and
   H, and fig. S3, G to I). Similar observations were made in the case of
   PARP1 mutant (PARP1^K498Q). However, the extent of DNA damage was
   notably reduced in the presence of HDAC5 mutant (HDAC5^S259/498A) or
   PARP1 mutant (PARP1^K498R). We also investigated the physiological
   implications of unresolved DNA damage. Our data suggest that HDAC5
   depletion or the presence of HDAC5 mutant (HDAC5^H893F) results in
   increasing G1 arrest and cell death (sub-G1 population) from the early
   period of genotoxic stress. Similar observations were made in the case
   of PARP1 mutant (PARP1^K498Q). However, no discernible G1 arrest or
   cell death was observed in the presence of HDAC5 mutant
   (HDAC5^S259/498A) and PARP1 mutant (PARP1^K498R) over the time course
   of genotoxic stress ([78]Fig. 3I and fig. S3, J to L). These results
   suggest that HDAC5-mediated deacetylation at the K498 site plays a
   pivotal role in determining PARP1-dependent DNA damage response.

Lys^521 deacetylation enables PARP1 to function as ATF4 coactivator

   To unravel the functional relevance of K521 acetylation status, we used
   a biochemical approach ([79]Fig. 4A, left). We observed that many
   proteins specifically interacted with a non-acetylable PARP1 mutant
   (PARP1^K521R) but not with acetyl-lysine mimic PARP1 mutant
   (PARP1^K521Q) upon genotoxic stress ([80]Fig. 4A, right). Among the
   most prominent interacting proteins, interaction with activating
   transcription factor 4 (ATF4) was notably reduced with acetyl-lysine
   mimic PARP1 mutant (PARP1^K521Q) but not so with wild-type PARP1 or
   acetyl-lysine mimic PARP1 mutant (PARP1^K498Q) ([81]Fig. 4B). The
   interaction was also observed with a non-acetylable PARP1 mutant
   (PARP1^K498R). These data indicate that Lys^521 acetylation status
   determines PARP1 interaction with ATF4. We further examined the
   kinetics of PARP1-ATF4 interaction under genotoxic stress conditions.
   We observed that during the early phase of genotoxic stress (12 hours),
   PARP1 coimmunoprecipitated with ATF4. However, upon prolonged stress,
   the interaction was lost. Reverse coimmunoprecipitation also
   corroborated these observations (fig. S4A). Similar results were
   obtained upon depletion of BRCA1 or BRCA2 (fig. S4, B and C). We
   further examined the interaction between PARP1 and ATF4 by performing
   PLA. Our results indicate that PLA foci were up-regulated at the early
   phase of genotoxic stress but declined upon prolonged stress
   concomitant to PARP1 acetylation ([82]Fig. 4, C and D). Substantially
   reduced foci were observed upon HDAC5 depletion or in the presence of
   deacetylase-dead HDAC5 mutant (HDAC5^H893F). The foci were also
   similarly diminished in the presence of acetyl-mimetic PARP1 mutant
   (PARP1^K521Q). However, increased PLA foci were observed in the
   presence of a non-acetylable PARP1 mutant (PARP1^K521R) and a
   constitutively nuclear HDAC5 mutant (HDAC5^S259/498A). Thus, our
   results suggest that HDAC5 determines PARP1-ATF4 interaction under
   genotoxic stress conditions. We further examined whether HDAC5
   regulates ATF4 levels. As has been previously reported, we also
   observed that ATF4 levels are induced by genotoxic stress ([83]Fig. 4E)
   ([84]21). However, HDAC5 depletion did not affect ATF4 levels.

Fig. 4. HDAC5-mediated deacetylation at Lys^521 modulates PARP1 coactivator
function.

   [85]Fig. 4.
   [86]Open in a new tab

   (A) Schematic representation of workflow to identify mutant PARP1
   binding proteins upon 5FU (10 μM) treatment for 12 hours in HT29 cells
   (left). Volcano plot of proteins interacting with PARP1 mutant
   (PARP1^K521R) (right). Created with [87]BioRender.com. (B) HT29 PARP1
   knockdown (PARP1kd; PARP1 shRNA) cells were transfected with EV or
   FLAG-tagged wild-type or mutant PARP1 constructs as indicated.
   Twenty-four hours after transfection, the cells were treated with 5FU
   (10 μM) for 12 hours. Immunoprecipitation followed by Western blotting
   for the indicated proteins was performed. (C and D) HT29 control
   (Control; Luc shRNA), HDAC5 knockdown (HDAC5kd; HDAC5 shRNA), and PARP1
   knockdown cells (PARP1kd; PARP1 shRNA) expressing PARP1 mutants and
   HDAC5 knockdown cells expressing HDAC5 mutants were treated with 5FU
   (10 μM) for indicated time points. PLA was performed (C). Scale bars,
   10 μm. The number of PLA foci per cell was quantified and plotted (D).
   Statistical analyses were done using two-way ANOVA (Bonferroni’s post
   hoc test). Error bars are means ± SD. **P < 0.01; ***P < 0.001. (E)
   HT29 control (Control) and HDAC5 knockdown (HDAC5kd) cells were treated
   with 5FU (10 μM) for indicated time points. Western blotting was then
   performed. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. (F) HT29
   control (Control; Luc shRNA), PARP1 knockdown cells expressing
   PARP1^K521R (PARP1^K521R), as well as PARP1 and activating
   transcription factor 4 (ATF4) double knockdown cells (PARP1 and ATF4
   shRNA) expressing PARP1^K521R (PARP1^K521R/ATF4kd) were treated with
   5FU and subjected to RNA sequencing analysis. Red and blue indicate
   up-regulation and down-regulation, respectively. (G) HT29 control
   (Control; Luc shRNA), HDAC5 knockdown (HDAC5kd), PARP1 knockdown cells
   expressing PARP1 mutants, and HDAC5 knockdown cells expressing HDAC5
   mutants were treated with 5FU (10 μM) for indicated time points. A
   reverse transcription quantitative polymerase chain reaction was
   performed. Statistical analyses were done using two-way ANOVA (Tukey’s
   post hoc test). Error bars are means ± SD of three independent
   experiments with triplicate samples. ***P < 0.001.

   Since ATF4 plays a pivotal role in stress response by regulating the
   expression of diverse genes ([88]22), we next examined the impact of
   PARP1-ATF4 interaction on the cellular transcription program. We
   observed that under genotoxic stress conditions, PARP1 K521 acetylation
   status determines the expression of 2107 genes ([89]Fig. 4F).
   Furthermore, ATF4 co-depletion results in down-regulation of 755 genes.
   Pathway enrichment analysis suggests that these genes are involved in
   cancer risk pathways including epithelial-to-mesenchymal transition
   (EMT)/metastasis and cell cycle regulation as well as metabolic
   pathways such as glutamine metabolism and amino acid biosynthesis
   ([90]Fig. 5A). To gain further insights, we examined the effect of
   HDAC5-mediated PARP1 deacetylation on the expression of key
   ATF4-regulated genes involved in metabolism (SLC1A5, GLS1, and GLUD1),
   proliferation (CCNB1), and EMT (MMP9). Our results suggest that during
   the early phase of genotoxic stress, the transcript levels of these
   genes were up-regulated, but upon prolonged stress, the levels declined
   concomitantly to the loss of PARP1-ATF4 interaction. Furthermore, HDAC5
   depletion or the presence of deacetylase-dead HDAC5 mutant
   (HDAC5^H893F) resulted in the down-regulation of these genes under
   genotoxic stress conditions ([91]Fig. 4G and fig. S5, B to D). On the
   other hand, in the presence of a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A), a notable increase in transcript levels of these
   genes was observed over the course of genotoxic stress. Thus, the
   presence of a non-acetylable PARP1 mutant (PARP1^K521R) robustly
   induces the transcript levels of these genes, while in the presence of
   acetyl-lysine mimic PARP1 mutant (PARP1^K521Q), expression of these
   genes was down-regulated. These results suggest that HDAC5 plays a key
   role in determining PARP1-dependent ATF4-mediated transactivation. To
   corroborate our findings, we further examined the role of PAPR1
   acetylation status in determining ATF4 recruitment at the promoter of
   these genes. We observed that upon genotoxic stress, increasing levels
   of ATF4 were recruited at the promoters of these genes. Furthermore,
   the absence of HDAC5 or PARP1 acetylation status did not affect ATF4
   recruitment (figs. S6A, left; and S7, A to C, left). To ascertain the
   presence of PARP1 in the transcription complex, we performed reChIP
   assay (fig. S6A, right; and S7, A to C, right). Our results suggest
   that increasing levels of PARP1 were detected at the promoters at the
   early phase of genotoxic stress which declined upon prolonged stress.
   In the absence of HDAC5 or in the presence of deacetylase-dead HDAC5
   mutant (HDAC5^H893F), PARP1 presence at the promoters was
   down-regulated under genotoxic stress conditions. On the other hand, in
   the presence of a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A), notably higher levels of PARP1 were observed at
   these promoters. Thus, high levels of a non-acetylable PARP1 mutant
   (PARP1^K521R) were detected at these promoters throughout the course of
   genotoxic stress, while no notable levels of acetyl-lysine mimic PARP1
   mutant (PARP1^K521Q) could be detected. Together, these results suggest
   that HDAC5 regulates PARP1-dependent ATF4 transactivation function.

Fig. 5. PARP1 K521 acetylation status determines metabolic response upon
genotoxic stress.

   [92]Fig. 5.
   [93]Open in a new tab

   (A) HT29 control (Control; Luc shRNA), HDAC5 knockdown (HDAC5kd; HDAC5
   shRNA), PARP1 knockdown (PARP1kd; PARP1 shRNA), PARP1 knockdown cells
   expressing PARP1^K521R (PARP1^K521R), as well as PARP1 and ATF4 double
   knockdown cells expressing PARP1^K521R (PARP1^K521R/ATF4kd) were
   treated with 5FU (10 μM) for 12 hours. The metabolites were extracted
   from the harvested cells and analyzed by liquid chromatography–mass
   spectrometry (LC-MS). Relative levels of specific metabolites
   normalized to cell number are shown in the heatmap. The heatmap depicts
   relative changes in intracellular metabolites. Orange and cyan indicate
   up-regulation and down-regulation, respectively. (B) Metabolic pathway
   enrichment analysis of (A). Metabolite data are representative of three
   independent experiments. (C to G) HT29 control (Control), HDAC5
   knockdown (HDAC5kd), PARP1 knockdown (PARP1kd), PARP1 knockdown cells
   expressing PARP1^K521Q (PARP1^K521Q) or PARP1^K521R (PARP1^K521R),
   PARP1 and ATF4 double knockdown cells expressing PARP1^K521R
   (PARP1^K521R/ATF4kd), as well as HDAC5 knockdown cells expressing
   HDAC5^S259/498A (HDAC5kd/HDAC5^S259/498A) were treated with 5FU (10 μM)
   for indicated time points. The cells were harvested, and metabolites
   were extracted. The relative amounts of the indicated metabolites were
   quantified using respective kits. Statistical analyses were done using
   two-way ANOVA (Bonferroni’s post hoc test). The data are representative
   of three independent experiments. Error bars are means ± SD. ***P <
   0.001. (H) HT29 control (Control), HDAC5 knockdown (HDAC5kd), PARP1
   knockdown (PARP1kd), as well as PARP1 knockdown cells expressing
   PARP1^K521Q (PARP1kd/PARP1^K521Q) or PARP1^K521R (PARP1kd/PARP1^K521R),
   PARP1 and ATF4 double knockdown cells expressing PARP1^K521R
   (PARP1^K521R/ATF4kd), and HDAC5 knockdown cells expressing
   HDAC5^S259/498A (HDAC5kd/HDAC5^S259/498A) were treated with 5FU (10 μM)
   for indicated time points. The cellular oxygen consumption rate (OCR)
   was then measured. The data are representative of three independent
   experiments.

PARP1-ATF4 promotes metabolic adaptation

   Since our transcriptomics data suggests that PARP1 K521 acetylation
   status determines the expression of several metabolic genes, we
   performed an unbiased analysis to comprehend its role in the metabolic
   response under genotoxic stress conditions. Among diverse pathways, we
   observed prominent enrichment in pathways associated with glutamine and
   amino acid metabolism ([94]Fig. 5, A and B). Glutamine has been
   reported to serve as a carbon source for tricarboxylic acid (TCA) cycle
   intermediates to support the energy requirement of cancer cells
   ([95]23). Upon HDAC5 or PARP1 depletion, the levels of several key
   intermediate metabolites involved in glutamine and amino acids
   metabolism declined under genotoxic stress conditions. However,
   increased levels of these metabolites were observed in the presence of
   a non-acetylable mutant (PARP1^K521R) but were down-regulated upon ATF4
   co-depletion. We further examined the effect of PARP1 K521 acetylation
   status on glutamine utilization and amino acid biosynthesis over the
   course of genotoxic stress. Our results indicated that the levels of
   glutamine-derived intermediates (Glutamine, Glutamate, a-Ketoglutarate,
   and Succinate) along with amino acids (Asparagine, Arginine, Proline,
   and Ornithine) were induced at early time points of genotoxic stress
   but the levels declined upon prolonged stress ([96]Fig. 5, C to G, and
   fig. S8, A to C). HDAC5 or PARP1 depletion resulted in down-regulation
   of these metabolites over the course of genotoxic stress. In the
   presence of a constitutively nuclear HDAC5 mutant (HDAC5^S259/498A),
   the levels of these metabolites were elevated, while reduced levels
   were detected in the presence of acetyl-lysine mimic PARP1 mutant
   (PARP1^K521Q) over the time course of genotoxic stress. On the other
   hand, in the presence of a non-acetylable PARP1 mutant (PARP1^K521R),
   increasing levels of these metabolites were observed which were
   down-regulated upon ATF4 co-depletion. To corroborate our findings, we
   also examined the effect of PARP1 K521 acetylation status on cellular
   energetics. OCR (oxygen consumption rate) levels were elevated at the
   early phase of genotoxic stress but declined upon prolonged stress
   ([97]Fig. 5H). A notable decline in OCR levels was also observed upon
   HDAC5 or PARP1 depletion under genotoxic stress conditions. Elevated
   OCR levels were observed in the presence of a constitutively nuclear
   HDAC5 mutant (HDAC5^S259/498A), while reduced levels were observed in
   the presence of acetyl-lysine mimic PARP1 mutant (PARP1^K521Q) over the
   time course of genotoxic stress. In addition, in the presence of a
   non-acetylable PARP1 mutant (PARP1^K521R), increasing OCR levels were
   observed which declined upon ATF4 co-depletion. These results suggest
   that during the early phase of genotoxic stress, PARP1-ATF4 interaction
   promotes metabolic adaptation, which is perturbed upon prolonged stress
   due to PARP1 acetylation at the K521 site and subsequent loss of
   interaction with ATF4.

HDAC5 augments PARP1 oncogenic potential

   We next investigated the effect of PARP1 acetylation status on
   malignant phenotype under genotoxic stress conditions. We observed that
   the depletion of HDAC5 or PARP1 resulted in reduced proliferation
   capacity, while the presence of a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A) increased proliferation capacity ([98]Fig. 6, A and
   B, and fig. S9, A to D). Proliferation capacity was also higher in the
   presence of non-acetylable PARP1 mutants, PARP1^K498R, PARP1^K521R, and
   PARP1^K498R/K521R, which was down-regulated upon treatment with
   pharmacological PARP1 inhibitor olaparib in the case of PARP1^K498R but
   not in the case of PARP1^K498R/K521R. More robust down-regulation was
   observed in the case of stronger PARP1 inhibitor talazoparib (fig. S9,
   E and F) ([99]19). We next examined the effect of HDAC5-mediated PARP1
   deacetylation on invasiveness and migration potential. Our results
   suggest that invasiveness and migration potential declined upon
   depletion of HDAC5 or PARP1 under genotoxic stress conditions but was
   up-regulated in the presence of a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A) ([100]Fig. 6, C and D, and fig. S10, A to D). In
   addition, in the presence of non-acetylable PARP1 mutants, PARP1^K498R,
   PARP1^K521R, and PARP1^K498R/K521R, invasiveness and migration
   potential were elevated, which was repressed upon olaparib treatment in
   the case of PARP1^K498R but not in the case of PARP1^K498R/K521R. The
   repression was more prominent in the case of talazoparib treatment
   (fig. S10, E and F).

Fig. 6. HDAC5-mediated deacetylation induces PARP1 oncogenic functions.

   [101]Fig. 6.
   [102]Open in a new tab

   (A and B) Indicated cells were treated with 5FU (4 μM). Cells were also
   subjected to olaparib cotreatment (5 mM) as indicated. Colonies were
   enumerated after 7 days. Representative images are shown (A). Results
   were plotted as a number of colonies formed after 7 days (B).
   Statistical analyses were done using one-way ANOVA (Tukey’s post hoc
   test). The data are representative of three independent experiments.
   Error bars are means ± SD. ***P < 0.001. (C) Indicated cells were
   treated with 5FU (10 μM). Cells were also subjected to olaparib
   cotreatment (5 mM) as indicated. In vitro invasion potential was
   measured. Error bars are means ± SD of three independent experiments
   with triplicate samples. ***P < 0.001. (D) Indicated cells were treated
   with 5FU (10 μM). Cells were also subjected to olaparib cotreatment (5
   mM) as indicated. The migration potential was then measured. Error bars
   are means ± SD of three independent experiments with triplicate
   samples. ***P < 0.001. (E) Indicated cells were subcutaneously injected
   into the right flank of nude mice. After 1 week of injection, the mice
   were administered 5FU (25 mg/kg) every alternate day. Olaparib was also
   coadministered at a dose of 10 mg/kg daily as indicated. Tumor volume
   was measured on the indicated days. The data are representative of
   three independent experiments (n = 5 mice per group). Error bars
   represent means ± SD from five individual mice. **P < 0.01; ***P <
   0.001. Statistical analyses were done using one-way ANOVA (Tukey’s post
   hoc test). (F) Tumor lysates [(E) above] were subjected to Western
   blotting. (G to K) Immunohistochemical analysis was performed on the
   tumor sections from (E) above (G). Scale bars, 50 μm. The relative
   intensity was quantified and represented as means ± SD of three
   independent experiments [(G) to (K)]. ***P < 0.001. Statistical
   analyses were done using one-way ANOVA (Tukey’s post hoc test).

   Since our data suggest that PARP1 induces the expression of genes
   associated with EMT in an ATF-dependent manner, we further investigated
   the effect on cellular morphology and EMT markers. We observed that
   under genotoxic stress conditions, the presence of a non-acetylable
   PARP1 mutant (PARP1^K521R) or a constitutively nuclear HDAC5 mutant
   (HDAC5^S259/498A) induced cellular morphological changes indicative of
   EMT with loss of cell-cell contacts and cell scattering (fig. S10G).
   The cellular morphological changes were corroborated by the reduced
   levels of epithelial marker E-cadherin and up-regulation of mesenchymal
   markers vimentin and fibronectin (fig. S10H). On the other hand, the
   presence of acetyl-lysine mimics PARP1 mutant (PARP1^K521Q), and
   depletion of HDAC5 or PARP1 did not induce EMT.

   We further investigated the effect of HDAC5-PARP1 interplay on
   tumorigenicity. Tumor growth was attenuated upon HDAC5 or PARP1
   depletion, while substantially larger tumors were observed in the
   presence of HDAC5 mutant (HDAC5^S259/498A) ([103]Fig. 6, E and F).
   Moreover, non-acetylable PARP1 mutants, PARP1^K498R, PARP1^K521R, and
   PARP1^K498R/K521R, promoted tumor growth, which was suppressed upon
   olaparib treatment in the case of PARP1^K498R but not in the case of
   PARP1^K498R/K521R. The inhibition of tumor growth was more prominent in
   the case of talazoparib treatment (fig. S11, A to C). The increased
   tumor size promoted by the HDAC5-PARP1 axis was corroborated by higher
   Ki67 and CD31 staining, which are indicative of higher proliferation
   rate and angiogenesis, respectively, along with a concurrent reduction
   in γH2AX and terminal deoxynucleotidyl transferase–mediated

   deoxyuridine triphosphate nick end labeling (TUNEL) staining indicative
   of reduced DNA damage and apoptosis, respectively ([104]Fig. 6, G to
   K). Together, our results suggest that HDAC5-mediated deacetylation
   plays a critical role in determining PARP1 oncogenic potential. Both
   the DNA repair and transactivation functions of PARP1 determined by the
   acetylation status of K498 and K521 sites, respectively, play a
   critical role in PARP1-mediated tumorigenesis. Thus, olaparib, a PARP1
   DNA damage response inhibitor, does not sufficiently inhibit
   tumorigenesis as PARP1 transactivation function remains unperturbed.

HDAC5 is critical for PARP1-induced aggressive tumor phenotype

   We next investigate the role of HDAC5 in PARP1-induced tumor malignancy
   in orthotopic colon cancer model. Our data suggest that abrogation of
   HDAC5 or PARP1 expression results in reduced tumor growth, while the
   presence of a constitutively nuclear HDAC5 mutant (HDAC5^S259/498A)
   promotes a notable increase in tumor size under genotoxic stress
   conditions ([105]Fig. 7, A and B). Furthermore, numerous metastatic
   nodules were detected in the lung and liver ([106]Fig. 7, C and D).
   Likewise, non-acetylable PARP1 mutants, PARP1^K498R, PARP1^K521R, and
   PARP1^K498R/K521R, induced robust tumor growth and metastasis, which
   was suppressed upon olaparib treatment in the case of PARP1^K498R but
   not in the case of PARP1^K498R/K521R. The mice with metastatic tumors
   also exhibited higher number of circulating tumor cells in these mice
   ([107]Fig. 7E). This was further corroborated by the reduced levels of
   epithelial marker E-cadherin and up-regulation of mesenchymal markers
   vimentin and fibronectin in the primary tumors, which is indicative of
   predisposition of these tumors to metastasize ([108]Fig. 7F). Thus, our
   data demonstrate that HDAC5 is a key determinant of PARP1-promoted
   tumor growth and metastasis. Since HRR proteins determine sensitivity
   to PARP inhibitors, we further examined the response to olaparib
   treatment in BRCA1/2-deficient orthotopic tumors ([109]24). We observed
   that repression of tumor growth and metastasis upon olaparib treatment
   in the case of PARP1^K498R was more robust in BRCA1/2-deficient tumors
   (fig. S12, A to F).

Fig. 7. PARP1 promotes aggressive tumor phenotype in an HDAC5-dependent
manner.

   [110]Fig. 7.
   [111]Open in a new tab

   (A) HT29^Luc2 control (Control; Luc shRNA); HDAC5 knockdown (HDAC5kd;
   HDAC5 shRNA); PARP1 knockdown (PARP1kd; PARP1 shRNA); PARP1 knockdown
   cells (PARP1kd; PARP1 shRNA) expressing PARP1^K498R
   (PARP1kd/PARP1^K498R), PARP1^K521R (PARP1kd/PARP1^K521R), or
   PARP1^K498R,521R (PARP1kd/PARP1^K498R,521R); and HDAC5 knockdown cells
   expressing HDAC5^S259/498A (HDAC5kd/HDAC5^S259/498A) were
   orthotopically injected into the cecum wall of nude mice. After 1 week,
   mice were administered 5FU (10 mg/kg) every alternate day. Olaparib was
   also coadministered at a dose of 10 mg/kg daily as indicated.
   Bioluminescence imaging was performed. (B) Bioluminescence
   quantification [(A) above]. The data are representative of three
   independent experiments using five individual mice per group. Error
   bars are means ± SD from five individual mice (n = 5 mice per group).
   Statistical analyses were done using two-way ANOVA (Tukey’s post hoc
   test). ***P < 0.001. (C and D) At the end of 5 weeks, the liver and
   lung from indicated mice [(A) above] were analyzed by ex vivo imaging
   (C) and hematoxylin and eosin staining (D). (E) At the end of 5 weeks,
   the blood from mice in (A) above was used to isolate genomic DNA for
   examining circulating tumor cells. The data are representative of three
   independent experiments using five individual mice per group. Error
   bars are means ± SD from five individual mice (n = 5 mice per group) *P
   < 0.05; **P < 0.01; ***P < 0.001. (F) At the end of 5 weeks, lysates of
   primary orthotopic tumors from (A) above were analyzed by
   immunoblotting for the indicated proteins. (G and H) Representative
   image of immunostaining of the indicated proteins in different grades
   of colon adenocarcinoma and matched normal adjacent tissue (NAT) (G).
   Quantitation of HDAC5, AcK498-PARP1, and AcK521-PARP1 levels (H).
   Statistical analyses were done using one-way ANOVA (Dunn’s multiple
   comparison test). Error bars are means ± SD. **P < 0.01; ***P < 0.001.

   Previous reports suggest that dysregulation of HDAC5 is associated with
   colon adenocarcinoma ([112]25). Hence, we next examined HDAC5 levels
   and PARP1 acetylation status in different grades of colon
   adenocarcinoma. We observed that HDAC5 levels were elevated in colon
   adenocarcinoma as compared to matched normal adjacent tissue. Moreover,
   with increasing grades of colon adenocarcinoma, HDAC5 levels were
   up-regulated, while AcK498-PARP1 and AcK521-PARP1 levels declined
   ([113]Fig. 7, G and H). Thus, our results suggest that dysregulation of
   HDAC5 and consequent perturbation in PARP1 acetylation status plays a
   critical role in determining colon cancer progression.

DISCUSSION

   PARP1 is an abundant nuclear protein with well-documented roles in both
   DNA repair and transcription. Previous studies suggest that the
   transactivation and repair functions of PARP1 could be modulated by
   posttranslational modifications including mono-ADP-ribosylation and
   acetylation. Sirt6 mono-ADP-ribosylates PARP1 thereby inducing its
   enzymatic activity and augmenting DNA repair under genotoxic stress
   conditions ([114]26). p300/CBP has been reported to acetylate PARP1 in
   response to inflammatory stimuli ([115]15). Acetylated PARP1 serves as
   a coactivator for nuclear factor κB (NF-κB)–dependent transcription.
   Histone deacetylase inhibitors have also been reported to trigger PARP1
   acetylation. Acetylation of PARP1 promotes chromatin trapping and
   reduces repair efficiency ([116]27). Our studies provide a concerted
   mechanism for the dynamic regulation of both DNA repair and
   transactivation functions of PARP1. Using a biochemical approach, we
   have established HDAC5 as the bona fide deacetylase determining PARP1
   acetylation status under genotoxic stress conditions. HDAC5-mediated
   PARP1 deacetylation at the K498 site augments repair function and
   prevents cell death. On the other hand, HDAC5-mediated deacetylation at
   the K521 site promotes PARP1 coactivator function in ATF4-mediated
   transactivation. Moreover, the functional consequences of deacetylation
   at these sites were mutually exclusive. In the absence of HDAC5 both,
   the sites were acetylated in a p300-dependent manner. Thus, our study
   establishes a preeminent role for acetylation status in determining
   PARP1’s functions.

   PARP1-dependant transcriptional regulation has been reported to play a
   key role in several human malignancies ([117]5). PARP1 regulates gene
   expression by functioning as part of enhancer/promoter binding
   complexes. Furthermore, PARP1 has been reported to serve as a
   coactivator of transcription factors including NF-κB and
   hypoxia-inducible factor 1α (HIF1α). PARP1 binds to NF-κB and is
   pivotal for the induction of target genes involved in tumor development
   ([118]28, [119]29). Likewise, PARP1 forms a complex with HIF1α
   involving direct interaction ([120]30). This complex triggers the
   expression of genes that promote angiogenesis and inhibit apoptosis to
   promote tumorigenesis. To delineate the role of acetylation status in
   determining its transactivation status, we examined the PARP1 binding
   proteins. Our data indicate that HDAC5-mediated deacetylation at
   Lys^521 site promotes PARP1 interaction with ATF4. ATF4 is a
   stress-induced transcription factor that is commonly up-regulated in
   several cancers. It induces the expression of genes involved in
   adaptive responses to stress conditions ([121]22). We provide further
   insights to establish a role for PARP1 in ATF4-mediated induction of
   genes involved in diverse pathways including proliferation, metastasis,
   and metabolism to promote survival under stress conditions and to
   sustain rapid tumor growth. Thus, our findings add to the repertoire of
   transcription factors involved in PARP1-mediated transcriptional
   regulation.

   PARP1 activated in response to genotoxic insult catalyzes PARylation,
   which uses nicotinamide adenine dinucleotide (NAD^+) as substrate.
   NAD^+ serves as a cofactor for several enzymes involved in cellular
   energy metabolism. Thus, PARP1 promotes a metabolic shift to oxidative
   phosphorylation over glycolysis to maintain homeostasis and promote
   cell survival ([122]31). Furthermore, transketolase, a key enzyme of
   the pentose phosphate pathway, is PARylated by PARP1 upon DNA damage
   ([123]32). PARylation inhibits transketolase enzymatic activity, which
   boosts ribose-5-phosphate levels required for DNA biosynthesis, thereby
   augmenting the repair process. In addition, the PARP1 transactivation
   function also determines metabolic flux. PARP1 interacts with
   transcription factors, including peroxisome proliferator–activated
   receptor α, to induce the expression of genes involved in lipid
   metabolism ([124]33). Thus, inhibition of PARP1 alleviates lipid
   accumulation in mouse models of fatty liver. However, the role of PARP1
   in metabolic rewiring in tumor cells is not well explored. Our study
   unravels PARP1’s function in determining glutamine metabolism.
   PARP1-ATF4 induces the expression of metabolic genes that promote the
   glutamine anaplerotic pathway ([125]34) to maintain cellular
   energetics. Glutamine also serves as the nitrogen source for amino acid
   and nucleic acid biosynthesis. Thus, PARP1-mediated transactivation is
   critical for metabolic adaptation to support rapid proliferation (fig.
   S13A). In conclusion, our findings highlight the significance of
   acetylation status in determining PARP1 oncogenic potential.
   HDAC5-mediated deacetylation at Lys^498 provides a survival advantage
   by preventing the accumulation of damaged DNA upon genotoxic insult,
   while deacetylation at Lys^521 modulates a transcriptional program that
   augments rapid proliferation and metabolic reprogramming. Thus, our
   study delineates the role of the PARP1-HDAC5 axis in determining
   malignant phenotype.

MATERIALS AND METHODS

Cell lines and culture conditions

   HT29 and SW480 cells were cultured in Dulbecco’s modified Eagle’s
   medium (DMEM) supplemented with 10% fetal bovine serum (FBS;
   Invitrogen), penicillin (100 U/ml), and streptomycin (100 μg/ml) at
   37°C. Human normal colon epithelial cells CCD 841 CoN were cultured in
   Eagle’s minimum essential medium supplemented with 10% FBS. The cell
   lines were obtained from the American Type Culture Collection.
   Recombinant adenoviruses were amplified, and titration was performed as
   previously reported ([126]35). Cells were cultured to ∼50 to 70%
   confluency followed by infection with recombinant adenovirus at a
   multiplicity of infection of 10 to 20. Adenovirus expressing green
   fluorescent protein was used as a negative control. To induce genotoxic
   stress, cells were grown to ∼50% confluency and then treated with 5FU
   (Sigma-Aldrich) or oxaliplatin (Sigma-Aldrich).

Plasmids and shRNA

   The Flag-tagged HDAC5 double mutant (HDAC5^S259A/S498A) was a gift from
   R, Shaw (Addgene, plasmid #32218). The PARP1 construct was a gift from
   T. Muir (Addgene, plasmid #111575). HDAC5^H893F, PARP1^K498Q,
   PARP1^K521Q, PARP1^K498R, PARP1^K521R, and PARP1^K498,521R mutant
   constructs were generated by site-directed mutagenesis (GCC Biotech,
   India). For the in vitro GST pull-down assay, HDAC5 was cloned in the
   pET-28a(+) vector, and PARP1 was cloned in pGEX-4T-1 to generate His
   and GST fusion proteins. For domain mapping, FLAG-tagged HDAC5 and
   PARP1 full-length and different domains were cloned in pcDNA3.1 vector
   to generate FLAG fusion proteins. Transfection was performed using
   Lipofectamine 3000 (Invitrogen) for HT29 and SW480, and Lipofectamine
   2000 (Invitrogen) for CCD 841 CoN cells according to the manufacturer’s
   instructions. In transient transfection experiments, plasmid DNA was
   kept constant with an empty vector. Recombinant adenovirus expressing
   HDAC5-HA-FLAG, PARP1-HA, and PARP1^K521R-HA was generated, as described
   previously ([127]36).

   Constructs expressing short hairpin RNA (shRNA) were generated using
   the psiRNA-DUO plasmid (Invivogen) which allows independent expression
   of two shRNAs. The shRNAs targeting human HDAC5, PARP1, and ATF4 were
   designed against the 3′ untranslated region of the transcript and hence
   cannot target the ectopically expressed gene. The sequences used have
   been provided in table S1.

Determination of acetylation status of proteins

   Biochemical analysis of acetylation was performed, as described
   previously ([128]37). Briefly, acetylated peptides were enriched using
   anti–acetyl-lysine antibody (Cell Signaling Technology). The
   immunoaffinity-purified acetylated peptides were analyzed by liquid
   chromatography–tandem mass spectrometry (LC-MS/MS; Thermo Fisher
   Scientific).

Western blot and immunoprecipitation

   Cells were lysed in ice-cold lysis buffer containing 20 mM tris·HCl (pH
   7.4), 5 mM EDTA, 10 mM Na[4]P[2]O[7], 100 mM NaF, 2 mM Na[3]VO[4], 1%
   NP-40, 1 mM phenylmethylsulfonyl fluoride, and 1× protease inhibitor
   mixture (Roche). Subcellular fractions were prepared using a
   Subcellular Protein Fractionation Kit (Thermo Fisher Scientific)
   following the manufacturer’s instructions. SDS–polyacrylamide gel
   electrophoresis (SDS-PAGE) was performed using equal amounts of
   proteins for each sample and transferred to a PVDF membrane
   (Millipore). All blots were imaged using a chemiluminescent Western
   blot imaging system (Azure Biosystems). The antibodies are listed in
   table S2. A polyclonal antibody specific for the mouse PARP1 acetylated
   at lysine 498 was generated (Abgenex Pvt. Ltd.). Briefly, rabbits were
   immunized with the acetylated mouse PARP1 peptide
   [PVEVVAPRGK(AcK)SGAALS]. Antisera from the immunized rabbits were
   affinity purified using the unacetylated peptide followed by the
   acetylated peptide. To detect various acetylated forms of PARP1, cell
   lysates were immunoprecipitated with PARP1 antibodies and then
   subjected to Western blotting with acetyl-specific antibodies. To
   detect PARylation, cell lysates were immunoprecipitated with PARP1
   antibodies and then subjected to Western blotting with PAR-specific
   antibodies.

   Immunoprecipitations were performed using 500 μg of cell extracts
   pre-treated with MNase, as previously described ([129]38), and
   incubated with the indicated antibodies. Western blots were performed
   subsequently, as described earlier. For identifying PARP1 binding
   proteins, HT29 cells were infected with recombinant adenovirus
   expressing hemagglutinin (HA)-tagged PARP1^K521Q or PARP1^K521R.
   Twenty-four hours after infection, nuclear extracts were prepared.
   Precleared lysates were then immunoprecipitated with anti-HA
   antibody–conjugated agarose beads (Santa Cruz Biotechnology) and were
   eluted with HA peptide (Sigma-Aldrich). The eluate was resolved by
   SDS-PAGE and visualized by silver staining. The bands were excised and
   analyzed by mass spectrometry (ITSI Biosciences, USA).

In vitro deacetylation assay

   An in vitro deacetylation assay was performed, as described previously
   ([130]39). Briefly, in vitro deacetylation reactions were performed
   using human recombinant HDAC5 (Sigma-Aldrich) and acetylated PARP1
   peptides (Sigma-Aldrich). The reaction mixture was run on an LC-MS/MS
   System (Thermo Fisher Scientific) and the data were analyzed by
   Proteome Discoverer software. Acetylated PARP1 peptides used in the
   assay are as follows: AcK498: VAPRGK(AcK)SGAAL, AcK505:
   SGAALSK(AcK)KSKG, AcK508: KSK(AcK)GQVKEEGIN, AcK521: NKSEK(AcK)RMKLTL,
   AcK524: RMK(AcK)LTLKGGAA.

Alkaline comet assay

   A total of 1 × 10^3 cells/ml were mixed with 1% low melting point
   agarose, transferred onto slides, and allowed to settle. The slides
   were then immersed in lysis solution and subjected to electrophoresis
   as reported ([131]40). Slides were stained with SYBR Gold (Invitrogen,
   USA). Imaging was done using a Zeiss ApoTome microscope and the images
   were analyzed with the OpenComet plugin for ImageJ software. In each
   experiment, 100 cells were scored per sample.

Proximity ligation assay

   The PLA was performed using a Duolink in situ PLA kit (Sigma-Aldrich)
   according to the manufacturer’s instructions. Slides were
   counterstained with 4′,6-diamidino-2-phenylindole (DAPI) UltraCruz
   mounting medium (Santa Cruz Biotechnology). Imaging was done using a
   Zeiss ApoTome microscope and the images were analyzed with the ZEN Blue
   software. The following primary antibodies were used: HDAC5 (Santa Cruz
   Biotechnology), PARP1 (Santa Cruz Biotechnology), pADPr (Santa Cruz
   Biotechnology), Mre11 (Cell Signaling Technology), and ATF4 (Abcam).

Cell cycle analysis

   BrdU (5-bromo-2-deoxyuridine) incorporation was measured as described
   previously ([132]41). The cells were analyzed by flow cytometry using
   FACScan (BD Biosciences). Fluorescence-activated cell sorting (FACS)
   data were analyzed using Flowjo software.

RNA-seq and analysis

   Total RNA was extracted using TRIzol reagent (Invitrogen) following the
   manufacturer’s instructions. Transcriptome sequencing was performed on
   the Illumina platform (Neuberg Supratech, India). Quality assessment of
   the raw fastq reads of the sample was performed using FastQC v.0.11.9
   ([133]42). The processed reads were aligned to the STAR-indexed Homo
   sapiens (GRCh38) genome using STAR aligner v 2.7.9a. The differentially
   expressed genes were identified using DESeq2 ([134]43).

Metabolite profiling

   To determine metabolic profiles, metabolites were extracted and
   analyzed as described previously ([135]44). Briefly, metabolites were
   extracted on dry ice with chilled 80% methanol. The extract was dried
   using a speed vacuum concentrator and resuspended in 80 ml of water.
   The samples were run on a Q-TOF Synapt G2 mass spectrometry (Waters)
   coupled to UPLC. MassLynx 4.1 software was used for data acquisition,
   data handling, and instrument control. MZmine 2 software was used for
   analysis. The relative quantification process was done by comparing the
   signal intensity or peak area of the specific metabolite relative to
   the control.

Quantification of Metabolites

   For targeted quantification of specific intracellular metabolites, 2 ×
   10^5 cells were seeded in six-well plates for 24 hours before treatment
   and then treated with 5FU (Sigma-Aldrich) for indicated time points.
   The cells were harvested and suspended in phosphate-buffered saline
   (PBS). The kits used for the assays are described below. The amount of
   specific metabolite within a sample measured using kits was plotted
   relative to the control.

Glutamine quantification assay

   The abundance of intracellular total Glutamine content was measured
   using a Glutamine Assay Kit (Elabscience) following the manufacturer’s
   protocol and normalized to cell number. The Gln content was calculated
   by measuring the absorbance value at 450 nm.

Glutamate quantification assay

   Total glutamate content was measured using a Glutamate
   Colorimetric/Fluorometric Assay Kit (Caymen Chemicals) as per the
   manufacturer’s protocol and normalized to cell number. The absorbance
   was measured at 570 nm.

Alpha-ketoglutarate quantification assay

   Total alpha-ketoglutarate content was measured using α-Ketoglutarate
   Assay Kit (Caymen Chemical) according to the manufacturer’s protocol
   and normalized to cell number. The absorbance was measured at 570 nm.

Succinate quantification assay

   Total succinate content was measured using EnzyChrom Succinate Assay
   Kit (Bioassay systems) according to the manufacturer’s protocol and
   normalized to cell number. The absorbance was measured at 570 nm which
   correlates with the concentration of succinate in the sample.

Asparagine quantification assay

   Total asparagine content was measured using the Asparagine Assay Kit
   (Abcam) according to the manufacturer’s protocol and normalized to cell
   number. The fluorescence signal was measured at Ex/Em = 538/587 nm.

Arginine quantification assay

   Total arginine content was measured using the Arginine Assay Kit
   (Abcam) according to the manufacturer’s protocol and normalized to cell
   number. The colorimetric signal absorbance was measured at 450 nm.

Proline quantification assay

   Total proline content was measured using Proline Assay Kit
   (Elabscience) according to the manufacturer’s protocol and normalized
   to cell number. The concentration of proline can be calculated by
   measuring the absorbance at 520 nm.

Ornithine quantification assay

   Total ornithine content was measured using Ornithine Assay Kit (Abcam)
   according to the manufacturer’s protocol and normalized to cell number.
   The fluorescence signal was measured at Ex/Em = 538/587 nm.

Oxygen consumption rate measurement

   OCR was determined using an XF24 flux analyzer (Seahorse Biosciences).
   OCR was analyzed under basal condition or in response to 5FU
   (Sigma-Aldrich) followed by injection with oligomycin, carbonyl cyanide
   p-trifluoromethoxyphenylhydrazone, and antimycin A, as described
   previously ([136]44).

RT-qPCR

   Total RNA extraction was carried out using TRIzol (Invitrogen). cDNA
   was synthesized using an iScript cDNA synthesis kit (Bio-Rad) as per
   the manufacturer’s instruction. Quantitative polymerase chain reaction
   (qPCR) was carried out using Maxima SYBR Green Master Mix (Fermentas)
   in an Eppendorf real-time PCR machine. 18S ribosomal RNA served as
   internal control across all samples. ΔΔCt method was used for the
   analysis of the reverse transcription qPCR (RT-qPCR) data. Error bars
   are means ± SD of three independent experiments with triplicate
   samples. Primer sequences are listed in table S3.

ChIP assay

   Chromatin immunoprecipitation (ChIP) was performed using a commercially
   available kit following the manufacturer’s instructions (Millipore).
   Briefly, 1 × 10^7 cells were fixed with formaldehyde and lysed with SDS
   lysis buffer. The cell lysates were then subjected to sonication to
   fragment the DNA to lengths between 0.2 and 1 kb. The samples were
   precleared using Protein A Agarose/Salmon Sperm DNA slurry. Control
   immunoglobulin G (IgG) (Santa Cruz Biotechnology) or ATF4 (Santa Cruz
   Biotechnology) antibody was added and incubated overnight at 4°C. The
   samples were subsequently incubated with fresh Protein A Agarose/Salmon
   Sperm DNA slurry for 2 hours. The precipitated chromatin complexes were
   eluted by incubating the beads with 500 μl of elution buffer (1% SDS
   and 0.1 M NaHCO[3]) for 30 min. Last, protein-DNA cross-links were
   reversed by heating at 65°C for 4 hours, and the immunoprecipitated DNA
   was analyzed by qPCR. For reChIP experiments, a part of the chromatin
   complexes immunoprecipitated with the anti-ATF4 antibody above was
   eluted by incubation for 30 min at 37°C in 100 μl of 10 mM
   dithiothreitol instead of elution buffer. Following centrifugation, the
   supernatant was diluted 20-fold with ChIP dilution buffer and
   re-immunoprecipitated with control IgG or PARP1 antibodies (Santa Cruz
   Biotechnology), and the ChIP protocol was repeated. Primers used for
   this analysis are provided in table S4.

Colony formation assay

   Cells were counted and 1000 cells were seeded in a six-well culture
   dish. After 7 days of incubation, plates were gently washed with 1×
   PBS, fixed in methanol, and stained with 0.1% crystal violet. Colonies
   were manually counted.

Transwell migration assay

   Cell migration was performed using the Cultrex cell migration assay kit
   (Trevigen), following the manufacturer’s instructions. Briefly, 1 × 10
   ^5 cells were seeded in serum-free DMEM in the upper chamber of a
   24-well Transwell plate, while the lower chamber contained DMEM medium
   with 10% (v/v) FBS. After 24 hours, the cells were collected in a cell
   dissociation solution containing 1 μM of calcein-AM. The plate was read
   using 485-nm excitation and 520-nm emission filters, and percentages of
   migrated cells were calculated from the standard curve established for
   respective cell lines.

Transwell invasion assay

   Cell invasion through basement membranes was assayed using the Cultrex
   Coat BME-coated cell invasion assay (Trevigen), as per the
   manufacturer’s instructions. At first, rehydration of the membranes was
   performed, and then the protocol described in the migration assay above
   was followed.

Xenograft studies

   All the animal studies were approved by the Institutional Animal Care
   and Use Committee of the National Institute of Immunology, New Delhi,
   India. Female nude mice (nu/nu) were subcutaneously injected with 4 ×
   10^6 cells/0.1 ml of DMEM. After 1 week of injection, 5FU was injected
   intraperitoneally at a dose of 25 mg/kg per mouse every alternate day.
   Mice were injected intraperitoneally with olaparib (Sigma-Aldrich) at a
   dose of 10 mg/kg daily or talazoparib (Sigma-Aldrich) at a dose of 2
   mg/kg daily. Tumor volume (in cubic millimeters) was measured with a
   slide caliper and calculated using the formula: (widest diameter ×
   smallest diameter^2)/2. At the end of the experiment, the mice were
   euthanized and the tumors were harvested. Tumor extracts were then
   prepared, followed by immunoblotting as described earlier. For
   immunohistochemical analysis, tumors were fixed in formalin, washed
   with PBS, and embedded in paraffin for sectioning. Five-millimeter
   sections were dewaxed and immunostained with Ki67 (Santa Cruz
   Biotechnology, sc-23900), CD31 (Santa Cruz Biotechnology, sc-376764),
   and γH2AX (Abcam, ab11174) antibody. For secondary detection,
   anti-mouse Alexa Fluor 555 (Molecular Probes) and anti-rabbit Alexa
   Fluor 546 (Molecular Probes) antibodies were used. TUNEL staining was
   performed using the TUNEL Assay Kit (Promega) as per the manufacturer’s
   protocol. Sections were counterstained using DAPI. Imaging was done
   using a Zeiss ApoTome microscope and the images were analyzed with the
   ZEN Blue software.

In vivo metastasis assay

   HT29 cells were stably transfected (pooled neomycin-resistant
   population) with pGL4.51[luc2/CMV/Neo] plasmid (Promega) to generate
   HT29^Luc2 cells. The cell suspension was injected orthotopically into
   the cecum wall of female nude mice. After 1 week of injection, 5FU was
   injected intraperitoneally at a dose of 25 mg/kg per mouse every
   alternate day. Mice were injected intraperitoneally with olaparib at a
   dose of 10 mg/kg daily for 4 weeks. For weekly in vivo bioluminescence
   imaging, the mice were anesthetized using ketamine (80 mg/kg) and
   xylazine (10 mg/kg) by intraperitoneal injection. The anesthetized mice
   were administered d-luciferin (150 mg/kg) in PBS intraperitoneally.
   Imaging was performed using the Spectral LAGO X imaging system, and
   image analysis was performed using Aura imaging software. At the end of
   the experiment, the mice were euthanized. The liver and lung were
   excised and ex vivo imaging was performed. Primary tumors were excised
   and Western blot analysis was performed as described earlier.
   Hematoxylin and eosin staining was performed, as described previously
   ([137]45). Genomic DNA was isolated from blood obtained from euthanized
   mice for measurement of circulating tumor cells using the DNeasy Blood
   and Tissue Kit (Qiagen). qPCR was performed using primers specific to
   the human Alu repeat sequence (Fwd 5′-ACGCCTGTAATCCCAGCACTT-3′; Rev.
   5′-TCGCCCAGGCTGGAGTGC-3′), while mouse actin served as the control (Fwd
   5′-GCTTCTTTGCAGCTCCTTCGTTG-3′; Rev. 5′-TTTGCACATGCCGGAGCCGTTGT-3′).

Tissue microarray

   Human colon adenocarcinoma and matched normal adjacent tissue sections
   were procured from US Biomax as tissue microarrays. Immunofluorescence
   was performed as described previously ([138]45). Briefly, the paraffin
   sections were dewaxed and incubated with HDAC5 (Santa Cruz
   Biotechnology, Sc-133106), AcK498-PARP1, and AcK521-PARP1 antibodies
   (CSB-[139]PA890185). Anti-mouse Alexa Fluor 488 (Molecular Probes,
   A11001), anti-rabbit Alexa Fluor 555 antibody (Molecular Probes,
   A21428), and anti-rabbit Alexa 546 (Molecular Probes) were used for
   secondary detection. Slides were mounted with ProLong Gold antifade
   reagent with DAPI (Molecular Probes) and imaging was performed using a
   Zeiss Apotome microscope. The images were analyzed using Zeiss ZEN Blue
   software and ImageJ software was used for quantitation. The average
   signal intensity (in arbitrary units, AU) from four random fields was
   used for the analysis.

Statistical analysis

   All experiments were conducted independently at least three times.
   Results were expressed as means ± SD. The sample size was chosen to
   give sufficient power for calling significance with standard
   statistical tests. Statistical analyses were performed by a standard
   two-tailed Student’s t test or one-way analysis of variance (ANOVA).
   ***P < 0.005 was considered significant.

Acknowledgments