Hyperinsulinemia promotes aberrant histone acetylation in triple negative breast cancer

Excess levels of insulin relative to glucose in the blood, or hyperinsulinemia, is considered to be a poor prognostic indicator for patients with triple negative breast cancer (TNBC). While this association has been recognized for some time, the mechanistic role of hyperinsulinemia in promoting TNBC remains unclear. We show that insulin treatment leads to genome-wide increase in histone acetylation, in particular at H3K9, through the PI3K/AKT/mTOR pathway in MDA-MB-231 cells. Genome-wide analysis showed that the increase in histone acetylation occurs primarily at gene promoters. In addition, insulin induces higher levels of reactive oxygen species and DNA damage foci in cells. In vivo, hyperinsulinemia also enhances growth of MDA-MB-231 derived tumors through increased histone acetylation. These results demonstrate the impact of hyperinsulinemia on altered gene regulation through chromatin and the importance of targeting hyperinsulinemia-induced processes that lead to chromatin dysfunction in TNBC.


Introduction 43
Triple negative breast cancer (TNBC) is a clinically aggressive subtype of breast cancer 44 that does not express estrogen receptor (ER), progesterone receptor (PR) or human 45 epidermal growth factor receptor 2 (HER2) ( kinase substrate, in rapamycin untreated cells ( Figure 1B, lanes 1 and 2). Rapamycin 119 pre-treatment, however, inhibited the phosphorylation of S6K without affecting AKT 120 phosphorylation confirming that rapamycin indeed inhibited mTOR kinase activity 121 ( Figure 1B, lanes 3 and 4). Both mTOR inhibition and PI3K inhibition, by rapamycin and 122 LY294002 treatment respectively, inhibited the H3K9ac increase induced by insulin 123 ( Figure 1B and 1C). To confirm that the insulin induced histone H3 acetylation is 124 chromatin-bound and not on newly synthesized or free histones, we performed 125 chromatin fractionation after insulin treatment followed by western blot analyses. 126 Results showed that insulin induced H3K9ac was exclusively chromatin bound ( Figure  127 S1A-B). with enhanced mitochondrial activity, we performed western blot analyses for TFAM and 133 ATP5D (a subunit of the ATP synthase complex/Complex V) after insulin treatment. 134 TFAM and ATP5D protein levels increase after 1h insulin treatment ( Figure 1D). Next, 135 we tested indicators of mitochondrial biogenesis and activity after insulin treatment. 136 Mitochondrial DNA content is an indicator of mitochondrial number and ATP levels are a 137 measure of mitochondrial activity in cells (Morita et al., 2013). Insulin increased the 138 mitochondrial DNA content ( Figure 1E) as well as ATP levels ( Figure 1F) in MDA-MB-139 231 cells, indicating an enhancement in mitochondrial biogenesis and activity. In 140 addition, we performed GC-MS to measure TCA cycle metabolites produced in the 141 mitochondria. We observed increased levels of lactate and TCA cycle intermediates 142 succinate, pyruvate, alpha-ketoglutarate, malate and citrate after 6h of insulin treatment 143 ( Figure S1C). These data suggest that insulin induces genome-wide increase in histone 144 acetylation, in particular H3K9ac, through the PI3K-AKT-mTOR pathway. 145

Insulin induces H3K9ac acetylation on promoter regions. 146
To characterize the genomic loci associated with increased histone acetylation after 147 insulin treatment, we performed quantitative ChIP-seq analyses as described in 148 (Orlando et al., 2014) where we spiked in Drosophila S2 cells with MDA-MB-231 cells 149 before the chromatin immunoprecipitation (ChIP) (see Methods). We observed an 150 increase in the number of reads aligning to the human genome in the 3h and 6h insulin 151 treated H3K9ac ChIPs indicating an increase in global histone acetylation levels ( Table  152 S1). By performing spike normalization (see Methods, Figure   Next, we annotated the H3K9ac peaks based on distance to the nearest RefSeq 156 annotated TSS. Results showed that ~46% of the H3K9ac peaks were promoter 157 proximal (~26% within 1kb and ~20% between 1kb-10kb of nearest TSS) ( Figure 2A).  These results show that insulin induces genome-wide increase in H3K9ac at promoter 171 regions of genes and thereby could be involved in transcriptional regulation. 172

Insulin induces H3K9ac acetylation on promoters of insulin-induced genes. 173
To further test whether the increase in H3K9ac enrichment levels correlate with gene 174 expression changes induced by insulin, we performed RNA-sequencing (RNA-seq) in 175 MDA-MB-231 cells untreated (UT) or treated for 3h and 6h with insulin. We quantified 176 changes in gene expression after 3h and 6h insulin treatment from RNA-seq data using 177 DESeq2 (Love et al., 2014). 207 and 384 genes exhibited significantly altered 178 expression in in 3h and 6h insulin treated cells respectively ( Figure 3A and 3B). Insulin 179 treatment induced metabolic pathways required for cellular growth such as ribosome 180 biogenesis, transcription, and splicing as well as known insulin regulated downstream 181 pathways related to ATP production and mTOR signaling ( Figure 3C). Insulin treatment 182 downregulated FOXO signaling genes as well as apoptosis inducing genes. 183 Interestingly, insulin treatment also downregulated genes involved in reactive oxygen 184 species (ROS) metabolism or scavenging as well as immune cell migration and 185 activation. Moreover, insulin upregulated several MYC (c-Myc) target genes and genes 186 related to zinc ion homeostasis in cells ( Figure 3C). These results indicate that insulin 187 induces cell growth and proliferation while also suppressing apoptosis. 188 We then compared the changes in H3K9ac enrichment on the promoter regions (TSS 189 ±1kb) of genes upregulated and downregulated by insulin. Upregulated genes showed a 190 larger increase in H3K9ac enrichment induced by insulin ( Figure 3E and 3G) at 3h and 191 6h respectively. However, genes downregulated at 3h also showed a modest but 192 significant increase in H3K9ac enrichment at their promoters ( Figure 3F). Interestingly, 193 genes downregulated after 6h insulin treatment did not show any significant enrichment 194 in H3K9ac signals ( Figure 3H). These results indicate that there is a genome-wide 195 increase in H3K9ac signal at all expressed genes after 3h insulin treatment. However, 196 at 6h the increase in H3K9ac signal is more specific to upregulated genes.  Increased NRF1 binding was also observed 6h post-insulin treatment, however, it was 233 lower than that at 3h ( Figure 4D-K) indicating an early response to insulin. NRF1 234 binding at promoters of NRF1 target genes could lead to increased histone acetylation 235 at these regions. These results indicate that NRF1 might regulate the metabolic 236 capacity of cancer cells by integrating metabolic inputs from the environment to 237 increase histone acetylation on chromatin that allow continuous transcription from these 238 genes. 239

Insulin induced reactive oxygen species (ROS) causes genome instability. 240
mTOR pathway induces mitochondrial biogenesis and activity that might lead to 241 increased ROS production through the electron transport chain. To investigate whether 242 insulin treatment induces ROS production in the cells, we measured ROS using a 243 fluorescent dye, CellROX green. Results showed that ROS was significantly increased 244 after 3h of insulin treatment and remained high at 6h ( Figure 5A). Increase in ROS 245 production could be deleterious to cells as the free radicals could cause DNA damage 246 and mutation. We measured DNA damage using the DNA damage marker g-H2AX in 247 cells treated with insulin using immunofluorescence assays. We observed that the 248 number of cells with g-H2AX foci and the number of g-H2AX foci per cell increased after 249 3h insulin treatment ( Figure 5B). Interestingly, the number of cells with g-H2AX foci 250 decreased after 6h indicating possible activation of repair pathways ( Figure 5B  Moreover, the insulin treatment performed in vitro is a short (3h-6h) and acute (100 nM) 303 exposure to insulin whereas tumors from Rag/MKR mice are exposed to chronic 304 genome-wide increases in chromatin-associated histone acetylation that was dependent 315 on the insulin mediated signaling through the PI3K-AKT-mTOR pathway. We used a 316 quantitative method of ChIP-seq (ChIP-Rx) to identify the regions associated with 317 changes in H3K9ac in response to insulin. We found genome-wide increases in H3K9ac 318 occupancy at gene promoters especially those that increased expression after insulin 319 treatment. However, insulin-induced increase in histone acetylation at gene promoters 320 was not always associated with an increase in gene expression indicating that 321 increased acetylation at these sites may have a distinct function. Our observation is 322 supported by a recent study investigating histone acetylation levels in response to high 323 glucose levels (Lee et al., 2018). Interestingly, it has been proposed that histone 324 acetylation may function as a capacitor for acetate/acetyl-CoA which could be utilized 325 as an energy source or to balance the intracellular pH based on cellular condition 326 (Kurdistani, 2014). Thermofisher Scientific). 409

Mitochondrial DNA and ATP measurement 410
For mitochondrial DNA quantification, genomic DNA and mitochondrial DNA was 411 extracted from insulin treated cells using DNeasy Blood and Tissue kit (Qiagen, Hilden, 412 Germany). Genomic and mitochondrial DNA were quantified by qPCR using 413 mitochondrial and genomic DNA specific primers; namely, cytochrome B and RPL13A 414 (Table S2).

Chromatin fractionation 428
Chromatin fractionation was performed using the method described in (Mendez & 429 Stillman, 2000). Briefly, cells were collected from six well plates after insulin treatment

ChIP-seq analyses 515
Sequencing reads from each library were aligned to a combined reference genome 516 (human + Drosophila) using bowtie (Langmead, Trapnell, Pop, & Salzberg, 2009). The 517 combined reference genome was generated as described in (Orlando et al., 2014). 518 Briefly, a combined genome sequence was created by concatenating the genome 519 sequences of human (hg19) and Drosophila (dm3). Next, custom bowtie indexes were 520 generated for the combined genome sequence using 'bowtie-build' command. Bowtie 521 alignment was done against the combined genome using parameters: -m 1 -e 70 -k 1 -n 522 2 --best --chunkmbs 200. About 6% of reads aligned to the dm3 genome and ~94% 523 aligned to the hg19 genome. The number of reads aligned to human and Drosophila 524 genome are reported in Table S1. We identified a union set of 40,222 and 5,716 525 H3K9ac peaks in the human and Drosophila cells respectively. We normalized peak 526 scores for the 40,222 human (hg19) peaks using hg19 aligned read counts (read count 527 normalization) ( Figure S2A). Moreover, we used Drosophila (dm3) aligned read counts 528 for normalizing peak scores (spike normalization) ( Figure S2B). Significantly, we 529 observed greater changes in H3K9ac levels on peaks in 3h and 6h insulin treated cells 530 after spike normalization as compared to read count normalization ( Figure S2C and D). 531 The effect of spike normalization was also evident in aggregate profiles of H3K9ac ±2kb 532 around annotated transcription start sites (TSSs) (Figure S2E and F). Overall spike 533 normalization led to better conformity between replicates and revealed global increase 534 in histone acetylation that could be quantified. 535

RNA-seq analyses 536
Total RNA was isolated from insulin treated cells or from tumors tissues using 537 NucleoSpin® RNA kit (Macherey-Nagel, Germany) with on-column DNase I digestion.  (Table S2). Relative gene expression between groups was determined using 2^-556 DDCt method after normalization with 18S levels. 557

Visualization of ChIP-seq data and additional analyses 558
Reads aligning to the hg19 genome were filtered out from the aligned bam files and 559 peaks were called for each library with respective Input using macs2 (Zhang et al.,560 2008) with broad peak calling option and q < 0.1 for broad regions and q < 0.05 for 561 narrow regions. Peaks were annotated using annotatePeaks.pl command in homer 562 (Heinz et al., 2010) and assigned to the nearest hg19 RefSeq gene TSS. Wiggle tracks 563 were generated using custom scripts, normalized by the number of dm3 aligned reads 564 and visualized on the UCSC Genome Browser. Heatmaps were generated using Java 565 TreeView (Saldanha, 2004) and aggregate profiles were made using deepTools 566 (Ramirez et al., 2016). Motif enrichment analysis was performed using 567 findMotifsGenome.pl command in homer. 568

Human samples 569
Blood samples were obtained from patients after an 8h fasting period following 570 institutional guidelines at the City of Hope (IRB no. 15418) in purple-top EDTA 571 vacutainer tubes after obtaining written informed consent. Insulin resistance was 572 identified by measuring Hemoglobin A1C (HbA1c) using HPLC method (Davis, 573 McDonald, & Jarett, 1978). HbA1C levels between 5.7-6.3 were used to define insulin-574 resistance. All individuals were within ages of 18-55. PBMCs were isolated from whole 575 blood using Ficoll-Paque method. Briefly, whole blood was diluted 1:1 with PBS 576 containing 2% FBS, layered on top of 15 ml Ficoll and spun down at 1200´g for 10 577 mins. The white buffy coat containing PBMCs were collected and washed twice with 578 PBS containing 2% FBS and spun down at 1200´g for 10 mins to remove platelets. 0.5 579 tumor growth was measured as previously described (Shlomai et al., 2017). At the end 595 of the study, tumors were dissected and flash frozen in liquid nitrogen for further 596 analysis. 597

Statistical analyses 598
Data are represented as mean and standard error of mean (Mean + SEM). Statistical 599 analyses were performed using GraphPad Prism 7.0 software (GraphPad Prism 600 Software Inc., San Diego, CA) and R. Normal distribution was confirmed using Shapiro-601 Wilk normality test before performing statistical analyses. For normally distributed data, 602 comparison between two means were assessed by unpaired two-tailed Student's t test 603 and that between three or more groups were evaluated using one-way analysis of 604 variance (ANOVA) followed by Tukey's post hoc test. In case of Student's t-test, F-test 605 was performed to check whether the variance in the groups compared were significantly 606 different. For data where significantly different variances were observed, t-test with 607 Welch correction was performed. For data that did not follow a normal distribution, 608 Mann-Whitney test was performed for comparison between two groups and Kruskal 609 Wallis test followed by Dunn's multiple comparisons test was performed for comparing 610 more than two groups. A p-value of < 0.05 was considered statistically significant. 611

Competing Interests 622
The authors declare that no competing interests exist.