The National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) and PubMed were queried using the terms zinc, TPEN (N,N,N’N’-tetrakis (–)[2-pyridylmethyl]-ethylenediamine), treatment, depletion, supplementation, and deficiency. A total of 10 zinc treatment and three zinc depletion datasets were identified, of which nine and two were analyzed, respectively, based on dataset quality and availability. Datasets (19–28) were analyzed using the R statistical environment (29) as outlined in Supplementary Figure S1. Briefly, array expression data was logged and quantile normalized as required followed by dataset-specific filtration based on log expression data. Gene count data from RNA sequencing platforms underwent dataset-specific filtration, followed by normalization and dispersion estimation using EdgeR. Differential gene expression was evaluated using the Empirical Bayes Statistics for Differential Expression (eBayes) for array data and the Generalized Linear Model Quasi-Likelihood F-Test (glmQLF) for sequencing data using Benjamini-Hochberg method multiple testing correction to control for false discovery rate (FDR) (30) as we have done before (31). Differential gene expression performed by Palmer et al. (32) and Tuomela et al. (33) were additionally used in our analysis. Cell types were examined separately in all datasets containing more than one cell type (GSE39316, GSE39330) except for dataset GSE25167 where HaCat and Sk Mel-28 cell transcriptomes (n=1 each) were combined to allow statistical comparison between mock and zinc treated samples. ZSGs were defined by a significant (p<0.05) ≥1.5-fold induction in gene transcription following zinc treatment or conversely, a ≤1.5-fold reduction following zinc depletion.

AH transcriptomic datasets GSE28619 (34), GSE143318 (35), GSE155907 (36), GSE142530 (37) and GSE94417 (38) were analyzed as above. Datasets GSE94397, GSE94399 and GSE103580 were merged and batch corrected using ComBat (https://doi.org/10.18129/B9.bioc.sva) on quantile normalized data (39). Data access to the DbGaP study phs003112.v1.p1 (40) was approved by the National Institute of Health (Project #31612). For DbGaP study, fastq files and sample metadata were downloaded using SRAtools (v3.0.0), Library sequencing quality was determined using FastQC (Babraham Bioinformatics). Illumina adaptor sequence and low quality read trimming (read pair removed if < 20 base pairs) was performed using Trim Galore! (Babraham Bioinformatics: www.bioinformatics.babraham.ac.uk/). STAR 52 was used to align reads to human genome hg38 using ENSEMBL gene annotations as a guide. Read counts data corresponding to ENSEMBL gene annotations were generated using HTSeq (41). Bioinformatics processing and analysis pipelines have been added to https://github.com/Scotch25/ZincALDManuscript.

Zinc-stimulated gene expression was collated from 11 datasets (19–26, 28, 32, 33), with ZSGs defined as being up-regulated ≥1.5 fold, p<0.05 (Supplementary Table S1). Genes that were up-regulated by zinc in ≥5 of 11 datasets were included in the preliminary zinc signature. A cutoff of 5 datasets, as opposed to 6 (representing >50% of datasets examined) was chosen to account for a lack of ZSG expression data within some datasets (Supplementary Table S1).

Blood samples were collected from healthy volunteers at Westmead Hospital and the Westmead Institute for Medical Research. Liver tissue was obtained during hepatocellular carcinoma resection or bariatric surgery at Westmead and Blacktown Hospitals, respectively. All work was performed in accordance with both the Declarations of Helsinki and Istanbul. Ethics approval was obtained from the Sydney West Area Health Service and University of Sydney (HREC2002/12/4.9 (1564) and HREC/17/WMEAD/552 (5450)). Informed consent was obtained for all subjects.

Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll Paque Plus (GE Healthcare) density gradient separation. PBMCs were cultured in RPMI medium containing 10% fetal calf serum at 37 °C with 5% CO₂. To generate macrophages, 5 x 10⁵ monocytes were isolated from PBMCs using CD14 microbeads and the AutoMacs Pre Separator (Miltenyi Biotec). Cells were differentiated in RPMI medium plus 10% fetal calf serum and 50 ng/mL M-CSF over 7 days, changing the media on day 3 (42).

Liver organoid-based cell monolayers were generated from human liver tissue collected from liver biopsies/resections. Tissue was minced and digested with collagenase IV (1mg/ml) for 30 minutes at 37 °C after which cells were pushed through a 100 µm cell strainer and centrifuged at 50 x g for 5 minutes. The cell pellet representing concentrated hepatocytes was harvested, cells were resuspended in expansion media and the 50 x g centrifugation step was repeated twice. Hepatocytes were resuspended in Matrigel and organoid expansion media was added (media formulations available in Supplementary Tables S2, S3). Following sufficient expansion of liver organoids (3–5 days post-splitting while organoids are actively expanding but still relatively small), organoids were removed from matrigel using cold PBS, then digested with TrypLE for 3 minutes at 37 °C. The single cell suspension was added to a collagen coated 48 well plate in expansion media until cells reached 80-90% confluency (Supplementary Figure S2). Expansion media was replaced with differentiation media for 3 days prior to treatment with zinc supplementation/depletion media. Organoid cultured were grown at 37 °C with 5% CO_2.

Huh-7 cells were cultured in DMEM medium containing 10% fetal calf serum at 37 °C with 5% CO₂. All experimental treatments were performed at ~80% confluence.

Zinc depletion of media was performed as previously described (27). Briefly, media was incubated at room temperature, rocking overnight with S100A12 zinc-binding resin at a ratio of 80 µl resin: 1 mL culture media. Resin was pelleted by centrifugation at 400 x g for 5 minutes and media was removed for immediate use. Following removal of culture media and a single wash with PBS, zinc-depleted (ZnDep) media was added to cells. For all zinc treatment and repletion experiments, 50 µM ZnSO₄ was added to media.

RNA was extracted using the Favorgen Total RNA kit and cDNA was synthesized using MMLV reverse transcriptase (Promega). Gene expression was measured using VIC and FAM conjugated Taqman probes or gene specific primers and TaqMan Fast Advanced MasterMix (Thermo Fisher Scientific). A full list of primers and probes is available in Supplementary Table S4. Genes were normalized to 18s ribosomal RNA (Applied Biosystems, 4319413E) or the housekeeping genes 36B4 or UBC using the ΔCt method.

Prior to collection, cells were treated with 5 µM of the zinc fluorophore Zinpyr-1 in PBS for 30 minutes at 37 °C. Organoids and Huh-7 cells were next treated with collagenase to generate single cell suspensions necessary for flow analysis. Cells were stained with the viability dye FVS700 (Becton Dickinson) and examined using the BD Bioscience LSR Fortessa Cell Analyzer or the Miltenyi Biotec MACSQuant Analyser 10. To examine individual PBMC populations, PBMCs were treated with FcR blocking reagent (Miltenyi Biotec) then labeled with the following panel: BUV395 CD3, PE-Cy7 CD56, BV711 CD14, APC-Cy7 CD19 and PE-CF594 CD11c. Flow cytometry data was analyzed using FlowJo v10.8.1.

Data was analyzed and figures generated in GraphPad Prism Version 8. Statistical tests used were performed based on the normality of the data (parametric versus non-parametric) and are indicated in figure legends along with the number of biological replicates. All in vitro experimentation was performed using a minimum of 2 experimental replicates and 2 technical replicates, and sample sizes have been included in figure legends. Functional annotation was performed using ConsensusPath-DB (43) and Gene Set Enrichment Analysis (GSEA) Software (44). Over-representation analysis was used to define biological processes and pathways associated with up-regulated (>1.5x) gene sets. Gene expression heat maps were generated using Morpheus software from the Broad Institute (https://software.broadinstitute.org/morpheus).

To identify genes that are consistently and robustly induced by zinc, we searched for publicly available transcriptomic datasets that performed zinc treatments. Of ten studies identified, eleven zinc-treatment datasets were suitable for statistical analysis (Table 1). Analyzed transcriptomes were heterogeneous with regard to microarray/sequencing technology used, cell type, zinc treatment (zinc salts and zinc nanoparticles), zinc concentration (25-250 µM) and treatment duration (4–24 h). As a result, the number of significantly up-regulated (≥1.5-fold induction, p<0.05) ZSGs ranged from <10 to >1000 (full list in Supplementary File 1).

Significantly up-regulated ZSGs identified in ≥5/11 datasets were first chosen to generate a preliminary transcriptomic zinc signature (Figure 1A). A total of 18 transcripts were identified, dominated by genes belonging to the metallothionein (MT) gene family. In addition, a set of genes responsible for DNA damage and stress responses (e.g. HSPA6, DDIT3) were up-regulated suggesting that some zinc treatment experiments were eliciting significant cell stress. Indeed, when examining ZSG induction among all datasets, only those experiments using high zinc concentrations above 150 µM (e.g. Palmer et al. (32), GSE60159 (23)) or zinc ionophores (GSE6960 (19)) stimulated a transcriptomic stress response (Figure 1B). Dataset-specific transcriptional responses were confirmed using functional annotation performed using ConsensusPath-DB (43). Cell stress pathways were enriched among datasets using high zinc concentrations, whereas the GSE99435 transcriptome was enriched for metal ion response pathways, with minimal stress pathway enrichment (Figure 1C). As serum zinc rarely climbs above 20 µM in vivo (45), it is unlikely to stimulate cell stress responses identified here. Consequently, all cell stress related transcripts were removed from the preliminary zinc signature. Even in the absence of datasets eliciting zinc-mediated cell stress, the nine remaining ZSGs remained highly up-regulated (Figure 1D).

We next examined ZSG promoters using the online tool PScan to identify over-represented transcription factor binding motifs that may indicate up-regulation by zinc. We input 4 lists of ZSGs including the nine gene zinc signature, as well as lists representing ZSGs present in ≥5 datasets (n=18), ≥4 datasets (n=57) and ≥3 datasets (n=119). The binding motif of metal transcription factor 1 (MTF-1) was significantly enriched in all gene lists, possessing binding sites in every promoter sequence within the nine gene signature (Supplementary Table S5). MTF-1 is directly activated by zinc and is a major driver of MT expression, as demonstrated by knockdown studies (21).

To verify the nine gene transcriptional zinc signature following zinc depletion, we examined ZSG expression in two datasets that used different methods of zinc removal: S100A12 resin-based zinc depletion (GSE108923) (27) or TPEN-mediated ion chelation (GSE99204) (28) (full list of DEGs in Supplementary File 2). Confirming our hypothesis, all measured zinc signature genes were down-regulated in response to S100A12 resin-based zinc sequestration in HEK293T cells and were stimulated following the addition of zinc (Figure 2A). In TPEN treated human keratinocytes, zinc signature gene transcripts were significantly reduced except for MT1F and SLC30A1 which were up-regulated (Figure 2B). In both datasets, cell stress marker expression (e.g., DDIT3, DNAJB1) was increased by zinc removal, suggesting that they are indeed not true ZSGs.

Functional annotation of transcriptomes was carried out to better understand the effect of zinc-deprivation in these datasets. S100A12-mediated zinc removal is highly specific (27), and resulted in minimal disruption of cell processes based on transcriptional responses. On the other hand, TPEN ion chelation, which is not zinc specific (46, 47), inhibited numerous homeostatic pathways including transcription, translation, metabolism and cell cycle (Figure 2C).

In order to resolve the inconsistent modulation of ZSGs (GSE99204) and absence of MT expression data (GSE108923) we performed zinc treatment and depletion experiments in vitro. To model the major hepatic cell populations, three in vitro models were used: Huh-7 hepatoma cells and primary liver organoid monolayers were used to model hepatocytes and peripheral blood mononuclear cells (PBMCs) to model resident and infiltrating immune cells. All cells were cultured in mock treated or S100A12 resin zinc depleted (ZnD) media for 24 h ± supplementation with 50 µM ZnSO₄. Intracellular zinc was quantified using the zinc fluorophore Zinpyr-1 and ZSGs quantified by qPCR. Compared to untreated Huh-7 cells, zinc depletion and zinc treatment resulted in a significant reduction and increase in cellular zinc content, respectively, as measured by Zinpyr-1 fluorescence (Figure 3A). Consistent with cellular zinc content, all ZSGs were significantly down-regulated following zinc depletion, apart from SLC30A1, and significantly induced following zinc treatment (Figure 3B). Genes demonstrating the strongest response to zinc concentration (e.g., MT1E, MT1G) were generally highest expressed in untreated Huh-7 cells (Figure 3C). Liver organoid intracellular zinc was modulated by zinc depletion and treatment (Figure 3D, p>0.05), however to a far lesser degree than Huh-7 cells. Similarly, liver organoid ZSG expression was less affected by zinc depletion and highly sensitive to zinc treatment (Figure 3E). Organoid ZSG expression was also unique, with SLC30A1 demonstrating the highest relative expression and MT1G below the limit of detection (Figure 3F).

Following zinc depletion and supplementation, PBMC populations were also examined by flow cytometry to assess intracellular zinc content (gating in Supplementary Figure S3). Untreated dendritic cells (DCs) and monocytes possessed larger intracellular zinc pools compared to lymphocyte populations, as demonstrated by higher Zinpyr-1 fluorescence (Figure 3G). Interestingly, only lymphoid cells were significantly affected by zinc depletion, showing a reduction in zinpyr-1 fluorescence (B cells and NK cells, p<0.05) (Figure 3H). Except for T cells, all cell populations exhibited increased zinpyr-1 fluorescence in response to zinc treatment, with monocytes demonstrating a potent zinc influx as compared to other cells. PBMC MT1A, MT1E, MT1G and MT1M were stimulated by zinc, whereas all genes except MT1M and SLC30A1 were significantly reduced by zinc depleted media (Figure 3I). Baseline ZSG expression exhibited very low MT1M expression as compared to other ZSGs (Figure 3J), perhaps reflecting the strong induction following zinc treatment.

To assess the zinc-specific induction of the ZSGs, PBMCs were additionally treated with 50 µM metal salts containing divalent cations calcium (Ca²⁺), copper (Cu²⁺) manganese (Mn²⁺), magnesium (Mg²⁺) and iron (Fe²⁺) in addition to zinc (Zn²⁺) (Supplementary Figure S4). In addition to zinc, copper treatment demonstrated a significant induction of multiple ZSGs.

To compare hepatic zinc status among different chronic liver diseases, transcriptomic datasets containing data from healthy and diseased liver tissue were queried (full list in Supplementary Table S6). To simplify our analysis, a zinc signature score was generated using the average normalized log counts per million (CPM, RNA sequencing platforms) or log intensity (microarray platforms) of the nine validated ZSGs. All 9 ZSGs were given identical weights within the zinc signature score to facilitate the use of the zinc signature across datasets and tissues where expression of individual ZSGs differs significantly, as demonstrated in Figures 3C, F, J. ZSG expression data was obtained from chronic hepatitis B virus (CHB, n=3) and chronic hepatitis C virus (CHC, n=1) infected, steatosis (n=5), non-alcoholic steatohepatitis (NASH, n=5), hepatocellular carcinoma (HCC, n=4), cirrhosis (n=5) and alcohol-related liver disease (ALD, n=4) liver tissue, and compared to healthy controls available within each dataset to generate a LOG2 fold change (FC) for each gene (Figure 4A). While chronic viral infection and steatosis did not affect ZSG expression, ZSGs decreased in response to chronic inflammation (e.g., NASH) and fibrosis, reaching a minimum in HCC tumor tissue. Among non-cancerous tissue, ALD and cirrhotic liver ZSG expression were most significantly reduced compared to healthy tissue, supporting a zinc deficient state in the liver.

To further investigate the four AH datasets outlined in Figure 4A, ZSG expression was examined from healthy and severe AH liver tissue transcriptomes: GSE28619 (n=7 normal, n=15 severe AH [Maddrey’s discriminant function >32]) (34), GSE143318 (n=5 normal, n=5 severe AH) (35), GSE155907 (n=4 normal, n=6 AH undergoing liver transplant) and GSE142530 (n=12 normal, n=10 severe AH) (37). Among the four datasets, all ZSGs were consistently down-regulated, with MT1F reaching significance in all datasets (Figure 4B).

We next queried an additional liver tissue dataset (accession phs003112.v1.p1 (40)) containing demographic and blood test results from the database of Genotypes and Phenotypes (dbGaP) to determine whether hepatic zinc status, as defined by the transcriptomic zinc signature score, correlates with any clinical parameters. Of the 79 liver transcriptomes contained within the dataset, 10 were classified as normal, 41 AH, 19 chronic HCV infection and 10 MAFLD. Consistent with datasets outlined in Figure 4A, the MAFLD zinc signature score was similar to healthy liver, whereas both acute and chronic AH demonstrated a significantly reduced zinc signature score (Figure 4C).

Individual zinc signature scores were next correlated against demographic information (age), blood test results (e.g., leukocyte count), liver function tests (e.g., AST) and predictive scores (e.g., model for end-stage liver disease [MELD]). Using the entire dataset, the zinc signature score demonstrated a strong positive correlation with platelet count, albumin (ALB) and sodium, as well as a similarly strong negative correlation with AST, ALP, TBIL and INR. Among the AH patient subgroup (n=41), only the negative correlation with AST remained significant (Figure 4D). Scatter plots depicting correlation between the zinc signature score and albumin, AST and INR (Figure 4E) highlight the disparity between patients with ALD and the remainder of the cohort. Among patients with ALD, a low hepatic zinc signature score was associated with low serum albumin and low platelet count (thrombocytopenia) and prolonged clotting time (INR) suggestive of poor liver function, as well as elevated AST, a marker of liver inflammation and hepatocyte death. Correlation and regression analysis data from Figure 4 is available in Supplementary File 3.

To determine the role of hepatic zinc in AH pathogenesis, three severe AH datasets of liver biopsy transcriptomes were queried: GSE94397 (n=71), GSE94399 (n=38) and GSE103580 (n=110). In addition, a merged, batch-corrected dataset was generated from all three datasets containing 219 samples. A hepatic zinc signature score was generated for every patient sample and was and was subsequently correlated against every other gene across all four datasets. Figure 5A demonstrates the top 50 positively correlated genes across all four datasets (Supplementary File 4). The zinc signature score was found to correlate positively with multiple genes belonging to members of the acute phase response to infection, including complement (C5, C9), serum amyloid proteins (SAA1, SAA2), and C-reactive protein (CRP). Gene set enrichment analysis of resulting correlation coefficient ranked lists (Figure 5B) demonstrated that the zinc signature score was associated with zinc homeostasis and response, as well as acute inflammatory responses including complement activation and the acute phase response. Interestingly, enrichment of bioenergetic and metabolic responses were also identified including ATP synthesis and oxidative phosphorylation, as well as amino acid/fatty acid catabolism. Samples from each dataset were next grouped into quartiles based on their zinc signature scores and compared (Figure 5C). The top quartile, representing patients with the highest zinc signature scores, were compared to the bottom quartile with the lowest zinc signature scores. Acute phase response genes orosomucoid 1 (ORM1), serum amyloid A1 (SAA1), C-reactive protein (CRP) and complement component 9 (C9) were consistently higher among patients with higher zinc signature scores. Similarly, protein catabolism genes such as glutamate dehydrogenase (GLUD1) and arginase 1 (ARG1) responsible for catabolism of glucogenic amino acids glutamic acid and arginine were consistently higher in all datasets, however no significant differences were observed.

Zinc influx into the liver occurs rapidly during the acute phase response, so we sought to determine whether hepatic zinc is truly a requirement for the production of acute phase reactants. The acute phase response is a response to infection initiated in the liver following hepatocyte stimulation by inflammatory mediators such as IL-6, IL-1β and TNF. Because macrophages are the primary mediators of the hepatic inflammatory response, we first sought to examine macrophage IL-6 and IL-1β production in response to zinc depletion. Monocyte-derived macrophages were cultured in mock, ZnD or ZnD + 50 μM ZnSO₄ media overnight, then simulated with 100 ng/ml Kdo2-Lipid A (KLA), the immunoreactive component of lipopolysaccharide (LPS) for 6 h. Macrophage intracellular zinc and MT1F expression were significantly reduced by zinc depleted media supporting a reduction in intracellular zinc (Figure 6A). KLA-induced IL6 and IL1B expression were significantly reduced following zinc depletion and restored with the addition of ZnSO₄ (Figure 6B). Secreted IL-6 was reduced in response to zinc depletion, whereas IL-1β remained unaffected.

IL-1β and IL-6 (20 ng/ml each) were next used to stimulate the acute phase response in both Huh-7 cells (Figure 6C) and liver organoid monolayers (Figure 6D). Both Huh-7 SAA1 and ORM1 expression were increased following 6 h IL-1β and IL-6 treatment, however only ORM1 was affected by cellular zinc status, demonstrating a reduction in expression in zinc depleted samples, even following the addition of zinc. SAA1 protein secretion was measured in Huh-7 culture media following cytokine stimulation but could not be detected. Consistent with Huh-7 findings, SAA1 transcript expression and protein secretion remained unaffected by zinc concentration in the organoid model. ORM1 could not be detected in organoid cultures, however the acute phase reactant C4BPA was similarly unaffected by zinc. Together, these data indicate that zinc deficiency in AH may dampen the initiation of the acute phase response by limiting the expression of inflammatory mediators in the liver.