Molecular and Metabolic Analysis of Arsenic-Exposed Humanized AS3MT Mice

Abstract

Background:

Chronic exposure to inorganic arsenic (iAs) has been associated with type 2 diabetes (T2D). However, potential sex divergence and the underlying mechanisms remain understudied. iAs is not metabolized uniformly across species, which is a limitation of typical exposure studies in rodent models. The development of a new “humanized” mouse model overcomes this limitation. In this study, we leveraged this model to study sex differences in the context of iAs exposure.

Objectives:

The aim of this study was to determine if males and females exhibit different liver and adipose molecular profiles and metabolic phenotypes in the context of iAs exposure.

Methods:

Our study was performed on wild-type (WT) 129S6/SvEvTac and humanized arsenic $+3$ methyl transferase (human AS3MT) 129S6/SvEvTac mice treated with $400\text{ppb}$ of iAs via drinking water ad libitum. After 1 month, mice were sacrificed and the liver and gonadal adipose depots were harvested for iAs quantification and sequencing-based microRNA and gene expression analysis. Serum blood was collected for fasting blood glucose, fasting plasma insulin, and homeostatic model assessment for insulin resistance (HOMA-IR).

Results:

We detected sex divergence in liver and adipose markers of diabetes (e.g., miR-34a, insulin signaling pathways, fasting blood glucose, fasting plasma insulin, and HOMA-IR) only in humanized (not WT) mice. In humanized female mice, numerous genes that promote insulin sensitivity and glucose tolerance in both the liver and adipose are elevated compared to humanized male mice. We also identified Klf11 as a putative master regulator of the sex divergence in gene expression in humanized mice.

Discussion:

Our study underscored the importance of future studies leveraging the humanized mouse model to study iAs-associated metabolic disease. The findings suggested that humanized males are at increased risk for metabolic dysfunction relative to humanized females in the context of iAs exposure. Future investigations should focus on the detailed mechanisms that underlie the sex divergence. https://doi.org/10.1289/EHP12785

Introduction

Chronic exposure to inorganic arsenic (iAs) is associated with numerous health complications, including but not limited to cancer, metabolic disease, and cardiovascular disease.^1–3 iAs, depending on the duration and dose, is considered a diabetogen, as numerous epidemiological studies show an association between increased type 2 diabetes (T2D) prevalence and iAs exposure.^4–6 However, the underlying mechanisms that cause metabolic disease, such as T2D, remain unclear.^4,7–9 Previous studies have shown that iAs affected murine pancreatic beta cell function via impaired insulin secretion,^10–13 a well-established feature of T2D. In addition, other studies have highlighted the statistical association of iAs exposure with insulin resistance in peripheral tissues and impairment of glucose homeostasis.^14–17 However, an important caveat to the previous in vivo mouse studies is the inherent differences in iAs metabolism between species.^18,19 iAs is metabolized by the conserved enzyme arsenic ($+3$ oxidation state) methyltransferase (AS3MT).^1,20–22 In brief, AS3MT metabolizes iAs in a series of methylation steps, first generating monomethylarsenite (MMAs^III) and then the final product dimethylarsenite (DMAs^III), which is ultimately excreted.¹ It is well established that humans are poor metabolizers of iAs compared to other mammals, particularly common experimental models like rodents.^18,23 In humans, iAs is metabolized at a slower rate,²⁴ allowing for iAs and its metabolites to persist in the liver longer and potentially escape into circulation and resist clearance.^25,26 This difference in iAs metabolism between rodents and humans makes it very challenging to translate the results of rodent exposure studies for human populations. To overcome this limitation in the field, Koller et al. recently developed a mouse model on the 129S6 background in which the mouse As3mt is replaced with the human ortholog.²³ This model, hereafter referred to as “humanized,” expresses human AS3MT and metabolizes iAs at a human-like rate.²³

Various studies highlight the impact of iAs exposure as a potent modulator of gene expression through multiple mechanisms. We and others have shown that in vitro iAs exposure alters microRNA (miR)^12,27 and gene expression²⁷ in pancreatic beta cells,^12,27,28 while others have highlighted this in in vivo studies in wild-type (WT) mice^10,11 and in human studies in areas with endemic iAs.²⁹ However, due to variation in iAs metabolism across species^18,19 and differences across studies in terms of iAs exposure dose and duration, cross-study comparisons are difficult. Utilizing our novel humanized line,²³ we designed a study in mice that more closely resembles iAs metabolism and circulating arsenical distribution in humans.

It is well established that there are inherent sex differences in metabolic syndrome onset in both humans^30–33 and mice.^8,34 Also, it has been reported that inorganic arsenic trioxide (ATO) can directly interact with 17-beta-estradiol (E2).³⁵ Potential sex divergence in metabolic disease onset and severity has been observed in the contexts of both nutritional stress (high-fat diet)³⁶ and arsenic exposure in wild-type mice.³⁷ Notably, female wild-type mice exposed to arsenic exhibited metabolic syndrome only after the removal of ovaries, which was then rescued with estradiol supplementation.^36,37 One important limitation of the latter study is that the experiments were performed in wild-type mice, which exhibit very different rates of iAs metabolism relative to humans. In this study, we sought to assess sex divergence in humanized mice exposed to iAs. As the liver and adipose are both key insulin targets,^38–41 we hypothesized that male humanized mice will exhibit impaired metabolic phenotypes and related molecular profiles in the liver and adipose, relative to their female humanized counterparts.

Methods

Animal Model and Study Design

The 129S6/SvEvTac mice expressing either the mouse As3mt or the human AS3MT (from here on called “humanized”) were used for this study. Humanized mice were generated as described previously.²³ In brief, a specific displacer construct was introduced into embryonic stem cells (ESCs) [the construct has homology arms to flanking genes Cyp17a1 and Cnm2 and contains a phosphoglycerate kinase (PGK)-neo cassette flanked by mutant loxP sites that was used as a selection marker to maintain ESCs]. ESCs then underwent Cre-mediated displacement. ESCs expressing the human AS3MT were injected into the blastocyst of WT 129S6/SvEvTac mice (Taconic Bioscience). Chimeric mice were bred with WT 129S6/SvEvTac mice (Taconic Bioscience) to maintain the human AS3MT in the population.

For our study, we leveraged tissues from WT or humanized 129S6/SvEvTac mice (both male and female; $n=5–6$ for each group) used in a previously published study.²³ Briefly, liver and gonadal adipose tissue were collected from $\sim 5.5–6.5$-month-old WT and humanized mice exposed to $400\text{ppb}$ iAs in drinking water for 4 wk. This dosage was selected based on the following rationale: a) According to the World Health Organization (WHO), numerous people worldwide are exposed to concentrations of iAs much higher than the safe ($>100\text{ppb}$) legal limit ($10\text{ppb}$)⁴²; b) $400\text{ppb}$ was shown to be sufficient to lead to a nonlethal accumulation of iAs and its metabolites in various tissues in our humanized mouse line²³; and c) T2D onset in humans is associated in areas with iAs exposure above $150\text{ppb}$.⁴

Body weight was collected before exposure began and immediately prior to sacrifice. Harvested tissue was flash frozen and stored at $-{80}^{\mathrm{o}}\mathrm{C}$. Serum blood was collected for fasting blood glucose (FBG), fasting plasma insulin (FPI), and homeostatic model assessment for insulin resistance (HOMA-IR).

Diabetes Indicators

After 6 h of fasting, mouse blood was collected via tail nick to measure FBG using a OneTouch Ultra 2 glucometer with OneTouch Ultra strips (LifeScan Inc.). Submandibular blood was collected using a heparinized capillary (Kimble) and plasma was isolated by centrifugation ($\mathrm{1,700}\times g\text{for}15\text{min}$ at 4°C) and stored at $-80°\mathrm{C}$ until analysis. FPI was measured using Crystal Chem Mouse Insulin Elisa kits according to the manufacturer’s instructions. HOMA-IR was calculated as:

$$\text{HOMA-IR}=[\left(\text{FBG},\text{in}\mathrm{mg}/\text{dL}\right)\times (\text{FPI},\text{in}\mathrm{\mu }\text{IU}/\text{ml})]/405.$$

RNA Isolation, Sequencing, and Downstream Analysis

Total RNA was isolated from frozen liver and gonadal adipose tissue using the Total RNA Purification kit (Norgen Biotek) and quantified using the NanoDrop 2000 (Thermo Fisher Scientific). RNA integrity was quantified using the 4200 TapeStation (Agilent Technologies). Isolated RNA was used to make libraries for both small RNA (smRNA)-sequencing and RNA-sequencing. SmRNA-sequencing libraries were prepared by the Genome Sequencing Facility of Greehey Children’s Cancer Research Institute at the University of Texas Health Science Center in San Antonio using the TriLink CleanTag Small RNA Ligation kit (TriLink Biotechnologies). Eleven to twelve libraries were sequenced per lane single-end 50× on the HiSeq3000 platform. RNA-sequencing libraries ($\text{polyA}+$) were prepared at Cornell’s Transcriptional Regulation and Expression Facility (TREx) using NEB Next Ultra II kits (New England BioLabs). Paired-end sequencing was performed at 20 million reads/sample on the NextSeq500 platform (Illumina).

Small RNA-sequencing reads were processed using miRquant 2.0.⁴³ In brief, the 3’ sequencing adapter was removed from the read, reads larger than 14 nt were aligned to the mouse genome (mmu9),⁴⁴ and aligned reads were quantified. Any reads aligning to microRNA (miR) loci were annotated according to miRbase (version 18). Differential miR expression was determined using DESeq2.⁴⁵ RNA-sequencing reads were aligned to the mouse genome (mmu10)⁴⁶ using STAR (version 2.4.2a),⁴⁷ and reads aligning to the transcriptome were quantified using Salmon.⁴⁸ Differential gene expression across treatment groups was determined using DESeq2.⁴⁵ The design used for smRNA-seq was ∼batch + condition, where condition is the sex of the 129S6/SvEvTac mice in each genotype. Principal component analyses (PCA) were performed on regularized-logarithm (rlog) transformed DESeq2 data for both smRNA- and RNA-seq datasets. Regularized-logarithm (rlog) transformation was used to account for gene count variance.⁴⁵ The Limma package function RemoveBatchEffect() was used to correct for batch effect. miRs were considered significantly altered if they met the following parameters: basemean $>500$, ${\log }_2\text{FoldChange}>1$ or $<-1$, and $p$-adjusted $<0.05$. For RNA-seq, the design was the same, but there was only one batch (design used was ∼condition). We also performed a similar analysis in which the condition was the genotypes (i.e., WT male mice compared against humanized male mice). We used the same filtering parameters to identify significantly altered genes as what we used for miRs, albeit with a lower threshold for ${\log }_2\text{FoldChange}$: basemean $>500$, ${\log }_2\text{FoldChange}>0.5$ or $<-0.5$, and $p$-adjusted $<0.05$. Normalized counts were determined using DESeq2’s internal count normalization (normalized.counts <- as.data.frame(counts(dds, normalized=TRUE))). In brief, DESeq2 performs an internal normalization where the geometric mean is calculated for each gene across all samples. The counts for a gene in each sample is then divided by this mean. The median of these ratios in a sample is the size factor for that sample.^45,49 The basemean is the average of the gene’s normalized counts divided by size factors across all samples.⁴⁵ Volcano plots were used to indicate significantly altered (p-adjusted $<0.05$) genes that had a ${\log }_2\text{FoldChange}>|1|$ miR or a ${\log }_2\text{FoldChange}>|0.5|$ for messenger RNAs (mRNAs). Both ${\log }_2\text{FoldChange}$ and the $p$-adjusted were determined using DESeq2.⁴⁵ All ${\log }_2\text{FoldChange}$ calculations were performed as follows unless otherwise indicated: male mice/female mice for both WT or humanized groups. Benjamini-Hochberg correction (default DESeq2 method) was used to calculate adjusted $p$-values.

Pathway, Transcription Factor Motif, and Gene Network Analysis

TargetScan (version 8.0)⁵⁰ was used to identify putative miR binding sites in genes of interest. We performed gene pathway analysis using Enrichr⁵¹ (we highlighted KEGG and Elsevier pathway enrichment). We used the tool ChEA3⁵² to identify transcription factor binding motifs that were enriched in differentially expressed genes. As described in the original publication,⁵² transcription factor (TF) scores are determined through Fisher’s exact T-test (FET). We therefore used a TF enrichment score threshold of 0.05 to identify significantly enriched TF motifs. To generate the gene network presented in this paper, we used GeneMANIA,⁵³ STRING,⁵⁴ and Cytoscape⁵⁵ to highlight a network of selected genes that were associated with altered metabolic phenotypes.

Protein Isolation, Quantification, and Western Blotting

Protein isolation was performed by pulverizing frozen liver tissue in radioimmunoprecipitation assay (RIPA) buffer (Sigma-Aldrich; product number R0278) with HALT Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific; catalog number 78440) following standard guidelines. Tissue was sonicated at $4°\mathrm{C}$ using the Biorupter Pico (diagenode) following company guidelines. In brief, each sonication cycle is 30 s on, 30 s off, and was stopped after 2–5 cycles for the liver samples. Protein content was quantified using the Pierce Microplate BCA protein assay kit (Thermo Fisher; product number 23252). Twenty micrograms of protein was loaded into a 12% acrylamide gel (Thermo Fisher; product number XP00125BOX) and run at 100V. Samples were transferred to a $0.2\mathrm{\mu }\mathrm{m}$ nitrocellulose membrane (BioRad; product number 1704270) using a wet transfer system (100V, 100 min). Blocked in 5% nonfat milk in Tris buffer saline and tween (TBST) and blotted for the following proteins: Klf11 (1:500; Abnova; product number H00008462-M03) and $\mathrm{\beta }\text{-Actin}$ (1:1,000; Cell Signaling; catalog number 4970S). Anti-mouse IgG, horseradish peroxidase (HRP)-linked (1:3,000; Cell Signaling; product number 7076S) was used for the detection of Klf11 and anti-rabbit IgG, HRP-linked (1:3,000; Cell Signaling, product number: 7074S) to detect $\mathrm{\beta }\text{-Actin}$. Protein levels were quantified using ImageJ⁵⁶ normalized to $\mathrm{\beta }\text{-Actin}$ using densitometry.

Arsenical Quantification

Concentrations and proportions of iAs and its methylated metabolites were determined in the liver and gonadal adipose. Homogenates of tissues were prepared in ice-cold deionized (DI) water (10% wt/vol) using Wheaton Potter-Elvehjem-style tissue grinders with a polytetrafluoroethylene (PTFE) pestle and Wheaton overhead stirrer apparatus (DWK Life Sciences). Tissue homogenates were then treated with 2% L-cysteine (Sigma-Aldrich) at room temperature for 1 h to reduce pentavalent As species to their trivalent counterparts prior to the analysis.⁵⁷ As previously described here,^57,58 cysteine-treated liver homogenates were analyzed by hydride-generation (HG) atomic absorption spectrometry (AAS) coupled with a cryotrap (CT). Homogenates of adipose tissue were analyzed by HG-CT-inductively coupled plasma-mass spectrometry (ICP-MS).⁵⁹ The limits of detection (LODs) for iAs, MMAs, and DMAs using HG-CT-AAS were 14, 8, and 20 pg, respectively, and for HG-CT-ICP-MS⁵⁸ ranged from 0.27 to 1.7 pg iAs.⁵⁹ Zero was imputed for measurements below the LOD.

Statistical Analysis

Significance was determined using the Mann-Whitney statistical test (nonparametric, unpaired, and two-tailed; GraphPad Prism version 9.5.1 for Windows; GraphPad Software, www.graphpad.com) unless explicitly stated otherwise. All correlations are reported with Pearson’s correlation coefficient (GraphPad Prism version 9.5.1 for Windows, GraphPad Software, www.graphpad.com).

Results

Measuring Arsenicals in Liver and Adipose Tissue

To confirm that humanized mice exhibit slower rates of iAs metabolism, we analyzed the levels of arsenic in the liver and gonadal adipose of a subset of mice used in a prior iAs exposure study²³ (Figure 1A; corresponding data for Figure 1 in Table S1). In liver, we found that humanized mice showed higher levels of iAs compared to WT mice and exhibited an accumulation of the intermediate MMAs. In adipose, we found that while iAs accumulation was not as profound as in the liver, MMAs were more abundant than DMAs, unlike the WT counterparts (Figure 1B).

Figure 1.

Measuring arsenic in liver and adipose tissue of humanized and wild-type mice. (A) Arsenical quantification in isolated liver of mice (WT male, $n=5$; WT female, $n=4$; humanized male, $n=6$; humanized female, $n=5$). (B) Arsenical quantification in isolated adipose of mice (WT male, $n=4$; WT female, $n=5$; humanized male, $n=5$; humanized female, $n=6$). Mann-Whitney statistical test (unpaired and two-tailed) was performed to determine significance (ns = not significant $p$-value $>0.05$; *p-value $<0.05$; **p-value $<0.005$). Error bars indicate $\text{mean}\pm \text{standard}\text{deviation}$. Data reported in Table S1. Note: DMAs, dimethylarsenite; iAs, inorganic arsenic; MMAs, monomethylarsenite; WT, wild type.

Assessing Liver and Adipose microRNA Profiles

MicroRNAs (miRs) are well-studied biomarkers of insulin resistance in liver and adipose tissue.^60–62 To define miR profiles in iAs-exposed “humanized” mice, we performed small RNA-seq on isolated liver and adipose. Read length distributions indicated high-quality library preparations across all samples, as seen by a sharp peak at the canonical miR length ($\sim 22$ nt) and almost no sample degradation (Figure S1; corresponding data for Figure S1 in Table S2). Principal component analysis (PCA) of the adipose miR profiles showed a modest separation by genotype in male mice (Figure S2A; corresponding data for Figure S2 in Table S3); however, we found that there is a much more pronounced stratification by sex for both WT and humanized mice (Figure 2A; corresponding data for Figure 2 in Table S4). Differential expression analysis using DESeq2 (see “Methods”) showed no upregulated miRs and 12 downregulated miRs in WT male vs. humanized male mice, and no differentially expressed miRs when comparing WT and humanized females (Figure S2B). When comparing wild-type males to wild-type females, we found that there were seven downregulated miRs and 12 upregulated in males (Figure 2B; Table S5 and S6). The same analysis in humanized mice revealed three downregulated miRs and five upregulated miRs in males vs. females (Figure 2B). Notably, miR-34a was uniquely upregulated in male humanized mice (Figure 2C,D; Table S6).

Figure 2.

Assessing microRNA expression profiles in adipose tissue of humanized and wild-type mice exposed to iAs. (A) PCA plot of rlog transformed miR expression in the adipose of iAs-exposed mice. All groups together (top), WT only (middle), humanized only (bottom). Two different sequencing batches are shown by circle and triangle, respectively. (B) Volcano plots showing DE miRs in adipose of male vs. female WT (left panel) and humanized (right panel) mice. miRs upregulated in males vs. females shown in red, downregulated shown in blue. miR-34a is indicated by the red arrow. (p-adjusted $<0.05$, log2FoldChange $<-1$ or $>1$, basemean $>500$). (C) Venn diagram of upregulated miRs in adipose of males vs. females in WT and humanized mice. (D) Expression (normalized counts) of adipose miR-34a shown separately for each sex in humanized (left) and WT (right) mice. DESeq2 performs an internal normalization where geometric mean was calculated for each gene across all samples. The counts for a gene in each sample is then divided by this mean. The median of these ratios in a sample is the size factor for that sample.⁴⁹ Significance determined by DESeq2 (**p-adjusted value $<0.05$). Whiskers represent the minimum and maximum values, the midline is the median, and the limits are the 25th and 75th percentiles. Data reported in Table S4. Note: DE, differentially expressed; iAs, inorganic arsenic; miR, microRNA; PCA, principal component analysis; WT, wild type.

We performed the same analysis of miRs in liver from WT and humanized male and female 129S6/SvEvTac mice. In this tissue, sample stratification by sex or genotype was less clear (Figure 3A and Figure S3A; corresponding data for Figure 3 in Table S7 and corresponding data for Figure S3 in Table S8). We did not identify any miRs with significantly different expression when comparing WT to humanized mice in either male or female groups (Figure S3B). However, in WT mice, we found that seven miRs were upregulated and one was downregulated in male vs. female mice (Figure 3B; Table S4). In humanized mice, we found three upregulated miRs and one downregulated miR in male vs. female mice (Figure 3B; Table S5). Again, we found that miR-34a was significantly and uniquely upregulated only in male humanized mice compared to humanized female mice (Figure 3C,D; Table S5).

Figure 3.

Assessing microRNA expression profiles in liver tissue of humanized and wild-type mice exposed to iAs. (A) PCA plot of rlog transformed miR expression in the liver of iAs-exposed mice. All groups together (top left), WT only (top right), humanized only (bottom). Two different sequencing batches are shown by circle and triangle, respectively. (B) Volcano plots showing DE miRs in liver of male vs. female WT (left) and humanized (right) mice. miRs upregulated in males vs. females shown in red and downregulated shown in blue. miR-34a is indicated by the red arrow. ($p$-adjusted $<0.05$, log2FoldChange $<-1$ or $>1$, basemean $>500$). (C) Venn diagram of upregulated miRs in liver of males vs. females in WT and humanized mice. (D) Expression (normalized counts) of liver miR-34a shown separately for each sex in humanized (left) and WT (right) mice. DESeq2 performs an internal normalization where geometric mean is calculated for each gene across all samples. The counts for a gene in each sample is then divided by this mean. The median of these ratios in a sample is the size factor for that sample.⁴⁹ Significance determined by DESeq2 (**p-adjusted value $<0.05$). Whiskers represent the minimum and maximum values, the midline is the median, and the limits are the 25th and 75th percentiles. Data reported in Table S7. Note: DE, differentially expressed; iAs, inorganic arsenic; miR, microRNA; PCA, principal component analysis; WT, wild type.

Defining Liver and Adipose Gene Expression

Next, we performed parallel gene expression (RNA-seq) studies on matched liver and adipose tissue samples. In adipose, PCA showed sample clustering by sex (Figure 4A; corresponding data for Figure 4 in Table S9), although this was more evident among the humanized mice (Figure 4A). In the WT group, differential gene expression analysis revealed 122 downregulated genes and 148 upregulated in males compared to females (Figure 4B). In humanized mice, we found 84 downregulated genes and 55 upregulated genes in males (Figure 4B). Among downregulated genes, 64 were unique to humanized mice, 102 were unique to WT mice, and only 20 were shared (Figure 4C). Within-sex, cross-genotype comparisons are also shown in Figure S4 (corresponding data for Figure S4 in Table S10). In humanized mice, we detected notable sex-based divergence in the gene expression patterns. Specifically, we identified Bmp2 and Socs2 as two uniquely and significantly downregulated genes in humanized male adipose tissue compared to humanized female adipose tissue (Figure 4D). Furthermore, using TargetScan,⁵⁰ we identified a putative miR-34a target site for Socs2 in its 3′-untranslated region (3′-UTR). Interestingly, both Socs2 and Bmp2 showed elevated levels in humanized females compared to their WT counterparts, highlighting a divergence in the effect of iAs on gene expression based on sex. We also identified Fgfr3 and Il-33 as uniquely and significantly upregulated genes in humanized female adipose tissue compared to humanized male adipose tissue (Figure 4E).

Figure 4.

Assessing gene expression profiles in adipose tissue of humanized and wild-type mice exposed to iAs. (A) PCA plot of rlog transformed gene expression in the adipose of iAs-exposed mice. All groups together (top), WT only (middle), humanized only (bottom). (B) Volcano plots showing DE genes in adipose of male vs. female WT (top) and humanized (bottom) mice. Genes upregulated in males vs. females are shown in red, and downregulated are shown in blue. ($p$-adjusted $<0.05$, log2FoldChange $<-0.5$ or $>0.5$, basemean $>500$). (C) Venn diagram of downregulated genes in adipose of males vs. females in WT and humanized mice. (D) Expression (normalized counts) of genes that promote insulin sensitivity. (E) Expression (normalized counts) of genes that are associated with metabolic health. DESeq2 performs an internal normalization where geometric mean is calculated for each gene across all samples. The counts for a gene in each sample is then divided by this mean. The median of these ratios in a sample is the size factor for that sample.⁴⁹ Significance determined by DESeq2 (**p-adjusted value $<0.05$). Whiskers represent the minimum and maximum values, the midline is the median, and the limits are the 25th and 75th percentiles. Data reported in Table S9. Note: DE, differentially expressed; iAs, inorganic arsenic; miR, microRNA; PCA, principal component analysis; WT, wild type.

We performed a similar whole-transcriptome analysis in liver tissue. PCA showed clear stratification by both sex and genotype (Figure 5A; corresponding data for Figure 5 in Table S11). Within-sex, cross-genotype comparisons yielded few genes with altered expression (Figure S5; corresponding data for Figure S5 in Table S12). Comparisons across the sexes showed in WT mice 370 significantly downregulated genes and 348 upregulated in males compared to females. In humanized mice, we detected 162 significantly downregulated genes and 288 upregulated in males compared to females (Figure 5B). Among the downregulated genes in males, we found that 284 genes were unique to WT mice, 76 genes were unique to humanized mice, and 86 were shared (Figure 5C). We performed KEGG pathway analysis on genes uniquely downregulated in males in WT or humanized (Figure 5D and Figures S6 and S7; corresponding data for Figure S6 in Table S13 and corresponding data for Figure S7 in Table S14), which showed that insulin and peroxisome proliferator-activated receptors (PPAR) signaling are the top-most enriched only in the humanized group (Figure 5D; top 30 pathways Figure S7A). Examples of genes in these pathways included Plin5^67,68 and Angptl4^69–72 (Figure 5E). We also performed Elsevier pathway analysis, which revealed T2D and maturity-onset diabetes of the young (Figure 5E; Klf11⁷³) among the most enriched only in the humanized group (Figure S7B). Liver Nr1h2, Ppp1r3g, and Pdk4 were also uniquely and significantly downregulated in humanized male mice (Figure 5F and Figure S8; corresponding data for Figure S8 in Table S15), though they were not different with respect to WT male mice as in the case of Angptl4, Klf11, and Plin5 (Figure 5E). We further examined trends at the protein level specifically for Klf11 (Figure 5G). Klf11 protein expression followed the same pattern observed at the transcript level—a significant difference in Klf11 levels was only detected in the humanized male vs. female comparison (Figure 5G).

Figure 5.

Assessing gene expression profiles in liver tissue of humanized and wild-type mice exposed to iAs. (A) PCA plot of rlog transformed gene expression in the liver of iAs-exposed mice. All groups together (top), WT only (middle), and humanized only (bottom). (B) Volcano plots showing DE genes in liver of male vs. female WT (left) and humanized (right) mice. Genes upregulated in males vs. females are shown in red, and downregulated are shown in blue. ($p$-adjusted $<0.05$, log2FoldChange $<-0.5$ or $>0.5$, basemean $>500$). (C) Venn diagram of downregulated genes in liver of males vs. females in WT and humanized mice. (D) Results of KEGG pathway enrichment analysis. Pathways are sorted by the −log10 of pathway enrichment adjusted $p$-value. (Top) Top 15 enriched pathways of downregulated genes in WT male vs. female mice. (Bottom) Top 15 enriched pathways of downregulated genes in humanized male vs. female mice. (E) Expression (normalized counts) of genes that promote insulin sensitivity. (F) Expression (normalized counts) of genes that are associated with metabolic health. Whiskers represent the minimum and maximum values, the midline is the median, and the limits are the 25th and 75th percentiles. DESeq2 performs an internal normalization where geometric mean is calculated for each gene across all samples. The counts for a gene in each sample is then divided by this mean. The median of these ratios in a sample is the size factor for that sample.⁴⁹ Significance determined by DESeq2 (**p-adjusted value $<0.05$). (G) Western blot for Klf11 and $\mathrm{\beta }\text{-Actin}$ in humanized male and female mice. The reason for the different appearance of the background around the Klf11 band and the $\mathrm{\beta }\text{-Actin}$ band is exposure time – Klf11 exposure time was 30 s and $\mathrm{\beta }\text{-Actin}$ was 1 s. Normalization of quantified protein determined by ImageJ analysis (see supplemental Figure 5C for Western blot). Klf11 was normalized to $\mathrm{\beta }\text{-Actin}$ (left). Student’s unpaired $t$-test was used to determine significance. A Mann-Whitney statistical analysis was also performed: $p\text{-value}=0.1000$. Error bars indicate $\text{mean}\pm \text{standard}\text{deviation}$. Each genotype had an $n=3$. Data reported in Table S11. Note: DE, differentially expressed; iAs, inorganic arsenic; miR, microRNA; PCA, principal component analysis; WT, wild type.

Determining Metabolic Phenotypes

Given the liver and adipose molecular data summarized above, we hypothesized that iAs-exposed humanized male mice would exhibit worse metabolic phenotypes than humanized females. We examined body weight, FBG, FPI, and HOMA-IR. Indeed, although there was no significant difference between WT males and females in any parameter of interest at the time of sacrifice (Figure 6A; corresponding data for Figure 6 in Table S16), we did find that humanized males had higher body weight, FBG, FPI, and HOMA-IR compared to their humanized female counterparts, indicative of insulin resistance only in the humanized males (Figure 6B–D).

Figure 6.

Determining metabolic phenotypes of wild-type and humanized mice exposed to iAs. On the day of sacrifice after 1 month of iAs exposure, mice were (A) weighed and serum was collected to determine (B) FBG and (C) FPI, and to measure (D) insulin resistance via HOMA-IR. The numbers of mice used are indicated below each bar (n). Mann-Whitney statistical test (unpaired and two-tailed) was performed to determine significance (ns, not significant; *$p$-value $<0.05$; **$p$-value $<0.005$). Whiskers represent the minimum and maximum values, the midline is the median, and the limits are the 25th and 75th percentile. Data reported in Table S16. Note: BW, body weight; FGB, fasting blood glucose; FPI, fasting plasma insulin; HOMA-IR, homeostatic model assessment for insulin resistance; iAs, inorganic arsenic; WT, wild type.

Correlating Molecular Markers with Metabolic Phenotypes

Given the altered metabolic phenotypes in humanized 129S6/SvEvTac mice, we next sought to assess the relationship between miR-34a, a well-known marker of insulin resistance,^74–76 T2D,^77–79 and obesity,^80–83 and the phenotypes of interest. We performed a Pearson’s correlation analysis between miR-34a levels in either the liver or adipose and FBG, FPI, or HOMA-IR (Figure S9; corresponding data for Figure S9 in Table S17). In adipose, we observed a significant positive correlation between miR-34a levels and both FPI (Figure S9A) and HOMA-IR (Figure S9B). Humanized males exhibited the strongest correlation (Figure S9E and Figure S9F). There did not appear to be any obvious relationship between miR-34a levels in the adipose and the relative abundance of arsenicals in adipose (Figure S10A; corresponding data for Figure S10 in Table S18). In the liver, we found that miR-34a levels significantly positively correlated with FBG (Figure S9C). Humanized males again showed the strongest correlation between miR-34a expression levels and FBG, though this relationship does not achieve statistical significance (Figure S9C). miR-34a levels also positively correlated with the percent iAs in the liver of humanized male mice but not the percent MMAs or DMAs (Figure S10B).

We next examined the relationship between the expression levels of the highlighted liver genes in Figures 5E,F and the observed metabolic phenotypes. Angptl4, Klf11, and Plin5 were significantly inversely correlated with FBG (Figure 7; corresponding data for Figure 7 in Table S19) but not FPI or HOMA-IR. Ppp1r3g was found to have a significant negative correlation with FBG levels (Figure 7D), while Nr1h2 did not (Figure 7E). We did not detect a significant correlation between either gene’s expression and FPI or HOMA-IR (Figure 7D,E). For all genes highlighted here, humanized males exhibited significantly lower expression than the humanized females (Figure 7). These data further highlight a diverging effect based on sex in iAs-exposed humanized mice (Figure 7F).

Figure 7.

Correlating molecular markers with metabolic phenotypes in WT and humanized 129S6/SvEvTac mice exposed to iAs. Correlation of FBG (left), FPI (middle), and HOMA-IR (right) with expression levels of (A) Klf11, (B) Angptl4, (C) Plin5, (D) Ppp1r3g, and (E) Nr1h2. R, Pearson’s correlation coefficient. Student’s unpaired $t$-test was used to determine significance. (F) Graphical depiction of working model (created with BioRender.com). Data reported in Table S19. Note: FGB, fasting blood glucose; FPI, fasting plasma insulin; HOMA-IR, homeostatic model assessment for insulin resistance; iAs, inorganic arsenic; WT, wild type.

Defining Transcriptional Networks of Sex Divergence

We also examined genes that were uniquely differentially expressed in the liver of humanized males vs. females for transcription factor motif enrichment. Specifically, utilizing ChEA3,⁵² we determined whether the up- or downregulated genes are over-represented with targets for specific transcription factors (Table S20). We then determined if any of the identified transcription factors had significantly altered expression according to our previous DESEq2 analysis (Figure 5B). This analysis revealed that three transcription factors, Klf11, Safb2, and Foxo1, were significantly downregulated in the liver tissue of humanized male mice compared to humanized female mice and exhibited a significant transcription factor motif enrichment score (Figure 8A). Putative targets for Klf11 included Ppp1r15a, Nfkbiz, Pdk4, Txnip, Angptl4, Celsr1, and Foxo1 (Table S20). Finally, we also constructed a gene network based on databases with information on known protein interactions, co-expression, and regulatory connections (“Methods”), suggesting that Klf11 is a candidate master regulator of the sex divergence in molecular profiles in humanized mice (Figure 8; corresponding data for Figure 8 in Table S21).

Figure 8.

Defining transcriptional networks of sex divergence. (A) Enrichment scores (y-axis) for TF binding motifs among genes downregulated in liver of males vs. females of humanized mice. The x-axis is DESeq2 adjusted $p$-value of difference in liver expression of each TF in males vs. females. Blue dots are transcription factors that are both significantly enriched (score $<0.05$) and have significantly altered gene expression (p-adjusted $<0.05$, log2FoldChange $<-0.5$, basemean $>500$). (B) Network from Cytoscape⁵⁵ integrating data from CHEA3, STRING, and GeneMANIA. Black nodes are the input genes (including Klf11) to build a network. Gray dashed lines represent CHEA3 predicted transcriptional interactions. Blue, orange dotted, and gray lines indicate GeneMANIA-based co-expression, shared protein domains, predicated interactions, and physical interactions, respectively. Pink lines represent protein-protein relationships mined from STRING. Biological process annotations derived from GeneMANIA. Data reported in Table S21. Note: TF, transcription factor.

Discussion

It is well established that mice and humans have differing metabolic rates of iAs.¹⁸ To our knowledge, this is the first report of As speciation in the adipose of mice with humanized iAs metabolism. Overall, the data convey that humanized mice metabolize iAs less efficiently than their WT counterparts. We show that humanized mice have significantly higher concentrations of MMAs in gonadal adipose depots relative to their WT counterparts exposed to the same level of iAs in drinking water. We did not find any major sex differences in iAs accumulation or iAs metabolism in humanized mice. We did find that there was significantly less DMAs in the adipose of humanized mice compared to their WT counterparts (Figure 1B). Arsenical distribution in the adipose and liver was consistent with less efficient metabolism in humanized mice compared to WT. Therefore, we believe the 129S6/SvEvTac expressing the human AS3MT variant is a more relevant model for iAs-associated diseases in humans.

A consistent finding across all molecular and phenotypic analyses was the sex-specific effect observed in the humanized mice. This effect would not have been observed had we used only WT mice in the study. In both tissues, miR-34a expression was significantly higher only in humanized males relative to females. This is of considerable interest because miR-34a is one of the best-established miR markers of insulin resistance.^76,80,82,84 miR-34a has known associations with metabolic dysfunction in both the liver and adipose, including obesity,^{76,82,85–87} T2D,^77,79 impaired insulin sensitivity,^{74,75,77,88,89} nonalcoholic fatty liver disease,⁹⁰ and impaired beige and brown fat function.⁸¹ miR-34a has been found to be aberrantly elevated in individuals with obesity^77,82,83,87 and T2D.⁷⁷ Furthermore, adipocytes have been shown to secrete miRs, including miR-34a,^82,85,86,91 which is also associated with impaired metabolic phenotypes.^38,82 For our study, it was only relevant that miR-34a is a well-established marker of insulin resistance, irrespective of whether it is the main driver. Altogether, these data show that miR profiles in the context of iAs exposure depend on sex and genotype, and most notably, a classic miR marker of insulin resistance (miR-34a) was uniquely elevated in humanized male mice exposed to iAs relative to humanized female mice.

It is noteworthy also that we found miR-196 family members to be significantly downregulated in adipose tissue of both humanized and WT males relative to females. It was shown that the loss of miR-196 resulted in significantly less effective browning in vitro and in vivo.⁹² Though outside the scope of this study, it will be of interest to determine whether iAs exposure impairs brown adipose function, possibly due in part to loss of miR-196, and if this effect is sex dependent.

Our RNA-seq analysis revealed a sex-based divergence in gene expression. We identified genes with altered expression that are associated with either metabolic health or metabolic impairment. Humanized female mice exhibited higher expression of genes that confer health, whereas humanized males exhibited lower expression of metabolically healthy genes. In adipose tissue, we identified that Bmp2 and Socs2, which are implicated in promoting adipose insulin sensitivity,^93,94 were significantly lower in humanized males relative to humanized females (and relative to WT counterparts). We also identified Fgfr3 and Il-33 as uniquely more highly expressed in humanized female mice relative to humanized males (and relative to WT counterparts). These genes code for proteins that have been shown to promote insulin sensitivity (Il-33^95,96) and metabolic health (Fgfr3^63–66).

In liver tissue, we identified Angptl4,^69–71 Klf11,^73,97 and Plin5,^67,68 which have important roles in promoting insulin sensitivity and glucose homeostasis, as genes that are significantly downregulated in humanized males compared to females. Angptl4 (or Fiaf) is a multifaceted protein that is associated with numerous metabolic diseases including T2D and obesity.⁷¹ It is expressed in several tissues in the body with specific functions in each. Angptl4 is also excreted to elicit systemic effects outside of the tissues in which it is expressed.⁷¹ Its most prominent role is in the regulation of peripheral lipid and glucose metabolism.^69,71 Mouse studies have shown that overexpression of Angptl4 in the liver can improve blood glucose homeostasis and is an important insulin-sensitizing agent but increases fatty liver.^70,72,98 Klf11 has an established role in regulating insulin synthesis in the pancreatic beta cell⁷³ and has also been shown to regulate gluconeogenesis and lipid metabolism in the liver of mice.^97,99 Klf11 expression in the liver was significantly reduced in diabetic mice and mice fed a high-fat diet.^97,99 In addition, overexpression of Klf11 ameliorated the fatty liver phenotype in mice.⁹⁹ Loss of Plin5 has been shown to lead to insulin resistance in both the liver and muscle.^67,68 Plin5 is important in regulating fatty acid flux, and its loss of expression also leads to an increase in fatty acid uptake and storage.⁶⁸ We found that, under the condition of iAs exposure, humanized male mice showed significantly less Angptl4, Klf11, and Plin5 expression, suggesting loss of insulin sensitivity (Figure 5E). Decrease in Angptl4, Klf11, and Plin5 was not seen in WT male mice exposed to iAs, further supporting the importance of studying humanized mice to observe the sex-specific effect of iAs on metabolic syndrome. Ppp1r3g and Nr1h2 were both elevated in humanized female mice (compared to males) but not WT female mice (compared to males) (Figure 5F). Nr1h2 (LXR-B) is a nuclear hormone receptor that is activated by cholesterol derivatives.¹⁰⁰ Nr1h2^101,102 signaling leads to increases in glucose metabolism, and Ppp1r3g is a part of the phosphatase complex that removes phosphate groups from glycogen synthase and thereby activates glycogen synthesis.^103–105 Furthermore, Nr1h2 plays an important role in the regulation of fatty acids and triglyceride production.^106,107 Loss of Nr1h2 resulted in liver steatosis and hyperinsulinemia in mice regardless of diet.¹⁰¹ Nr1h2 activation also had similar effects on insulin signaling, wherein there was a downregulation of gluconeogenic genes and an upregulation of glucose uptake genes.¹⁰² Elevated Nr1h2 may be working in tandem with insulin to promote glucose uptake in humanized female mice. Ppp1r3g is a regulatory subunit of protein phosphatase 1 (PP1).¹⁰⁵ In response to insulin signaling, hepatic Ppp1r3g is directly phosphorylated by protein kinase B (AKT). Phosphorylation of Ppp1r3 leads to the activation of the PP1 complex and enhances dephosphorylation of glycogen synthase, leading to more glycogen production and increased glucose clearance from circulation.¹⁰⁵ Furthermore, Ppp1r3g expression is uniquely downregulated under fasting conditions, and loss of Ppp1r3g expression resulted in the decrease of glycogen synthase activity, indicating a key role in glycogen production.¹⁰⁸ Our data shows that Ppp1r3g mRNA was significantly elevated in humanized female mice in iAs-exposed conditions, which is correlated with better metabolic outcomes compared to humanized male mice. Therefore, elevated Ppp1r3g in humanized female mice exposed to iAs may be playing a role in more efficient glucose clearance. Together, the data strongly indicate sex-based divergence in iAs-exposed humanized mice of molecular profiles associated with metabolic phenotypes. iAs-induced metabolic syndrome shows sex bias in disease onset.^32,109 There is some evidence that estrogen may be the mitigating factor in iAs-induced metabolic effects.^30,110,111 A proposed hypothesis is that estrogen signaling activates an alternative pathway leading to additional methyl donation and recycling.¹¹² However, in our humanized mouse model, we saw no significant difference in iAs metabolism between humanized male and female mice and, therefore, do not believe that iAs methylation efficiency between the humanized sexes is causing the observed sex-specific effects. Estrogen signaling has also been shown to affect glucose and lipid metabolism.^36,113 Ovariectomized mice fed a high-fat diet show glucose intolerance that can be remedied with estradiol supplementation.³⁶ The metabolic protective effect of estrogen was lost in estrogen receptor (ER-a) knockout mice³⁶ and was seen in ovariectomized mice exposed to iAs.³⁷ In ovariectomized iAs-exposed WT mice, there was significantly worse fasting blood glucose, fasting plasma insulin levels, impaired glycogen production, insulin secretion in isolated islets, and less adiponectin levels.³⁷ Metabolic impairment in estrogen-deficient mice was reversed with estradiol supplementation, further suggesting that estrogen plays a key protective role in iAs-induced metabolic dysfunction.³⁷ This suggests that estrogen may be important for enhancing glucose clearance and insulin sensitivity in the context of iAs exposure. However, as stated, the above study was performed in WT mice. To test properly the protective role of estrogen in iAs-induced metabolic dysfunction, future similar studies should be performed in the humanized mouse model.

We also analyzed transcription factor motif enrichment for genes that were uniquely downregulated in the liver of humanized male mice relative to humanized female mice. Interestingly, we found that Klf11, a gene we had identified as significantly downregulated in humanized male mice, had the second highest transcription factor binding site enrichment score in our downregulated genes unique to humanized male mice. This indicates that there may be a global disruption of the Klf11 regulatory network in humanized males, leading to worse metabolic phenotypes relative to humanized females. In this network, we highlight that Klf11 has predicted transcriptional interactions with Foxo1, Angptl4, and Pdk4, all important in energy regulation and metabolism.^70,114–116 Furthermore, this network shows an integration of glucose and lipid energy metabolism wherein Klf11 may play an important regulatory role in determining whether glucose or lipids are metabolized for energy usage. It is well established that the ability to switch between glucose and lipid utilization is a key feature of metabolic health. In our network analysis, we found that a critical node is Pdk4. While elevated Pdk4 is generally associated with T2D and other metabolic dysfunction,^117,118 Pdk4 has also been shown to be critical in regulating whether the cell is utilizing glucose or lipids for energy.^119,120 Therefore, it is possible that in humanized female mice exposed to iAs, Pdk4 gene expression is elevated to allow for tighter control of glucose and lipid usage.

Conclusions

In conclusion, our study highlights a sex divergence in metabolic phenotypes in iAs-exposed humanized (but not WT) mice. We propose a general model in which humanized male mice exposed to iAs exhibit molecular changes that increase the risk for insulin resistance. Also, in humanized female mice, numerous genes that promote insulin sensitivity and glucose tolerance in both the liver and adipose were elevated compared to humanized male mice. These findings suggest that humanized females are protected from metabolic dysfunction relative to humanized males in the context of iAs exposure. Further studies should be aimed at determining the underlying mechanism for this sex-divergent effect.

A limitation of this study lies in determining the mechanism behind iAs-induced metabolic dysfunction. Though we observed a relationship between our genes of interest (miR-34a, Klf11, Angptl4, etc.) and FBG, it was outside of the scope to perform detailed functional studies to further elucidate the roles these genes play in iAs-induced metabolic dysfunction. Future studies should be aimed at functionally characterizing these genes in iAs-induced metabolic impairment, particularly Klf11 and its potential master regulator role.