ADAR regulates APOL1 via A-to-I RNA editing by inhibition of MDA5 activation in a paradoxical biological circuit

Significance

APOL1-associated kidney disease occurs in some but not all individuals carrying two copies of APOL1 genetic risk variants. The factors leading to kidney disease are incompletely understood. Viral illnesses or other environmental triggers are likely needed to activate the production of APOL1 at toxic levels to cause kidney damage. We found that a type of RNA modification called adenosine-to-inosine editing, carried out by adenosine deaminase acting on RNA (ADAR), suppressed APOL1 gene expression mediated by inflammatory pathways. APOL1’s messenger RNA was notable for triggering inflammation if not modified by ADAR editing. A transgenic mouse model replicated the editing pattern of APOL1 messenger RNA. The findings indicate that ADAR functions as a brake to counter APOL1’s rapid production during acute inflammation.

Abstract

APOL1 risk variants are associated with increased risk of kidney disease in patients of African ancestry, but not all individuals with the APOL1 high-risk genotype develop kidney disease. As APOL1 gene expression correlates closely with the degree of kidney cell injury in both cell and animal models, the mechanisms regulating APOL1 expression may be critical determinants of risk allele penetrance. The APOL1 messenger RNA includes Alu elements at the 3′ untranslated region that can form a double-stranded RNA structure (Alu-dsRNA) susceptible to posttranscriptional adenosine deaminase acting on RNA (ADAR)–mediated adenosine-to-inosine (A-to-I) editing, potentially impacting gene expression. We studied the effects of ADAR expression and A-to-I editing on APOL1 levels in podocytes, human kidney tissue, and a transgenic APOL1 mouse model. In interferon-γ (IFN-γ)–stimulated human podocytes, ADAR down-regulates APOL1 by preventing melanoma differentiation-associated protein 5 (MDA5) recognition of dsRNA and the subsequent type I interferon (IFN-I) response. Knockdown experiments showed that recognition of APOL1 messenger RNA itself is an important contributor to the MDA5-driven IFN-I response. Mathematical modeling suggests that the IFN–ADAR–APOL1 network functions as an incoherent feed-forward loop, a biological circuit capable of generating fast, transient responses to stimuli. Glomeruli from human kidney biopsies exhibited widespread editing of APOL1 Alu-dsRNA, while the transgenic mouse model closely replicated the edited sites in humans. APOL1 expression in mice was inversely correlated with Adar1 expression under IFN-γ stimuli, supporting the idea that ADAR regulates APOL1 levels in vivo. ADAR-mediated A-to-I editing is an important regulator of APOL1 expression that could impact both penetrance and severity of APOL1-associated kidney disease.

Risk variants identified in APOL1, the gene encoding apolipoprotein L-1, are associated with increased risk of nondiabetic chronic kidney disease, focal segmental glomerulosclerosis (FSGS), and HIV-associated nephropathy (1–5). These APOL1 risk variants are found only in individuals with recent African ancestry (5). Despite their detrimental effects on the kidneys, these two variants likely increased in allele frequency over the past several thousand years because they conferred resistance to the trypanosomal infections causing African sleeping sickness (6). While one risk allele confers resistance to African trypanosomes but little increase in kidney disease, the presence of two risk alleles (high-risk genotypes) greatly increases the risk of kidney disease (1, 2, 4, 7–9).

High-risk APOL1 genotypes are incompletely penetrant. Most people with two risk alleles do not develop kidney disease. The factors that determine whether a carrier of two risk alleles will develop kidney disease remain unknown. Increased levels of APOL1 expression may trigger kidney disease or increase disease susceptibility in the setting of additional trigger(s) (10, 11). Higher APOL1 expression levels correlate with increased podocyte toxicity in both cell and animal models (10, 11). Moreover, therapeutically indicated treatment with interferon (IFN; a potent inducer of APOL1 gene expression) predisposed patients with two APOL1 risk alleles to develop collapsing glomerulopathy (11, 12). Thus, understanding the regulation of APOL1 expression may help determine the factors that lead to APOL1 kidney disease in individuals with the high-risk genotype.

While both regulation of gene transcription and posttranslational modification of proteins have large effects on protein levels, posttranscriptional regulation of messenger RNA (mRNA) can also have a substantial impact. APOL1 mRNA transcripts have a secondary structure that may contribute to regulation of APOL1 mRNA levels. The 3′ untranslated region (3′UTR) contains two inversely oriented Alu repeats that can form a 282-nucleotide-long Alu double-stranded RNA (dsRNA). Alu-dsRNAs formed by endogenous transcripts closely resemble pathogen-associated molecular patterns (PAMPs) and can trigger inflammatory responses when recognized as nonself-dsRNA by pattern recognition receptors (PRRs) (13–16). In order to regulate detection of self vs. nonself, the cell also contains RNA-modifying enzymes that alter dsRNA present in both sequences of viral origin and host-encoded RNA transcripts. These enzymes defend against viruses while also preventing self-dsRNA from activating inflammatory responses (14, 17).

Adenosine-to-inosine (A-to-I) RNA editing (referred to subsequently as RNA editing) is the most prevalent form of posttranscriptional modification in mammals (14, 17, 18). Adenosine deaminase acting on RNA (ADAR) enzymes convert A to I in dsRNA (13, 18–20). Inosines pair preferentially to cytosines, leading to a base pair mismatch that alters not only dsRNA sequence but also its structure. RNA editing can affect gene expression levels through multiple mechanisms, including alteration of microRNA binding sites, alteration of mRNA recognition by RNA binding proteins (such as human antigen R), and suppression of the IFN response to self-dsRNA via dsRNA sensor recognition (18, 21, 22).

Previous studies in nonrenal human tissues have shown that APOL1 mRNA transcripts are among the most highly RNA-edited transcripts in the genome (22–24). In the present study, we manipulated RNA-editing enzymes in primary human podocytes to better understand how ADAR and RNA editing modify APOL1 expression. We then modeled the interaction of IFN, ADAR, and APOL1 in a biological circuit and derived functional characteristics of the network motif. We searched for evidence of APOL1 mRNA editing in glomerular RNA sequencing (RNA-seq) data from proteinuric patients, mapping the overall frequency and distribution of APOL1 mRNA editing and exploring its relationship to APOL1 expression level and to clinical phenotype. Finally, we studied RNAseq data from the glomeruli of APOL1-transgenic mice to define the relationship between APOL1 and ADAR in vivo.

Results

ADAR Down-Regulates APOL1 Expression.

To test whether ADAR might be important in APOL1 regulation, we first used small interfering RNA (siRNA) to achieve knockdown of ADAR in primary human podocytes. We stimulated the podocytes with IFN-γ to induce APOL1 expression (Fig. 1A) (12). APOL1 was up-regulated by more than twofold in the ADAR knockdown vs. the scramble small interfering RNA control (siCtrl) condition. In contrast, knockdown of the closely related RNA-editing enzyme ADARB1 did not affect APOL1 expression levels (Fig. 1A). These results suggest that ADAR suppresses IFN-γ–induced up-regulation of APOL1 by over 50%.

Fig. 1.

ADAR suppresses APOL1 expression via inhibition of dsRNA recognition through the MDA5–MAVS–IRF3/IRF7 pathway. Primary human podocytes were transfected with siRNAs (50 nM) 48 h prior to treatment with IFN-γ and harvested 24 h later for isolation of RNA or protein. (A) APOL1 RT-qPCR after knockdown of ADAR or of ADARB1 vs. nontargeting siRNA (siCtrl). (B) Combined knockdown of ADAR with the dsRNA sensor MDA5 reversed the up-regulation observed with ADAR knockdown alone at the levels of RNA (RT-qPCR) and (C) protein (western blot). (D) Combined knockdown of MAVS and ADAR, which interacts with activated MDA5, reversed the effect of ADAR knockdown on APOL1 expression. Thus, ADAR knockdown up-regulates APOL1 through an MDA5–MAVS-dependent pathway. (E) RT-qPCR of APOL1 after combined knockdown of ADAR and IRF3, IRF7, or IRF3 + IRF7. Both transcription factors are the last step in the activation of IFN-I triggered by MDA5. (F) Western blot of the experiment of combined knockdown of IRF3, IRF7, and IRF3 + IRF7. (G and H) Primary human podocytes were transfected with enhanced green fluorescence protein (eGFP)-ADAR_p110, eGFP-ADAR_p150, and eGFP-control (EV) overexpressing plasmids. Overexpression of either isoform of ADAR, ADAR p110, or ADAR p150 led to reduced APOL1 expression levels in podocytes stimulated with IFN-γ for 24 h to up-regulate APOL1. (H) Immunoblot showing ADAR protein levels of the different constructs. The higher molecular weight of the ADAR overexpression constructs compared with endogenous ADAR (EV) reflects the eGFP tag. APOL1 expression was reduced in the setting of ADAR overexpression. Results are representative of at least three biological replicates. *P = 0.01 to 0.05; **P = 0.001 to 0.01; ***P = 0.0001 to 0.001; ****P < 0.0001; ns, not significant, P ≥ 0.05.

We hypothesized that in the setting of ADAR knockdown, unedited Alu-dsRNAs encoding both APOL1 and other Alu-containing transcripts may activate innate immunity via PRRs, such as melanoma differentiation-associated protein 5 (MDA5; IFIH1) or retinoic acid–inducible gene I (RIG-I; DDX58), triggering type I (IFN-α/β) interferon (IFN-I) response and leading to additional APOL1 up-regulation (12–14, 25, 26). To test this hypothesis, we knocked down MDA5 or RIG-I in combination with ADAR. Combined knockdown of ADAR and MDA5 reversed the up-regulation of APOL1 to baseline control levels of APOL1 mRNA and protein (Fig. 1 B and C). We then tested combined knockdown of ADAR and the mitochondrial antiviral-signaling protein (MAVS), the protein immediately downstream of MDA5 that links dsRNA recognition with IFN-I up-regulation via activation of TANK (TRAF Family Member Associated NFKB Activator) binding kinase 1 (TBK1), interferon regulatory factor 3 (IRF3), and IRF7. Simultaneous knockdown of MAVS and ADAR similarly reversed the up-regulatory effect of ADAR knockdown alone, indicating that ADAR’s effect on APOL1 expression operates through an MDA5–MAVS-dependent pathway (Fig. 1D). To further interrogate the pathway, we performed knockdowns of the transcription factors IRF3 and IRF7 in combination with knockdown of ADAR. The combined small interfering adenosine deaminase acting on RNA (siADAR)/small interfering interferon regulatory factor 3 (siIRF3)/siIRF7 condition resulted in IFN-γ–induced expression levels of APOL1 mRNA and protein lower than produced by siADAR alone (Fig. 1 E and F).

Since ADAR knockdown led to increased APOL1 expression, we proceeded to test if ADAR overexpression would reduce APOL1 expression in cultured primary human podocytes with or without IFN-γ stimulation. We overexpressed the enhanced green fluorescence protein (eGFP)-tagged ADAR isoforms p110 (constitutive) and p150 (IFN inducible) in podocytes. Next, podocytes were stimulated with IFN-γ for 24 h to up-regulate APOL1. Podocytes overexpressing ADAR exhibited decreased APOL1 expression as compared with empty vector (EV)–transfected podocytes (Fig. 1 G and H). No differences in APOL1 expression were observed between ADAR isoforms p110 and p150. Our data indicate that ADAR acts as a repressor of APOL1 expression, despite their shared status as IFN-stimulated genes.

Modeling of IFN–APOL1–ADAR Interaction as a Biological Circuit.

We observed that IFN exerts two opposing effects on APOL1: direct stimulation of APOL1 and indirect repression of APOL1 via ADAR up-regulation and RNA editing. The interaction between IFN, APOL1, and ADAR is, therefore, best defined by a common type of biological circuit known as an incoherent feed-forward loop (FFL) (Fig. 2 A, i) (27–29). The term “incoherent” describes the fact that one arm of the FFL (direct APOL1 activation by IFN) exerts an effect opposite to that of the parallel arm (indirect APOL1 repression by IFN via ADAR) (30). Incoherent FFLs contrast with coherent FFLs, in which both arms exert consonant stimulatory or repressive effects. Both incoherent and coherent FFLs differ from feedback loops, in which a gene regulates its own activity.

Fig. 2.

APOL1 expression as a function of ADAR modeled with an incoherent FFL circuit design. (A, i) In the incoherent FFL network motif, (A, i-ii) gene X(IFN) activates Y(ADAR) and Z(APOL1), while Y represses Z. (A, iii) The Z output is defined by an AND logic, in which both X and Y are required for Z stimulation. (B) Graphical representation of the model in conditions of (B, a) baseline ADAR expression (β = 1.6) and (B, b) ADAR knockdown (β = 0.025). APOL1 dimensionless concentration doubles from 0.32 to 0.64 with a change in β from 1.6 to 0.025. This model maintains the mean ratio found experimentally (B, c), an approximately twofold up-regulation in the setting of ADAR knockdown, between IFN-stimulated control vs. ADAR knockdown. (C) Relation between normalized concentrations of ADAR and APOL1 for two β-values. For β = 0.025 (blue dots), the system operates as a simple Hill function for activation, where APOL1 concentration is independent of ADAR. In this case, APOL1’s maximum concentration is reached at steady state without pulse generation or accelerated response (note the linear correlation). For β = 1.6, APOL1 reaches maximum concentration at a higher level than steady state, which configures an accelerated and pulse-like response, characteristic of this type of circuit. (D) Comparison of an incoherent FFL vs. an unregulated response illustrates the behavior of an accelerated response. The incoherent FFL reached half of the steady-state (T_1/2) level before the unregulated response.

We applied the incoherent FFL (type 1) to model the biological circuit formed by IFN, APOL1, and ADAR (Fig. 2 A, ii). In the incoherent FFL, a gene regulator X (IFN) activates Z (APOL1) but also activates Y (ADAR), a repressor of Z. We then chose the AND logic gate for the incoherent FFL model (Fig. 2 A, iii), where both X and Y are needed to regulate Z (as opposed to an OR logic gate when activation only requires X or Y but or both).

We considered an initial concentration of zero for X, Y, and Z. Then, the concentration of X was changed from an input of zero to one as a step function. After this activation step, the concentrations of Y and Z follow Eqs. 1 and 2:

$$\frac{dY}{dt}=\beta \cdot \frac{{\left(\kappa X\right)}^{n_{XY}}}{1+{\left(\kappa X\right)}^{n_{XY}}}-Y$$ [1]

and

$$\frac{dZ}{dt}=\gamma \cdot \frac{X^{n_{XZ}}}{(1+X^{n_{XZ}})(1+Y^{n_{YZ}})}-\gamma \cdot Z$$ [2]

where:

$$\beta =\frac{{\beta }_Y}{{\gamma }_Y{\kappa }_{YZ}}; \kappa =\frac{{\kappa }_{XZ}}{{\kappa }_{XY}} and \gamma =\frac{{\gamma }_Z}{{\gamma }_Y}.$$

To simulate steady-state ADAR repression of APOL1, k (production) and $\gamma$ (degradation/dilution) were both set to the value one. In other words, the same amount of IFN activates ADAR and APOL1, even as the degradation–dilution rate is assumed to be the same for both gene products. We used an exhaustive algorithm to obtain β-parameters that approximate the experimental ratio of APOL1 expression observed in human podocytes (IFN stimulated; siCtrl vs. siADAR). The lowest β returned by the algorithm equaled 1.60 for the baseline condition (Fig. 2B). The ADAR knockdown condition was simulated by setting β tending to zero (β = 0.025). When β tends to zero, ADAR concentration tends to zero, and APOL1 rises to a dimensionless concentration of 0.65 at steady state (Fig. 2 B, b), approximately twofold higher than the baseline control steady state (dimensionless value = 0.32), accurately modeling the IFN, ADAR, and APOL1 interaction at steady state on primary human podocytes (Fig. 2 B, c).

We then plotted the dimensionless concentration of ADAR vs. APOL1 to visualize the relationship in both β-values (0.025 and 1.60) (Fig. 2C). APOL1 and ADAR were directly correlated until ADAR approached the dimensionless concentration of one, even though ADAR represses APOL1.

The two main characteristics of an incoherent FFL are the abilities to generate pulses and to accelerate responses (31). In Fig. 2D, we compare the response time of an unregulated gene vs. an incoherent FFL–regulated gene targeting the same steady state. The response time to half the steady-state concentration, or $T_{1/2}$, is shorter for the incoherent FFL–regulated gene at the same target concentration.

Using biological circuit principles, we mathematically modeled the interaction of IFN–APOL1–ADAR and derived characteristics of APOL1 regulation. Taken together, our incoherent FFL model indicates that APOL1’s biological circuit is capable of pulse generation and accelerated responses, advantageous characteristics that likely play an important role in APOL1 biological function and regulation.

Unedited APOL1 mRNA Binds to MDA5, Which, in Turn, Activates IFN-I.

APOL1 is only one of a large number of protein-coding genes containing two or more exonic Alu repeats capable of forming dsRNAs and therefore, subject to RNA editing (32). APOL1 is also one of the most highly edited genes in the transcriptome (33). APOL1’s AluSc and AluY repeats and their close positioning (244 nucleotides apart) favor dsRNA formation and consequently, editability (Fig. 3A). The increase in APOL1 expression caused by ADAR knockdown could reflect a combination of two potential mechanisms: 1) Transcriptome-wide unedited Alu-dsRNAs activate IFN-I via MDA5, and/or 2) unedited APOL1 Alu-dsRNA itself is recognized by MDA5 and can amplify the innate immune response via IFN-I as an mRNA species. In either case, Alu-dsRNAs would trigger IRF3 and IRF7 activation, IFN-I up-regulation, and consequent up-regulation of APOL1 expression.

Fig. 3.

APOL1 mRNA binds MDA5 and functions as an inflammatory mRNA to activate IFN-I. (A) RNA secondary structure modeling of APOL1’s 3′UTR, where AluSc and AluY are in reverse orientation. A long dsRNA structure is predicted at the Alu repeats region (sections I and II). (B) Primary human podocyte cell lines were treated with IFN-γ for 24 h prior to harvesting for RNA immunoprecipitation (RIP). RIP lysates were then incubated with anti-immunoglobulin G (IgG) or anti-ADAR antibodies and magnetic bead purified for RNA, followed by complementary DNA generation and reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR). Western blot (WB) of input and anti-ADAR antibody condition (Left). RT-qPCR of APOL1 (Right). Fold enrichment is represented as natural log (ln). (C) Primary human podocytes were transfected with either scramble siRNA or ADAR siRNA to compare RIP enrichment of APOL1 mRNA at baseline RNA editing levels (scramble) vs. low editing levels (siADAR). Western blot of input and anti-MDA5 immunoprecipitates (Left). RT-qPCR of APOL1 was performed, indicating fold enrichment (Right). (D–G) Primary human podocytes were transfected with siRNAs, treated with IFN-γ 48 h posttransfection, and then, harvested 24 h later. (D) ADAR knockdown strongly up-regulated IFN-β and (E) IRF7 mRNA expression, indicating IFN-I activation as expected. Dual knockdown of ADAR and APOL1 blunted the up-regulation of IFN-β and IRF7, indicating decreased activation of IFN-I through dsRNA recognition. (F) RT-qPCR of APOL1 and (G) ADAR demonstrates mRNA knockdown efficiency for the same experiment (D–G). Results are representative of at least three biological replicates. (H) Motif analysis of APOL1’s promoter region evidenced motifs for IRF7 (core match 0.956) and IRF3 (core match 0.994), both downstream effectors of MDA5 activation (46, 47). *P = 0.01 to 0.05; **P = 0.001 to 0.01; ***P = 0.0001 to 0.001; ****P < 0.0001; ns, not significant, P ≥ 0.05.

We used RNA immunoprecipitation (RIP) to assess whether APOL1 mRNA can indeed bind to the RNA binding proteins ADAR and MDA5. As expected, given the extensive signature of RNA editing, APOL1 mRNA showed robust enrichment in the anti-ADAR vs. anti-immunoglobulin G (IgG) control lysates under IFN-γ treatment, demonstrating direct ADAR–APOL1 mRNA interaction (Fig. 3B).

To investigate whether MDA5 interacts with APOL1, we hypothesized that APOL1 mRNA would be enriched in the ADAR knockdown RIP lysate condition since MDA5 would more likely bind unedited dsRNA. We performed RIP with anti-MDA5 and control anti-IgG antibodies under both control (scramble siRNA) and ADAR knockdown conditions. APOL1 mRNA was significantly enriched in the anti-MDA5 ADAR knockdown condition vs. the anti-MDA5 control condition (Fig. 3C). Within the ADAR knockdown condition, APOL1 mRNA was also significantly enriched when compared with anti-IgG (Fig. 3C).

After establishing that MDA5 binds to APOL1 mRNA, we investigated the quantitative importance of the APOL1 Alu-dsRNA in activating innate immunity in the setting of IFN-γ–mediated APOL1 up-regulation. We compared the effect of combined knockdown of ADAR plus APOL1 vs. ADAR knockdown alone. IFN-I activity was quantified by measuring IFN-β transcript levels because as previously described, IFN-β is up-regulated upon ADAR knockdown, indicating IFN-I activation via dsRNA recognition (Fig. 3D) (13, 34). Surprisingly, the combined knockdown of ADAR plus APOL1 showed a reduction in IFN-β expression by more than 75% compared with ADAR knockdown alone, indicating that the APOL1 mRNA itself is a major contributor to MDA5 activation leading to IFN-I induction (Fig. 3D). Levels of IRF7 mRNA, another indicator of IFN-I activation, were also significantly reduced in the ADAR plus APOL1 combined knockdown condition (Fig. 3E). In contrast to IRF7, IRF3 and TBK1 expression levels were not significantly altered by ADAR knockdown (SI Appendix, Fig. S1 K and L).

APOL1 and ADAR knockdown levels were statistically similar to unstimulated controls (Fig. 3 F and G). The results suggest that unedited APOL1 Alu-dsRNA may serve as an inflammatory mRNA, driving expression of other IFN-stimulated genes through IFN-I induction. Moreover, APOL1’s promoter region possesses transcription factor binding motifs for both IRF3 and IRF7, which potentially allow for direct APOL1 up-regulation by MDA5 activation (Fig. 3H). The results confirm direct interaction among APOL1 mRNA, ADAR, and MDA5. Moreover, the results demonstrate that APOL1’s mRNA contributes to dsRNA sensor activation (MDA5), leading to IFN-I induction in the setting of ADAR knockdown and functioning as an inflammatory mRNA.

APOL1 Undergoes Extensive A-to-I Editing at the 3′UTR Alu-dsRNA in Human Glomeruli.

To study the extent of APOL1 editing in human kidney tissue in vivo, we analyzed RNA editing of APOL1 mRNA transcripts in microdissected human glomeruli from kidney biopsies using the Nephrotic Syndrome Study Network (NEPTUNE) dataset (9). We analyzed 278 glomerular RNA-seq samples for editing in the region of the terminal exon of APOL1 (Dataset S1). To obtain editing values for each editing site, we utilized REDItools to measure the A-to-I editing ratio (inosines/total adenosines + inosines) per chromosomal coordinate. Comprehensive editing analysis of APOL1’s terminal exon revealed widespread RNA editing restricted to the Alu repeat region, which adopts the predicted secondary structure of dsRNA (Fig. 3A). We identified 138 single-nucleotide editing sites with over 1% A-to-I editing ratio in APOL1’s AluSc and AluY (Fig. 4A and Dataset S1). No editing sites were observed outside the predicted AluSc and AluY regions, including the regions encoding the G1 and G2 risk variant genotypes.

Fig. 4.

A-to-I editing is limited to the paired Alu repeats of APOL1’s 3′UTR in kidney glomeruli. (A) Box plot representing median A-to-I editing rates in 278 samples of kidney glomeruli from patients with nephrotic syndrome. The x axis shows chromosome 22 coordinates of Alu repeat regions (AluSc: Chr 22: 36,266,096 to 36,266,380 [Upper] and AluY: Chr 22: 36,266,624 to 36,266,924 [Lower]). The median editing ratio per sample per coordinate was obtained with the REDItools software (median of inosine/adenosine+inoside [total]). The results are filtered for common single nucleotide polymorphisms (SNPs), the presence of editing in other tissues at the same site, and read depth of over 10 reads per site. (B) Alu editing index (AEI) quantification by base pair change. The AEI metric calculates a weighted average of all editable adenosines over the Alu repeat regions of APOL1’s last exon, which outputs one index per sample. Adenosine to guanosine (A-to-G) represents A-to-I editing, as the sequencing machinery detects inosines as guanosines (inosines pair preferentially with cytosines). (C) APOL1 3′UTR AEI of low–risk genotype vs. high–risk genotype patients. (D) Pearson correlation between APOL1 expression and ADAR expression of the NEPTUNE dataset. (E) Pearson correlation of APOL1 expression vs. APOL1 AEI. ns, not significant, P ≥ 0.05.

Read coverage at Alu elements is lower than for neighboring regions given the repetitive nature of Alus and their abundance across the human genome. A read may thus map to multiple sites (mapping ambiguity) (33, 35, 36). To account for low coverage, we applied the Alu editing index (AEI) metric, which calculates a weighted average of A-to-I mismatches over Alu repeat regions (37). The predominant type of RNA editing was A to I, with a mean editing index per sample of 4.4 (SD ± 1.9) (Fig. 4B). AEI values were similar across high-risk and low-risk genotype samples (Fig. 4C).

We then evaluated the relationship between APOL1 expression and ADAR expression. APOL1 expression was weakly correlated with ADAR expression (r² = 0.29, P ≤ 0.001) (Fig. 4D). The APOL1 AEI values had a slightly higher correlation coefficient with APOL1 transcript abundance (r² = 0.47, P = 8.3e-16) (Fig. 4E).

The human glomerular RNA-editing data evidenced extensive A-to-I editing limited to the 3′UTR Alu-dsRNA. ADAR expression and RNA-editing index were observed to be directly correlated with APOL1, consistent with the incoherent FFL modeling.

RNA Editing in Glomeruli of an APOL1-BAC Transgenic Mouse Model Closely Replicated Human Editing at the APOL1 3′UTR.

In order to further investigate in vivo regulation of APOL1 under acute IFN stimuli, we studied the activity of APOL1 expression, RNA editing, and Adar1 expression in isolated glomeruli from an APOL1 mouse model. Our previously characterized transgenic mouse model expresses the APOL1 G0 and G2 alleles via a bacterial artificial chromosome (BAC), and therefore, it includes APOL1’s 3′UTR and noncoding regulatory regions up to 50 kilobases upstream and downstream of the APOL1 gene. Robust expression of APOL1 is achieved with IFN-γ stimuli, and the mouse model mirrors the human kidney disease with albuminuria development on G2 risk genotype mice but not on G0 mice (38). The functional domains of the human gene ADAR (The National Center for Biotechnology Information [NCBI] GeneID 103) are conserved in the mouse gene Adar1 (NCBI GeneID 56417), the only dsRNA-editing enzyme known in mice (21, 39).

RNA sequencing was performed on isolated glomeruli from APOL1-BAC transgenic mice at baseline (day 3, sham injected) and at day 3 post–IFN-γ stimuli. A total of 12 baseline controls (G0 n = 6, G2 n = 6) and 18 (G0 n = 9, G2 n = 9) IFN-stimulated mice were sequenced. Adar1 expression displayed over twofold up-regulation upon IFN-γ stimuli, replicating the amplitude of ADAR up-regulation in primary podocytes (Fig. 5A). APOL1 up-regulation was also equivalent to the human cell-based experiments (range from 15- to 40-fold up-regulation) (Fig. 5B). There were no significant APOL1 or Adar1 expression differences between G0 and G2 within baseline or within the IFN-γ condition (Fig. 5 A and B).

Fig. 5.

A-To-I editing in a transgenic APOL1-BAC mouse model closely replicates human RNA editing. (A) Normalized counts of ADAR expression in the control group (sham injected) and IFN-γ–stimulated mice. (B) Normalized counts of APOL1 comparing control and IFN-γ stimuli. (C) Venn diagram of editing sites compared across the human dataset (NEPTUNE), IFN-stimulated mice, and baseline control mice. (D) Spearman correlation of the median editing ratio between human vs. mouse matched by coordinate in (D, i) baseline and (D, ii) IFN-γ–treated conditions. Blue shaded areas represents the 95% CIs calculated with 1,000 bootstrap resamples. (E) AEI comparison between human and mouse in both control and IFN conditions. An undersampling step was performed to level the number of samples between mouse and human datasets. We used an unsupervised cluster-based method to maintain the essential characteristics of the dataset. The Mann–Whitney U test was used to calculate the difference between the two groups. (F) Comparison of median editing rates per common sites (112) between humans and baseline mice (bars represent median and interquartile). An undersample step was applied to the human dataset. Human–mouse coordinate-to-coordinate comparison was done with the Mann–Whitney U test followed by the Benjamini–Hochberg correction. Results are represented below the x axis as −ln(false discovery rate [FDR]) (natural log [ln]). The significance was set at −ln(FDR) > 3 (dashed line; FDR = 0.05).There was no statistically significant difference between human and baseline mouse median editing values. ****P < 0.0001; ns, not significant, P ≥ 0.05.

Analysis of the RNA-editing sites with REDIT tools in mouse glomeruli revealed an editing signature at the APOL1 3′UTR that mirrored that of the human glomerular RNA-seq, with 118 of the 138 human editing sites matching the corresponding human counterpart (Fig. 5C). The median human A-to-I editing ratio matched by site best correlated with the baseline condition mice (r² = 0.69) vs. IFN-γ stimulated (r² = 0.62) (Fig. 5D). We then performed AEI analysis to compare values between human and mice datasets as a global weighted average per sample of the APOL1 terminal exon. We applied an undersample step to the human dataset given the sample size difference (mice n = 30 vs. human n = 278) and used an unsupervised cluster-based method to maintain the essential characteristics of the dataset. AEI at the APOL1 3′UTR was statistically similar in human vs. baseline mice, while IFN-γ–treated mice had a significantly higher AEI value (P value < 0.001) (Fig. 5E).

Given that control (baseline, non-stimulated) mice better approximated human RNA editing according to AEI metric and RNA editing per site correlation values, the control group was used for a pairwise comparison of the editing sites by coordinate between mouse and undersample human datasets (Fig. 5F). Remarkably, all common editing sites between baseline mice and the human dataset were statistically (Mann–Whitney U test) similar after Benjamini–Hochberg correction [Fig. 5F, −log(FDR)].

The transgenic APOL1-BAC mouse model closely reproduced the human kidney editing profile, while APOL1 and Adar1 expression changes with IFN-γ were comparable with the primary podocyte experiments, further validating the model.

IFN-γ–Treated APOL1-BAC Transgenic Mice Displayed Higher A-to-I Editing Rates, and APOL1 Correlated Negatively with Adar1 Expression.

Once we established that APOL1 editing in the transgenic APOL1-BAC mouse model closely replicated the human editing signature, we focused on investigating the differences between baseline and IFN-γ condition editing among mice. RNA-editing rates on all common sites, with editing rate of at least 1%, were significantly higher in IFN-γ–treated mice as compared with baseline (Fig. 6A). We then applied the AEI metric to obtain a global APOL1 editing value to compare across conditions and genotypes. Mice treated with IFN-γ (n = 18) had significantly higher AEI than controls (n = 12) (Fig. 6B). We then split the mice by genotype, and there were no observable differences between G0 (n = 6) and G2 (n = 6) genotypes at baseline or at the IFN-γ–treated condition (G0 n = 9, G2 n = 9) (Fig. 6C).

Fig. 6.

IFN-γ–treated mice displayed globally increased A-to-I editing rates as compared with controls. APOL1 and Adar1 expressions were inversely correlated in the IFN-treated mice group in contrast to controls. (A) A-to-I editing ratios per coordinate over the AluSc and AluY paired by condition (baseline, blue; IFN stimuli; orange). Medians and interquartiles are represented on the bar plots. The significance was calculated with the Mann–Whitney U test followed by the Benjamin–Hochberg correction. −ln(FDR) (natural log [ln]) is represented on the x axis; a significant −ln(FDR) was considered above three (dashed line; FDR = 0.05). (B) AEI compared across baseline and IFN conditions. (C) AEI compared across baseline and IFN conditions further sorted by genotype subgroups. P values were calculated with the Mann–Whitney U test. (D) Spearman correlation analysis of AEI by APOL1 expression in baseline (blue) and IFN (orange) conditions. (E) Spearman correlation between Adar1 and APOL1 expression for baseline and IFN conditions. Shaded areas represent CIs of 95%. ***P = 0.0001 to 0.001; ****P < 0.0001; ns, not significant, P ≥ 0.05.

Given the striking differences in RNA editing under control and IFN-γ treatment, we asked the question of whether the correlation between APOL1, Adar1 expression, and RNA editing would change under baseline vs. IFN-γ–stimulated states.

We first characterized the correlation of APOL1 expression as normalized counts vs. the AEI metric measured over APOL1’s 3′UTR. As observed in human glomeruli, APOL1 expression was positively correlated to the AEI metric (Spearman of 0.433 baseline and 0.318 IFN; P < 0.2 and P < 0.2, respectively) (Fig. 6D). We then analyzed APOL1 vs. Adar1 expression as normalized counts. Under baseline condition, a positive correlation was observed, while under IFN-γ stimuli, the correlation was reversed at a negative value (Spearman: baseline = 0.517, P < 0.1; IFN-γ = –0.643, P < 0.01) (Fig. 6E).

The transgenic mouse model allowed us to detect differential editing between baseline and IFN-γ stimuli both globally (AEI) and per editing site (REDItools). Moreover, under an acute IFN-γ stimuli, an inverse correlation between APOL1 and Adar1 expression was observed as in cell-based experiments and further supported the incoherent FFL model circuit during an acute IFN stimuli. The baseline and IFN-γ treatment conditions revealed patterns not previously detected in the human RNA-seq glomerular dataset.

Discussion

APOL1 nephropathy likely requires the combination of a high-risk genotype and elevated APOL1 expression. The presence of inversely oriented Alu repeats in the APOL1 3′UTR prompted us to ask if the A-to-I editing enzyme, ADAR, might influence APOL1 expression. We observed that ADAR reduces APOL1 expression by limiting activation of the MDA5–MAVS–TBK1–IFN-I pathway. Moreover, we found evidence that the dsRNA structure of unedited APOL1 mRNA may itself act as a PAMP to trigger an innate immune response via MDA5 binding. We then quantified APOL1–Alu-dsRNA editing in glomeruli of patients with proteinuric kidney disease and in a transgenic APOL1-BAC mouse model, which closely reproduced editing patterns in humans. APOL1 expression was correlated with ADAR expression in mice at baseline but then flipped to a stronger inverse correlation after acute stimulation with IFN. Our results show that the IFN–ADAR–APOL1 network operates as an incoherent FFL; both APOL1 and ADAR are initially stimulated by IFN, with ADAR eventually acting to reduce total APOL1 levels by limiting the immunogenicity of the Alu-dsRNA repeats in APOL1 and other Alu-dsRNA–containing genes.

Our experimental data in primary human podocytes support the role of ADAR as a suppressor of APOL1 expression during IFN responses. We demonstrated an APOL1 regulatory pathway, in which Alu-dsRNA recognition by the PRR MDA5 activates MAVS, leading to downstream activation of IRF3 and IRF7, generation of IFN-I, and up-regulation of APOL1 (Figs. 5, 7). ADAR overexpression interrupts this cycle and decreases APOL1 levels by suppressing dsRNA activation of PRRs. ADAR knockdown led to amplification of this cycle via increased unedited Alu-dsRNA that promoted increased IFN-I activation.

Fig. 7.

APOL1 regulation via A-to-I editing and the MDA5–MAVS–TBK1–IFN-I pathway. (A) Under physiologic conditions, ADAR-mediated A-to-I editing inhibits MDA5 binding to endogenous (or self-) dsRNAs by disrupting the dsRNA structure. (B) During certain conditions of acute innate immune response, IFN-stimulated genes, including APOL1, may be rapidly up-regulated. If the ADAR-mediated editing rate does not match the resulting increase in transcribed dsRNAs, MDA5 will recognize endogenous dsRNA as PAMPs by binding to APOL1’s Alu-dsRNA and other genes harboring Alu repeats, leading to downstream activation of MAVS, TBK1, IRF3/IRF7, and IFN-I. (C) Activation of the MAVS complex is followed first by activation of the protein kinase TBK1 and subsequently, by phosphorylation of IRF3 and IRF7, which then dimerize (homo-/heteroform) and translocate into the nucleus leading to (D) IFN-I up-regulation and potential direct (E) APOL1 up-regulation. (F) APOL1 up-regulation creates an FFL, in which more unedited mRNA is available for MDA5 binding. Lastly, IFN-I up-regulates APOL1, further amplifying its signal. In summary, RNA editing suppresses APOL1 by decreasing recognition of APOL1 dsRNA and of transcriptome-wide dsRNA as foreign RNA/PAMPs, thus reducing IFN-I signaling and direct transcriptional up-regulation of APOL1 by IRF3 and IRF7 (dashed arrows). * A-to-I editing event; P = phosphorilation.

We also demonstrated, via combined knockdown of APOL1 and ADAR in the setting of IFN-γ stimulation, that the APOL1 mRNA transcript itself can contribute to the inflammatory response by stimulating IFN-I production. We demonstrated that APOL1 mRNA directly interacts with MDA5 and that this interaction is enhanced in the setting of ADAR knockdown and prevention of the A-to-I editing that normally limits the immunogenicity of endogenous dsRNA. Therefore, APOL1 mRNA appears to function as an inflammatory signaling molecule, amplifying the IFN-I response, although other dsRNA species most likely also contribute to varying degrees.

The type 1 incoherent FFL, as shown with IFN–APOL1–ADAR, is one of the most common network motifs found in nature (31). The two main characteristics of the incoherent FFLs are accelerated up-regulation and pulse generation. Both properties are advantageous for innate immune molecules with potential toxic effects if overexpressed for a prolonged period, allowing for fast up-regulation and down-regulation. The nuanced manifestations of such pulses and responses in individual patients may be one of the keys to understanding the incomplete penetrance of APOL1 nephropathy. We observed the incoherent FFL by manipulating ADAR and APOL1 in cells and then tested for evidence of the motif in APOL1 transgenic mice treated with IFN. In the basal state in mice, there was a weak positive correlation between APOL1 and ADAR expression. After treating the mice with IFN, we observed a “flip” in correlation between APOL1 and ADAR from positive to negative, reproducing the inverse relationship observed in cultured podocytes.

Our studies in human glomeruli confirmed that widespread RNA-editing events are occurring in the APOL1 3′UTR, demonstrating the biological relevance of A-to-I editing in human kidney disease. The correlation between APOL1 and ADAR in the human glomeruli was positive, similar to the APOL1 transgenic mice prior to acute IFN stimulation. The positive correlation observed in the human dataset likely reflects the chronic response to low-level inflammation rather than the acute response as seen in the mice treated with IFN. Alternatively, the human data may reflect the fact that both APOL1 and ADAR are up-regulated by low-level IFN, but there is too much heterogeneity (genetic background, disease type, inflammatory milieu, and timing of biopsy) in the human biopsy samples to observe the suppressive effect of ADAR on APOL1 levels in the chronic inflammatory state.

The transgenic mouse model incorporating noncoding APOL1 regulatory elements served as an important tool to study RNA editing and APOL1 transcript regulation in vivo. The transgenic APOL1-BAC mouse model allowed us to investigate unique differences in the RNA-editing landscape at baseline and the acute inflammatory state. Higher editing rates were observed on day 3 IFN-γ–treated mice in over 98% of editing sites. The editing signature in common sites (112 of 138 human sites) was highly similar to humans, providing an excellent model for future studies of posttranscriptional modification and editing of APOL1.

Our data demonstrating APOL1 3′UTR Alu-dsRNA recognition by MDA5 complement the findings of Okamoto et al. (40) that APOL1 mRNA is also recognized by the dsRNA-activated protein kinase R (PKR). In that instance, however, it was the APOL1 coding region plus the early 3′UTR (not including the Alu-dsRNA) that provoked a dsRNA inflammatory response. The data suggest that APOL1 mRNA may play multiple important but complex roles in APOL1 biology beyond those recognized to date both directly and secondarily through alteration of APOL1 protein levels. While the data of Okamoto et al. (40) showed differences in PKR activation that differed based on the APOL1 genotype, the sequence we examined is shared by all APOL1 genotypes. Moreover, we did not observe any evidence of RNA editing in the protein coding region, including at the G1 and G2 loci forming a dsRNA, likely due to the short extension of the dsRNA formed at this region. Understanding the variable penetrance of APOL1 kidney disease among individuals with the high-risk genotype will require not comparison of low– vs. high–disease risk genotypes but rather further elucidation of the regulation of APOL1 gene expression.

In another study addressing APOL1 regulation utilizing human AB8/13 podocytes, Davis et al. (41) found that nucleosome-associated double-stranded DNA (nsDNA) fragments were implicated in activating APOL1 via DNA sensor activation. Cyclic guanosine monophosphate-adenosine monophosphate (GMP-AMP) synthase and IFN-inducible protein 16, both important DNA sensors, triggered the expression of APOL1 and IFN-β through STING and IRF3 along with the secondary IFN-I receptor IFNAR and STAT1 signaling, further increasing APOL1 expression. Although cross-talk between dsRNA-sensing and nsDNA-sensing pathways via STING has been proposed, the authors found that RIG-I–like receptors were not likely to be involved in APOL1 activation through the STING pathway. Further experiments under ADAR knockdown conditions need to be performed to explore the cross-talk hypothesis.

It is interesting to speculate how the regulatory elements we observe might operate. IFNs have a powerful effect on both APOL1 expression and on the pathogenesis of APOL1 nephropathy. APOL1 likely evolved as an innate immune protein component of the antiviral response. The presence of Alu-dsRNA may have provided an evolutionary advantage that can lead to faster and higher-amplitude APOL1 expression response and, more importantly, faster down-regulation to avoid toxicity. In addition, the multiple IFN pathways associated with APOL1 and their redundancy (PKR, nsDNA–STING–IRF3 fragments, and now, dsRNA–MDA5–IFN-I) suggest not only the importance of APOL1 to the immune system but also a broader role for APOL1 as an innate immune molecule.

Our data on the role of ADAR-mediated RNA editing in the etiology of APOL1 kidney disease expose gaps in our understanding of APOL1 transcriptional regulation and inspire questions for investigation. Although our human glomerular sample represents a rich source of data, it is not yet large enough to make strong correlations between editing and specific glomerular phenotypes or to understand the more complex relationships between APOL1 risk genotype, editing, inflammatory state, and disease phenotype. The transgenic APOL1-BAC mouse model offers the opportunity to combine mechanistic questions and in vivo relevance, with the caveat that neither APOL1 nor Alu repeats are natively present in mice. ADAR’s role in APOL1-associated kidney disease likely lies in regulating innate immunity in order to decrease APOL1 peak levels as well as steady-state concentration. Excessive innate immune activation or lack of suppression may be the “second hit” needed to trigger disease in high–risk genotype patients. Therapeutic interventions could target the incoherent FFL to reduce APOL1 expression during an inflammatory response by either increasing the expression of ADAR (earlier and higher amplitude) or inhibiting the downstream activators of the dsRNA pathway, such as MDA5, MAVS, or TBK1. Future studies will benefit from access to glomerular RNA-seq data from healthy kidneys and from kidneys of patients with extrarenal systemic inflammatory disease. These efforts will help us further understand how APOL1 is regulated and by extension, whether an individual at risk for APOL1 nephropathy does or does not progress to overt disease.

Materials and Methods

Cell Culture.

Primary human podocytes (Celprogen) were cultured in proprietary podocyte culture media (Celprogen) containing 10% FBS in a 5% carbon dioxide/95% air-humidified atmosphere at 37 °C. Cells of passages 2 to 5 were trypsinized, passaged with 0.05% trypsin–EDTA, and plated on six-well plates. Cells were allowed to grow for 24 h to a confluency of 60 to 80%, and then, they were transfected with siRNAs. After 48 h of transfection, the cells were stimulated with IFN-γ (R&D Systems) at 10 ng/mL, harvested after 24 h, lysed in lysis buffer, and then, prepared for electrophoresis (SDS-PAGE) and immunoblotting or RNA extraction.

siRNA Transfection.

Pooled siRNAs comprising four sequences (siGENOME-SMARTpool ON-TARGETplus; Dharmacon) were transfected at 25 nM per well in the presence of 5 μL/well of RNAiMAx in a six-well plate format (Invitrogen by Thermo Fisher Scientific). Media were changed 24 h posttransfection.

Immunoblotting.

Immunoblotting was performed as previously described (42). Homogenized cell lysates were separated by SDS-PAGE and transferred to polyvinylidene difluoride membranes (Biorad). The membranes were then blocked with 3% bovine serum albumin in Tris buffered saline and incubated with primary antibodies followed by horseradish peroxidase–conjugated secondary antibodies (Santa Cruz). Membranes were developed with Super Signal West Dura (Thermo Fisher Scientific) and imaged with a FluorChem E imager (ProteinSimple; mouse monoclonal anti-APOL1 antibody [catalog no. AMAB90532; Sigma Milipore], rabbit monoclonal anti-MDA5 [D74E4; catalog no. 5321; Cell Signaling Technology], and mouse monoclonal anti-ADAR [catalog no. sc-73408; Santa Cruz]).

RNA Isolation and Real-Time qPCR.

RNA was isolated per the Qiagen RNeasy Kit protocol (Qiagen). Complementary DNA synthesis was performed with SuperScript IV VILO (Invitrogen by Thermo Fisher Scientific). Real-time qPCR was performed with the QuantStudio 6 Flex platform and the TaqMan Universal PCR Master Mix (Applied Biosystems by Thermo Fisher Scientific). The TaqMan gene expression assay for β-actin mRNA was used for normalization. Relative expression was calculated via the 2^–ΔΔCT method.

ADAR Overexpression.

ADAR overexpression plasmids pmGFP-ADAR1-p110 (Addgene plasmid 117928) and pmGFP-ADAR1-p150 (Addgene plasmid 117927) were transfected into primary human podocytes (Celprogen) with Lipofectamine 3000 on six-well plates. An EV plasmid was generated by introduction of an early stop codon downstream of pmGFP in the plasmid pmGFP-ADAR1-p110, so that GFP would be expressed in the absence of ADAR expression (Phusion site mutagenesis kit; Thermo Fisher Scientific). Transfection efficiency was verified with fluorescent microscopy. IFN-γ added 24 h posttransfection (the control group underwent simple media change at the same time point), and cells were harvested at the 48-h time point for downstream RNA or protein isolation.

I-FFL Modeling.

The mathematical model (Eqs. 1 and 2) is based on the Hill equation with dimensionless variables. The nondimensionalization reduces the number of parameters that vary independently. The k constant is the required amount of X (IFN) to stimulate Y (ADAR) and Z (APOL1). The variable $\gamma$ is the degradation–dilution rate of Y and Z, which affects both terms of Eq. 2. $n_{XY}$ is the steepness of the Hill function for Y activated by X; it defines how quickly the concentration of Y reaches the steady-state value. The second term of Eq. 1 represents the dilution due to the cell growth and the degradation of Y. $n_{xz}$ is the steepness of the Hill function for Z activated by X, while $n_{yz}$ is the steepness of the Hill function for Z repressed by Y.

Eq. 1 defines how the concentration of Y varies over time (t) determined by its rate of production (k) and of degradation–dilution ($\gamma$). In Eq. 1, β is the steady-state value of Y. In Eq. 2, the rate expression of Z is defined by its production and degradation–dilution ($\gamma$). In contrast to Y production, which depends only on the concentration of X (Eq. 1), Z depends on concentrations of both X and Y. In this case, a higher Y value lowers the concentration of Z.

The Biocircuits and SciPy packages were used to solve the equations (43, 44). Matplotlib was used to plot the results (45).

Human RNA-seq Data.

Participants in the NEPTUNE were recruited and consented at the time of their first clinically indicated biopsy for suspicion of primary proteinuric glomerular disease (9). Biopsy-based diagnoses included minimal change disease, FSGS, membranous nephropathy, and other glomerulopathies. An extra biopsy core designated for research was manually dissected and isolated into tubulointerstitial and glomerular compartments per the established protocol. Manual glomerular microdissection captured glomeruli with an open Bowman’s space. The glomerular expression studies were thus disproportionately enriched in transcriptomes of functioning glomeruli over those of globally sclerosed glomeruli. Glomerular RNA-seq data from NEPTUNE were used both to quantitate A-to-I editing at the APOL1 3′UTR and to measure the expression of APOL1 mRNA.

A-to-I Editing.

RNA sequencing reads in the form of FASTQ files underwent quality control, trimming of adaptors (trimgalore/0.4.5), and they were mapped to the reference genome GRCh38 using Spliced Transcript Alignment to a Reference (STAR) aligner (star/2.7.3a). We focused the analysis to reads mapping to the terminal APOL1 exon (Chr 22: 36,265,151 to 36,267,611), which includes the 3′UTR and Alu repeats. REDItools 2.0 was used to identify editing sites in mapped BAM files. To account for editing levels and to collate with REDIportal (http://srv00.recas.ba.infn.it/atlas/index.html), we adapted the python script getOverallEditing.py available on REDItools scripts (https://github.com/BioinfoUNIBA/QEdit/blob/master/scripts/getOverallEditing.py) to calculate the editing level by the chromosome coordinate for each sample (source code: https://github.com/guilhermetabordaribas/ADAR_Regulates_APOL1_via_A2I_RNA_Editing/blob/main/codes_for_analytics.ipynb). For all analysis based on REDItools editing rate measurements, we considered edited positions supported by a read depth of at least 10 reads and greater or equal than 1% of the editing level.

To account for differences in read coverage between samples, we used the Alu Editing Indexer AEI tool (37). The AEI metric is defined by the ratio of A-to-G mismatches to the total coverage of adenosines over Alu repeats, where the majority of the A-to-I editing occurs, providing a weighted average of the editing across a sample or chromosomal region: in this case, APOL1’s last exon. The resulting AEI value per sample was then used in the correlation analysis with gene expression (i.e., APOL1, ADAR) and editing level calculation between the bases A to C, A to G, A to T, C to A, C to G, and C to G.

Statistical Analysis.

Gene expression was quantified with HTSeq, followed by the variance-stabilizing transformation from DESeq2. We next corrected batch effects with Combat and quantile-normalized expression. A-to-I editing levels between samples with and without the APOL1 high-risk genotype were compared by Wilcoxon test. Pearson correlations were used to compare AEI with gene expression.

Human podocyte data are shown as mean ± SD of independent replicates (n ≥ 3) and were analyzed with ANOVA using GraphPad Prism version 8 (GraphPad Software). A P value of <0.05 was considered statistically significant.

Transgenic APOL1-BAC Mouse Model Experiments.

Transgenic APOL1 mice were housed and cared for using our protocol as previously described (38). Mice were selected for RNA-seq based upon a variety of factors. We attempted to select an equal number of male and female mice for each genotype between 6 and 13 wk old with a low day 3 albumin/creatinine ratio (<500 μg/mg) measured using previously described methods. Glomeruli tissue was harvested from mice 3 d after administration of the IFN-γ plasmid as previously described (38). RNA isolation was performed on extracted glomeruli using the Qiagen RNeasy kit (Qiagen). RNA was sent to Azenta for QC and next-generation RNA sequencing on the Illumina HiSeq Series with paired-end 150–base pair reads, with an average of 25 million reads per sample. The deconvoluted reads were returned for analysis.

Mouse RNA-seq Data Analysis.

The adapter sequences were trimmed using trim-galore version 0.6.6, and the trimmed reads were aligned using the STAR aligner version 2.7.9. After alignment, the 3′UTR was truncated from each bam file using samtools version 1.13 for analysis with REDItools and the AEI as described above to match the exact human chromosomal coordinates used with the NEPTUNE dataset. Coverage was calculated using the samtools version 1.13 coverage tool. Gene counts were obtained from the nontruncated bam files using featureCounts version 2.0.0 of the Subread package and normalized using the DESeq2 package in R version 4.1.2. A Shapiro–Wilk test for normality was performed before using the Spearman method for correlation in R. The AEI index was fit to normalized APOL1 counts using the QR decomposition method in R, and the ADAR-normalized counts were fit to APOL1-normalized counts using the same method.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information. Mouse sequence reads can be found on the NCBI BioProject under accession # PRJNA887438 (48). The data for NEPTUNE (human editing data) is in three studies uploaded to NCBI GEO with the following accession numbers: GSE68127, GSE68126, and GSE68125 at https://www.ncbi.nlm.nih.gov/gds/?term=apol1%20sampson (49).