Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization study.

Chronic kidney disease (CKD) and atrial fibrillation are both major burdens on the health care system worldwide. Several observational studies have reported clinical associations between CKD and atrial fibrillation; however, causal relationships between these conditions remain to be elucidated due to possible bias by confounders and reverse causations. Here, we conducted bidirectional two-sample Mendelian randomization analyses using publicly available summary statistics of genome-wide association studies (the CKDGen consortium and the UK Biobank) to investigate causal associations between CKD and atrial fibrillation/flutter in the European population. Our study suggested a causal effect of the risk of atrial fibrillation/flutter on the decrease in serum creatinine-based estimated glomerular filtration rate (eGFR) and revealed a causal effect of the risk of atrial fibrillation/flutter on the risk of CKD (odds ratio, 9.39 per doubling odds ratio of atrial fibrillation/flutter; 95% coefficient interval, 2.39-37.0; P = 0.001), while the causal effect of the decrease in eGFR on the risk of atrial fibrillation/flutter was unlikely. However, careful interpretation and further studies are warranted, as the underlying mechanisms remain unknown. Further, our sample size was relatively small and selection bias was possible.

Introduction

Chronic kidney disease (CKD) is a major global burden, and 1.5% of the total deaths worldwide were attributed to CKD in 2012 according to the World Health Organization [1]. CKD is principally caused by diabetes, hypertension, and glomerulonephritis. The comorbidities of CKD include anemia, bone disease, cancer, and cardiovascular diseases [1]. Atrial fibrillation (AF) is the most common arrhythmia and is a major burden on the health care system worldwide, as this condition can cause ischemic stroke and cardiac dysfunction [2]. Several observational studies have reported clinical associations between CKD and AF [3-8]. However, the causal relationship between CKD and AF remains to be elucidated as traditional observational studies lacking randomization designs are typically prone to bias due to various factors including confounders and reverse causations [9].

Mendelian randomization (MR) is an epidemiological method that mimics the design of randomized controlled studies using single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs) and is used to examine the causal effects of a risk factor on the outcome of interest. As genetic variants such as SNPs are randomly assigned at conception according to Mendel’s law, MR studies are not influenced by confounders or reverse causations and can overcome the limitations of observational studies [9]. In this study, we conducted bidirectional two-sample MR analyses using publicly available summary statistics of genome-wide association studies (GWASs) to investigate causal associations between the risk of atrial fibrillation/flutter (AF/F) and the change in serum creatinine-based estimated glomerular filtration rate (eGFR) or the risk of CKD for the first time in the European population.

Methods

Study design and data sources

We performed bidirectional two-sample MR analyses that included (1) an MR analysis estimating the causal effect of the risk of AF/F (binary data) on the change in eGFR (continuous data), (2) an MR analysis estimating the causal effect of the risk of AF/F (binary data) on the risk of CKD (binary data), and (3) an MR analysis estimating the reverse causal effect of the change in eGFR (continuous data) on the risk of AF/F (binary data). All analyses were conducted using the TwoSampleMR package (version 0.5.5) in R software (version 4.0.3) [10]. A P-value below 0.017 (0.05/3 by Bonferroni correction) was considered statistically significant and a P-value between 0.017 and 0.05 was considered suggestively significant in the three MR analyses. A P-value below 0.05 was considered statistically significant in the MR-PRESSO (Mendelian Randomization Pleiotropy RESidual Sum and Outlier) global test and outlier test.

For eGFR and CKD datasets, summary statistics were available from the GWAS meta-analysis performed by the CKD Genetics (CKDGen) consortium [11]. The dataset for eGFR used continuous data of log (eGFR) and included 567,460 participants of European ancestry [11]. Serum creatinine assays were described in the GWAS study [11]. GFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration equation on adults (> 18 years of age) and the Schwartz formula on individuals who were 18 years old or younger, respectively [11]. The dataset of CKD (defined as eGFR < 60 ml/min/1.73m2) used binary data of log odds ratio (OR) and included 41,395 cases and 439,303 controls of European ancestry [11]. These two datasets are publicly available from the “Wuttke et al. 2019 publication files” uploaded in the CKDGen Meta-Analysis Data as the file names “20171017_MW_eGFR_overall_EA_nstud42.dbgap” and “CKD_overall_EA_JW_20180223_nstud23.dbgap”, respectively [12].

For the AF dataset, summary statistics were available from the largest GWAS meta-analysis published by Nielsen et al. that included 60,620 AF cases and 970,216 controls of European ancestry [13]. However, Nielsen’s GWAS meta-analysis included the deCODE study with 13,471 AF cases and 358,161 controls. In contrast, the CKDGen GWAS meta-analysis also included the deCODE study with 15,939 CKD cases and 192,362 controls [11]. If we use Nielsen’s GWAS meta-analysis, at most 22.2% (13,471 out of 60,620) of the AF cases may overlap with participants in the CKDGen GWAS meta-analysis. Sample overlap in cases between the exposure and outcome datasets can lead to substantial bias in the causal estimate of MR studies in the direction of both the null and the observational association [14]. Therefore, we obtained another dataset from a GWAS meta-analysis in UK Biobank performed by MRC IEU [15] that used binary data and included 5,669 AF/F cases of International Classification of Diseases (ICD)-10 code I48 and 457,341 controls in the European population. This dataset was publicly available from the MRC IEU Open GWAS database [16] and from MR-Base [17], as GWAS-ID of “ukb-b-964.” For example, GWAS datasets in the UK Biobank by MRC IEU (“ukb-b-19953” for body mass index [BMI] and “ukb-b-223” for smoking) were also used in another MR analysis [18]. As the CKDGen GWAS meta-analysis did not include UK Biobank participants [19], there was no apparent sample overlap between the exposure and outcome datasets.

As all data used in the present study were derived from publicly available summary-level GWAS datasets and no individual-level data were used, additional ethical approval and patient consent were not necessary.

Selection of instrumental variables

In the MR analysis, SNPs from the exposure dataset were used as IVs. IVs must satisfy the following three assumptions: the IVs are associated with the exposure (IV assumption 1); the IVs affect the outcome only via exposure (IV assumption 2); the IVs are not associated with measured or unmeasured confounders (IV assumption 3) [20].

For the causal effect of the exposure on the outcome, the SNPs were selected from the exposure GWAS summary data as IVs by clumping together all SNPs that were associated with the exposure trait at a genome-wide significance threshold (P < 5.0×10−8) and were not in linkage disequilibrium (LD) (r2 < 0.001, and distance > 10,000 kb) with the other SNPs. Moreover, the bidirectional MR analysis depends on an assumption that the SNPs used as IVs do not overlap or are not in LD between the exposure and the outcome [9, 21]. When SNPs overlapped or were in LD, the SNPs (if any existed) were excluded from the MR analysis [22]. The summary statistics of each SNP were extracted from both the exposure and outcome datasets and then harmonized. When an exposure SNP was not available in the outcome dataset, we used a proxy SNP (if any existed) with high LD (r2 > 0.8) in combination with the exposure SNP. Palindromic SNPs exhibiting an intermediate minor allele frequency > 0.42 were excluded from the analyses [20].

To evaluate the strength of the exposure IVs, we calculated the F-statistic of each SNP using the following formula: F-statistic = R2×(N-2)/(1-R2), where R2 is the variance of the phenotype explained by each genetic variant in exposure, and N is the sample size. R2 was calculated using the following formula: R2 = 2×(Beta)2×EAF×(1-EAF)/[2×(Beta)2×EAF×(1-EAF) + 2×(SE)2×N×EAF×(1-EAF)], where Beta is the per-allele effect size of the association between each SNP and phenotype, EAF is the effect allele frequency, and SE is the standard error of Beta [23]. IVs with an F-statistic below 10 were considered as weak instruments [24]. Moreover, we used the MR Steiger filtering method that was implemented in the TwoSampleMR package to infer the causal direction of each SNP on the hypothesized exposure and outcome [25]. If a SNP has a causal effect of the exposure on the outcome, the SNP used as an IV should be more predictive of the exposure than the outcome [26]. When a SNP was more predictive of the outcome than the exposure, the SNP (if any existed) was excluded from the MR study as a sensitivity analysis [26, 27].

Two-sample Mendelian randomization

Wald ratio estimates the causal effect for each IV, and this value was calculated as the ratio of Beta for the corresponding SNP in the outcome dataset divided by Beta for the same SNP in the exposure dataset [20]. Our main approach was to conduct a meta-analysis of each Wald ratio according to the inverse variance weighted (IVW) method to estimate the overall causal effect of the exposure on the outcome. For the IVW method, we used a multiplicative random-effects model when Cochran’s Q statistic (as described below) was significant (P < 0.05) [28]. Otherwise, a fixed-effects model was used. Additionally, we conducted sensitivity analyses using the weighted median method, the MR-Egger regression method, the weighted mode method, the MR-PRESSO global and outlier tests, and leave-one-out sensitivity analysis. The weighted median method provides a valid causal estimate when more than half of the instrumental SNPs satisfy the IV assumptions [29]. The MR-Egger regression method is used to assess horizontal pleiotropy of IVs. When IV assumption 2 is violated, horizontal pleiotropy occurs, and the MR-Egger regression intercept is non-zero with statistical significance [29]. The weighted mode method forms clusters of individual SNPs and estimates the causal effect from the largest cluster [29]. The MR-PRESSO global and outlier tests investigate if there are outlier SNPs that possess variant-specific causal estimates that differ substantially from those of other SNPs [30]. Leave-one-out sensitivity analysis was conducted to assess the reliability of the IVW method by removing each SNP from the analysis and re-estimating the causal effect [30]. Heterogeneity was also measured among the causal estimates across all SNPs in the IVW method by calculating Cochran’s Q statistic and the corresponding P-value. Low heterogeneity provides more reliability for causal effects [31]. Moreover, among all the SNPs used as IVs for the exposure datasets, we searched for SNPs associated with P < 5.0×10−8 with possible pleiotropic effects on other diseases and traits using the web-tool PhenoScanner (version 2) [32, 33].

Results

Estimating the causal effect of the risk of AF/F on the change in eGFR

First, we investigated the causal effect of the risk of AF/F on changes in eGFR. The exposure dataset for AF/F included 20 instrumental SNPs with rs651386 excluded by LD clumping. As the 20 SNPs were all identified for the outcome GWAS datasets of eGFR, we used no proxy SNPs. The characteristics of the 20 SNPs are listed in S1 Table in S1 File. The F-statistic of every instrument was > 30, thus indicating that there was no weak instrument bias. In bidirectional two-sample MR study, two sets of instrumental SNPs for both traits should not be in LD with each other [9, 21]. We confirmed that none of the 20 SNPs used as IVs for the AF/F dataset overlapped or were in LD with the 144 SNPs used as IVs for the eGFR dataset (see S2 Table in S1 File). For harmonization, rs7853195 was excluded as it was palindromic (“palindromic” was “TRUE” in S1 Table in S1 File) with intermediate allele frequencies (“ambiguous” was “TRUE” in S1 Table in S1 File). All the remaining 19 SNPs were more predictive of the exposure (the risk of AF/F) than the outcome (the change in eGFR) (“Steiger direction” was “TRUE” in S1 Table in S1 File). The MR results are shown in Table 1, Figs 1 and S1. The IVW method using a multiplicative random-effects model suggested that the risk of AF/F may have decreased eGFR (beta for log [eGFR] per log OR of AF/F, -0.0721; SE, 0.0585) [19]; however, the effect was not significant (P = 0.23). The results of both the weighted median and weighted mode methods revealed consistent results with suggestive significance. The MR-Egger intercept indicated little evidence of horizontal pleiotropy. MR-PRESSO global and outlier tests indicated that rs796427 was an outlier SNP (P < 0.0038), as suggested by the funnel plot (S2 Fig). When we excluded rs796427 from the IVW method using a fixed-effects model, the risk of AF/F decreased eGFR with statistical significance (Table 1). Here, we used a fixed-effects model for the IVW method, as Cochran’s Q statistic for the IVW method indicated low heterogeneity after we excluded rs796427 (Cochran’s Q statistic, 24.3; P = 0.11).

Table 1

MR results of the effects of AF/F on the change of eGFRcr and the risk of CKD.

Exposure traits	Outcome traits	Number of SNPs	IVW method	Weighted median method	MR-Egger regression method		Weighted mode method	Heterogeneity (IVW)	MR-PRESSO global test	Outlier-corrected IVW
			Beta	Beta	Beta	Intercept	Beta	Cochran’s Q		Beta
			(SE)	(SE)	(SE)	(SE)	(SE)			(SE)
			P-value	P-value	P-value	P-value	P-value	P-value	P-value	P-value
AF/F	eGFRcr	19	-0.0721	-0.121	-0.137	0.000185	-0.134	43.6		-0.0955
			(0.0585)	(0.0570)	(0.130)	(0.000328)	(0.0584)			(0.038)
			0.23	0.035	0.31	0.58	0.034	< 0.001	0.002	0.012
AF/F	CKD	19	3.23	3.86	4.62	-0.00397	3.84	23.0		Not Available
			(1.01)	(1.45)	(2.51)	(0.00638)	(1.63)
			0.001	0.008	0.084	0.54	0.03	0.19	0.22

Abbreviations: AF/F, atrial fibrillation/flutter; CKD, chronic kidney disease; eGFR, serum creatinine-based estimated glomerular filtration rate; IVW, inverse variance weighted; MR, Mendelian randomization; SE, standard error; SNPs, single nucleotide polymorphisms.

Fig 1

Scatter plot for estimating the risk of AF/F on the change in eGFR.

Each black point representing a SNP is plotted in relation to the effect size of the SNP on the exposure (x-axis) and on the outcome (y-axis) with corresponding standard error bars. The slope of each line corresponds to the causal estimate using IVW (light blue), weighted median (light green), MR-Egger regression (blue), and weighted mode (green) method.

Estimating the causal effect of the risk of AF/F on CKD risk

Next, we investigated the causal effect of the risk of AF/F on the risk of CKD, as the MR analyses indicated a causal effect of the risk of AF/F on the decrease in eGFR when an outlier SNP was excluded. The exposure dataset for AF/F was the same as described above (S1 Table in S1 File). All 19 SNPs were also identified in the outcome GWAS datasets of CKD, and were more predictive of the exposure (the risk of AF/F) than the outcome (the risk of CKD) (“Steiger direction” was “TRUE” in S1 Table in S1 File). The MR results are shown in Table 1, Figs 2 and S1. The IVW method using a fixed-effects model revealed that the risk of AF/F was significantly associated with a higher risk of CKD (OR of CKD per log OR of AF/F, 25.3; 95% coefficient interval [CI], 3.51–183.0; P = 0.001) [19], thus suggesting that the OR of CKD was 9.39 per doubling OR of AF/F (95% CI, 2.39–37.0). For interpretation purposes, we multiplied beta by log (2) (= 0.693) and then exponentiated this value [34]. Other MR methods also provided consistent results although the weighted mode method indicated only suggestive significance (Table 1). Leave-one-out sensitivity analysis demonstrated that the significance disappeared when rs6843082 was excluded from the IVW method (S3 Fig); however, the MR-PRESSO global test revealed that rs6843082 was not an outlier (Table 1). Cochran’s Q statistic for the IVW method indicated low heterogeneity and reliability of the causal effect. PhenoScanner identified four SNPs associated with possible pleiotropic effects on other diseases and traits as shown in S1 Table in S1 File (rs35176054 on height, rs6843082 on cardioembolic stroke and ischemic stroke, rs796427 on hand grip strength, arm impedance, and years of educational attainment, and rs879324 on cardioembolic stroke and ischemic stroke). We excluded these four SNPs from the IVW method using a fixed-effects model, and then we obtained a comparable result to that of the original IVW method (OR of CKD per log OR of AF/F, 33.2; 95% CI, 2.07–532.2; P = 0.013). Here, we used a fixed-effects model for the IVW method, as Cochran’s Q statistic for the IVW method indicated low heterogeneity after we excluded the four SNPs (Cochran’s Q statistic, 17.1; P = 0.25).

Fig 2

Scatter plot for estimating the causal effect of the risk of AF/F on the risk of CKD.

Estimating the reverse causal effect of the change in eGFR on the risk of AF/F

Finally, we investigated the reverse causal effect of the change in eGFR on the risk of AF/F. The exposure dataset of eGFR included 308 instrumental SNPs. A total of 158 SNPs were excluded by LD clumping, and nine SNPs were not identified for the outcome GWAS datasets for AF/F. However, three SNPs were detected as proxy SNPs (rs140124 for rs131263, rs2293579 for rs61897431, and rs147726416 for rs75625374, as shown in S2 Table in S1 File). As a result, 144 SNPs in the exposure dataset of eGFR were used as IVs. The characteristics of the 144 SNPs are listed in S2 Table in S1 File. The F-statistic of every instrument was > 28, thus indicating that there was no weak instrument bias. During harmonization, five SNPs (rs10865189, rs154656, rs55929207, rs8096658, and rs8474) were excluded due to the observation that they were palindromic (“palindromic” was “TRUE” in S2 Table in S1 File) with intermediate allele frequencies (“ambiguous” was “TRUE” in S2 Table in S1 File). All the remaining 139 SNPs were more predictive of the exposure (the change in eGFR) than the outcome (the risk of AF/F) (“Steiger direction” was “TRUE” in S2 Table in S1 File). The MR results are shown in Table 2, Figs 3 and S1. The IVW method using a multiplicative random-effect model revealed that the change in eGFR was not significantly associated with the risk of AF/F (OR of AF/F per unit change in log[eGFR], 0.996; 95% CI, 0.980–1.013; P = 0.66) [11]. None of the other methods showed any causal effects (Table 2). As the MR-PRESSO global and outlier tests indicated that two SNPs were outliers (P < 0.0278 for rs3775932, P < 0.0278 for rs4656220), as was suggested by the funnel plot (S2 Fig), we excluded them from the IVW method using a multiplicative random-effect model. The result remained insignificant (OR, 0.99967; 95% CI, 0.985–1.015; P = 0,97). Moreover, PhenoScanner identified 89 SNPs associated with possible pleiotropic effects on other diseases and traits among 144 SNPs as shown in S2 Table in S1 File. We excluded all 89 SNPs from the IVW method using a multiplicative random-effect model. The result remained insignificant (OR, 1.005; 95% CI, 0.980–1,031; P = 0.69). Here, we used a random-effects model for the IVW method, as Cochran’s Q statistic for the IVW method indicated high heterogeneity after we excluded the 89 SNPs (Cochran’s Q statistic, 100.4; P = 0.015).

Table 2

MR results of the effect of the change of eGFRcr on the risk of AF/F.

Exposure traits	Outlier traits	Number of SNPs	IVW method	Weighted median method	MR-Egger regression method		Weighted mode method	Heterogeneity (IVW)	MR-PRESSO global test	Outlier-corrected IVW
			Beta	Beta	Beta	Intercept	Beta	Cochran’s Q		Beta
			(SE)	(SE)	(SE)	(SE)	(SE)			(SE)
			P-value	P-value	P-value	P-value	P-value	P-value	P-value	P-value
eGFRcr	AF/F	139	-0.0038	-0.000766	0.0142	-0.0000699	-0.00393	264.8		-0.00033
			(0.00852)	(0.00996)	(0.0215)	(0.0000763)	(0.0135)			(0.00769)
			0.66	0.94	0.51	0.36	0.77	< 0.001	< 0.001	0.97

Abbreviations: AF/F, atrial fibrillation/flutter; CKD, chronic kidney disease; eGFR, serum creatinine-based estimated glomerular filtration rate; IVW, inverse variance weighted; MR, Mendelian randomization; SE, standard error; SNPs, single nucleotide polymorphisms.

Fig 3

Scatter plot for estimating the causal effect of the change in eGFR on the risk of AF/F.

Discussion

To the best of our knowledge, this is the first MR analysis to estimate causal associations between the risk of AF/F and the change in eGFR or the risk of CKD, and we made two novel discoveries in the European population. First, our study suggested a causal effect of the risk of AF/F on the risk of CKD. Second, the causal effect of the change in eGFR on the risk of AF/F was unlikely. However, careful attention must be paid to interpreting these results as our sample size of the AF/F dataset was relatively small.

Several observational studies have reported a higher prevalence of AF in a dose-dependent manner as kidney function decreased [3-8]. For example, the cross-sectional CRIC study in the US reported that the OR of prevalent AF was 1.35 (95% CI, 1.13–1.62) in subjects with eGFR < 45 ml/min/1.73m2 compared to those with eGFR > 45 ml/min/1.73m2 [3]. The cross-sectional REGARDS study in the US reported an OR of AF that was 2.86 (95% CI, 1.38–5.92) in CKD Stage 4–5 compared to no CKD [4]. However, few observational studies have reported a causal effect of the prevalence of AF on the risk of CKD. The prospective Niigata preventive medicine study in Japan reported that among subjects without hypertension or diabetes during a mean follow-up of 5.9 years, the development of kidney dysfunction was 16.6 incidences per 1000 person-years (95% CI, 13.0–20.2) in subjects with baseline AF, and only 5.2 (95% CI, 5.0–5.3) in those without [5]. Our MR results were consistent with those of the Niigata preventive medicine study regarding the causal association of AF prevalence with kidney dysfunction. However, our MR study did not support the causal effect of kidney dysfunction on AF prevalence that was previously reported by several observational studies. Although our sample size of the AF/F dataset was relatively small, this discrepancy may be partly due to the knowledge that observational studies lacking randomization designs are generally prone to bias resulting from various factors including confounders and reverse causations [9]. The possibility of confounding the association between AF and CKD is always present in observational studies [35]. Consistent with our MR results suggesting a lack of causal effect of the eGFR decrease on the risk of AF/F, an MR study revealed that the urine albumin adjusted for creatinine as a proxy for kidney function did not exert a significant causal effect on the outcome of AF (beta, 0.105; 95% CI, -0.064–0.274; P = 0.23) [36].

The mechanisms by which AF/F may cause kidney dysfunction remain unknown. Several risk factors are shared between AF and CKD, including elevated inflammation and an activated renin-angiotensin-aldosterone system (RAAS) [35]. Kidney dysfunction can occur when AF/F causes systemic inflammation and RAAS activation, but the reverse may also be true. As a possible mechanism, a decrease in cardiac output due to AF/F may cause pre-renal failure leading to chronic kidney dysfunction. Additionally, thromboembolism due to AF/F may cause renal artery occlusion [5]. Our search using PhenoScanner determined that two of our exposure SNPs of AF/F (rs6843082 and rs879324) also effected cardioembolic stroke traits (P < 5.0×10−8), which is one of the AF/F comorbidities [37]. This supports the idea that AF/F could cause kidney dysfunction via thromboembolism. However, further studies are required to elucidate the precise mechanisms involved.

Our MR estimate scale of OR of CKD per doubling OR of AF/F may be too large and the corresponding 95% CI width was very wide (OR, 9.39; 95%CI, 1.99–44.2). One possible reason for this observation was the relatively small sample size of the AF/F dataset that was one of the major limitations of this study, as we could not use the largest GWAS meta-analysis due to substantial sample overlap. In general, the genetic instruments used in MR studies can estimate the lifetime effect, and this may explain the larger estimates compared to observational studies [38]. We believe that given that the primary aim of our MR study was to assess if the exposure had a causal effect on the outcome, estimating the size of the causal effect was less important [30]. On the other hand, we could not detect, if any existed, the causal effect of the change in eGFR on the risk of AF/F probably because the scale of the OR was very small. The F-statistic of every SNP > 28 indicated that there was no weak instrument bias (S2 Table in S1 File).

Selection bias was also one of the major limitations of this study. Our MR Steiger filtering method inferred the causal direction of all 19 SNPs used as IVs for the AF/F datasets on the exposure (the risk of AF/F) and outcome (the risk of CKD). However, our instrumental SNPs for the risk of AF/F were selected from the same GAWS dataset as used for subsequent analyses, as is often the case with typical MR studies. We did not use three-sample MR design (three non-overlapping GWASs: selection dataset, exposure dataset, outcome dataset) in the present study [39]. Then, our instrumental SNPs could not be regarded as random samples because they were selected at a genome-wide significant threshold (P < 5.0×10−8) [40]. The double use of the same sample for SNP/IV selection and estimation was subject to selection bias and horizontal pleiotropy that could invalid the causality, resulting in inflated type 1 error rates and excessive false positives in the MR Steiger method [21, 40, 41]. Therefore, careful interpretation of the causality is warranted in our MR study.

There are other limitations in our study. The causal effect of the risk of AF/F on the decrease in eGFR was indicated only when an outlier SNP was excluded. The GWAS dataset “ukb-b-964” with ICD-10 code I48 included cases with atrial flutter in addition to AF, although so did the DiscovEHR study that was included in Nielsen’s GWAS [13]. Our analysis was based on populations of European ancestry, and the findings are unlikely generalized to other populations. Conversely, the lack of possible sample overlap between the exposure and outcome datasets was a strength of our study that allowed us to avoid substantial bias.

In conclusion, our MR analyses suggested a causal effect of the risk of AF/F on the decrease in eGFR and revealed a causal effect of the risk of AF/F on the risk of CKD. Conversely, the reverse causal effect of the decrease of eGFR on the risk of AF/F was unlikely. However, careful interpretation and further studies are warranted, as the sample size was relatively small and selection bias was possible.

STREGA reporting recommendations, extended from STROBE statement.

(DOC)

Click here for additional data file.

Forrest plots.

(a) Forrest plot for estimating the risk of AF/F on the change in eGFR. (b) Forrest plot for estimating the causal effect of the risk of AF/F on the risk of CKD. (c) Forrest plot for estimating the causal effect of the change in eGFR on the risk of AF/F. Each black point represents the causal estimate of each SNP on the outcome per increase in the exposure, and red points show the combined causal estimates using IVW and MR-Egger regression methods with horizontal lines denoting 95% confidence intervals.

(PPTX)

Click here for additional data file.

Funnel plots.

(a) Funnel plot for estimating the risk of AF/F on the change in eGFR. (b) Funnel plot for estimating the causal effect of the risk of AF/F on the risk of CKD. (c) Funnel plot for estimating the causal effect of the change in eGFR on the risk of AF/F. Each black point representing an SNP is plotted in relation to the estimate of the exposure on the outcome (x-axis) and the inverse of the standard error (y- axis). Vertical lines show the combined causal estimates using IVW (light blue) and MR- Egger regression (blue) methods.

(PPTX)

Click here for additional data file.

Leave-one-out sensitivity analyses.

(a) Leave-one-out sensitivity analysis for estimating the risk of AF/F on the change in eGFR. (b) Leave-one- out sensitivity analysis for estimating the causal effect of the risk of AF/F on the risk of CKD. (c) Leave-one-out sensitivity analysis for estimating the causal effect of the change in eGFR on the risk of AF/F. Each black point represents the combined causal estimates on the outcome per increase in the exposure using IVW methods with horizontal lines denoting 95% confidence intervals after removing the corresponding SNP from the analysis.

(PPTX)

Click here for additional data file.

Contains all the supporting tables.

(XLSX)

Click here for additional data file.

23 Jun 2021

PONE-D-21-11069

Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample Mendelian randomization study

PLOS ONE

Dear Dr. Yoshikawa,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

After external review, both reviewers raised some concerns about the methodology. Please respond to each of the comments and revise accordingly.

Please submit your revised manuscript by 22 Aug 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Jie V Zhao

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

3. Please include captions for *all* your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: No

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This research studies the causal effect of atrial fibrillation/flutter on chronic kidney

disease. It is a standard application of two-sample Mendelian randomization method implemented in the R package TwoSampleMR.

The manuscript is well written. The description is clear and well laid out.

My main concern is on the way the IV SNPs are selected. A recent study has established that using the same data that are used to select the IVs in the subsequent analysis can lead to biased results (Wang and Han, 2021). Any comments on how your results are affected by SNPs selection?

Furthermore, in the original paper by Bowden et al (2015) and an extension by Sue and Pan (2020), the selection of IV SNPs are such that they are significant in both the exposure GWAS AND the trait GWAS. However, your IV SNPs seem to be selected based only on the exposure GWAS, not on the trait GWAS. Am I correct? If I am, do you have any comments?

Some minor comments:

1. In lines 118-119, the R^2 is expressed as a fraction. The term "EAFx(1-EAF)" appears in both the numerator and the denominator. Shouldn't term be cancelled out?

2. The figures in the text and the appendices are difficult to read.

References:

Wang, K. and Han, S., 2021. Effect of selection bias on two sample summary data based Mendelian randomization. Scientific reports, 11(1), pp.1-8.

Bowden, J., Davey Smith, G. and Burgess, S., 2015. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International journal of epidemiology, 44(2), pp.512-525.

Xue, H. and Pan, W., 2020. Inferring causal direction between two traits in the presence of horizontal pleiotropy with GWAS summary data. PLoS genetics, 16(11), p.e1009105.

Reviewer #2: Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample Mendelian randomization study

By Masahiro Yoshikawa and Kensuke Asaba

Reviewer Comments

The Authors conducted a Mendelian Randomization (MR) study to assess causality of atrial fibrillation/flutter (AF/F) and estimated glomerular filtration rate (eGFR)/chronic kidney disease (CKD), both directions. While state-of-the-art methods are essentially applied I have serious issues about methods and their presentations. Without their resolution this presentation could be misunderstood and incorrect procedure promoted. Still, the results might be valid but to ensure this up to a certain point proper methodology needs to be applied/reported. In the following I distinguish major and minor points:

Major points:

1 Data source: CKDGen GWAS on eGFR

In this large meta-analysis, contributing studies derived residuals for log(eGFR) which then were used as outcome in the GWAS. It is incorrect to assume that this trait has SD of 1. This assumption was already incorrectly made by another study on which the Authors rely. Please revise.

2 Palindromic SNPs

Authors define palindromic SNPs with minor allele frequency (MAF) >0.42 and excluded thus some of the otherwise eligible SNPs. However, the Authors overlooked that a palindromic SNPs is also one with an effect allele frequency (EAF) – as presented in Supplementary Tables 1 and 2 - of, for example, 0.55. Since there are several SNPs with an EAF in the range of 0.5 to 0.58, exclusion of SNPs is incomplete. The analysis should be thus revised.

3 Estimation of R² and F-statistics

Especially regarding R², I am not familiar with this presentation of formula. The reference provided by the Authors is also only an application that in turn point to a reference of Palmer et al 2012. Tracing that paper, however, does not provide the explanation on the formula. While the formula might be correct (I know presentations that look similar to a certain degree), it would be good to see who this formula proposed and how it was derived.

The main reason for my search was that a proper R² incorporates the variance of the outcome. In case such as in a 2-sample MR study the variance is often not available why usually some simplifying assumption such as SD=1 is applied. Since this is mostly likely not true for eGFR (see my first comment) and since I cannot deduce any other approximation of the variance of the outcome in the presented formula, I am wondering how reliable the estimates of R² and thus of the F-statistics are.

As – in the worst case – the strengths of instruments could be overestimated, this aspect needs clarification and discussion.

4 Post-hoc power calculations

As discussed by many authors/statisticians (e.g. Dziak et al 2020, PMID: 32523323), post-hoc power calculations as presented in this paper are not valid. This part needs to be deleted and the topic differently approached.

Minor points:

1 Since both directions are assessed and the standard phrasing is that of a “bidirectional” MR study it should be stated in the title as well as in the introduction.

2 Please provide phenotype definitions as used in the underlying studies.

3 Please add further information on SNPs in Supplementary Tables 1 and 2, e.g., marking palindromic SNPs, outliers and otherwise noticeable SNPs as well as adding Phenoscanner information.

4 Please do Phenoscanner look up for all SNPs. Please be cautious in your wording of “unrelated traits”. Sometimes relation may not be so obvious.

5 In presentation of results on the evaluation of the causal relationship of AF/F on eGFR, the Authors state that they used a multiplicative random effects model because of observed heterogeneity. Since the results on IVW estimate is presented above, this statement needs clarification if the presented result is already from multiplicative random effects model, also in contrast to all other presented results from IVW analysis. Was this the only instance?

6 The Authors state in the Discussion that estimates may be too large and CIs were too wide and these present a major limitation of the study. However, the reported results present maybe a consequence of some limitations but do not present a limitation themselves. Please revise.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

20 Aug 2021

Dear Dr. Jie V Zhao,

We would like to sincerely thank you very much for the time and effort you have dedicated to providing insightful feedback on ways to strengthen our manuscript entitled “Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample (bidirectional) Mendelian randomization study” under consideration for publication in PLOS ONE.

First, we would like to sincerely apologize for our major mistake and to correct it.

Among 178 SNPs used as IVs for the change of eGFR in the original version of Supplementary Table 2, a total of 34 SNPs were not associated with eGFR at a genome-wide significance threshold (P = 5E-08) and they should have been excluded from the subsequent analyses (for example, rs10197255, P = 4.7E-07 in the original version of Supplementary Table 2).

To satisfy IV assumption 1 (the IVs are associated with the exposure), we should have selected SNPs as IVs that were associated with the exposure trait with P < 5.0×10-8 from the exposure GWAS summary data.

This error was due to the default setting of the clump_data function in the TwoSampleMR package (https://mrcieu.github.io/TwoSampleMR/reference/clump_data.html).

We originally selected the SNPs used as IVs for the change of eGFR from Supplementary Table 4 (ST4) in the GWAS study by CKDgen (Please refer to Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/).

Some of the 308 SNPs listed in ST4 were not associated with eGFR at a genome-wide significance threshold (that is, P > 5.0×10-8) based on populations of European ancestry.

And the default setting of the clump_data() function was as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 1,

clump_p2 = 1,

pop = "EUR"

)

Therefore, the 34 wrong SNPs with P > 5.0×10-8 were not excluded from our study by LD clumping.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the original version of the Result session.)

Now, we have changed the setting of the clump_data() function as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 5E-08,

clump_p2 = 1,

pop = "EUR"

)

And we have excluded a total of 158 SNPs by LD clumping including the 34 wrong SNPs.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the revised version of the Result session.)

Although nine SNPs have not been identified for the outcome GWAS datasets for AF/F, three SNPs were detected as proxy SNPs.

Then, we have used a total of 144 SNPs for estimating the causal effect of the change in eGFR on the risk of AF/F. (144 SNPs = 178 SNPs - 34 wrong SNPs.)

The 144 SNPs have been listed correctly in the revised version of Supplementary Table 2.

Fortunately, the result of causality has been unchanged (the change in eGFR was not significantly associated with the risk of AF/F) though odds ratio, 95% CI, and P-value have been slightly changed.

Incorrect: OR, 0.997; 95% CI, 0.981-1.012; P = 0.67.

Correct: OR, 0.996; 95% CI, 0.980-1.013; P = 0.66.

(Please see lines 246-247 in the revised manuscript.)

Incorrect: OR, 1.00081; 95% CI, 0.987-1.015; P = 0.91.

Correct: OR, 0.99967; 95% CI, 0.985-1.015; P = 0.97.

(Please see lines 251-252 in the revised manuscript.)

Please also refer to Table 2 in the revised manuscript.

Second, we would like to correct one more mistake.

IV assumption 1 was not "the IVs are associated with the outcome" but "the IVs are associated with the exposure".

(Please see line 105 in the revised manuscript.)

The errata above are also described in the response to each Reviewer.

Third, we would like to invite Dr. Tomohiro Nakayama to be a co-author of our paper.

He contributed substantially to our work by interpreting data, revising the manuscript, and reviewing it critically during this revision process.

Finally, our responses to each Reviewer’s comment are described below in a point-to-point manner.

In addition, we have submitted the revised manuscript as well as the document file with tracked changes to highlight the revisions in red color.

We hope that our manuscript will be acceptable for publication in PLOS ONE.

Sincerely,

Masahiro Yoshikawa, M.D., Ph.D.

Corresponding author.

Address; Division of Laboratory Medicine, Department of Pathology and Microbiology, Nihon University School of Medicine, Oyaguchi-kamicho 30-1, Itabashi, Tokyo, Japan.

Phone; 81-3-3972-8111

e-mail: myosh-tky@umin.ac.jp

To Reviewer #1

We would like to sincerely thank you very much for the time and effort you have dedicated to providing insightful feedback on ways to strengthen our manuscript entitled “Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample (bidirectional) Mendelian randomization study” under consideration for publication in PLOS ONE.

First, we would like to sincerely apologize for our major mistake and to correct it.

Among 178 SNPs used as IVs for the change of eGFR in the original version of Supplementary Table 2, a total of 34 SNPs were not associated with eGFR at a genome-wide significance threshold (P = 5E-08) and they should have been excluded from the subsequent analyses (for example, rs10197255, P = 4.7E-07 in the original version of Supplementary Table 2).

To satisfy IV assumption 1 (the IVs are associated with the exposure), we should have selected SNPs as IVs that were associated with the exposure trait with P < 5.0×10-8 from the exposure GWAS summary data.

This error was due to the default setting of the clump_data function in the TwoSampleMR package (https://mrcieu.github.io/TwoSampleMR/reference/clump_data.html).

We originally selected the SNPs used as IVs for the change of eGFR from Supplementary Table 4 (ST4) in the GWAS study by CKDgen (Please refer to Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/).

Some of the 308 SNPs listed in ST4 were not associated with eGFR at a genome-wide significance threshold (that is, P > 5.0×10-8) based on populations of European ancestry.

And the default setting of the clump_data() function was as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 1,

clump_p2 = 1,

pop = "EUR"

)

Therefore, the 34 wrong SNPs with P > 5.0×10-8 were not excluded from our study by LD clumping.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the original version of the Result session.)

Now, we have changed the setting of the clump_data() function as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 5E-08,

clump_p2 = 1,

pop = "EUR"

)

And we have excluded a total of 158 SNPs by LD clumping including the 34 wrong SNPs.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the revised version of the Result session.)

Although nine SNPs have not been identified for the outcome GWAS datasets for AF/F, three SNPs were detected as proxy SNPs.

Then, we have used a total of 144 SNPs for estimating the causal effect of the change in eGFR on the risk of AF/F. (144 SNPs = 178 SNPs - 34 wrong SNPs.)

The 144 SNPs have been listed correctly in the revised version of Supplementary Table 2.

Fortunately, the result of causality has been unchanged (the change in eGFR was not significantly associated with the risk of AF/F) though odds ratio, 95% CI, and P-value have been slightly changed.

Incorrect: OR, 0.997; 95% CI, 0.981-1.012; P = 0.67.

Correct: OR, 0.996; 95% CI, 0.980-1.013; P = 0.66.

(Please see lines 246-247 in the revised manuscript.)

Incorrect: OR, 1.00081; 95% CI, 0.987-1.015; P = 0.91.

Correct: OR, 0.99967; 95% CI, 0.985-1.015; P = 0.97.

(Please see lines 251-252 in the revised manuscript.)

Please also refer to Table 2 in the revised manuscript.

Second, we would like to correct one more mistake.

IV assumption 1 was not "the IVs are associated with the outcome" but "the IVs are associated with the exposure".

(Please see line 105 in the revised manuscript.)

Our responses to the Reviewer’s comments

#1. My main concern is on the way the IV SNPs are selected. A recent study has established that using the same data that are used to select the IVs in the subsequent analysis can lead to biased results (Wang and Han, 2021). Any comments on how your results are affected by SNPs selection?

Thank you very much for your suggestion.

We have carefully read the study by Wang and Han. They considered the two-sample MR Steiger method in their study and referred to the study by Hemani et al [26].

Hemani et al. introduced the method using the R programming language and implemented it in the MR-Base [26].

Some studies [27, 28] performed the MR Steiger filtering method with the reference to the study by Hemani et al [26].

(https://mrcieu.github.io/TwoSampleMR/reference/steiger_filtering.html).

Therefore, we have performed the MR Steiger filtering method using steiger_filtering function in the TwoSampleMR package to infer the causal direction of each SNP on the hypothesized exposure and outcome.

As a result, all 19 SNPs used as IVs for the risk of AF/F were more predictive of the exposure (the risk of AF/F) than the outcome (the change in eGFR) (“Steiger direction” was “TRUE” in the revised version of Supplementary Table 1).

(Please also see lines 127-133, 174-176, 207-208, and 242-244 in the revised manuscript.)

However, as Reviewer pointed out, using the same data as used to select the IVs in the subsequent analysis can lead to biased results.

Our instrumental SNPs for the risk of AF/F were selected from the same GAWS dataset as used for subsequent analyses. Then, our instrumental SNPs could not be regarded as random samples because they were selected at a genome-wide significant threshold (P < 5.0×10-8) [41].

This could cause selection bias and invalid the causality, resulting in inflated type 1 error rates and excessive false positives in the MR Steiger method.

Therefore, selection bias is one of major limitations, and careful interpretation of the causality is warranted in our study.

We have reflected these comments, in addition to the next comments, to the revised manuscript. Please refer to the next point #2.

(References)

26. Hemani G, Tilling K, Davey Smith G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. PLoS Genet. 2017;13: e1007081.

27. Treur JL, Demontis D, Smith GD, Sallis H, Richardson TG, Wiers RW, et al. Investigating causality between liability to ADHD and substance use, and liability to substance use and ADHD risk, using Mendelian randomization. Addict Biol. 2021;26: e12849.

28. Zheng J, Brion MJ, Kemp JP, Warrington NM, Borges MC, Hemani G, et al. The Effect of Plasma Lipids and Lipid-Lowering Interventions on Bone Mineral Density: A Mendelian Randomization Study. J Bone Miner Res. 2020;35: 1224-1235.

41. Wang K, Han S. Effect of selection bias on two sample summary data based Mendelian randomization. Sci Rep. 2021;11: 7585.

#2. Furthermore, in the original paper by Bowden et al (2015) and an extension by Sue and Pan (2020), the selection of IV SNPs are such that they are significant in both the exposure GWAS AND the trait GWAS. However, your IV SNPs seem to be selected based only on the exposure GWAS, not on the trait GWAS. Am I correct? If I am, do you have any comments?

Thank you very much for providing these insights, and as Reviewer pointed out, our instrumental SNPs were selected based only on the exposure GWAS.

As Xue and Pan described in their study [22], bi-directional MR works depends on one critical, most often unknown, assumption: two sets of valid SNPs/IVs used in the two directions. That is, if an SNP primarily influences X, but influences Y only through X, then it should be used as an IV to infer X → Y in the first step; at the same time, it cannot be used as an IV to infer Y → X.

Similarly, as Davey Smith G and Hemani G [21] described,

“In bi-directional MR, if trait A causes trait B, then the instrument, ZA, will be associated with both A and B.

However, a second instrument specific to trait B, ZB, will be associated with trait B, and not with trait A.

This method is only valid on the condition that the two instruments are not marginally associated with each other (e.g. there is no LD between instruments for A and B).”

Then, a study [23] excluded SNPs used as IVs for one trait that were in LD with the significant SNPs for the other trait by referring to the study by Davey Smith G and Hemani G [22].

Therefore, we checked whether our instrumental SNPs for the risk of AF/F overlapped or were in LD with those for the change in eGFR, and confirmed that there were no overlap or LD between the two sets of SNPs/IVs.

(Please see lines 112-115 and 169-172 in the revised manuscript.)

However, as Reviewer pointed out, our instrumental SNPs for the risk of AF/F were selected from the same GAWS dataset as used for subsequent analyses. We did not use three-sample MR design (three non-overlapping GWAS: selection dataset, exposure dataset, outcome dataset) [40].

As Xue and Pan described [22], the double use of the same sample for SNP/IV selection and estimation can cause possible selection bias due to wide-spread horizontal pleiotropy. Similarly, Bowden et al. [42] described that, if genetic variants are chosen due to their association with the exposure in the dataset under analysis, then the association with the exposure is likely to be overestimated, and the association with the outcome could also then be overestimated due to confounding.

Therefore, also in our study, this double use of the same sample for SNP/IV selection and estimation was subject to selection bias and horizontal pleiotropy that could invalid the causality, resulting in inflated type 1 error rates and excessive false positives.

We have reflected these comments, in addition to the previous comments (please refer to the previous point #1), to the revised manuscript as follows.

Selection bias was also one of the major limitations of this study. Our MR Steiger filtering method inferred the causal direction of all 19 SNP used as IVs for the AF/F datasets on the exposure (the risk of AF/F) and outcome (the risk of CKD). However, our instrumental SNPs for the risk of AF/F were selected from the same GAWS dataset as used for subsequent analyses, as is often the case with typical MR studies. We did not use three-sample MR design (three non-overlapping GWAS: selection dataset, exposure dataset, outcome dataset) in the present study [40]. Then, our instrumental SNPs could not be regarded as random samples because they were selected at a genome-wide significant threshold (P < 5.0×10-8) [41]. The double use of the same sample for SNP/IV selection and estimation was subject to selection bias and horizontal pleiotropy that could invalid the causality, resulting in inflated type 1 error rates and excessive false positives in the MR Steiger method [22, 41, 42]. Therefore, careful interpretation of the causality is warranted in our MR study.

(Please see lines 319-331 in the revised manuscript.)

Please also refer to lines 31 and 344 in the revised manuscript.

(References)

21. Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet. 2014;23: R89-98.

22. Xue H, Pan W. Inferring causal direction between two traits in the presence of horizontal pleiotropy with GWAS summary data. PLoS Genet. 2020;16: e1009105.

23. Wang K, Ding L, Yang C, Hao X, Wang C. Exploring the Relationship Between Psychiatric Traits and the Risk of Mouth Ulcers Using Bi-Directional Mendelian Randomization. Front Genet. 2020;11: 608630.

40. Zhao Q, Chen Y, Wang J, Small DS. Powerful three-sample genome-wide design and robust statistical inference in summary-data Mendelian randomization. Int J Epidemiol. 2019;48: 1478–1492.

41. Wang K, Han S. Effect of selection bias on two sample summary data based Mendelian randomization. Sci Rep. 2021;11: 7585.

42. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44: 512-25.

Some minor comments:

#3. In lines 118-119, the R^2 is expressed as a fraction. The term "EAFx(1-EAF)" appears in both the numerator and the denominator. Shouldn't term be cancelled out?

We agree with Reviewer on this point.

In fact, we cancelled out 2×EAF×(1-EAF) when we calculated R2.

As Reviewer #2 pointed out, our reference was inappropriate. This formula was originally derived from the study by Sim H et al. (Please refer to S1 Text in PLoS One. 2015;10: e0120758. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4405269/)

#4. The figures in the text and the appendices are difficult to read.

We would like to apologize for any inconvenience we have caused. We have made Figure 1 with a resolution of 600 dpi. Moreover, we have extended Supplementary Figures 1c and 3c longitudinally as much as we can.

To Reviewer #2

We would like to sincerely thank you very much for the time and effort you have dedicated to providing insightful feedback on ways to strengthen our manuscript entitled “Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample (bidirectional) Mendelian randomization study” under consideration for publication in PLOS ONE.

First, we would like to sincerely apologize for our major mistake and to correct it.

Among 178 SNPs used as IVs for the change of eGFR in the original version of Supplementary Table 2, a total of 34 SNPs were not associated with eGFR at a genome-wide significance threshold (P = 5E-08) and they should have been excluded from the subsequent analyses (for example, rs10197255, P = 4.7E-07 in the original version of Supplementary Table 2).

To satisfy IV assumption 1 (the IVs are associated with the exposure), we should have selected SNPs as IVs that were associated with the exposure trait with P < 5.0×10-8 from the exposure GWAS summary data.

This error was due to the default setting of the clump_data function in the TwoSampleMR package (https://mrcieu.github.io/TwoSampleMR/reference/clump_data.html).

We originally selected the SNPs used as IVs for the change of eGFR from Supplementary Table 4 (ST4) in the GWAS study by CKDgen (Please refer to Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/).

Some of the 308 SNPs listed in ST4 were not associated with eGFR at a genome-wide significance threshold (that is, P > 5.0×10-8) based on populations of European ancestry.

And the default setting of the clump_data() function was as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 1,

clump_p2 = 1,

pop = "EUR"

)

Therefore, the 34 wrong SNPs with P > 5.0×10-8 were not excluded from our study by LD clumping.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the original version of the Result session.)

Now, we have changed the setting of the clump_data() function as follows:

clump_data(dat,

clump_kb = 10000,

clump_r2 = 0.001,

clump_p1 = 5E-08,

clump_p2 = 1,

pop = "EUR"

)

And we have excluded a total of 158 SNPs by LD clumping including the 34 wrong SNPs.

(Please refer to the paragraph of "Estimating the reverse causal effect of the change in eGFR on the risk of AF/F" in the revised version of the Result session.)

Although nine SNPs have not been identified for the outcome GWAS datasets for AF/F, three SNPs were detected as proxy SNPs.

Then, we have used a total of 144 SNPs for estimating the causal effect of the change in eGFR on the risk of AF/F. (144 SNPs = 178 SNPs - 34 wrong SNPs.)

The 144 SNPs have been listed correctly in the revised version of Supplementary Table 2.

Fortunately, the result of causality has been unchanged (the change in eGFR was not significantly associated with the risk of AF/F) though odds ratio, 95% CI, and P-value have been slightly changed.

Incorrect: OR, 0.997; 95% CI, 0.981-1.012; P = 0.67.

Correct: OR, 0.996; 95% CI, 0.980-1.013; P = 0.66.

(Please see lines 246-247 in the revised manuscript.)

Incorrect: OR, 1.00081; 95% CI, 0.987-1.015; P = 0.91.

Correct: OR, 0.99967; 95% CI, 0.985-1.015; P = 0.97.

(Please see lines 251-252 in the revised manuscript.)

Please also refer to Table 2 in the revised manuscript.

Second, we would like to correct one more mistake.

IV assumption 1 was not "the IVs are associated with the outcome" but "the IVs are associated with the exposure".

(Please see line 105 in the revised manuscript.)

Our responses to the Reviewer’s comments

Major points:

#1. Data source: CKDGen GWAS on eGFR

In this large meta-analysis, contributing studies derived residuals for log(eGFR) which then were used as outcome in the GWAS. It is incorrect to assume that this trait has SD of 1. This assumption was already incorrectly made by another study on which the Authors rely. Please revise.

We would like to apologize for our mistake and thank you very much for pointing it out.

As described in the CKDgen GWAS study that we referred to for the eGFR dataset, "exp(β) can be interpreted as the OR for the disease per unit change in log(eGFR)".

(Please see the Methods section, the paragraph entitled "Genetic risk score analysis in the UK Biobank dataset" in Nat Genet. 2019;51: 957-972,

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/).

Therefore, we corrected the effect size of the eGFR change on the AF/F risk as follows.

Incorrect: OR of AF/F per 1-SD higher log[eGFR]

Correct: OR of AF/F per unit change in log[eGFR]

Please see line 247 in the revised manuscript.

#2. Palindromic SNPs

Authors define palindromic SNPs with minor allele frequency (MAF) >0.42 and excluded thus some of the otherwise eligible SNPs. However, the Authors overlooked that a palindromic SNPs is also one with an effect allele frequency (EAF) – as presented in Supplementary Tables 1 and 2 - of, for example, 0.55. Since there are several SNPs with an EAF in the range of 0.5 to 0.58, exclusion of SNPs is incomplete. The analysis should be thus revised.

Please accept our apologies. We did not explain it clearly.

Supplementary Tables 1 and 2 originally included palindromic SNPs that were excluded from the subsequent analysis.

We have marked all the excluded palindromic SNPs "TRUE" in the "palindromic ambiguous" columns in the revised version of Supplementary Tables 1 and 2.

Please also see lines 173-174, and 242 in the revised manuscript.

#3. Estimation of R² and F-statistics

Especially regarding R², I am not familiar with this presentation of formula. The reference provided by the Authors is also only an application that in turn point to a reference of Palmer et al 2012. Tracing that paper, however, does not provide the explanation on the formula. While the formula might be correct (I know presentations that look similar to a certain degree), it would be good to see who this formula proposed and how it was derived.

The main reason for my search was that a proper R² incorporates the variance of the outcome. In case such as in a 2-sample MR study the variance is often not available why usually some simplifying assumption such as SD=1 is applied. Since this is mostly likely not true for eGFR (see my first comment) and since I cannot deduce any other approximation of the variance of the outcome in the presented formula, I am wondering how reliable the estimates of R² and thus of the F-statistics are.

As – in the worst case – the strengths of instruments could be overestimated, this aspect needs clarification and discussion.

We sincerely apologize that our reference was inappropriate.

As Reviewer pointed out, the variance of the sex- and age-adjusted log(eGFR) residuals was assumed to be 0.016 (not = 1) as described in the CKDgen GWAS study (Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/).

However, we are sorry but we were not sure of the variance of the risk of AF/F.

Therefore, we could not use the formula (1) but used the formula (2) to estimate R² as follows:

(1); R² = (Beta)²×2×(EAF)×(1-EAF)/variance.

(2); R² = 2×(Beta)²×EAF×(1-EAF)/[2×(Beta)²×EAF×(1-EAF) + 2×(SE)²×N×EAF×(1-EAF)].

The formula (2) that we used to estimate R² for each SNP was originally derived from the study by Sim H et al. (Please refer to S1 Text in PLoS One. 2015; 10(4): e0120758. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4405269/).

For example, using the formula (2), R² of rs10159261 (the first row and the "O" column in the revised version of Supplementary Table 2) for the exposure (the change of eGFR) (Beta = -0.003774, SE = 0.00038, EAF = 0.31, N = 525153) was calculated as follows.

R² = 2×(-0.003774)²×0.31×(1-0.31)/[2×(-0.0038)²×0.31×(1-0.31) + 2×(0.00038)²×525153×0.31×(1-0.31)] = 0.0001877885.

R² of the other SNPs for the exposure datasets were also calculated using the formula (2) in the revised version of Supplementary Table 1 and 2.

On the other hand, we performed the MR Steiger filtering method using steiger_filtering function in the TwoSampleMR package

(https://mrcieu.github.io/TwoSampleMR/reference/steiger_filtering.html) to infer the causal direction of each SNP on the hypothesized exposure and outcome.

(Please see lines 127-133, lines 174-176, lines 207-208, and lines 242-244 in the revised manuscript.)

Please also refer to point #1 for Reviewer #1.

The steiger_filtering function automatically calculated R² for the exposure and outcome dataset.

When the SNPs for the change of eGFR were used as the exposure, the results of steiger_filtering function were as follows (e.g. the first 10 SNPs were shown):

SNP rsq.exposure

1 rs10159261 1.878319e-04

2 rs10254101 5.305158e-04

3 rs1028455 5.570221e-05

4 rs1047891 5.453993e-04

5 rs1055256 9.377984e-05

6 rs10821905 1.343213e-04

7 rs10846157 1.206184e-04

8 rs10857788 1.129092e-04

9 rs10865189 9.680874e-05

10 rs11062167 2.595056e-04

Although we are not sure how the steiger_filtering function estimated R² values, the R² values estimated by the steiger_filtering function were almost the same as those by the formula (2) as shown in the revised version of Supplementary Table 2 (for example, R² of rs10159261 by the formula (2) and by steiger_filtering function were 0.0001877885 and 1.878319e-04, respectively, as shown above).

Therefore, we suppose that our R² estimation by the formula (2) and then F-statistics did not differ from the true values very much, and that our MR study was unlikely to suffer from weak instrument bias.

Please see lines 238-239 and 317-318 in the revised manuscript.

#4. Post-hoc power calculations

As discussed by many authors/statisticians (e.g. Dziak et al 2020, PMID: 32523323), post-hoc power calculations as presented in this paper are not valid. This part needs to be deleted and the topic differently approached.

We quite agree with Reviewer on this point.

As Reviewer pointed out, we should have run the power calculations prior to conducting the analyses.

We have deleted them totally (including Table 3) from our study in the revised manuscript.

We suppose that our MR study was unlikely to suffer from weak instrument bias although our sample size of the AF/F dataset was relatively small.

In the MR study estimating the causal effect of the change in eGFR on the risk of AF/F, we suppose that we could not detect the causal effect, if any existed, because the scale of the OR (0.996) was very small.

We have reflected these comments to the revised manuscript as follows.

On the other hand, we could not detect, if any existed, the causal effect of the change in eGFR on the risk of AF/F probably because the scale of the OR was very small. The F-statistic of every SNP > 28 indicated that there was no weak instrument bias (S2 Table).

(Please see lines 315-318 in the revised manuscript.)

Minor points:

#1. Since both directions are assessed and the standard phrasing is that of a “bidirectional” MR study it should be stated in the title as well as in the introduction.

We agree with Reviewer on this point.

We have made it clear that our MR study is bidirectional thorough the manuscript including the title.

(Please see lines 3, 20, 50, and 58 in the revised manuscript.)

#2. Please provide phenotype definitions as used in the underlying studies.

Thank you very much for your suggestion.

We have added phenotype definition that was described in the CKDgen GWAS (Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/) as follows:

Serum creatinine assays were described in the GWAS study. GFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration equation on adults (> 18 years of age) and the Schwartz formula on individuals who were 18 years old or younger, respectively.

(Please see lines 72-75 in the revised manuscript.)

The definition of CKD (defined as eGFR < 60 ml/min/1.73m2) and AF/F (ICD-10 code I48) was originally described in the manuscript.

(Please see lines 75-76 and 92-93 in the revised manuscript.)

#3. Please add further information on SNPs in Supplementary Tables 1 and 2, e.g., marking palindromic SNPs, outliers and otherwise noticeable SNPs as well as adding Phenoscanner information.

Thank you very much for your suggestion.

We added the further information about the SNPs as well as “Steiger direction” in the revised version of Supplementary Table 1 and 2 as follows:

“Palindromic ambiguous”, the SNP was excluded if “TRUE.” (Please also see lines 173-174, and 242 in the revised manuscript);

“MR-PRESSO outlier”, outlier SNP was marked “Yes.”;

“Sample size”;

“Possible pleiotropic effects of the SNP for the exposure on other diseases and traits”;

“Steiger direction”, if “TRUE”, the SNP used as IV for the exposure was more predictive of the exposure than the outcome.

(Please also refer to the point #1 for Reviewer #1 as for “Steiger direction”).

#4. Please do Phenoscanner look up for all SNPs. Please be cautious in your wording of “unrelated traits”. Sometimes relation may not be so obvious.

We would like to appreciate this important suggestion.

We rephrased “unrelated to” as follows:

Incorrect: SNPs associated with P < 5.0×10-8 with pleiotropic effects on additional traits unrelated to AF/F.

Correct: SNPs associated with P < 5.0×10-8 with possible pleiotropic effects on other diseases and traits.

(Please see lines 158, 220-221, and 253 in the revised manuscript.)

Then, we have performed PhenoScanner serrch for 144 instrumental SNPs for the change in eGFR as well as re-performed for 20 instrumental SNPs for the risk of AF/F.

(Please see lines 157-159 in the revised manuscript.)

We found four SNPs for the AF/F dataset and 89 SNPs for eGFR dataset.

The results were shown in the revised version of Supplementary Table 1 and 2.

When we excluded the four SNPs from the IVW method to estimate the causal effect of the risk of AF/F on CKD risk, we obtained a comparable result to that of the original IVW method (OR of CKD per log OR of AF/F, 25.3 vs 33.2; 95% CI, 3.51-183.0 vs 2.07-532.2; P = 0.001 vs 0.013).

(Please see lines 220-226 in the revised manuscript.)

Please also see lines 301-304 in the revised manuscript.

Similarly, when we excluded the 89 SNPs from the IVW method to estimate the causal effect of the change in eGFR on the AF/F risk, we obtained a comparable result to that of the original IVW method (OR of CKD per log OR of AF/F, 0.996 vs 1.005; 95% CI, 0.980-1.013 vs 0.980-1,031; P = 0.66 vs 0.69).

(Please see lines 252-256 in the revised manuscript.)

#5. In presentation of results on the evaluation of the causal relationship of AF/F on eGFR, the Authors state that they used a multiplicative random effects model because of observed heterogeneity. Since the results on IVW estimate is presented above, this statement needs clarification if the presented result is already from multiplicative random effects model, also in contrast to all other presented results from IVW analysis. Was this the only instance?

We would like to apologize for any confusion we have caused and to appreciate the suggestion.

In a similar way to the MR study conducted by Dr. Jie V Zhao et al [29], we re-analyzed the IVW methods thorough our study using a multiplicative random-effects model when Cochran’s Q statistic was significant (P < 0.05). Otherwise, a fixed-effects model was used.

We have stated it clearly in the Methods and Result section of the revised manuscript (Please see lines 139-141, 177, 184-187, 209, 224-225, 226-229, 245, 251, 255, and 256-258 in the revised manuscript.)

(Reference)

29. Zhao JV, Schooling CM. Effect of linoleic acid on ischemic heart disease and its risk factors: a Mendelian randomization study. BMC Med. 2019;17: 61.

Accordingly, when we excluded an outlier SNP (rs796427) by MR-PRESSO from our MR analysis estimating the causal effect of the risk of AF/F on the change in eGFR, Cochran’s Q statistic for the IVW method indicated low heterogeneity (24.3, P = 0.11).

Therefore, we re-analyzed the IVW method using a fixed-effects model.

As a result, Beta (ORs) remained unchanged, but SE got smaller (from 0.0455 to 0.038) with significant significance (P = from 0.036 to 0.012 < 0.017).

We corrected it in the Result session (lines 183-187, and 203-205), the Discussion session (line 333), and in Table 1 of the revised manuscript.

Similarly, in the analysis estimating the causal effect of the risk of AF/F on CKD risk, Cochran’s Q statistic for the IVW method indicated low heterogeneity (23.0, P = 0.19) as shown originally in Table 1.

Therefore, we re-analyzed the IVW method using a fixed-effects model.

As a result, Beta (ORs) remained unchanged, but SE got smaller (95% CI got narrower) with significant significance.

Then we corrected the result as follows.

Incorrect: OR of CKD per log OR of AF/F, 25.3; 95% CI, 2.71-236.9; P = 0.0046.

Correct: OR of CKD per log OR of AF/F, 25.3; 95% CI, 3.51-183.0; P = 0.001.

Incorrect: OR of CKD was 9.39 per doubling OR of AF/F (95% CI, 1.99-44.2).

Correct: OR of CKD was 9.39 per doubling OR of AF/F (95% CI, 2.39-37.0).

(Please see lines 27 and 209-212 in the revised manuscript.)

Please also refer to Table 1 in the revised manuscript.

#6. The Authors state in the Discussion that estimates may be too large and CIs were too wide and these present a major limitation of the study. However, the reported results present maybe a consequence of some limitations but do not present a limitation themselves. Please revise.

We quite agree with Reviewer on this point.

The relatively small sample size of the AF/F dataset was one of our major limitations.

We have rephrased this part as follows:

Our MR estimate scale of OR of CKD per doubling OR of AF/F may be too large and the corresponding 95% CI width was very wide (OR, 9.39; 95%CI, 1.99-44.2). One possible reason for this observation was the relatively small sample size of the AF/F dataset that was one of the major limitations of this study, as we could not use the largest GWAS meta-analysis due to substantial sample overlap.

(Please see lines 307-311 in the revised manuscript.)

Please also see lines 270 and 285-286 in the revised manuscript.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

1 Sep 2021

PONE-D-21-11069R1

Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization study

PLOS ONE

Dear Dr. Yoshikawa,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please refer to the comments raised by the reviewers and revise accordingly.

Please submit your revised manuscript by 1 Oct 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Jie V Zhao

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have adequately addressed my comments.

Figure 1 is still hard to read. Please revise.

Reviewer #2: Causal effect of atrial fibrillation/flutter on chronic kidney disease: A two-sample Mendelian randomization study

By Masahiro Yoshikawa, Kensuke Asaba and Tomohiro Nakayama

Reviewer Comments

I thank the Authors for their thorough job addressing all my comments. Explanations were provided, corrections were made. Only one last point I still do not understand and concerns my previous point on palindromic SNPs. While Authors’ explanation are understandable I still have trouble:

In Supplementary Table 1, one SNP (rs7853195) is marked as ‘palindromic ambiguous’ and is mentioned to be excluded from the analysis in the Results (line 173). This can be retraced by the fact that the allele frequency of the effect allele is 0.42 in both GWAS of exposure and outcome. However, there is another SNP (rs4245712) in that table that is not marked as ‘palindromic ambiguous’ but has allele frequency of the effect allele of 0.55 (exposure) and 0.56 (outcome). Since MAF is >0.42 (definition of palindromic SNPs, lines 119-120) I do not understand why this SNP was not excluded. In Supplementary Table 2, there are also such instances (e.g., rs1055256). What do I miss? Did the Authors used a different source than the GWAS to define palindromic SNPs or incorporated some additional criteria?

Otherwise, I am quite satisfied with the presentation of the project.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

8 Sep 2021

Dear Dr. Jie V Zhao,

We would like to express our sincere gratitude and appreciation for the opportunity to submit a revision of our manuscript entitled "Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization study".

Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Reference 18 is not "Forthcoming" but "Online ahead of print" and we revised it (please see line 397 in the newest version of the manuscript).

Now we have confirmed that our reference list is correct without retracted papers.

And our responses to each Reviewer’s comment are described below in a point-to-point manner.

We thank you very much for handling this paper and look forward to hearing from you again.

Sincerely,

Masahiro Yoshikawa, M.D., Ph.D.

Corresponding author.

Address; Division of Laboratory Medicine, Department of Pathology and Microbiology, Nihon University School of Medicine, Oyaguchi-kamicho 30-1, Itabashi, Tokyo, Japan.

Phone; 81-3-3972-8111

e-mail: myosh-tky@umin.ac.jp

To Reviewer #1

Our responses to the Reviewer’s comments

The authors have adequately addressed my comments.

We would like to express our sincere gratitude and appreciation for your insightful suggestions.

We have learned a lot, and we believe that our manuscript has been significantly improved thanks to Reviewer #1.

Figure 1 is still hard to read. Please revise.

Could you please click "Click here to access/download;Figure;Fig1.tif " on the top right of Revised Figure 1 page to download TIF file with a resolution of 600 dpi?

To Reviewer #2

Our responses to the Reviewer’s comments

I thank the Authors for their thorough job addressing all my comments. Explanations were provided, corrections were made.

We would like to express our sincere gratitude and appreciation for pointing out our mistakes and faults. We also thank you very much for insightful suggestions.

We believe that our manuscript has been significantly improved thanks to Reviewer #2.

Only one last point I still do not understand and concerns my previous point on palindromic SNPs. While Authors’ explanation are understandable I still have trouble: In Supplementary Table 1, one SNP (rs7853195) is marked as ‘palindromic ambiguous’ and is mentioned to be excluded from the analysis in the Results (line 173). This can be retraced by the fact that the allele frequency of the effect allele is 0.42 in both GWAS of exposure and outcome. However, there is another SNP (rs4245712) in that table that is not marked as ‘palindromic ambiguous’ but has allele frequency of the effect allele of 0.55 (exposure) and 0.56 (outcome). Since MAF is >0.42 (definition of palindromic SNPs, lines 119-120) I do not understand why this SNP was not excluded. In Supplementary Table 2, there are also such instances (e.g., rs1055256). What do I miss? Did the Authors used a different source than the GWAS to define palindromic SNPs or incorporated some additional criteria?

Thank you very much for your careful reading and suggestion, and we sincerely apologize that our tables were still hard to read.

In the same way as MR-Base platform does, we have marked the palindromic SNPs "TRUE" in the "palindromic" column and the excluded SNPs "TRUE" in the "ambiguous" column in the newest version of Supplementary Tables 1 and 2. Please also see lines 174-175 and 242-243 in the newest manuscript.

For example, rs4245712 in Supplementary Table 1 has effect allele frecency (EAF) of 0.55 for exposure and 0.56 for outcome (i.e. minor allele frequency (MAF) > 0.42), but it is not a palindromic SNP ("palindromic" is "FALSE"). Therefore, it was not excluded.

In Supplementary Table 2, EAF of rs4820324 for exposure is, more precisely, 0.5801 (Please see row 308 and column N in ST4 in the GWAS study by CKDgen, Nat Genet. 2019;51: 957-972, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6698888/). And EAF of rs4820324 for outcome is 0.580514. Both are slightly > 0.58 (i.e. MAF < 0.42), and therefore it was not excluded from our analyses.

On the other hand, rs7853195 in Supplementary Table 1 in our manuscript (EAF is 0.421733 for exposure, slightly > 0.42) was excluded properly.

Otherwise, I am quite satisfied with the presentation of the project.

We really appreciate your help in improving our manuscript.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

8 Oct 2021

PONE-D-21-11069R2Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization studyPLOS ONE

Dear Dr. Yoshikawa,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Nov 22 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Jie V Zhao

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Figure 1 is still hard to read. I don't think the author's response that "Could you please click "Click here to access/download;Figure;Fig1.tif " on the top right of Revised Figure 1 page to download TIF file with a resolution of 600 dpi?" is acceptable. If one chooses to print out a hard copy to read, then where to click?

Reviewer #2: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

9 Nov 2021

To Reviewer #1

Figure 1 is still hard to read. I don't think the author's response that "Could you please click "Click here to access/download;Figure;Fig1.tif " on the top right of Revised Figure 1 page to download TIF file with a resolution of 600 dpi?" is acceptable. If one chooses to print out a hard copy to read, then where to click?

We would like to apologize for any inconvenience.

We have separated original Figure 1(a-c) into revised Figures 1-3 to increase the resolution.

We sincerely hope that new Figures could meet with your criteria.

Submitted filename: Response to Reviewers_PlosOne.docx

Click here for additional data file.

23 Nov 2021

Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization study

PONE-D-21-11069R3

Dear Dr. Yoshikawa,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Jie V Zhao

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

29 Nov 2021

PONE-D-21-11069R3

Causal effect of atrial fibrillation/flutter on chronic kidney disease: A bidirectional two-sample Mendelian randomization study

Dear Dr. Yoshikawa:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Jie V Zhao

Academic Editor

PLOS ONE