Comparison of cardiovascular outcomes and cardiometabolic risk factors between patients with type 2 diabetes treated with sodium-glucose cotransporter-2 inhibitors and dipeptidyl peptidase-4 inhibitors: a meta-analysis

Abstract

Aims 

Prevention of cardiovascular outcomes is a goal of the management of patients with type 2 diabetes mellitus patients as important as lowering blood glucose levels. Among the various glucose-lowering agents, the effects of sodium-glucose cotransporter-2 inhibitors (SGLT-2Is) and dipeptidyl peptidase-4 inhibitors (DPP-4Is) on cardiovascular outcomes have become the focus of recent researches.

Methods and results 

A systematic search was performed through several online database. All studies that compared the effects of SGLT-2Is and DPP-4Is on cardiovascular outcomes and cardiometabolic risk factors were reviewed. A total of 30 studies were included. Compared with DPP-4Is, SGLT-2Is treatment reduced the risk of stroke [risk ratio (RR) = 0.80; 95% confidence interval (CI), 0.76–0.84], myocardial infarction (RR = 0.85; 95% CI, 0.81–0.89), heart failure (RR = 0.58; 95% CI, 0.54–0.62), cardiovascular mortality (RR = 0.55; 95% CI, 0.51–0.60), and all-cause mortality (RR = 0.60; 95% CI, 0.57–0.63). In addition, SGLT-2Is presented favourable effects on hemoglobinA1c, fasting plasma glucose, systolic blood pressure, and diastolic blood pressure. The differences in blood lipids were also compared.

Conclusion

Sodium-glucose cotransporter-2 inhibitors are superior to DPP-4Is in terms of cardiovascular outcomes. Sodium-glucose cotransporter-2 inhibitors bring more benefits with respect to the cardiometabolic risk factors.

SGLT-2 inhibitors, DPP-4 inhibitors, Cardiovascular outcomes •Type 2 diabetes

Introduction

Type 2 diabetes mellitus (T2DM), a major cause of morbidity and mortality worldwide, is rising in prevalence.¹ It is well known that T2DM is closely associated with both cardiovascular microvascular and macrovascular complications.^2,3 Patients with T2DM have more than twice the incidence of cardiovascular events, such as myocardial infarction (MI) and stroke, and cardiovascular mortality has become the leading cause of death in T2DM patients. In addition, patients with T2DM confer an increased incidence of heart failure (HF) and those patients substantially worsen prognosis.⁴ Furthermore, it has been reported that glucose fluctuation is associated with endothelial cell damage, which is the first stage of atherosclerosis and a predictor of cardiovascular diseases.⁵ Although there is early indication of benefit,⁶ studies have not provided a favourable impact of intensive glucose-lowering therapy on improving cardiovascular outcomes in patients with T2DM.^7,8 Thus, prevention of cardiovascular outcomes is a goal of the management of patients with T2DM as important as lowering blood glucose levels. Since 2008, all new anti-hyperglycaemic agents must assess the cardiovascular safety according to the guidance of the US Food and Drug Administration (FDA).⁹ Among the various glucose-lowering agents, sodium-glucose cotransporter-2 inhibitors (SGLT-2Is) and dipeptidyl peptidase-4 inhibitors (DPP-4Is) have become the focus of recent researches.

Some SGLT-2Is improve cardiovascular outcomes in patients with T2DM. The EMPA-REG OUTCOME trial reveals a significant favourable effect of empagliflozin on hospitalization for HF, all-cause mortality, and major adverse cardiovascular events (MACE: cardiovascular mortality, non-fatal MI, and non-fatal stroke) compared with placebo. But there are no significant differences in prevalence of MI and stroke.¹⁰ The CANVAS programme reports a significant reduction in the rate of MACE and hospitalization for HF among patients treated with canagliflozin compared with placebo. The incidences of all three components of the primary outcomes have trend to be reduced, but do not have significances. The differences between canagliflozin and placebo in all-cause mortality and hospitalization for HF are not regarded to be significant.¹¹ However, DECLARE-TIMI 58 trial¹² and VERTIS trial¹³ reveal neutral effect on the rate of MACE between patients treated with SGLT-2Is (dapagliflozin and ertugliflozin, respectively) and placebo. On the other hand, several trials have investigated the cardiovascular safety of DPP-4Is. It has been reported that patients treated with DPP-4Is have lower risk of stroke.¹⁴ However, according to the results of EXAMINE,¹⁵ SAVOR-TIMI 53,¹⁶ TECOS trials,¹⁷ and CARMELINA,¹⁸ DPP-4Is show neutral effect on the rate of MACE compared with placebo.

Recently, a direct comparison of the effects on cardiovascular outcomes between SGLT-2Is and DPP-4Is has been addressed. However, there are some contradictory conclusions. In EMPRISE study, patients with T2DM treated with empagliflozin are at a lower risk of hospitalization for HF compared with T2DM patients treated with sitagliptin.¹⁹ In another study, SGLT-2Is are associated with a lower risk of HF compared with DPP-4Is, but not the rate of MACE.²⁰ A multi-database retrospective cohort study reveals that there is significant reduction of the rate of MACE among T2DM patients treated with SGLT-2Is.²¹ In CVD-REAL Nordic, dapagliflozin exerts neutral effects on all components of MACE compared with DPP-4Is.²² A network meta-analysis providing indirect comparative evidence suggests that SGLT-2Is present favourable effects on the rate of HF, all-cause death, and cardiovascular mortality compared with DPP-4Is. However, no significant differences in the rate of MI and stroke are noticed between SGLT-2Is and DPP-4Is treatments.²³ In order to provide a more comprehensive conclusion and a basis for the selection of second-line agents in T2DM patients, we collected the latest available data for a meta-analysis of clinical studies to directly compare the effects of SGLT-2Is on cardiovascular outcomes to those of DPP-4Is. In addition, we clarified the efficacy of SGLT-2Is vs. DPP-4Is for improving cardiometabolic risk factors, including hemoglobinA1cbA1c level, body weight, blood pressure, and lipid profile in patients with T2DM.

Methods

This study was registered in PROSPERO, with registration No. CRD42020214933.

Search strategy

A literature search in PubMed, Web of Science, EMBASE, the Cochrane library, and China National Knowledge Infrastructure until 20 October 2020 was conducted using following search terms: ‘diabetes’, ‘diabetes mellitus’, ‘sodium glucose co-transporter 2 inhibitor’, ‘SGLT-2 inhibitor’, ‘dapaliflozin’, ‘empagliflozin’, ‘ipragliflozin’, ‘canagliflozin’, ‘dipeptidyl peptidase 4 inhibitor’, ‘DPP-4 inhibitor’, ‘sitagliptin’, ‘saxagliptin’, ‘teneligliotin’, ‘alogliptin’, ‘anagliptin’, ‘linagliptin’, ‘vildagliptin’, ‘evogliptin’, and ‘gemigliptin’, alone or in combination, without language restriction.

Inclusion and exclusion criteria

Eligible trials were included in the meta-analysis according to the following inclusion criteria: (i) type of participants: patients (≥18 years old) in each study who have been diagnosed as T2DM and (ii) type of study: all studies that compared the effects of SGLT-2Is and DPP-4Is as monotherapy or combination therapy with other hypoglycaemic drugs on cardiovascular outcomes and cardiometabolic risk factors. Exclusion criteria include (i) study design: reviews, case reports, comments, letters, and abstracts; (ii) type of participants: patients <18 years old, pregnant women, and animals; and (iii) insufficient information concerning evaluation rates.

Outcomes

The primary outcomes evaluated in this study were cardiovascular outcomes including MI, stroke, HF, cardiovascular mortality, and all cause of mortality. MI, stroke, and HF were defined on the basis of primary discharge diagnosis codes. Cardiovascular mortality was defined as in-hospital death or out-of-hospital death with a cardiovascular diagnosis, without documentation of other life-threatening disease. All cause of mortality defined as death from any cause. The secondary outcomes were cardiometabolic risk factors, including change level (relative to baseline) of hemoglobinA1c (HbA1c), fasting plasma glucose (FPG), body weight, systolic blood pressure (SBP), diastolic blood pressure (DBP), lipid indices [high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), cholesterol, triglycerides), body mass index (BMI), and estimated glomerular filtration rate (eGFR).

Study selection

After removing duplicates, two independent investigators (S.W. and Z.Z.) screened abstracts and titles to confirm that they fulfilled the inclusion criteria. When they felt the titles or abstracted was potentially useful, full text of articles would be retrieved and considered by three authors (S.W., T.W., and Z.Z.). When discrepancies occurred between investigators, all authors (S.W., T.W., Z.Z., P.J., X.L., and M.D.) would compare the result and identify the disagreements. Final decision was made based on the agreement of all authors. Other potentially relevant studies mentioned in reference lists of included studies were assessed in the same manner.

Data extraction

The data extracted included last name of the first author, publish year, geographical region, sample size (N), percentage of male patients (%), mean age (years), portion of different drugs use, outcomes, and research type. Two authors (S.W. and T.W.) independently extracted data by using a predefined data extraction form. When they completed separately, Z.Z. compared two data sets and identified any possible discrepancies. Finally, three authors (S.W., T.W., and Z.Z.) would resolve all disagreements by discussion and result in a final consensus. We employed the Newcastle-Ottawa scale (NOS) for quality assessment of included cohort trials. The NOS scores of 1–3, 4–6, and 7–9 indicated low, intermediate, and high quality, respectively. The Cochrane risk of bias tool was employed to assess the quality of randomized controlled trials (RCTs). Three independent authors (S.W., T.W., and Z.Z.) assessed quality of trials included in meta-analysis according to predefined tools. When they completed, P.J. compared three data sets and identified discrepancies occurred between authors. The reasons would be identified and a final decision was made based on the agreement of all authors (S.W., T.W., Z.Z., P.J., X.L., and M.D.).

Statistical analysis

The risk ratio (RR) and weighted mean difference (WMD) were employed to compare dichotomous variables and continuous, respectively. 95% confidence interval (CI) was used in reports of all results. The effect estimates of outcomes were pooled by using fixed-effect models. When significant heterogeneity was detected, a random effect model was conducted. Heterogeneity was assessed by using the I² value and I² >50% was considered significant. The possible publication bias of outcomes was assessed by Egger’s test and Begg’s test (P<0.10). There were many different decisions of trials involved in meta-analysis, and it was important to perform a sensitivity analysis to explore the impact of different decisions on overall results. One sensitivity analysis may explore the impact of excluding or including studies in meta-analysis. In our analysis, the sensitivity analyses were made by excluding one study at a time to observe the change of effects for outcomes. If the analysis was robust, then there should be little change in the overall outcomes. If the overall results changed, there was an indication that the result may needed to be interpreted with caution. All statistical analyses were performed with the STATA 12.0 statistical software package (StataCoporation, College Station, TX, USA).

Results

Selection of included studies and study characteristics

Figure 1 presented the screening and selection process. A total of 1461 relevant articles were identified by searching online databases. After removing duplicates and assessing each study according inclusion and exclusion criteria, 30 studies were included in the meta-analysis.^{19–22,24–49} The characteristics of the included trials are summarized in Supplementary material online, Table S1. Among these studies, 13 were RCTs, 16 were retrospective studies, and the last one was prospective studies. These studies focused on comparisons of SGLT-2Is (empagliflozin, canagliflozin, dapagliflozin, ertugliflozin, etc.) with DPP-4Is (sitagliptin, linagliptin, saxagliptin, vildagliptin, alogliptin, etc.), in combination with other antidiabetic drugs or as new used drug. The results of our quality assessment were presented in Supplementary material online, Tables S1 and S2.

Figure 1

PRISMA diagram of study selection.

Primary outcomes

In term of cardiovascular outcomes, SGLT-2Is treatment reduced the risk of MI by 15% compared with DPP-4Is (RR = 0.85; 95% CI, 0.81–0.89). Treatment with SGLT-2Is also lowered the risk of stroke by 20% (RR = 0.80; 95% CI, 0.76–0.84). The use of SGLT-2Is was closely associated with a lower incidence of HF (RR = 0.58; 95% CI, 0.54–0.62). Furthermore, significantly reduced risks of cardiovascular mortality (RR = 0.55; 95% CI, 0.51–0.60) and all-causes mortality (RR = 0.60; 95% CI, 0.57–0.63) were observed among patient treated with SGLT-2Is than that among patients treated with DPP-4Is (Figure 2).

Figure 2

Meta-analysis of cardiovascular outcomes: myocardial infarction, stroke, heart failure, cardiovascular mortality, and all-cause mortality.

Secondary outcomes

Compared with DPP-4Is, SGLT-2Is significantly reduced the level of HbA1c (WMD = −0.07%; 95% CI, −0.13 to −0.01%; WMD = −1.90 mmol/mol; 95% CI, −3.18 to −0.61 mmol/mol) and FPG (WMD = −7.39 mg/dL; 95% CI, −9.73 to −5.06 mg/dL). In addition, SGLT-2Is brought benefit with respect to the body weight (WMD = −2.16 kg; 95% CI, −2.42 to −1.90 kg) (Figure 3). SGLT-2Is treatment was associated with lower BMI (WMD = −0.94 kg/m²; 95% CI, −1.14 to −0.74 kg/m²), SBP (WMD = −4.22 mmHg; 95% CI, −5.66 to −2.77 mmHg), and DBP (WMD = −2.26 mmHg; 95% CI, −2.99 to −1.53 mmHg) levels compared with DPP-4Is. The differences in blood lipids were also compared. No significant difference was observed in triglycerides (WMD =1.39 mg/dL; 95% CI, −2.58 to 5.37 mg/dL) between SGLT-2Is and DPP-4Is treatment. However, SGLT-2Is treatment was associated with higher levels of cholesterol (WMD =3.96 mg/dL; 95% CI, 0.21–7.72 mg/dL), HDL (WMD =3.55 mg/dL; 95% CI, 2.90–4.20 mg/dL), and LDL (WMD =4.63 mg/dL; 95% CI, 1.73–7.54 mg/dL). The pooled result showed the effect of SGLT-2Is and dipeptidyl peptidase-4 inhibitors treatment on eGFR is similar (WMD =0.86 mL/min/1.73m²; 95% CI, −0.42 to 2.14 mL/min/1.73m²) (Figure 4).

Figure 3

Meta-analysis of cardiometabolic risk factors: HbA1c, fasting plasma glucose, and body weight.

Figure 4

Meta-analysis of cardiometabolic risk factors: SBP, DBP, HDLcholesterol, LDL cholesterol, triglycerides, cholesterol, BMI, and eGFP.

Publication bias and sensitivity analysis

Publication bias was noted in the following outcomes: cardiovascular mortality, HbA1c (mmol/mol), and triglycerides (Supplementary material online, Table S3). More studies should be included in further analysis to reduce publication bias. Our sensitivity analysis revealed no meaningful differences in the outcomes except for the pooled results of HbA1c (%) and LDL (Supplementary material online, Figure S1).

Discussion

In this meta-analysis, data from 30 trials were analysed and we draw the following conclusions. Compared with DPP-4Is, SGLT-2Is were associated with a lower risk of cardiovascular outcomes, including MI, stroke, HF, cardiovascular, and all-cause mortality. Sodium-glucose cotransporter-2 inhibitors brought more benefits with respect to the control of HbA1c, FPG, and bodyweight. Sodium-glucose cotransporter-2 inhibitors could significantly reduce the level of BMI, SBP, and DBP. However, SGLT-2Is were associated with higher level of cholesterol, HDL and LDL. SGLT-2Is had no favourable effects on eGFR level.

DPP-4Is exert their glucose-lowering effects by inhibiting the enzyme that degrades two incretin hormones in gut, glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide, followed by stimulating insulin secretion and reducing glucagon secretion.⁵⁰ DPP-4Is have occupied an important place in the treatment of T2DM, because they are not associated with weight gain and hypoglycaemia.⁵¹ In addition, they can be used in T2DM patients with chronic kidney disease due to their renoprotective effect by suppressing eGFR decline.⁵² The cardiovascular benefits of DPP-4Is have also been extensively studies in preclinical settings. After ischaemic/reperfusion injury, DPP-4Is demonstrated significant improvement in left ventricle recovery and reduction in infarct size.^53,54 In a rat model, DPP-4Is treatment significantly improved cardiac function recovery from HF and attenuated cardiac remodelling, cardiomyocyte apoptosis, and pulmonary congestion.⁵⁵ DPP-4Is also exerted favourable effects on renal function, stroke volume, and heart rate in a pig HF model.⁵⁶ Some favourable effects on cardiac function have been observed in clinical setting. T2DM patients with coronary artery disease treated with DPP-4Is demonstrated a sustained improvement in myocardial performance during dobutamine stress and a significant reduction in myocardial stunning.⁵⁷ PROLOGUE trial also proved the beneficial actions of DPP-4Is on echocardiographic parameters of diastolic function.⁵⁸ Taken together, these data presented that DPP-4Is have potential beneficial action on cardiovascular outcomes, which is inconsistent with other clinical studies, including EXAMINE,¹⁵ SAVOR-TIMI 53,¹⁶ TECOS trials,¹⁷ and CARMELINA¹⁸ trials. The reasons for these disappointing results are still unknown, but several hypotheses have been proposed.^59,60 First of all, the purpose of these studies is to prove safety by comparing non-inferiority with placebo. Second, the duration of these trials is too short to show difference in cardiovascular effects, especially in T2DM patients with advanced cardiovascular diseases. Third, complex pathophysiology of vascular damage in T2DM patients, which combines many risk factors. Fourth, the mild and moderate increase in GLP-1 levels after DPP-4Is treatment, which might explain why the patients do not get beneficial effects. Finally, we cannot exclude that these are some unknown deleterious effects of DPP-4Is due to their many other substrates, resulting in out of target effects.⁶¹

It is well known that the kidney facilitates elimination and reabsorption of glucose, playing an important role in glucose homeostasis. The majority of filtered glucose could be reabsorbed through SGLT2 that is located in the early proximal tubule of the kidney. It has been reported that an enhanced expression of SGLT2 is noticed in T2DM patients, increasing renal glucose reabsorption.⁶² Thus, SGLT-2Is enhance glycosuria and reduce hyperglycaemia by selectively inhibited SGLT2, independently of insulin. Along with the anti-hyperglycaemic effect, cardiovascular benefits of SGLT-2Is have also been noticed in preclinical settings. In diabetic mice, SGLT-2Is therapy attenuated the cluster of differentiation (CD36) and expression of lectin-like ox-LDL receptor-1 on macrophages, resulting in inhibition of macrophage foam cell formation to exert anti-atherogenic effect.⁶³ Another study also demonstrated that SGLT-2Is treatment significantly ameliorated cardiac inflammation and fibrosis, vascular dysfunction, and coronary arterial remodelling in type 2 diabetic mice. In addition, these beneficial actions might be associated with the attenuation of oxidative stress in cardiovascular tissues.⁶⁴ These potential beneficial action on cardiovascular outcomes of SGLT-2Is has also been confirmed in some large clinical trials comparing SGLT-2Is with placebo, including EMPA-REG OUTCOME¹⁰ and CANVAS¹¹ trials. Although SGLT-2Is do not result in significantly lower rate of MACE, significantly lower rate of cardiovascular death, and hospitalization of HF are observed in DECLARE-TIMI 58.¹² However, there is not a clear explanation about why these results in recent VERTIS trial do not reach significance between SGLT-2Is and placebo groups.¹³

In the present meta-analysis, we firstly and directly compare cardiovascular effects between SGLT-2Is and DPP-4Is therapy. The pooled results confirmed that SGLT-2Is are superior to DPP-4Is in terms of cardiovascular outcomes, including MI, stroke, HF, cardiovascular, and all-cause mortality, which is consistent with a network meta-analysis.⁶⁵ Several underlying mechanisms have been proposed to explain the cardiovascular benefits of SGLT-2Is. Improvement of ventricular loading conditions is one of the main mechanisms. SGLT-2Is therapy can trigger natriuresis and glucosuria by inhibiting SGLT-2 in the proximal tubule, which ensures favourable osmatic diuresis. SGLT-2 resorbs about 5% of the sodium under normal conditions. In the setting of chronic hyperglycaemic conditions, this capacity will increase, leading to more pronounced effects on sodium homeostasis and resulting in a significantly reduction of preload.^66,67 In addition, SGLT-2Is decrease afterload through reducing blood pressure and improving vascular function.⁶⁸ Our results noticed that SGLT-2Is could reduce the level of both SBP and DBP compared with DPP-4Is, which may be attributable to the excretion of sodium and water as well as the suppression of renal renin–angiotensin system activation because of enhance delivery of sodium to the juxtaglomerular apparatus.³⁶ Furthermore, SGLT-2Is may induce vasodilatation by improving endothelial function and arterial stiffness indices due to activation of voltage-gated potassium channels and protein kinase G.^69,70 Improvements in cardiac energy metabolism and bioenergetics is another hypothesis, secondary to enhanced cardiac efficiency, and cardiac output.⁷¹ A shift in fuel energetics has been suggested as a protective mechanism for HF. SGLT-2Is are known to induce ketogenesis, including β-hydroxybutyrates, in patients with T2DM.⁷² β-hydroxybutyrates is regarded as an energy-efficient ‘superfuel’, which means it offers an alternative and less expensive myocardial fuel source in patients with diabetes, resulting in improving cardiac metabolism.⁷¹ SGLT-2Is also mediate a uricosuric effect via the GLUT9 transporter and cause secretion of uric acid, resulting in a 10–15% reduction in plasma uric acid level.⁷³ It is known that increased plasma uric acid level is associated with cardiovascular complications.⁷⁴ Another postulated mechanism is lowering myocardial intracellular sodium concentration by inhibiting Na⁺/H⁺ exchanger 1 isoform in the cardiomyocyte with a secondary decrease in intracellular calcium and an increase in mitochondrial calcium.⁷⁵ In fact, increased myocardial intracellular sodium is early hallmarks of HF, while increased mitochondrial calcium prevented HF in porcine models.⁷⁶ In addition, SGLT-2Is also affect recognized cardiovascular outcomes risk factors. A weight reduction results from calorie loss due to glucosuria is observed after SGLT-2Is therapy, which is consistent with our result and also be confirmed in another meta-analysis.⁷⁷ Our result also shows that SGLT-2Is are also associated with a stronger reduction in HbA1c and FPG level, which is also consistent with previous study.⁷⁸ When we removed Sun’s and Roden’s trial, the pooled results of HbA1c (%) was −0.05% (95% CI, −0.13 to 0.03%) and −0.06% (95% CI, −0.13 to 0.002%), respectively. Although the tendencies were similar with the overall outcomes, there were no significances. Sun’s and Roden’s trials included patients with HbA1c > 10%, which were excluded by most of remaining trials. We proposed that the effect of SGLT-2Is and DPP-4Is on HbA1c may be associated with the baseline level of HbA1c, but it still needs to be confirmed further. In addition, Sun’s trial was focused on a single hospital, representativeness should be questioned. Therefore, a more robust result should be got by including more trials. Both DPP-4Is and SGLT-2Is participate in the regulation of lipid profile in T2DM patients, but the mechanisms have not been fully understood. When we removed Schernthaner’s trial, the pooled result of LDL was 3.5 mg/dL (95% CI, −0.43 to 7.5 mg/dL). It was noted that Schernthaner’s study was a 52 weeks randomized trials, while duration of the remaining trials was shorter than 24 weeks. Therefore, the different effects of SGLT-2Is and DPP-4Is on LDL level may be associated with application timepoints. However, we could not have robust result due to lack of evidence.

Limitations

Our meta-analysis has several limitations. First, due to the limited number of clinical studies, it is difficult for us to do subgroup analysis for each agent. Second, the duration of follow-up is not consistent between studies, but it is noted the heterogeneity of these results is very low. Finally, the effects of other antidiabetic agents cannot be excluded.

Conclusion

SGLT-2Is are superior to DPP-4Is in terms of cardiovascular outcomes, including MI, stroke, HF, cardiovascular, and all-cause mortality. SGLT-2Is brought more benefits with respect to the cardiometabolic risk factors. SGLT-2Is also participate in the regulation of lipid profile. Our findings provide a basis for the selection of second-line agents in T2DM patients and are potentially useful in regarding the cardiovascular effectiveness of SGLT-2Is in routine clinical practice.

Supplementary material

Supplementary material is available at European Journal of Preventive Cardiology online.

Funding

The National Natural Science Foundation of China under contract (Nos. 82071250, 31672290, 81670730), Open Sharing Fund for the Large-scale Instruments and Equipments of Central South University (No. CSUZC2020043), Innovation and entrepreneurship education reform research project of Central South University(2019CG052), Postgraduate education and teaching reform project of Central South University (2021JGB105), and Biochemistry Maker Space of Central South University (2015CK009).

Conflict of interest: none declared.

Data Availability Statements

The data underlying this article are available in the article and in its online supplementary material.

Author notes

†

Siwen Wang and Ting Wu contributed equally to this work.