Relationship of OCT-Based Diabetic Retinal Neurodegeneration to the Development and Progression of Diabetic Retinopathy: A Cohort Study

Abstract

Purpose: To evaluate the relationship between diabetic retinal neurodegeneration (DRN), as quantified by optical coherence tomography (OCT), to the development of diabetic retinopathy (DR), progression of DR, and development of proliferative DR (PDR).

Methods: This was a prospective cohort study, including 385 eyes with no DR or nonproliferative DR at baseline. The thicknesses of the macular ganglion cell-inner plexiform layer (m-GCIPL), macular retinal nerve fiber layer, and peripapillary RNFL (p-RNFL) were measured using Cirrus OCT (Carl Zeiss Meditec, Dublin, CA, USA). DR outcomes were determined from macula- and optic disc–centered fundus photographs, following the modified Airlie House classification system. Cox proportional hazards models were used to estimate hazard ratio (HR) adjusting for age, mean arterial blood pressure, diabetes mellitus duration, HbA1c, diabetic kidney disease, axial length, OCT signal strength, and disc area (for p-RNFL only).

Results: After a median follow-up of 6.2 years (range 5.0–7.7 years), 79 eyes developed DR, 99 eyes developed DR progression, and 38 eyes developed PDR. Thinner mean and sectorial m-GCIPL thicknesses were significantly associated with higher risk of DR development, with HRs ≥ 1.373 (1.023–1.843), except for the superonasal and superotemporal sectors. Similar to DR development, thinner m-GCIPL thicknesses were significantly associated with DR progression and PDR development, with HRs ranging from 1.306 (1.094–1.559) to 2.331 (1.524–3.566). Additionally, the inclusion of inferior m-GCIPL thickness significantly improved the predictive discrimination for DR development (C statistics: 0.661 vs. 0.705, P < 0.001), and DR progression (C statistics: 0.704 vs. 0.729, P < 0.001), as well as inferotemporal m-GCIPL for PDR development (C statistic: 0.917 vs. 0.930, P < 0.001) beyond established risk factors.

Conclusions: OCT measurements that elucidate DRN may enhance prognostic identification and predictive discrimination of DR development, DR progression, and PDR development beyond established risk factors.

Diabetic retinopathy (DR) is the most common microvascular complication of diabetes mellitus (DM).^1,2 In a recent meta-analysis, the number of adults worldwide with DR is estimated to be around 103.12 million and is projected to increase to 160.50 million by 2045.³ Considering the substantial financial, societal, and mental health impacts associated with the growing DM population, biomarkers that can predict and facilitate early DR diagnosis may promote timely intervention to prevent vision loss.⁴ Conventionally, “DR” focuses on the clinical manifestations of retinal vascular complications and diabetic macular edema (DME). There is ongoing effort to update the classification of diabetic retinal disease (DRD) to consider both the vascular and neuronal aspects of DRD, based on advances in retinal imaging, microperimetry, and electroretinography.^5,6 In fact, DRD is now defined as a highly specific neurovascular disease involving the retina by the American Diabetes Association.⁷ 

Diabetic retinal neurodegeneration (DRN) features neural apoptosis and glial activation secondary to DM.⁸ It is an important clinical manifestation of DRD, as DRN development can precede vascular lesions on structural, functional, and molecular levels, suggesting it may allow early diagnosis of DRD development and progression.^9–13 Moreover, the presence of DRN may limit visual improvement in eyes with concurrent DME despite successful treatment of the edema.¹⁴ Optical coherence tomography (OCT) is a promising tool in assessing structural changes related to DRN.^12,15 For example, previous studies have reported that reduced inner retinal layers are associated with increasing DR severity.^16–18 Additionally, during a four-year observation, Sohn et al. demonstrated a progressive reduction in mean macular ganglion cell-inner plexiform layer (m-GCIPL) thickness and macular retinal nerve fiber layer thickness (m-RNFL) of −0.25 µm/y and −0.29 µm/y, respectively, in 45 DM participants with no to minimal DR.¹³ Kim et al. also found that baseline mean m-GCIPL thickness was an independent risk factor for DR progression among 87 eyes of DM patients followed up to four years.¹⁹ Nevertheless, the relationships between the structural alterations of DRN and other DR endpoints, namely, DR development and PDR development, remain largely unknown. Besides, whether other OCT measurements of the neural retina, such as different sectors of m-GCIPL thickness (e.g., inferior or superior), m-RNFL thickness, and peripapillary RNFL (p-RNFL) thickness can also exhibit similar or even stronger associations with DR endpoints is also unknown. 

To address these gaps, this study was designed to investigate the relationships between mean and sectors of OCT measurements and DR endpoints (i.e., DR development, DR progression, and PDR development) in a cohort of patients with DM. We also examined whether OCT measurements provide additional predictive value for DR outcomes beyond the previously reported risk factors while adjusting for OCT measurement confounders. Finally, we identified which sector has the strongest association among significant OCT measurements. 

Material and Methods

Study Subjects

This was a prospective observational analysis of OCT measurements for a cohort of participants with DM. Participants were consecutively recruited from the Chinese University of Hong Kong Eye Centre from July 2015 to June 2018 and were followed for at least five years.^15,20 All participants attended their second visit six months after the baseline examination. Participants without DR or only mild non-proliferative diabetic retinopathy (NPDR) were followed annually thereafter, whereas those with moderate or severe NPDR were followed every six months. The study was conducted in accordance with the 1964 Declaration of Helsinki and was approved by the Hong Kong Kowloon Central Research Ethics Committee, with informed consent obtained. 

Inclusion criteria included (1) age older than 18 years; (2) diagnosis of type 1 or type 2 DM; and (3) eyes with no DR or with NPDR at baseline. Exclusion criteria were (1) eyes with PDR or DME at baseline; (2) eyes with a history of panretinal photocoagulation or focal laser treatment; (3) eyes with a history of cataract surgery or other intraocular surgery within six months before recruitment; (4) eyes with ungradable OCT scans defined as (a) signal strength less than 5; (b) inaccurate segmentation of retinal layers; (c) blurry images; (d) signal loss; and (e) poor centration; (5) eyes with ungradable color fundus photographs; and (6) eyes with any ocular conditions other than DR (e.g., epiretinal membrane, glaucoma, retinal vein occlusion, or neovascular age-related macular degeneration) at baseline or during follow-ups. Both eyes were enrolled if eligible. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline. 

Spectral-Domain OCT Measurements

All participants underwent OCT scan using Cirrus OCT (Carl Zeiss Meditec, Dublin, CA, USA) with a central wavelength of 840 nm, an acquisition speed of 27,000 A-scans/s, and axial and lateral resolutions of 5 µm and 20 µm in tissue, respectively. The schematic diagrams of OCT measurements are shown in Supplementary Table S1. The mean and sectorial m-GCIPL and m-RNFL were measured from a macular cube scan (512 × 128 B-scans, 6 × 6 mm²) centered at the fovea within an elliptical annulus (inner vertical and horizontal axes of 1.0 mm and 1.2 mm, respectively; outer vertical and horizontal axes of 4.0 mm and 4.8 mm, respectively). The built-in algorithm provided thicknesses for mean, superior, superonasal, superotemporal, inferior, inferonasal, and inferotemporal m-GCIPL and m-RNFL, respectively. The p-RNFL thickness was measured from the optic disc cube scan (200 × 200 B-scans, 6 × 6 mm²) positioned around the optic disc in a circle of 3.46 mm in diameter. The built-in algorithm subsequently yielded the thicknesses for mean, superior, inferior, nasal, temporal, and 12 clock-hour p-RNFL. The 12 clock-hour thicknesses were labeled in a clockwise direction for the right eye and a counterclockwise direction for the left eye, with the 3 o'clock position as nasal, the 6 o'clock position as inferior, the 9 o'clock position as temporal, and the 12 o'clock position as superior. 

Definition of Endpoints

Macula- and optic disc-centered fundus photographs were obtained using a nonmydriatic retinal camera (TRC 50DX; Topcon Optical Company, Tokyo, Japan) after pupil dilation with 0.5% tropicamide and 0.5% phenylephrine. We used the modified Airlie House classification scale from the Early Treatment Diabetic Retinopathy Study (ETDRS) for DR grading from baseline throughout follow-ups.²¹ This classification provided a 15-step DR severity scale.²² DR development was defined as the presence of DR with ETDRS level 15 or above during follow-up among those with ETDRS level 10 at baseline. DR progression was defined as an increase of 2 or more steps in severity level compared with baseline in those who had level 10 to level 53 at baseline.²³ PDR development was defined as the presence of PDR (i.e., level 61 or above) during follow-up in those who had no DR or NPDR at baseline. 

Two masked graders (ZT and TN) determined the ETDRS level and documented other abnormalities (e.g., dense cataract) observed in fundus photographs. Intra-grader reliability and inter-grader reliability were assessed by comparing their results with a senior grader (DY) before grading the data included in the study. Cohen's kappa scores for inter-grader reliability were 0.901 and 0.872, whereas intragrader reliability scores were 0.981 and 0.977, respectively. Uncertain cases were further reviewed by a panel of retina specialists for final adjudication. 

Assessment of Risk Factors

The risk factors related to DR outcomes including age, duration of DM, glycated hemoglobin (HbA1c), diabetic kidney disease (DKD), mean arterial blood pressure (MABP), body mass index (BMI), baseline DR severity, and associated factors of OCT measurements (i.e., axial length [AL], OCT signal strength, and disc area)^24,25 were included in multivariable analyses. Age was defined as the age at baseline recruitment. Duration of DM was defined as the period between the date of baseline assessment and the date of diagnosis first recorded by a physician in the patient's file or hospital electronic record. Each patient's medical record was reviewed at each visit for the most recent fasting blood tests, including HbA1c and serum creatinine. The estimated glomerular filtration rate was calculated from serum creatinine based on the equation developed by the Modification of Diet in Renal Disease Study Group.²⁶ DKD was defined as estimated glomerular filtration rate less than 60 ml/min/1.73m². MABP was calculated as diastolic pressure plus one-third pulse pressure (systolic blood pressure minus diastolic blood pressure). BMI was calculated as body weight (in kilograms) divided by squared body height (in meters). AL was measured using non-contact partial coherence laser interferometry (IOL Master, Carl Zeiss Meditec, Dublin, US). Refractive status was obtained with an autorefractor (ARK-510A, Nidek Co., Ltd., Japan). Spherical equivalent (SE) was defined as sphere plus half the negative cylinder. High myopia was defined as the SE refractive error of an eye being less than −6.0 D.²⁷ OCT signal strength and disc area were automatically acquired from Cirrus HD-OCT. 

Statistical Analysis

In the analysis, each eligible eye was regarded as a unit. A generalized linear mixed model was used to compare the baseline OCT measurements with inter-eye correlation adjusted using participant as a random effect. Cox proportional-hazards models were used to examine the relationship between baseline OCT measurements to the risk of DR development, DR progression, and PDR development, adjusting for the aforementioned risk factors. A shared frailty model following a gamma distribution was used to adjust for inter-eye correlation.²⁸ Schoenfeld's global test was used to verify that the proportional hazards assumption was not violated. The additional predictive values of OCT measurements were determined by the C-statistic using the method described by Steyerberg et al.²⁹ We compared the C-statistics between multivariable models with and without the inclusion of OCT measurements. The likelihood ratio test was used for the model comparison. A P value < 0.05 was considered statistically significant. All statistical analyses were performed using R (version 3.5.3, R Foundation for Statistical Computing) and Python 3.12.1. 

Results

There were 1031 eyes with no DR or with NPDR out of 520 participants at baseline in the present study. Two hundred seventy-three participants were lost to follow-up due to the following reasons: withdrawal (60.8%, mainly during the COVID-19 period), death (30.0%), anti-VEGF injection (4.0%), loss of contact (3.3%), glaucoma (1.1%), and emigration (0.7%) (Fig. 1). Two hundred forty-seven participants completed at least five years of follow-up. After excluding eyes that met the exclusion criteria, we finally included 385 eyes from 215 participants with a median follow up of 6.2 years (range 5.0–7.7 years) in the final analysis. Specifically, 79 (44.4%) out of 178 eyes developed DR, 99 (25.7%) out of 385 eyes developed DR progression, and 38 (9.9%) out of 385 eyes developed PDR during the follow-up period. The baseline demographic characteristics of the included and excluded eyes/participants are presented in the Table. The comparison of baseline OCT measurements between included and excluded eyes is reported in Supplementary Figure S1. 

Figure 1.

 

Study flowchart. PRP, pan-retinal photocoagulation.

Table.

 

Baseline Demographic Characteristics of the Study Population With at Least Five Years of Follow-Up

 

Cox Proportional Hazards Models

Figure 2 shows the relationship of baseline m-GCIPL, m-RNFL, and p-RNFL thickness to the risk of DR development. Regarding the m-GCIPL in the multivariable analyses, except for the superonasal and superotemporal sectors, the remaining sectors exhibited significant associations with DR development, with HRs ranging from 1.373 (95% confidence interval [CI], 1.023–1.843) to 1.651 (95% CI, 1.246–2.187) (Fig. 2). In contrast, for the m-RNFL analyses, only thinner inferior m-RNFL thickness showed a significant association with DR development (HR, 1.348; 95% CI, 1.001–1.816) in the multivariable analyses. Regarding the p-RNFL analyses, HRs per 1-SD reduction of superior and inferior p-RNFL thicknesses were 1.441 (95% CI, 1.064–1.950) and 1.390 (95% CI, 1.064–1.950) in the multivariable analyses, respectively. Furthermore, we observed that thinner p-RNFL-1, p-RNFL-7, and p-RNFL-12 thicknesses were significantly associated with DR development in the multivariable models as shown in Supplementary Figure S2. The remaining clock-hour sectors did not exhibit any significant associations. 

Figure 2.

 

Relationships of baseline OCT measurements to the risk of DR development. The multivariable models were adjusted for baseline age, mean arterial blood pressure, diabetes mellitus duration, HbA1c, diabetic kidney disease, axial length, OCT signal strength, and disc area (disc area for p-RNFL only).

Figure 3 shows the relationship of baseline m-GCIPL, m-RNFL, and p-RNFL thicknesses to the risk of DR progression. Thinner mean and all sectorial m-GCIPL thicknesses were significantly associated with DR progression in both univariable and multivariable models, with HRs ranging from 1.277 (95% CI, 1.002–1.628) to 1.566 (95% CI, 1.285–1.909). Moreover, thinner mean, superonasal, inferior, and inferonasal m-RNFL thicknesses were significantly associated with DR progression in the multivariable models with HRs ranging from 1.279 (95% CI, 1.020–1.604) to 1.414 (95% CI, 1.149–1.739). Other sectorial m-RNFL did not present significant associations with DR progression. Furthermore, only thinner inferior p-RNFL thickness was associated with DR progression (HR, 1.236 [95% CI, 1.005–1.521]). In Supplementary Figure S2, p-RNFL-6, p-RNFL-7, and p-RNFL-12 were significantly associated with DR progression in both univariable and multivariable models. 

Figure 3.

 

Relationships of baseline OCT measurements to the risk of DR progression. The multivariable models were adjusted for baseline age, mean arterial blood pressure, diabetes mellitus duration, HbA1c, diabetic kidney disease, axial length, OCT signal strength, DR severity, and disc area (disc area for p-RNFL only).

Figure 4 shows the relationship of baseline m-GCIPL, m-RNFL, and p-RNFL thicknesses to the risk of PDR development. Thinner mean and all sectorial m-GCIPL thicknesses were associated with PDR development, with HRs ranging from 1.500 (95% CI, 1.131–1.989) to 2.331 (95% CI, 1.524–3.566). Regarding m-RNFL, thinner mean, superonasal, inferior, inferonasal, and inferotemporal m-RNFL thicknesses demonstrated significant associations with PDR development, with HRs ranging from 1.439 (95% CI, 1.010–2.050) to 1.863 (95% CI, 1.307–2.656). However, none of the p-RNFL measurements were significantly associated with PDR development, as shown in Figure 4 and Supplementary Figure S2. 

Figure 4.

 

Relationships of baseline OCT measurements to the risk of PDR development. The multivariable models were adjusted for baseline age, mean arterial blood pressure, diabetes mellitus duration, HbA1c, diabetic kidney disease, axial length, OCT signal strength, DR severity and disc area (disc area for p-RNFL only).

Predictive Discrimination

Figure 5 demonstrates the discriminative performance of the multivariable model with and without OCT measurements included for identifying DR outcomes. The model without OCT measurements achieved C-statistics of 0.661, 0.704, and 0.917 (i.e., dashed lines), respectively. The inclusion of OCT measurements significantly improved the discriminative performance, as shown in Figure 5. Specifically, the inclusion of inferior m-GCIPL achieved the highest C-statistics of 0.705 for DR development and 0.729 for DR progression among all OCT measurements, whereas the inclusion of inferotemporal m-GCIPL achieved the highest C-statistic of 0.930 for PDR development, as shown in Figure 5. 

Figure 5.

 

The discriminative performance of the multivariable model with and without the OCT measurements for identifying DR development, DR progression, and PDR development. The dash lines represented 0.661, 0.704, and 0.917 for DR development, DR progression, and PDR development, respectively. They represented the C-statistics that were obtained from the models without OCT measurement included. Asterisk indicates the significant difference of C-statistics between model without OCT measurement and with OCT measurement.

Subgroup Analysis of Eyes With Non-High Myopia and Eyes With High Myopia

Finally, we found that the significant associations with DR outcomes barely maintained in eyes with high myopia. Specifically, of the significant associations found in Figure 2, only the inferior m-GCIPL thickness remained significant in relation to the DR development (Supplementary Table S2). Additionally, of the significant associations found in Figures 3 and 4, only mean and inferior m-RNFL thicknesses remained significant associations with the DR progression and PDR development, respectively (Supplementary Tables S3 and S4). 

Discussion

We demonstrated that reduced retinal neuronal and axonal layers as measured by OCT, particularly in the macular region at baseline, were significantly associated with higher risks of DR development, DR progression, and PDR development during a median follow-up of 6.2 years. These associations were independent of previously established risk factors. Moreover, the inclusion of OCT measurements significantly improved the predictive discrimination for DR outcomes. 

Currently, once DR has developed, its progression is often foreseeable, resulting in visual loss or blindness if individuals are left undetected or untreated. The current effective treatments (i.e., anti-VEGF and retinal laser surgery) primarily focus on the advanced stages of DR (i.e., PDR and DME). Despite the effectiveness of these treatments, procedure-induced pain, worsening of macular edema, scarring, or retinal pigment epithelium atrophy are inevitable in some patients. Therefore identifying individuals who are at higher risks of DR development, DR progression, and PDR development is vital for early intervention to protect the retina from the deleterious effects of DM. Our study provided longitudinal evidence to support that OCT measurements can be considered as surrogates of DRN and could potentially be incorporated into an updated DRD staging classification to enhance DR management. 

Our findings are consistent with previous studies that suggested DRN may precede the microvascular changes characteristic of DR (i.e., DR development)¹³ and thinning of m-GCIPL is an independent risk factor for the progression of DR.³⁰ The first new finding is that we observed that DRN-induced thinning primarily leads to generalized thinning in m-GCIPL thickness that involves all sectors, followed by localized thinning in m-RNFL. This differs from thinning induced by other neurodegenerative disorders, which mainly occurs in the localized optic disc area. For example, Alzheimer's disease and glaucoma are characterized by preferential p-RNFL thinning in the superior and inferior quadrants, whereas Parkinson's disease involves preferential thinning in the temporal p-RNFL.³¹ Second, inferior m-GCIPL and inferotemporal m-GCIPL can improve predictive discrimination for DR development, DR progression, and PDR development. Retinal ganglion cells (RGCs) in the inferior hemiretina of the macular area project to the inferior quadrant of the optic disc, whereas the superior hemiretina mainly projects to the temporal quadrant. Also, the optic disc generally lies above the horizontal meridian.³² In other words, RGCs in the inferior macula have to travel longer to reach the optic disc compared to their superior counterparts. Given that the typical neuropathy associated with diabetes is length-dependent in nature,³³ this may imply the farthest nerve endings in the retina are where symptoms might develop first or are worse. In other words, RGCs in the inferior and inferotemporal areas are the two regions that are relatively vulnerable under the DRN compared to other regions. More studies are warranted to validate our findings. 

The second new finding is that thinner macular OCT measurements, especially m-GCIPL, were significantly associated with higher risk of PDR development. A multicenter study from diabetic eye screening programs reported the 10-year visual outcomes for patients who were referred with PDR. They found that 14.7% of patients developed moderate vision loss (i.e., best correct visual acuities [BCVA] of less than 70 ETDRS letters score) and 24% of PDR required vitrectomy.³⁴ Similar visual outcomes were also observed in the ETDRS study.³⁵ Therefore many studies have made great efforts to investigate how to identify patients who have a higher risk of developing PDR to prevent vison loss. Nwanyanwu et al.³⁶ recruited 4617 subjects with NPDR at baseline and found that every 1-point increase in HbA1c level was associated with a 14% (adjusted HR, 1.14 [95% CI, 1.07–1.21]) increased hazard of developing PDR. A similar finding was also observed in a meta-analysis, demonstrating that increased HbA1c had adjusted HRs ranging from 1.14 (95% CI, 1.07–1.21) to 1.43 (95% CI, 1.23–1.67), with one study reporting a nonsignificant result.³⁷ Except for systemic risk factors, Klein et al.³⁸ demonstrated that retinal venular curvature tortuosity is associated with risk of PDR (adjusted odds ratio [OR], 1.15 [95% CI, 1.01–1.30]). Our study provided new findings, suggesting that generalized thinner m-GCIPL is related to a higher risk of PDR development. Specifically, a 1-SD reduction in inferotemporal m-GCIPL thickness was associated with a 2.33-fold increased hazard of developing PDR. 

Myopia is a well-known protective factor against DR development,³⁹ DR progression,³⁹ and vision-threatening DR development⁴⁰ and is also associated with thinner retinal thicknesses (i.e., GCIPL and RNFL).⁴¹ Additionally, a population-based study also demonstrated that increasing severity of myopia was associated with a decrease in the odds ratio of DR.⁴² In our subgroup analysis stratified by eyes with non-high myopia and eyes with high myopia in relation to DR outcomes, we found that OCT measurements were barely maintained significant associations with DR outcomes among eyes with high myopia. A possible explanation is that eyes with high myopia have thinner retinal layers, and thus DM-induced thinning might have a limited effect on elongated retina, thereby having little impact on the DR outcomes. 

Currently, there is a growing appeal to update the diabetic retinal disease staging system as the traditional ETDRS scale only evaluates the vascular component of DR without regard to the role of the neural retina.^43,44 Our current study provides important evidence that the inclusion of OCT measurements, which reflect neural degenerative changes, can confer a higher degree of certainty in prognosticating the outcomes of DR.⁴⁵ Regarding the potential clinical implications, once thinning of m-GCIPL, m-RNFL, and p-RNFL in a related sector is identified at the time of DM diagnosis, ophthalmologists may recommend a more frequent follow-up plan, motivate lifestyle modifications, and consider exercising more aggressive glycemic control to reduce patients’ risk of developing DR. Given the availability of potential neuroprotective drugs, such as glucagon-like protein 1 (GLP-1),⁴⁶ and pigment epithelium derived factor,^47,48 which target neural abnormalities, our study might provide insights for clinical trials to use retinal neuronal thickness (e.g. m-GCIPL) in the most susceptible sectors as one of the clinical endpoints to evaluate the efficacy of treatment. 

The strengths of this study include a prospective study design, adjustment of known OCT measurement confounders (i.e., axial length, OCT signal strength, and disc area) and a relatively frequent follow-up schedule along with a longer follow-up duration, which improved the power of analysis. We acknowledge several limitations in our study. First, we only used macula- and optic disc–centered retinal fundus photographs to define DR severity and DR outcomes. Second, the measurement of retinal layers comprises a proportion of non-neuronal tissue components (e.g., stromal tissue).^49,50 Discriminating neuronal from non-neuronal components is not yet feasible with the latest OCT devices. Third, some study subjects withdrew from our study during the COVID-19 period, which may have introduced bias into our analysis. Fourth, we did not include the development of DME as one of the outcomes in the current study, as the edematous thickness can affect the accurate measurement of OCT thickness to some extent. Fifth, although medication usage may influence DR progression,⁵¹ we did not collect such data from our study subjects and include in our statistical analysis to account for potential bias. 

In conclusion, thinner OCT measurements, particularly in the macular region at baseline, were significantly associated with higher risks of DR development, DR progression, and PDR development. These OCT measurements significantly improve predictive discrimination for detecting DR outcomes, particularly in eyes without high myopia. 

Acknowledgments

The authors thank Thomas W. Gardner (Ophthalmology & Visual Sciences, University of Michigan Medical School, USA) for reviewing and commenting on the manuscript. 

Supported by CUHK Direct Grant (Project Code: 4054419 & 4054487) 

Disclosure: Z. Tang, None; D. Yang, None; T.X. Nguyen, None; S. Zhang, None; D. Fang, None; V.T.T. Chan, None; C.C. Tham, None; E.H. Sohn, None; K.K. Tsang, None; C.Y.K. Wong, None; V.W.K. Hui, None; A.H.Y. Yu, None; J.T.W. Lam, None; C.K.M. Chan, None; T.Y.Y. Lai, None; S.K.H. Szeto, None; C.Y. Cheung, None