Foveal eversion patterns in diabetic macular edema

The aim of the present study was to describe foveal eversion patterns in diabetic macular edema (DME) and to assess their relationship with the course of the disease and the outcome. The study was designed as prospective, observational, with two years of follow-up. DME patients were divided in two groups, one treated by combined anti-VEGF injections and dexamethasone (DEX) implants, and the other treated by fluocinolone acetonide (FAc) implant with additional anti-VEGF retreatments if needed. Main outcome measures were foveal eversion prevalence, foveal eversion patterns, best-corrected visual acuity (BCVA), central macular thickness (CMT), structural OCT metrics, number of intravitreal injections. One hundred and forty-six eyes (146 patients; 80 males; mean age 67 ± 8 years) affected by already treated DME, with 84 eyes treated with anti-VEGF/DEX treatments (mean of 10 ± 3 injections) and 62 treated with FAc implant. Looking at the treatments administered before the inclusion into the study, 84 eyes (58%) were treated with anti-VEGF injections, whereas 62 eyes (42%) underwent a combination of anti-VEGF and corticosteroids implants. DME eyes showed statistically significant improvements of LogMAR BCVA and CMT over the 2-year follow-up. Foveal eversion was found in 83 eyes (57%), categorized as follows: Pattern 1a (16;19%); Pattern 1b (22;27%) and Pattern 2 (45;54%). BCVA improvement was detected in all the subgroups, excepting for Pattern 2, which showed also significantly worse structural OCT parameters. Pattern 1b and Pattern 2 were characterized by significantly higher prevalence of persistent DME (64% and 89% of cases, respectively). Foveal eversion patterns were correlated with progressively worse DME outcome. Foveal eversion may be associated to the loss of foveal homeostasis, with consequent poor response to intravitreal treatments and worse DME outcome.

www.nature.com/scientificreports/ and any ophthalmic or systemic disease potentially affecting the results of the study, follow-up interval major than 4 months within the entire observational study window. A two-year follow-up was planned for all the eyes included.
All the patients underwent complete ophthalmologic examination, including best corrected visual acuity (BCVA) evaluated by standard ETDRS charts, anterior and posterior segment slit lamp evaluation, and Goldmann applanation tonometry. Structural OCT examination (Spectralis HRA, Heidelberg Engineering; Heidelberg, Germany) was performed at each follow-up visit, with radial, raster and dense scans with a high number of frames (ART > 25), and enhanced depth imaging (EDI) to highlight choroidal structures. Follow-up examinations were planned according to ophthalmologists' discretion and treatment strategy.
We analyzed eyes with and without foveal eversion, understood as the complete convex profile of the central fovea, separately. Taking cases displaying foveal eversion first, we considered three different patterns. Pattern 1 was defined as the presence of foveal eversion with mixed reflectivity, intraretinal, vertical material, developing in the context of the hyporeflective cystic space. We further subdivided Pattern 1, considering the columnar organization of the mixed reflectivity vertical signal (1a) or multiple fan-like lines (1b). Pattern 2 was defined as the presence of foveal eversion with only a hyporeflective intraretinal cystic signal. The DME categorization is shown in Fig. 1. Two independent graders (AA, AS) measured the horizontal thickness of the vertical mixed reflectivity material, assessing the thickness of all the mixed reflectivity lines included in the central 750 µm for both Pattern 1a and Pattern 1b. The grading of foveal eversion patterns was mandatorily done on a horizontal structural OCT b-scan image, centered to the fovea. DME eyes never showing foveal eversion over the entire study window were included in the "no foveal eversion" group. DME eyes were treated with anti-VEGF (Ranibizumab 0.5 mg), dexamethasone (DEX) implant (Ozurdex ® ; Allergan inc.), or fluocinolone acetonide (FAc) 0.19 mg intravitreal drug-delivery system (ILUVIEN ® ; Alimera Sciences, Inc., Alpharetta, GA, USA), at the ophthalmologists' discretion. Anti-VEGF injections and DEX implants were administered following a pro-re-nata regimen. The criterion employed to switch the treatment was the OCT-based evidence of poor response to the first molecule (arbitrarily considering DME reduction < 25%). FAc implant was administered according to the Italian guidelines, consisting of the following criteria: refractory DME, previously treated with at least one DEX implant, with no evidence of "cortico-responder"-related phenomena, in mandatory pseudophakic eyes. FAc-treated eyes might undergo further anti-VEGF treatment if displaying structural OCT evidence of poor response (in this case too, regarded as DME reduction < 25%).  www.nature.com/scientificreports/ To make all the eyes reliably comparable in terms of image quality and visual changes, being bound by the Italian criteria for FAc implant, we considered mandatory pseudophakia for eyes treated by anti-VEGF/DEX too. Likewise, we chose to exclude naïve DME eyes to further guarantee a reliable statistical comparison between anti-VEGF/DEX and FAc implant eyes.
The main outcome measure was the assessment of the relationship between presence/absence of foveal eversion, and the clinical course of DME. The secondary outcome was the assessment of the relationship between each foveal eversion pattern (1a, 1b and 2), and both the final outcome (DME resolving or recurrence) and the presence of persistent DME, understood as DME enduring throughout the 24-week examination 15 .
We considered the parameters measured at baseline and at the last follow-up. Two independent blinded graders performed all the measurements (AA, AS). Interclass correlation coefficient (ICC) was measured to assess the agreement between the two graders through a two-way random-effects model. For the statistical analyses we included the following variables: age, gender, systemic arterial hypertension, type and duration of diabetes mellitus (DM), glycate hemoglobin values (HbA1c), DR type (NPDR/PDR), previous history of vitrectomy, previous panretinal photocoagulation (PRP), previous nature and number of treatments (anti-VEGF, intravitreal corticosteroids), baseline features (BCVA and CMT), retinal and/or choroidal HF > 15, ERM, DRIL, ELM/EZ status (normal/interrupted/absent) and SRF. For the FAc-treated eyes, we also considered the presence of any additional anti-VEGF treatments (yes/no) and their number, administered at the ophthalmologists' discretion.
All the statistical analyses were performed by means of the SPSS software package (SPSS, Illinois, USA). In our statistical models, we evaluated the normality distribution of each variable with frequency histograms and quantile-quantile plots. Descriptive statistics of continuous variables were reported as mean ± standard deviation, whereas frequency and proportions were described as categorical variables. Continuous variables were statistically evaluated using the two-tailed T-test. One-way ANOVA analysis assessed the differences between eyes displaying foveal eversion (considering all the foveal eversion patterns separately) and eyes without foveal eversion. The multiple testing required the introduction of Bonferroni correction in order to provide multiple comparisons. Tau-Kendall correlation analysis was adopted to assess the relationship between the various parameters. Statistical significance was set at p < 0.05.

Results
Overall, 146 eyes of 146 patients affected by DME (80 males; mean age 67 ± 8 years) were recruited for the study. Eighty-four DME eyes (45 males; mean age 67 ± 6 years) were treated with anti-VEGF injections and DEX implants, whereas 62 eyes (35 males; mean age 68 ± 8 years) were treated with FAc implant.
The mean number of intravitreal injections before inclusion in the study was 12 ± 3. The first DME diagnosis was performed 2 ± 1 years before our baseline visit. DME eyes treated by anti-VEGF/DEX received a mean of 11 ± 4 injections over the entire follow-up. Sixty-nine out of 84 eyes (82%) underwent treatment, switching over the 2-year follow-up. Twenty-one out of 62 eyes (34%) in the FAc-treated group required additional anti-VEGF treatments (3 ± 1 injections). Clinical data are shown in Table 1. DME eyes showed statistically significant improvements of LogMAR BCVA and CMT over the 2-year followup (p < 0.05) ( Table 1). No significant differences were detected looking at each administered drug (p > 0.05).
We found foveal eversion in 83 out of 146 eyes (57%). DME eyes with and without foveal eversion started with similar baseline BCVA values (p > 0.05) ( Table 2). No significant differences among subgroups were found looking at DM Type, DR Type and previous treatments (all p > 0.05). No significant difference was detected in the number of intravitreal treatments administered in each subgroup (p > 0.05). DME eyes showing foveal eversion were categorized as follows: Pattern 1a (23, 32%) (  (16, 19%); Pattern 1b (22; 27%) and Pattern 2 (45, 54%) at the end of the follow-up. One case of DME without foveal eversion is shown in Fig. 5. Age and disease duration were similar among groups (all p > 0.05). Pattern 2 showed the worst baseline BCVA (p < 0.05), whereas Pattern 1a, Pattern 1b and eyes without foveal eversion started with similar BCVA values (p > 0.05). All the subgroups, except Pattern 2, underwent statistically significant BCVA improvement at the end of the follow-up (p < 0.01). The entire cohort of eyes also showed significant CMT recovery (p < 0.01). The distribution of all the other parameters are extensively reported in Tables 2 and 3. It is worth noting that the disappearance of the EZ/ELM was significantly higher in Pattern 2 (p < 0.01). SRF had disappeared in all eyes by the end of the follow-up. The percentage of eyes showing resolved DME was significantly higher in eyes without foveal eversion (p < 0.01). Most of the eyes included in Pattern 1a and foveal eversion-free subgroups were characterized by recurrent DME (p < 0.01), whereas most of the eyes included in Pattern 1b and Pattern 2 featured persistent DME (64% and 87% of cases, respectively; p < 0.01).
In the three patterns of foveal eversion correlated with Hb1Ac level, we found that the higher the Hb1Ac value, the higher the foveal eversion pattern (Tau-Kendall coeff. 0.455; p < 0.001). Foveal eversion patterns showed negative correlation with the thickness of the mixed reflectivity vertical materials, Pattern 1a being the thickest (Tau-Kendall coeff. 0.732; p < 0.001). Furthermore, we found that the thinner the mixed reflectivity material, the worse the DME prognosis (looking at DME resolution, recurrence, or persistence) (Tau-Kendall coeff. 0.644; p < 0.001). Moreover, foveal eversion patterns significantly correlated with EZ/ELM status (Tau-Kendall coeff. 0.402; p < 0.001). As expected, the latter significantly correlated with BCVA values (Tau-Kendall coeff. 0.536; p < 0.001). The presence of SRF showed a moderate correlation with DME persistence (Tau-Kendall coeff. 0.398; p < 0.01); however, it had no significant influence on foveal eversion patterns (p > 0.05). Similarly, we found no significant correlations between foveal eversion patterns and all the other systemic and structural OCT parameters investigated (all p > 0.05).
There was high inter-grader agreement in the foveal eversion pattern classification (ICC 0.95; p < 0.01). The overall ICC regarding the other parameters analyzed was 0.94 (range 0.89-0.96; p < 0.01). www.nature.com/scientificreports/ Consecutive follow-up examinations, starting from a DME without foveal eversion, evolving into a DME with foveal eversion, prompted us to put forward the possible pathogenic mechanism shown in Fig. 6.

Discussion
Foveal eversion is a relatively new structural OCT biomarker and was previously related to worse DME outcomes with higher prevalence of persistent DME 15,16 . In the present study, we carried out a more thorough investigation of this parameter, in an attempt to further categorize DME eyes according to their morphological features. We detected three different patterns of foveal eversion. Pattern 1 was characterized by the presence of vertical, transretinal, mixed reflectivity material, associated with a cystoid edema, organized as a single thicker columnar unit (Pattern 1a) or in multiple thinner lines (Pattern 1b). Pattern 2 was characterized by the absence of this mixed reflectivity material, with only a hyporeflective cystoid intraretinal signal. Although the overall prevalence of persistent DME reported in our study (almost 50% of the cases) was in line with previous investigations 15,17 , the presence of foveal eversion was significantly associated with persistent DME, reaching the highest prevalence in Pattern 1b (64% of the cases) and in Pattern 2 (87% of the cases). In contrast, Pattern 1a and eyes without foveal eversion were mainly characterized by the recurrence of DME. www.nature.com/scientificreports/ In our previous investigation, we speculated about the eversion of the foveal profile's possible association with Muller cell impairment 16 . The theoretical basis for this hypothesis lies in the fact that Muller cells are a key structural and homeostatic component of the human retina 18,19 , being directly involved in DME pathogenesis and responsible for the morpho-functional integrity of the foveal region 18,[20][21][22] . In the current study, we hypothesize that Muller cells might be variably impaired, as clinically expressed by three different patterns of foveal eversion. In particular, the absence of the vertical transretinal mixed reflectivity material characterizing Pattern 2, interpreted as the total disruption of foveal Muller cells, was significantly associated with the worst morpho-functional outcome. Conversely, the presence of this vertical signal, detected either as thick column (Pattern 1a) or thinner lines (Pattern 1b) might be interpreted as the structural OCT sign of Muller cells partial sparing. The overall thickness of the vertical transretinal signal was significantly associated with better morpho-functional outcome. The impairment of the foveal region occurred independently of the administered drug, the duration of the disease and the number of injections. This might mean that other factors in the multifaceted DME pathogenesis might be responsible for the progressive loss of foveal integrity 23 . The eversion of the fovea might be the consequence of gliotic damage involving the Muller cells 22 , together with a different inflammatory profile typifying this category of DME eyes 24 . The most intriguing aspect of the present investigation is that different patterns of foveal eversion may allow to identify the prognostic information about the patient's morpho-functional recovery after treatments. In this perspective, foveal eversion might represent a feasible OCT-based retinal biomarker, potentially offering the basis for an efficient and personalized DME management.
However, we must acknowledge that further, multidisciplinary studies are needed in order to draw more definite conclusions about the role of foveal eversion in DME pathogenesis.
We are aware that our study labors under several limitations, related firstly to the relatively low number of DME eyes and the strict follow-up. Furthermore, our investigation was conducted exclusively on eyes with a history of DME; no dedicated information about naïve DME is available. We made this choice so we could include eyes treated by FAc implant, and thus try to provide a comprehensive scenario of foveal eversion's role in DME treated with all the current therapeutic strategies. Future prospective studies should focus on assessing foveal eversion's features, as well as its role in naïve DME eyes. Moreover, since most of the eyes underwent treatment switching, we were unable to provide data about the part played by foveal eversion in the decisions regarding treatment switching. Furthermore, future interventional trials are warranted to evaluate the effect of customized treatments, based on foveal eversion detection, on the clinical outcome of DME. Lastly, we proposed a possible pathogenic evolution of DME from a condition in which foveal eversion is absent to one in which it is present. However, since our considerations were mainly based on structural OCT follow-ups, histological validation is required to draw more definite conclusions about the progressive development of foveal eversion in DME.
In conclusion, we confirmed the role of foveal eversion as a feasible and useful biomarker requiring validation in a DME clinical setting. The everted fovea is likely related to the impairment of the Muller cells, with consequent loss of foveal homeostasis and worse DME outcome. In the interest of designing ever more optimized and personalized treatment strategies, foveal eversion may be considered a useful biomarker in DME eyes. Further studies are needed to provide a more thorough assessment of the significance of foveal eversion in the pathogenesis of DME. Table 2. Foveal eversion sub-analysis in DME eyes and clinical parameters. DME eyes are divided accordingly with the presence and type of foveal eversion, independently from the treatments. BCVA best corrected visual acuity, DME diabetic macular edema.