Ocular diseases of high socio-economic relevance such as macular edema secondary to diabetic retinopathy or retinal vein occlusion, are associated with elevated expression of vascular endothelial growth factor-A (VEGF-A) in the vitreous, which increases the permeability of retinal endothelial cells in vivo and in vitro1,2,3,4,5,6,7. Therapeutic options targeting VEGF-A include intravitreal injections of the Fab fragment ranibizumab (Lucentis) or of brolucizumab (Beovu), a single-chain variable fragment recently approved for treatment of age-related macular degeneration and diabetic macular edema8,9,10,11. Both therapeutic proteins bind to all relevant splice variants of VEGF-A with high affinities12,13,14. Although these therapies are generally well tolerated, rare but severe adverse effects, i.e. retinal vasculitis or retinal vein occlusion, have been observed after repetitive intravitreal injections of Beovu15,16,17,18,19,20. The processes leading to these undesirable effects still remain unclear despite the potential roles of humoral and cellular immune response and particularly of specific anti-drug antibodies being broadly discussed18,19,20. In this context, it is very important to know whether Beovu itself might impair the barrier formed by the monolayer of retinal endothelial cells which are responsible for maintaining a tight inner blood-retina-barrier in vivo.

Using the well-established in vitro model of immortalized microvascular endothelial cells from the bovine retina (iBREC), developed, validated and maintained in our laboratory, we investigated whether extended exposure to Beovu, the pharmaceutical formulation of brolucizumab, harms the stable barrier formed by these cells5,6,21. Depending on the nature of the interfering agent, disturbances of the tight barrier formed by a confluent monolayer of primary or immortalized retinal endothelial cells of various, including human origins correlate with reduced expression or appearance at the plasma membrane of the tight junction (TJ) proteins claudin-1 and claudin-5, but can also be due to down-regulation of the fibronectin receptor subunit CD29/integrin β1 and the tetraspanin CD9/TSPAN294,5,6,7,22,23,24. TJ-proteins together with adherens junction (AJ) proteins, e.g. vascular endothelial cadherin (VEcadherin), regulate paracellular flow, whereas the fibronectin receptor, formed by the subunits CD29/integrin β1 and CD49e/integrin α5, in complex with CD9/TSPAN29 mediates adhesion of iBREC to the extracellular matrix25,26,27,28. In addition to its capability to block VEGF-A-stimulated proliferation of endothelial cells (EC), brolucizumab prevents and at least transiently reverts the VEGF-A-induced dysfunction of the barrier formed by iBREC to a similar extend as ranibizumab4,13,14,29. To evaluate whether blocking VEGF-A-induced signal transduction per se impairs the integrity of the barrier formed by iBREC, we studied the effects of nintedanib or tivozanib which at the indicated concentrations specifically inhibit the tyrosine kinase activities of VEGF receptors (VEGFR) VEGFR1 (nintedanib: 34 nM, tivozanib: 30 nM), VEGFR2 (nintedanib: 21 nM, tivozanib: 6.5 nM) or VEGFR3 (nintedanib: 13 nM, tivozanib: 15 nM), respectively30,31. At concentrations of 10 nM treatment with both inhibitors indeed prevent and at least transiently revert the VEGF-A165-induced dysfunction of the barrier formed by an iBREC monolayer5,6,22,32. The VEGF-A-binding Fab fragment ranibizumab (Lucentis®) was included in our analyses as an already widely used reference protein not significantly affecting retinal vessels after its intravitreal application.


Treatment with Beovu lowered the stability of the barrier formed by iBREC

A confluent monolayer of iBREC—cultivated in cell culture medium adapted to the special needs of microvascular EC—was exposed to Beovu (final concentration: 1 mg/ml brolucizumab, Fig. 1a, b) or Lucentis (final concentration: 100 µg/ml ranibizumab, Fig. 1a, c) for up to seven days. The concentrations were chosen to correspond to those that can be reached in vivo by intravitreal injection8,9. As a measure of barrier stability, we determined the cell index of iBREC cultivated on gold electrodes, a method which allows the reliable detection of even subtle and transient changes: A high cell index is indicative of a tight barrier and—vice versa—an impairment of the integrity of the barrier is associated with a decline of the cell index5. Only Beovu (Fig. 1b, left panel), but not Lucentis (Fig. 1c, left panel), induced a slight, but significant transient decline of cell index values early after addition to the cultivated iBREC, followed by a more pronounced and sustained lowering effect, becoming evident about 1.5 days later (Fig. 1b, right panel). The extend of the Beovu-induced disturbance varied, often but not always depending on the lot of the pharmaceutical formulation. In contrast, cell index values of Lucentis-treated iBREC remained stable, reflecting those of control cells (Fig. 1c, right panel). Blocking VEGF-A165-induced signaling through specific inhibition of the VEGFR2 with 10 nM nintedanib (Fig. 1d, upper panel) or with 10 nM tivozanib (Fig. 1e) also did not lead to similarly low cell index values during prolonged exposure; interestingly, even slightly higher cell index values were observed with 10 nM nintedanib. Treatment of the cells with the very high concentration of 100 nM nintedanib did not change the cell index at all (Fig. 1d, lower panel).

Figure 1
figure 1

Beovu lowered the cell index of iBREC. iBREC were cultivated on gold electrodes until confluency was reached before (a), (b) Beovu (n = 7), (a), (c) Lucentis (n = 6), (d) nintedanib (n = 7), (e) tivozanib (n = 3), or phosphate-buffered saline (control, blue curves; n ≥ 6) was added and the cell index recorded continuously as a measure of barrier function. Cell index values were normalized in relation to those measured immediately before addition of the VEGF-A antagonists and are shown as means with standard deviations. Data were analyzed with two-way ANOVA followed by Sidak’s multiple comparison test. (a), (b) Beovu significantly lowered the cell index values transiently early after its addition and persistently from ~ 1.5 days of exposure until the end of the experiment over seven days. (a), (c) Cell index values were not different in the presence of Lucentis. (d) Nintedanib (10 nM) significantly enhanced cell index values late after its addition, whereas the higher concentration (100 nM) had no effect. (e) Cell index values were also not changed by 10 nM tivozanib.

Beovu did not induce cellular stress or an inflammatory response

The concentrations of cytokines expressed by microvascular EC and associated with disturbance of their barrier (i.e. VEGF-A, interleukin (IL)-8 or TNFα), cellular stress (i.e. IL-6), or with an inflammatory response (i.e. IL-1β) in cell culture supernatants of iBREC treated with Beovu for two hours (→ early decline of the cell index) or thirty hours (→ persistently lowered cell index values) were measured, but their concentrations were always below the minimal amounts detectable by the ELISAs used (Table 1). Even after prolonged exposure for several days, the cells did not secrete IL-6, IL-8, TNFα or VEGF-A (Table 1).

Table 1 Measured cytokine concentrations in cell culture supernatants.

iBREC expressed less claudin-1 after long-term treatment with Beovu

In order to better understand how exposure of iBREC to Beovu resulted in lower cell index values, we measured expression of proteins identified as determinants of barrier stability. Obvious candidate proteins that might play a role were those involved in the regulation of paracellular flow (i.e. claudin-1, claudin-5, and VEcadherin; Fig. 2a, c) and those mediating interaction of the cells with the extracellular matrix (i.e. CD9/TSPAN29, CD29/integrin β1 and CD49e/integrin α5; Fig. 2b, c). Effects of Beovu or Lucentis on their expression were analyzed after the cells had been exposed to the VEGF-A-binding proteins for one day, five or eight days. Treatment with both VEGF-A antagonists resulted in similarly changed claudin-1 expression measured one day and five days after their addition: After one day the levels were significantly lower, but significantly higher after five days compared to those of control cells (Fig. 2a, top panels). However, longer exposure (for eight days) to Beovu led to slightly, but significantly lower amounts of claudin-1 compared to those of control cells or cells similarly treated with Lucentis. Interestingly, exposure of iBREC to Beovu for five days also resulted in stronger expression of VEcadherin (Fig. 2a, bottom panels) and claudin-5 (Fig. 2a, middle panels), whereas only the expression of the latter was also higher after treatment with Beovu for the same period. In contrast, after cultivation of iBREC for only one day or as long as eight days with Beovu or Lucentis the expression levels of VEcadherin and claudin-5 were not different to those of control cells. Expression of CD49e/integrin α5 (Fig. 2b, top panels), CD9/TSPAN29 (Fig. 2b, middle panels) and CD29/integrin β1 (Fig. 2b, bottom panels) was not found to be changed in Lucentis-exposed iBREC at any time point investigated, but those of CD49e/integrin α5 were significantly lower and those of CD29/integrin β1 significantly higher in iBREC treated with Beovu for one day.

Figure 2
figure 2

Long-term treatment with Beovu but not with Lucentis resulted in lower expression of tight junction protein claudin-1. After having reached a confluent monolayer, iBREC were exposed to Beovu or Lucentis, and cell extracts were prepared one day, five or eight days later for Western blot analyses. To compare antigen-specific signals from extracts of inhibitor-treated cells to the hypothetical value of 1.0 of normalized signals from control experiments we used the Wilcoxon signed rank test, and the Mann–Whitney U test for comparing signals from extracts of inhibitor-treated cells. Scatter blots show means and standard deviations; one dot represents the analyte-specific signal from one of multiple independent Western blot analyses. (a) Claudin-1 levels were similarly changed after exposure to both VEGF-A antagonists for one day or five days. After treatment with Beovu for eight days, expression of claudin-1 was significantly lower compared to control cells or Lucentis-treated iBREC. Amounts of VEcadherin were higher after exposure to Beovu for five days, those of claudin-5 after treatment with both VEGF-A inhibitors for the same period of time. (b) After treatment with Beovu for one day, levels of CD49e/integrin α5 were significantly lower and those of CD29/integrin β1 significantly higher compared to those of control cells. (c) Representative cropped images of Western blot analyses. For original images, please refer to supplemental materials (Fig. S1).

We also investigated by immunofluorescence stainings whether treatment of iBREC with Beovu or Lucentis for five days changed the subcellular localization of proteins involved in the regulation of paracellular flow (Fig. 3). Neither Beovu nor Lucentis markedly changed the prominent presence of VEcadherin and claudin-5 at the plasma membrane, but fractions of cells with claudin1-specific staining of the plasma membrane were significantly smaller after their exposure to Beovu (98.4 ± 2.4% for control; 95.4 ± 3.8% for Lucentis, 89.9 ± 7,7% for Beovu with p = 0.0004 compared to control and p = 0.0141 compared to Lucentis).

Figure 3
figure 3

Exposure of iBREC to Beovu but not to Lucentis changed plasma membrane localization of TJ-protein claudin-1. A confluent monolayer of iBREC was treated with Beovu or Lucentis for five days before cells were fixated, and claudin-1 (upper panel), claudin-5 (middle panel) or VEcadherin (lower panel; all in red) were visualized by immunofluorescence stainings with specific antibodies. Lucentis did not markedly change the presence of any of the investigated proteins at the plasma membrane. Whereas subcellular localization of claudin-5 or VEcadherin remained largely unchanged also in the presence of Beovu, the staining specific for claudin-1 was weaker after exposure of the cells to the VEGF-A antagonist.

Polysorbate-80 but not polysorbate-20 affected the cell index, an indicator of iBREC barrier integrity

We also investigated whether components of the pharmaceutical formulation of Beovu might affect the behavior of iBREC. Polysorbate-80 and polysorbate-20—components of Beovu or Lucentis, respectively—seemed obvious candidate substances and we determined their possible effects on the cell index of an iBREC monolayer at concentrations close to 0.0002% polysorbate-80 or 0.0001% polysorbate-20 that can be reached by intravitreal injection of 50 µl of the medications. At the lowest tested concentration of 0.0001%, exposure of the cells to polysorbate-80 resulted in a slight but significant decline of the cell index visible ~ 38 h after its addition (Fig. 4a) and this effect increased considerably when higher concentrations of polysorbate-80 were applied. In contrast, in the investigated range of concentrations polysorbate-20 did not affect cell index values (Fig. 4b). Western blot analyses also revealed that significantly less claudin-1 or VEcadherin were expressed by iBREC exposed to 0.002% or 0.0002% polysorbate-80 for two days (Fig. 5a, c); levels of claudin-5 were lower only after treatment with 0.0002% polysorbate-80. Interestingly, expression of claudin-1 and claudin-5 remained stable after similar treatment of the cells with polysorbate-20, but VEcadherin was then strongly down-regulated (Fig. 5b, c).

Figure 4
figure 4

Polysorbate-80 lowered the cell index of iBREC. (a) Polysorbate-80 (n ≥ 6) or (b) polysorbate-20 (n ≥ 6) were added to a confluent monolayer of iBREC cultivated on gold electrodes and the cell index was recorded continuously as a measure of barrier function. Cell index values were normalized in relation to those measured immediately before addition of the detergents and are shown as means with standard deviations. Data were analyzed with two-way ANOVA followed by Sidak’s multiple comparison test. (a) Concentrations of polysorbate-80 ≥ 0.0001% significantly and persistently lowered cell index values. (b) Polysorbate-20 had no effect on the cell index.

Figure 5
figure 5

Less claudin-1 and VEcadherin were expressed after exposure of iBREC to polysorbate-80. iBREC were treated with (a) polysorbate-80 or (b) polysorbate-20 for two days before cells were harvested for preparation of cell extracts and subsequent Western blot analyses. We used the Wilcoxon signed rank test to compare antigen-specific signals from extracts of polysorbate-treated cells to the hypothetical value of 1.0 of normalized signals from control experiments and the Mann–Whitney U test for comparing signals from extracts of polysorbate-exposed cells. Scatter blots show means and standard deviations; one dot represents the analyte-specific signal from one of multiple Western blot analyses. (a) Levels of claudin-1 and VEcadherin were significantly lower after exposure to polysorbate-80. (b) Treatment with polysorbate-20 led to significantly lower expression of VEcadherin, but did not change the amount of claudin-1. (c) Representative cropped images of Western blot analyses (for original images see Fig. S2).


To evaluate the potential relevance of an alternative, non-immunological explanation for the observed adverse effects of Beovu, we studied its impact on the retinal endothelium itself using the well-established in vitro model of immortalized bovine retinal endothelial cells (iBREC). Although of non-human origin, these cells respond to various stimuli in a very similar way as their human counterparts21,33,34,35,36. VEGF-A165 also similarly increases the permeability of primary human or bovine retinal EC and of iBREC, always accompanied by lower expression of TJ-protein claudin-1 and higher levels of the plasmalemma vesicle-associated protein4,5,7,29,32,37,38. Accordingly, brolucizumab (Beovu) and ranibizumab (Lucentis) prevent and revert the VEGF-A165-induced dysfunction of the barrier formed by iBREC4,5,29. This cell line is also free of contaminating cells of other types potentially interfering with characteristic behavior of the retinal EC—most importantly in this context—free of cells of the immune system21. iBREC were always cultivated in a culture medium adapted to the special needs of microvascular EC supplemented with fetal bovine serum ensuring optimal conditions that allow to reveal specific responses23. Confirming appropriate cultivation, iBREC do not secrete the marker of cellular stress IL-6 under these conditions (see Table 1)39. Because adverse effects of Beovu are typically observed after multiple injections when efficient inhibition or even complete blocking of the growth factor can be assumed, we decided to measure its potential effects in the absence of VEGF-A. To mimic the situation of typical patients constantly exposed to the drug, we not only assessed short-term effects but focused on a potential impact of prolonged treatment of the endothelial cells for several days.

Monitoring the cell index of iBREC cultivated on gold electrodes allows detection of subtle and transient changes of the formed barrier due to de-regulated paracellular flow and compromised adhesion of the cells5,6,22,40,41. iBREC showed a bi-phasic response to Beovu: A transient weak, albeit significant lowering of the cell index evident within a few hours was followed by a more pronounced and persistent decline, indicating that the barrier consisting of the cell monolayer was then less compact and likely also more penetrable. Because similar changes were not observed in analogous experiments with Lucentis, nintedanib or tivozanib, blocking of VEGF-A-induced signaling—even for a long time—seems not to be harmful to retinal EC in general. Other ocular cell types, such as cells of the retinal pigment epithelium, Muller cells or retinal ganglion cells, also tolerate inhibition of VEGF-A-induced signaling even over extended periods of time6,42,43,44. That nintedanib affected the activity of a protein kinase involved in the regulation of cellular permeability but not yet identified as a target of this inhibitor seems to be the most reasonable explanation for the observed elevated cell index during prolonged exposure of iBREC to it (Fig. 1d).

The molecular basis of the observed barrier disturbance by Beovu remains unclear, but activation of signaling pathways involved in the regulation of permeability of (retinal) endothelial cells and triggered by TNFα, IL-6, or IL-8 is rather unlikely, as Beovu-exposed iBREC did not secrete any of these cytokines45,46,47. Also, relevant amounts VEGF-A were not detected in cell culture supernatants of unchallenged or Beovu-treated iBREC, which confirms our previous finding that confluent retinal EC do not secrete this growth factor23. We therefore conclude that the low cell index values observed in the presence of Beovu are not caused by the action of residual bovine VEGF-A165 not completely inactivated by binding to the VEGF-A antagonist. Although the exact binding affinity of brolucizumab to bovine VEGF-A has not been determined, strong complex formation can be assumed due to the high homology of human and bovine VEGF-A.

Immunofluorescence stainings revealed that iBREC exposed to Beovu or Lucentis for five days still expressed a confluent monolayer and that presence at the plasma membrane of claudin-5 and VEcadherin was not affected. Results of our protein expression studies suggest a multiphasic response of the cells to Beovu: After exposure of iBREC to Beovu for one day, expression levels of the subunits of the fibronectin receptor CD49e/integrin α5 and CD29/integrin β1 had changed in opposite ways and it could well be that as a consequence of the unbalanced stoichiometry of the two subunits, less of the dimer is generated and thereby adhesion of the cells to the extracellular matrix (on the gold electrodes) is weakened27. Expression of low amounts of the TJ-protein claudin-1—correlating with a leaky iBREC monolayer—was observed after eight days of exposure to Beovu, also pointing to de-regulated paracellular flow4,5. However, the surprising observation that claudin-1 expression was found to be similarly changed by both VEGF-A antagonists one day and five days after their addition but only in the case of Lucentis normalized three days later, suggests a more complex interaction between the cells and the inhibitors. That de-regulated expression of VEcadherin was evident only in an intermediate phase of exposure for five days to Beovu supports this assumption. However, less claudin-1 located at the plasma membrane of iBREC exposed to Beovu for five days in spite of its generally increased expression might cause the observed low cell index as it obviously precedes the overall decrease of claudin-1 expression during prolonged exposure to Beovu.

The surfactants polysorbate-80 and polysobate-20, components of the formulations of Beovu or Lucentis, respectively, are ethoxylated sorbitans containing the same total number of twenty polyoxyethylene units—mostly esterified to either oleic acid or lauric acid, respectively; commercial preparations usually are mixtures also containing certain amounts of esters with other fatty acids48. Such surfactants improve stability of proteins and prevent formation of aggregates, but their degradation—either by residual enzymatic activity or by autooxidation—may result in precipitation of the therapeutic proteins in formulations48,49,50. Intravitreal injection of polysorbate-80 into the rabbit eye did not result in ocular inflammation, but possible disturbances of the inner blood-retina barrier might have been disregarded51. Therefore, it is of interest that polysorbate-80 weakly, albeit significantly, destabilized the barrier formed by iBREC even at concentrations achievable by intravitreal injection of Beovu: A significantly decreased cell index was accompanied by lower expressions of claudin-1 and VEcadherin, consistently indicating dysfunction of the barrier4,5,6,7,25,26. In contrast to the effects of polysorbate-80, the barrier formed by iBREC remained stable in the presence of polysorbate-20: Although less VEcadherin was then expressed, constantly high cell index values and stable expression of claudin-1 and claudin-5 were observed. These results of in vitro experiments might be of particular relevance when Beovu is injected into eyes with a relatively low volume of vitreous fluid or after repeated intravitreal injections with short intervals. Anti-drug antibodies against brolucizumab have been described and they appear to be more common than those against ranibizumab18,20. Intravitreally injected brolucizumab might cross the inner blood-retina barrier—possibly further promoted by the surfactant polysorbate-80—more easily, thereby leading to a higher systemic concentration and an increased risk of inducing anti-drug antibodies. These or already existing IgG cross-reacting with the VEGF-antagonist might also more likely pass the barrier to contribute to adverse reactions in the eye. Such or other interactions of the intravitreally injected drugs with systemic processes would have to be taken into consideration in investigations of possible changes of the retinal endothelium induced directly by Beovu or Lucentis based on animal models. On the other hand, an in vitro-model consisting only of retinal EC to study changes of the blood-retina barrier can be considered an over-simplification of the neurovascular unit in which other cell types also play a role52. Co-culture models with retinal EC, retinal pericytes and glia cells, to be used in future studies, at least bear a greater resemblance to the in vivo situation24.

Nevertheless, our findings clearly indicate that Beovu, but not Lucentis can harm the barrier formed by a monolayer of retinal EC in vitro. An induced dysfunction of the retinal endothelium itself might provide an additional plausible and novel explanation for the rare but severe adverse events observed after repeated intravitreal injections of Beovu.


Reagents, antibodies, brolucizumab and ranibizumab

Lucentis (10 mg/ml ranibizumab in 10 mM histidine-HCl, 10% α,α-trehalose dihydrate, 0.01% polysorbate-20, pH 5.5) and Beovu (120 mg/ml brolucizumab in 10 mM sodium citrate, 0.02% polysorbate-80, 5.8% sucrose, pH ~ 7.2) were purchased from Novartis Pharma GmbH (Nuremberg, Germany). Nintedanib or tivozanib (Selleckchem, Absource Diagnostics, Munich, Germany) were dissolved in dimethyl sulfoxide (#D4540, Merck, Darmstadt, Germany) to result in final solvent concentrations below 0.05% in the cell culture medium which did not affect the morphology or behavior of iBREC5,6. Primary and secondary antibodies used for Western blot analyses are listed with all relevant information in Table 2.

Table 2 Primary and secondary antibodies used.

Cultivation of iBREC and treatment with effectors

Experiments were performed with confluent monolayers of telomerase-immortalized microvascular EC from bovine retina (iBREC), which have been established, characterized and maintained in our laboratory21. iBREC were cultivated at 37 °C and 5% CO2 on surfaces coated with fibronectin (Corning, Amsterdam, The Netherlands or Merck, Darmstadt, Germany) in Endothelial Cell Growth Medium MV (ECGM; Promocell, Heidelberg, Germany) containing 0.4% Endothelial Cell Growth Supplement, 90 µg/ml heparin, 10 ng/ml human epidermal growth factor (hEGF), 100 nM hydrocortisone, 5% fetal bovine serum (all supplements from Promocell) and 300 µg/ml geneticin (Gibco, Thermo Fisher Scientific), as described in detail elsewhere5,23,53. The authenticity of the pericyte-free iBREC was regularly confirmed by analyses of the expression of proteins specific for microvascular EC.

Two days before the VEGF-A-binding protein formulations Beovu or Lucentis were added to result in final concentrations of 1 mg/ml brolucizumab or 100 µg/ml ranibizumab, respectively, the cell culture medium had been completely replaced by ECGM lacking hEGF but containing 1 µg/ml fibronectin. After one day, five or eight days of exposure, cells were harvested for preparation of whole cell extracts. Alternatively, cells were exposed to 0.0002% or 0.002% polysorbate-20 (suitable for cell culture, 53.4% lauric acid, #P2287, Merck) or polysorbate-80 (suitable for cell culture, 74.2% oleic acid, #P4780, Merck) for two days. Depending on the nature of the investigated cytokine or growth factor, cell culture supernatants were analyzed after treatment for two and 30 h, five or eight days5,23,53. In control experiments, cells were processed in exactly the same way only without the effector(s) investigated.

Western blot analyses

After separation of whole cell extracts obtained from ~ 105 cells by SDS polyacrylamide gel electrophoresis and transfer to a protein-binding membrane, proteins of potential relevance were determined by Western blot analyses with specific antibodies (see Table 2)5,23,53. Chemiluminescence signals were directly scanned with the imaging system Fusion Pulse TS (Vilbert Lourmat, VWR, Darmstadt, Germany) resulting in bright bands on dark background as described in detail elsewhere5,23,53. Scanned peak volumes of the protein-specific bands (more than five replicates for each condition and time point) were determined with EvolutionCapt software (Version 17.01; Vilbert Lourmat), standardized in relation to those of actin in the very same sample, and normalized to those obtained from similarly processed control cells (= 1)5,23,53. In the presented scatter plots, which also show means and standard deviations, data from multiple Western-blot experiments performed with several independently prepared cell extracts were pooled and each dot represents a single signal from one Western blot analysis.

Cell index measurements

As a measure of stability of the barrier formed by an iBREC monolayer, we continuously determined the cell index of these cells cultivated on gold electrodes by electric cell-substrate impedance measurements with the microelectronic biosensor systems for cell-based assays xCELLigence RTCA DP (Acea, OLS, Bremen, Germany) as previously described5,6,53. Briefly, impedance was measured between gold electrodes in each individual well of an E-Plate 16 PET (Agilent, OLS) and expressed as the unit-free parameter cell index CI = (Zi-Z0)/15 Ω (RTCA Software 2.0, Acea) with Zi being the impedance measured at an individual time point and Z0 the impedance read at the start of the experiment5,54. iBREC (~ 104 cells) were placed in a fibronectin-coated well and cultivated until a high and constant cell index indicative of a confluent monolayer (CI ~ 18; cell index measurements every 15 min) was reached three to four days later. After complete replacement of the cell culture medium with ECGM lacking hEGF but containing 1 µg/ml fibronectin, the cell index was measured every 15 min for two days, before Beovu (final concentration of brolucizumab: 1 mg/ml), Lucentis (final concentration of ranibizumab: 100 µg/ml), nintedanib or tivozanib (final concentrations of both inhibitors: 10 nM or 100 nM) were added. The cell index was then determined every two minutes for 90 min, every five minutes for 62 h, and every 15 min until the end of the experiments. Alternatively, polysorbate-20 or polysorbate-80 was added (final concentrations: 0.0001%, 0.0003%, 0.001% or 0.003%) to confluent iBREC cultivated in ECGM lacking hEGF but containing 1 µg/ml fibronectin for two days, and the cell index was monitored over three days as described above. All experiments were performed at least three times and data were then obtained from three to eight individual wells for each condition and time point, and recorded cell index values were normalized in relation to those measured immediately before addition of the inhibitors (RTCA Software Pro 2.3.4 (Basic), Agilent). The results were analyzed and converted to graphs showing means and standard deviations with Graph Pad Prism 6 (Graph Pad Software, San Diego, USA)5,23,53.

Immunofluorescence staining

After exposing confluent monolayers of iBREC—cultivated on fibronectin-coated two-chamber slides (x-well PCA Tissue Culture Chambers; Sarstedt, Nuembrecht, Germany)—to Beovu or Lucentis for five days as described above, cells were fixated in methanol for 7.5 min at -20 °C. Slides were blocked in 10% ImmunoBlock (Roth)/phosphate-buffered saline without Ca2+ and Mg2+ (PBSd) for one hour before they were incubated with specific antibodies against claudin-1 (rabbit polyclonal; Aviva Systems Biology via Biozol, Eching, Germany, #ARP33623_P50; 4 µg/ml), claudin-5 (see Table 2; 2.5 µg/ml) or VEcadherin (see Table 2; 1:100) for one hour and subsequently with goat F(ab’)2 fragments (coupled to AlexaFluor595) directed against rabbit IgG, H + L chains (#A11072, Invitrogen, Thermo Fisher Scientific; 1:500) for 30 min; all antibodies were diluted in 1% Immunoblock/PBSd. Then cells were embedded in ProLong Gold Antifade Mountant (Thermo Fisher Scientific) which contains 4’,6-diamidino-2-phenylindole (DAPI) to stain the nuclei. Cells were examined by fluorescence microscopy (DM4000B, LAS X, Leica, Wetzlar, Germany) and portions of cells with claudin-1-specific plasma membrane staining were counted in at least twelve randomly chosen microscopic fields containing ~ 30 cells/field5,23,29.

Determination of cytokines by ELISA

The concentrations of VEGF-A, TNFα, IL-1β, IL-6 or IL-8 were determined in undiluted cell culture supernatants of iBREC treated with Beovu by ELISA, using the kits listed in Table 123,54. We processed triplicate or quadruplicate samples according to the manufacturers’ instructions and measured the analyte-dependent absorbance at 450 nm (reference wavelength: 570 nm) 10–20 min after addition of the stop solution with an Infinite 200Pro spectrophotometer controlled by Tecan i software (Tecan, Crailsheim, Germany); standard curves were always generated in parallel to the analyses of samples54.

Statistical analyses

We used the Wilcoxon signed rank test to compare antigen-specific Western blot signals from effector-treated cells to the hypothetical value of 1.00 of normalized signals from control cells, and the Mann–Whitney U test to compare those of differently treated cells. The Wilcoxon signed rank test makes allowances for the variability of the values obtained from controls although they appear without standard deviations (SD = 0). To analyze data from cell index measurements, the two-way ANOVA followed by Sidak’s multiple comparison test was used. Differences resulting in p-values below 0.05 were considered significant. All statistical analyses were performed with Graph Pad Prism 6; means and standard deviations are provided as numbers, graphs or in scatter plots.

Ethical approval

The authors have no ethical conflicts to disclose. This article does not contain any results of studies with human participants or animals.