*2.13. Protein Corona Structure and Composition*

The two-wavelength surface plasmon resonance (SPR) measurements and calculations, LC–MS/MS data acquisition, and sample preparation were conducted as described previously [28,29]. The LC gradients used for human plasma and porcine vitreous samples were 1 h and 3 h, respectively. The 500 µL liposome injection volume per replicate corresponds to ~16 <sup>×</sup> <sup>10</sup><sup>12</sup> particles, with ~13.2 <sup>×</sup> <sup>10</sup><sup>9</sup> immobilized liposomes on the sensor with ~55 nm<sup>2</sup> of active surface area. Each SPR replicate run accounts for the measured number of immobilized liposomes on the sensor before the determination of corona thicknesses, which assumes random sequential adsorption [28,29]. Raw data from earlier publications in human plasma with neutral 100 nm ICG liposomes (F1–F3) [28] were integrated with new data for 100 nm HA liposomes with and without ICG (F4–F5) and 50 nm ICG liposomes with and without PEG. Raw data from porcine vitreous studies with the 50 nm ICG liposomes (F6–F7) [29] were integrated with new data for the optimized pegylated formulation F2 and the HA liposome F4 in porcine vitreous humor. The non-functional liposomes without ICG were not tested in vitreous humor.

Upon integration of the two datasets, principal component analysis excluded the presence of significant batch effects. As such, LC–MS/MS raw data were processed and median-normalized without additional batch effect corrections. Protein identification and quantification of protein groups (LC–MS/MS data processing) were carried out with the MaxQuant v. 1.6.1.0 [34], with the UniProtKB FASTA file either for *Sus scrofa* (40,701 protein and 23,223 gene entries) or for *Homo sapiens*(86,725 protein and 20,605 gene entries), to which 245 commonly observed contaminants and all reverse sequences were added. Differential abundance and hierarchical clustering analyses for relative enrichment were carried out with Perseus v. 1.6.5.0 [35]. Protein identifications with non-zero intensity values in at least three samples in plasma and vitreous humor were retained for comparisons and annotated with pI, gene names, and sequences. Abundances were log2-transformed, and protein identifications were classified as "only identified by site". Reverse sequences and potential contaminants were filtered out. When applicable, a protein's median intensity (after normalization) across all samples where it was identified was used to rank and select high abundance (physiological expression or corona-enriched)

proteins. Missing intensity values of unidentified protein groups were replaced with random numbers to mimic low abundance measurements, using the input from the normal distribution function in Perseus. After removal of outliers, 2–4 replicates of all formulations were retained for multiple sample testing.

ANOVA with a Benjamini–Hochberg false discovery rate (FDR) of 0.05 on the top 20 proteins from the assigned groups (source, HC, SC) was followed by two-sample Student's *t*-tests with Benjamini–Hochberg FDR correction. To determine the protein physicochemical properties, gene names for the majority protein within each mass spec-identified protein group and their default protein sequences were obtained from UniProtKB. Unmapped identifications were excluded, and the same protein name was consistently retained for genes that encode multiple proteins. The same sequences were used to calculate the theoretical pI, molecular weight, and grand average of hydropathy (GRAVY) using the the ExPASy server ProtParam tool (https://web.expasy.org/protparam/) or the GRAVY Calculator (http://www.gravy-calculator.de/). The sums of the aromatic residues (phenylalanine, tyrosine, tryptophan, with and without histidine) were calculated from the sequences using the LEN function in MS Excel that returns the number of characters in a text string. For analysis of common gene subsets without accounting for enrichment, the HC Intersection in plasma and the HC Union in vitreous humor were used. Additional comparisons between gene sets were conducted using Venn diagram software (http://bioinformatics.psb.ugent.be/webtools/Venn/), followed by GeneMANIA (http://genemania.org) for functional analysis.
