Comment on Hu et al. Single-Cell Transcriptomics Reveals Novel Role of Microglia in Fibrovascular Membrane of Proliferative Diabetic Retinopathy. Diabetes 2022;71:762–773

We are writing this letter to discuss the article by Hu et al. (1), which reported that retinal microglia are involved in fibrosis in fibrovascular membranes of proliferative diabetic retinopathy. Unfortunately, when we reanalyzed the online reposited data (GSE165784), we were unable to reproduce the microglial subpopulation claimed by these authors.

To analyze the data set, we first removed cells with <500 or >5,000 unique features and cells with >25% mitochondrial RNA. We then used well-known markers to identify the cell populations in the fibrovascular membranes object of the study, and, as shown in Fig. 1A, we found endothelial cells (CLDN5⁺ and WVF⁺), stromal cells (ACTA2⁺, CSPG4⁺, and PDGFRB⁺), and immune cells (CD163⁺, CD68⁺, FCER1A⁺, and CCL5⁺) (mainly macrophages).

These data are in contrast with the original conclusion in the article by Hu et al., where the authors claimed the main cell population was microglia. One salient error in the authors’ analysis was their choice of microglia markers TREM2, C1QA, GPR34, CX3CR1, GPNMB, LIPA, FABP5, MRC1, SELENOP, IGF1, and FCN1, which are all macrophage markers (see Fig. 1 in Hu et al.).

In Fig. 1B, we explored profibrogenic molecules in the samples (COL1A1, COL1A2, COL5A1, LUM, THBS2, FBLN2, and CTGF), which were prominent in the stromal cluster but not in the other cell types, contradicting the authors’ conclusion about the microglial cell compartment. In their article, they also claimed that extracellular matrix genes were upregulated in the cell compartment defined as microglia (see Figs. 1 and 4 in Hu et al.); we were unable to reproduce these results.

To further analyze their data set, we reclustered the stromal cells separately and identified four clusters including pure pericyte and myofibroblast clusters. In addition, we confirmed a unique cluster that we recently reported as pericytes transitioning to myofibroblast (2), which express pericyte markers such as MCAM and CSPG4 in combination with profibrotic markers such as COL5A1 (Fig. 1C).

When we reclustered the immune cells, we found NK cells (KLRB1⁺, KLRD1⁺, and GZMA⁺), T lymphocytes (CD4⁺), dendritic cells (LST⁺, CLEC10A⁺, FLT3⁺, and FCER1A⁺), and macrophages (CD68⁺, CD163⁺, and CD14⁺), as shown in Fig. 1D. We also identified SPP1⁺ activated macrophages. In this focused immune cell analysis, we detected very low expression level of specific microglial markers such as TMEM119 and P2RY12, which raises questions about the identity of the population from which the conclusion of the original manuscript was derived.

Finally, when we analyzed the fibrogenic molecules in the immune cell clusters, we found that these molecules were not expressed in any of the clusters (Fig. 1D).

In conclusion, reanalysis of the authors’ original data set showed three major flaws that we believe completely invalidate their conclusions.

• 1.

  The sample contains only low expression of microglial markers, but the authors used the wrong labels (largely macrophages) to demonstrate “microglia” within the immune cells.

• 2.

  Upon examination of the entire immune cell compartment, we were unable to replicate profibrogenic properties in any cell cluster, further invalidating their conclusions.

• 3.

  Using the authors’ original data, we show that the stromal compartment is the main contributor to fibrogenesis in the proliferative diabetic retinopathy fibrovascular membranes, which confirms the findings in our recent publication (2).