diff --git a/content/100.figure-table-legends.md b/content/100.figure-table-legends.md
index 72f0049..f7699ef 100644
--- a/content/100.figure-table-legends.md
+++ b/content/100.figure-table-legends.md
@@ -103,11 +103,12 @@ All remaining columns give the number of libraries (as designated with `(L)`) wi
 ![**Figure S1. Results from benchmarking `alevin-fry` and `CellRanger` performance.**](https://raw.githubusercontent.com/AlexsLemonade/scpca-paper-figures/main/figures/compiled_figures/pngs/figure_s1.png?sanitize=true){#fig:figs3 width="7in"}
 
 Each panel compares metrics for six representative ScPCA libraries, including three single-cell and three single-nuclei suspensions, obtained when processing data with both `alevin-fry` and `CellRanger`.
-Panels B-D consider only cells present in both the `alevin-fry` and `CellRanger` output.
 
 A. Runtime in minutes (top row) and peak memory in GB (bottom row) for six ScPCA libraries processed with `alevin-fry` and `CellRanger`.
 Processing with `alevin-fry` was consistently faster and more memory-efficient compared to processing with `CellRanger`.
 
+Panels B-D show only cells present in both the `alevin-fry` and `CellRanger` output.
+
 B. Comparison of mean gene expression values for six ScPCA libraries processed with `alevin-fry` and `CellRanger`, shown on a log-scale.
 Each point is a gene, and only genes detected in at least 5 cells are shown.
 $R^2$ values shown in the top left corner of each panel reflect broad agreement in mean gene expression values between platforms.