The present study did not reprocess raw CEL files from GEO. Instead, analyses were based on the processed expression matrix and differential-expression results provided in the uploaded online analysis report associated with GSE206848. GEO describes this dataset as human synovium expression profiling by array and lists platform GPL570, Affymetrix Human Genome U133 Plus 2.0 Array [8]. The dataset contains normal, OA, and RA synovium samples. In this revision, the exact samples used by the uploaded online report were explicitly listed to improve reproducibility (Table 1).

Table 1. Sample definition and status in the uploaded online analysis report.

Normal synovium was defined according to GEO sample labels beginning with NS and described as non-OA/normal synovium. RA samples were not appropriate for an OA-versus-normal analysis. They were included in the uploaded online report as the two test samples, which is why the present revision avoids presenting the results as OA-specific. To support an OA-specific conclusion, the analysis should be rerun using GSM6265697-GSM6265703 as the OA group and excluding GSM6265704-GSM6265705.

According to the documentation of the uploaded online report, processed expression values were retrieved from GEO through GEOquery and analyzed using the limma package in R [9][10]. The present revision evaluated and interpreted the processed expression matrix reported by that workflow rather than generating a new expression matrix from raw microarray CEL files. Because the data were from Affymetrix GPL570 and the plotted sample values were on a typical microarray log2 scale, the expression matrix was treated as log2-transformed or already transformed processed expression data. Quantile normalization was used in the online workflow to align the distributions of expression values across samples. This step should not be described as removing batch effects unless an explicit batch variable and batch-correction method are used.

Probe annotation followed the GPL570 platform annotation file. Probes without gene annotation were removed. Where multiple probes mapped to the same gene symbol, the probe with the highest expression signal was retained as the representative probe, as described in the online report.

Differential expression analysis was performed using limma with group as the explanatory variable [10]. The design and contrast can be written as follows:

group <- factor(c(rep("ref", 7), rep("test", 2)), levels = c("ref", "test"))

design <- model.matrix(~0 + group)

colnames(design) <- c("ref", "test")

contrast.matrix <- makeContrasts(test_vs_ref = test - ref, levels = design)

fit <- lmFit(exprSet, design)

fit2 <- contrasts.fit(fit, contrast.matrix)

fit2 <- eBayes(fit2)

DEGs were screened using |log2FC| > 1 and nominal P < 0.05. Adjusted P values were calculated using the Benjamini-Hochberg method. Because most representative genes did not pass FDR < 0.05, nominally significant genes are described as exploratory or representative genes rather than confirmed significant biomarkers.

Box plots were used to assess sample expression distributions after normalization. PCA was used to visualize global expression differences between groups. Volcano plots summarized the relationship between log2FC and statistical significance. A heatmap displayed the expression patterns of selected DEGs and hierarchical clustering across samples; the heatmap visualization followed standard approaches for complex expression matrices [11].

As shown in Figure 1, the median expression values of the nine selected samples were broadly aligned after normalization, suggesting that the distributions were comparable. Figure 2 shows that the reference and test groups were separated in the reduced-dimensional PCA space. These observations support the use of the normalized matrix for exploratory differential expression analysis. However, comparable box-plot medians do not prove that batch effects were removed.

Figure 1. Box plot of normalized signal intensity in the selected samples. The medians are broadly comparable after normalization. The analysis included 7 normal (reference) and 2 RA (test) samples.

Figure 2. PCA plot of the selected samples. The reference and test groups show separation in the reduced-dimensional space. The analysis included 7 normal (reference) and 2 RA (test) samples.

Using |log2FC| > 1 and nominal P < 0.05, the online report identified 1,082 genes in the selected comparison (Table 2). The number of genes changed substantially under different fold-change thresholds, indicating that the results are sensitive to threshold choice. Figure 3 summarizes the genes that passed the nominal screening threshold in the volcano plot.

Table 2. Number of genes identified under different nominal screening thresholds.

Figure 3. Volcano plot showing genes that passed the nominal screening threshold in the selected comparison. The analysis included 7 normal (reference) and 2 RA (test) samples.

Representative genes from the original table are shown in Table 3. Because only CYP2C19 reached FDR significance among these listed genes, the other entries should be interpreted as nominally different exploratory genes in this selected comparison. The heatmap of selected DEGs is shown in Figure 4.

Table 3. Representative genes listed in the selected comparison from the online report.

Figure 4. Heatmap of selected DEGs. The red annotation bar represents the two test samples used by the online report. The analysis included 7 normal (reference) and 2 RA (test) samples.