RNA-seq data compared to qRT-PCR

Question

[SeoG]

Hello!

We performed qRT-PCR on a specific gene and subsequently conducted bulk RNA-seq on the same samples.

In the qRT-PCR results, the gene showed over 2-fold higher expression in the case group compared to the control. However, in the RNA-seq analysis performed using Illumina, there was almost no difference, and the expression level of the case group was slightly higher than that of the control.

What could be the potential issues causing this discrepancy?

The read counts in the count matrix were similar across samples, and the overall read counts were consistent. Additionally, the data were normalized using DESeq2, and no outliers were observed in PCA or MDS plots.

The count matrix was generated through the following pipeline: FASTQ - > Trimming (Cutadapt) - > Alignment (HISAT2) - > Gene Mapping (featureCounts).

The quality at each step was good, so no additional QC was performed.

Thanks for your answer.

[ATpoint]

qPCR is often misinterpreted in my opinion. Statistics do not take into account the overall range of the Ct values. For example, fold changes are always higher and noisier, the more lowly a gene is expressed, meaning the higher the Ct values are. RNA-seq in contrast is aware of the non-equal relationship between variance and expression and takes this into account in the statistical analysis. Also, qPCR is often poorly-controlled, for example using a single "housekeeping" gene. In contrast, RNA-seq normalizes using thousands of genes, so I would generally trust RNA-seq more, unless qPCR is done well, with good experimental replication, validation primers of good effficiency and normalized against a good panel of reliable reference genes.

[kai]

have you try convert counts to RPKM or FPKM or TPM

Answer 1

[Kevin Blighe]

Hello.

The discrepancy between your quantitative reverse transcription polymerase chain reaction (qRT-PCR) results and ribonucleic acid sequencing (RNA-seq) results for this gene may arise from several factors.

First, qRT-PCR targets a specific transcript region via primers, which may amplify particular isoforms or variants. In contrast, RNA-seq typically quantifies expression at the gene level by summing reads across all exons, potentially diluting differences if multiple isoforms exist. Verify if your gene has alternative splicing by checking Ensembl or RefSeq annotations.

Second, normalization methods differ. qRT-PCR often relies on one or a few housekeeping genes, which can introduce bias if their expression varies between case and control groups. RNA-seq, via DESeq2, uses a median-of-ratios approach across thousands of genes, providing more robust normalization. This may explain why RNA-seq shows minimal difference.

Third, expression levels matter. If the gene is lowly expressed, qRT-PCR cycle threshold (Ct) values may be high, leading to noisier fold changes due to amplification variability. RNA-seq accounts for count variance in its statistical model, reducing such noise. Examine your raw Ct values; if they exceed 30, the qRT-PCR fold change may be unreliable.

Fourth, technical aspects in qRT-PCR, such as primer efficiency (ideally 90-110 percent) and validation against multiple reference genes, are critical. Poor controls can inflate differences. RNA-seq, with its pipeline (Cutadapt, HISAT2, featureCounts), appears solid based on your description.

To investigate, convert your RNA-seq counts to transcripts per million (TPM) for comparison, as TPM better approximates relative expression:

# In R, using edgeR or similar
library(edgeR)
tpm <- cpm(countMatrix, log=FALSE, normalized.lib.sizes=TRUE) / (rowSums(countMatrix) / 1e6)
# Then compare mean TPM between groups

Also, plot raw counts for the gene across samples to confirm trends.

Overall, as per ATpoint, RNA-seq results are generally more trustworthy for genome-wide analyses unless qRT-PCR is rigorously validated.

Kevin