الانتقال إلى المحتوى الرئيسي

كتابة تعليق

مراجعة PREreview لComparative Transcriptomic Analysis of Gene Expression Patterns in Neutrophils from Rheumatoid Arthritis and COVID-19

نُشرت
DOI
10.5281/zenodo.24274872
الترخيص
CC BY 4.0

Summary

This preprint analyses public RNA-seq datasets from GEO to compare neutrophil gene expression in rheumatoid arthritis (RA) and COVID-19. The authors used DESeq2 to find differentially expressed genes (DEGs) and assessed Hallmark gene sets using fgsea and Gene Ontology using topGO. The authors suggest that RA neutrophils display an interferon and antiviral signature that partially overlaps with COVID-19 and provide a set of genes that change in opposite directions between the diseases. The authors suggest that this is dependent on which COVID-19 comparison is performed.

Strengths

  • The main strength is that the focus of the study is defined, and the datasets utilised are described in an accessible and traceable manner (Table 1).

  • The study design is sound, and each disease is first compared with its respective healthy controls. The authors then compare the resultant gene lists, which avoids batch effects between laboratories and the biases of batch-correction tools (Methods).

  • A second COVID-19 comparison is performed to test the first result. The authors state that in order to assess if a given gene is distinct to RA, it must differ from RA in both COVID-19 comparisons.

  • The authors state that their results are hypothesis-generating. They also state that Gene Ontology (GO) over-representation does not by itself indicate the direction of change, that statements drawn from the literature were not tested by their analysis, and that the findings should not be extrapolated to severe disease.

Concerns

  1. Multiple-testing problem: The Methods say that for this study, DEGs are defined as genes with an absolute log2 fold change ≥2 and p-value < 0.05. The DEG definition does not use adjusted p-values. The Results state 11,355 DEGs were observed in RA, and of these, only 90 met the log2 fold change criterion. It is unclear, therefore, what criterion defined the larger set of DEGs. Because thousands of genes are tested at once, some will pass p < 0.05 by chance alone, and a nominal p-value therefore risks many false positives. The authors could define DEGs using adjusted p-value (the padj column shown in Supplementary Fig. S1), and report how the 90-gene and 420-gene sets change. This would strengthen the downstream comparisons.

  2. Structure of the RA dataset: The GEO record for GSE116899 indicates 7 RA and 5 normal controls were included. The analysed data, however, includes 30 samples, of which 24 are RA and 6 are normal controls. These include technical replicates and both synovial-fluid and peripheral-blood neutrophils. The Limitations section acknowledges this, but the Methods do not discuss how this was handled. Collapsing technical replicates and modelling patient and site, or analysing each site separately, would likely help distinguish between disease and tissue effects.

  3. Study-level confounding in cohort 2: The severe versus convalescent comparison combines severe samples from GSE272381 with convalescent samples from GSE227116. Disease stage is associated with study identity, and no adjustment for batch effects is provided. According to Table 1, the two studies also describe different sample types (neutrophils and low-density granulocytes vs. blood samples). As both studies include healthy controls (Methods), it should be possible to perform within-study comparisons (as performed in the primary analysis).

  4. Inference about shared and different genes: A gene that is significant in one dataset and not in the other does not necessarily indicate that the diseases are different. Of the 90 RA DEGs, 73 are explained by 46 same-direction and 27 opposite-direction genes. The classification of the remaining 17 is not provided. The thresholds of “not significant” (p ≥ 0.1, or p ≥ 0.2 if fold change > 2) are also not justified. No Hallmark set achieves the adjusted p < 0.05 in cohort 1. The rank axis in Fig. 5B extends to about 70 versus about 28,000 in Fig. 2B, suggesting that the COVID-19 enrichment uses a smaller ranked list. Clarifying this would help readers judge the comparability of the two analyses. A more reasonable approach would be to perform a test of effect-size differences between datasets, such as a heterogeneity or interaction test, instead of comparing significance calls.

Minor points

  • In Figs 3A, 5A, 6A, 6B, 9A and 9B, the p-value labels for the top bars are cut off at the right edge of the panel, and so the most significant values cannot be seen properly. Some long term names in Figs 3A and 9A are also truncated, with an ellipsis.

  • The background gene set used by topGO is not defined, and the GO p-values appear to be not adjusted. The Declarations state funding, competing interests, author contributions, and code availability as “to be completed.” Providing the complete DEG and enrichment result tables, along with the analysis code, would help readers reproduce the results.

Overall assessment

The authors use publicly available data and frame their study with a clearly defined question. As is, the lists of converging and diverging genes can at best be read as preliminary. There are four major changes needed in order to assess the gene lists in Figs 8 and 11 with more confidence: adjusting the DEG calls, handling replicates and the source of tissue, within-study contrasts of cohort 2, and a direct comparison of effect sizes between the two diseases.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.

يمكنك كتابة تعليق على مراجعة PREreview هذه لـ Comparative Transcriptomic Analysis of Gene Expression Patterns in Neutrophils from Rheumatoid Arthritis and COVID-19.

قبل أن تبدأ

سنطلب منك تسجيل الدخول باستخدام ORCID iD الخاص بك. وإذا لم يكن لديك ORCID iD، يمكنك إنشاء واحد.

ما هو معرّف ORCID iD؟

ORCID iD هو معرّف فريد يميّزك عن جميع الأشخاص الذين يحملون الاسم نفسه أو اسمًا مشابهًا.

ابدأ الآن