Correction to: Nature Immunology https://doi.org/10.1038/s41590-021-01052-7. Published online 18 October 2021.

It has been called to our attention that our previous statistical analysis of our transcriptomic data of myeloid cell bone marrow progenitors in the offspring of exposed or non-exposed mice can be improved with a more stringent statistical analysis to identify differentially expressed genes (DEGs). The main improvement in the analysis would come if one filtered out lowly expressed genes or applied specific methods such as independent hypothesis weighting (IHW) or logFC shrinkage. We agree that those statistical models would reduce the number of false-positive results in the statistical test. In our published analysis—keeping in mind our biological question, a study on the transgenerational effect of acute infection—we applied a less stringent cutoff to maximize the number of genes to be further investigated, as we expected that relevant genes might have a relatively low fold change and expression. Having this in mind, we corroborated our results with several in vivo experiments supporting our findings.

Nevertheless, considering improved and more stringent statistical analyses and in order to improve the robustness of our analysis, we performed resequencing of the samples that were used in the study, as well as a set of reanalyses. First, we increased the sequencing depth of the dataset to assess the impact of sequencing depth on the reported results. Second, we included the above-mentioned stringent filtering and estimation modeling steps in our analysis and performed DEG and functional enrichment analysis.

Increasing sequencing depth did not markedly improve or worsen the results (47 DEGs in common monocyte progenitors (cMoPs) and 79 DEGs in Ly6Chi monocytes using the published analysis script, 39 DEGs in cMoPs and 46 DEGs in Ly6Chi monocytes when applying a fold-change cutoff of 1.5). As expected, when using more stringent filtering criteria on lowly expressed genes in each cell type and applying IHW and logFC shrinkage on the merged dataset including both the original and new sequencing data, the number of DEGs was substantially reduced (3 DEGs in cMoPs and 1 DEG in Ly6Chi monocytes when filtering on TPM, 20 DEGs in cMoPs and 15 DEGs in Ly6Chi monocytes when filtering on counts), in contrast to the initial larger number of genes reported in the original manuscript (12 DEGs in cMoPs, 53 DEGs in Ly6Chi monocytes). Batch correction of the data for an unknown batch variable led to an increase in the number of DEGs (12 DEGs in cMoPs and 14 DEGs in Ly6Chi monocytes when filtering on TPM, 58 DEGs in cMoPs and 52 DEGs in Ly6Chi monocytes when filtering on counts).

Within the cMoP dataset, a very important candidate gene survived all filtering and correction steps in all datasets, Ssc4d (scavenger receptor cysteine-rich family member with four domains). Ssc4d belongs to an ancient and highly conserved group of cell surface and/or secreted proteins, involved in the development of the immune system and the regulation of both innate and adaptive immune responses. Recently, this gene was established as a broad-spectrum pattern recognition receptor for bacteria and protozoan parasites1. However, one should consider that, while this transcript has been assigned coding status in mouse, there is no proof yet of protein-coding activity for the equivalent human transcript, and therefore the interspecies relevance of this protein remains to be demonstrated. In Ly6Chi cells, Wfdc21 (whey acidic protein/four-disulfide core domain 21) was identified to be differentially expressed, but the function of Wfdc21 is poorly known, and future studies on its role in host defense are warranted in order to identify its true biological relevance.

In conclusion, a more stringent analysis of our transcriptome data without taking low effect sizes into consideration revealed, as expected, that the number of DEGs associated with differences in previous exposure to pathogens was markedly reduced. Importantly, this does not change the biological and immunological conclusions of our study, which are based on four different models of in vivo transgenerational infection. Independently, the question remains of how this information is transmitted to the bone marrow compartment, so that transgenerational inheritance can be achieved. The new analysis reported here argues that the transcriptional changes at the level of bone marrow progenitors are more subtle than originally reported and most likely do not represent the most important molecular substrate of the immunological changes observed in the second generation of mice. A more comprehensive assessment of the epigenetic processes previously reported as important pathways of genetic inheritance (for example, in worms and other species), such as DNA methylation, RNA transport and histone modification, should be conducted. Collectively, these open questions warrant more in-depth studies to decipher the molecular mechanisms behind the cross-generational transmission of host resistance described in the study.

Scripts and count tables of the updated analysis on the resequencing data are publicly available on GitHub (https://github.com/schultzelab/Transmission-of-trained-immunity).