Reply to Walter and Reyfman: Transcriptomic Analysis of Alveolar Immune Cells in Acute Respiratory Distress Syndrome: To Lump or to Split?

From the Authors:

We appreciate Drs. Walter and Reyfman’s correspondence regarding our study (1). We agree that immune cell heterogeneity—particularly alveolar macrophage (AM) diversity—likely plays a key role in the disease pathogenesis of acute respiratory distress syndrome (ARDS). Recent studies by our group (2) and others (3, 4) have used single-cell approaches to better characterize alveolar immune subtypes in ARDS and animal models of acute lung injury. However, we caution against solely relying on “splitting” approaches such as single-cell RNA sequencing to understand the pathobiology of complex human syndromes. Highly granular approaches performed on limited numbers of subjects may not capture the diversity of clinical phenotypes that exist in critical illness, and there remain significant technical and computational limitations (5) regarding single-cell approaches.

Critical care translational studies rely on analyzing data from relatively large patient cohorts to overcome external confounders that can bias results such as variation in clinical interventions, timing in the onset of risk factors, and baseline genetic diversity. The complexity and cost of single-cell approaches currently limit the number of samples that can be practically analyzed. In addition, many important genes are not captured with commonly used single-cell RNA sequencing platforms because of the limited depth of sequencing coverage and amplification bias (6). For example, Myd88 (myeloid differentiation primary response 88) and Tlr9 (toll-like receptor 9) are two important macrophage effector genes that were not detected in a recent single-cell RNA sequencing experiment identifying AM subtypes in an animal model of acute lung injury (3).

Our bulk microarray approach was inclusive of 18,415 unique genes; however, we did not identify any differentially expressed genes in AMs from subjects with good versus poor clinical outcomes after adjustment for multiple hypothesis testing. We concur with Walter and Reyfman that because our bulk approach did not capture the relative contributions that specific AM subtypes made to the overall transcriptional signal, our ability to detect differentially expressed genes may have been weakened. Despite this limitation, our bulk transcriptomic approach has advanced our understanding of AM function in ARDS by identifying AM-specific genetic programs that were associated with good versus poor outcomes. Future work is needed to identify the AM subtypes that might be responsible for the bulk transcriptional signatures we identified in our clinical cohort.

We believe that “lumping” and “splitting” approaches are complementary in furthering our understanding of the pathobiology of syndromes such as ARDS and sepsis. Analytical approaches such as cellular deconvolution (7) may be able to bridge bulk transcriptomic datasets and clinical cohorts like ours with highly granular single-cell datasets to fully leverage the strengths of both lumping and splitting approaches to understand the mechanisms of complex human syndromes.