Biomarker Axes, Detection Limits, and Endotype Stability in CRS: Reply to Katkuri et al. and Hong

1.

To the Editor:

We thank Katkuri et al. and Hong for their insightful letters responding to our study on the effects of appropriate medical therapy (AMT) for chronic rhinosinusitis [1, 2, 3]. Both letters affirm the principal finding that AMT preferentially attenuates Type 2 (T2) and eosinophil‐associated inflammatory signals alongside improvements in endoscopic, radiographic, and patient‐reported outcomes, while raising important methodological and interpretative questions to which we are pleased to address.

2. PCA Interpretation

Both letters raise the important question of whether the principal component analysis (PCA)‐derived “severity” (PC‐1) and “endotype” (PC‐2) axes represent genuinely independent dimensions, noting that ECP, a downstream T2 effector, loaded heavily in two dimensions. We clarify that while PCA identifies dimensionally orthogonal “components” of inflammation, individual biomarkers can load on multiple components. For example, ECP loaded on both PC‐1 and the T2 direction of PC‐2; it loaded PC‐1 more heavily. In contrast, IL‐5 and IL‐13 loaded the T2 direction of PC‐2 predominantly. It is also important to note that our nominal “severity” component was based on empirical observations that proinflammatory biomarkers loaded in a unidirectional manner, consistent with general inflammatory burden in this dataset. Katkuri et al. suggest that neutrophil markers might better represent an eosinophil‐independent axis. While we acknowledge we did not have an unequivocally neutrophil‐derived biomarker, we find in a separate surgical dataset that neutrophil mediators such as HNE and FCGR3B similarly load both the “severity” and Type 1/3 components on PCA. Interpreting our PCA, the most endotype‐independent biomarker Katkuri et al. request, one may conclude that the macrophage‐derived biomarker MIP1a does not weight the “endotype” PC‐2 while heavily weighting the “severity” PC‐1.

Katkuri et al. point out that the persistence of T1/3 biomarkers despite clinical improvement aligns with emerging evidence that T1/3‐skewed CRS may be relatively corticosteroid‐resistant and may require alternative management strategies. Dorismond et al., also using a PCA‐based cluster analysis of mucus biomarkers among CRSwNP patients, found that a mixed T1/3 endotype characterized by IFN‐g and IL‐17a had the shortest time‐to‐polyp recurrence, with similar time‐to‐oral steroids and time‐to‐biologic therapy as other endotypes [4]. Stein et al. similarly found that persistent mucosal discharge or edema following endoscopic sinus surgery was associated with a T1 or T3 endotype [5]. These findings underscore the challenges of T1/3 driven disease and highlight the need for alternative therapeutic strategies for this subgroup.

Hong additionally raises questions about whether the baseline PCA structure remains stable after treatment. We performed a post‐AMT PCA that was not included in our original publication, which yields a component structure closely resembling the pre‐AMT organization and magnitude across the biomarkers of interest (Figure 1).

FIGURE 1.

Primary component analysis (primary dimension 1 vs. 2) of biomarkers post‐AMT. Purple = severity; green = T2 inflammation, and red = T1/3 inflammation.

3. Assay Detection Limits

Hong questions whether the absence of T1/3 change could reflect floor effects rather than true biological stability. We acknowledge an omission; our original manuscript did not explicitly describe handling of below‐detection values. In concordance with previous studies, values below the lower limit of detection (LLD) were imputed at one‐half the published LLD for each biomarker and not censored. This places samples with biomarker values below the LLD in an appropriate rank order. Since our analyses utilized Wilcoxon rank‐sum statistics, which operate on rank order, they are less skewed by floor effects than parametric tests. In further response to this point, we performed a McNemar's test with Benjamini–Hochberg correction to evaluate for significant differences in samples falling below the LLD. IL‐5 was more frequently below the LLD post‐AMT (p < 0.01), but no significant change in below‐LLD frequency was observed for any other biomarkers included IFN‐g and IL‐1b. We advance that these findings support the premise that AMT primarily affects T2 inflammation.

4. Subdomain Outcome Analyses

Hong proposes subdomain‐level analyses of SNOT‐22 and CRS‐PRO to identify more granular biomarker‐system mappings. However, as both instruments were developed and validated as total score measures, we evaluated the instruments as designed and discerned whether these total scores were related to the biomarker changes. While subdomain analyses may provide clearer understanding of endotype‐subdomain relationships, these would require multiple‐testing corrections, substantially reducing power in a 51‐patient cohort. Domain‐level biomarker mapping is an interesting direction for future work in larger, endotype‐stratified cohorts.

In summary, we appreciate these constructive exchanges with Hong and Katkuri et al. We look forward to further studies in this realm.

Sincerely,

Asher C. Park, BS; Brooke N. Gleason, BS; Eli Stein, MD; Saied Ghadersohi, MD; Atsushi Kato, PhD; Stephanie Shintani‐Smith, MD MS; David B. Conley, MD; Kevin C. Welch, MD; Robert C. Kern, MD; Bruce K. Tan, MD MS

Department of Otolaryngology—Head and Neck Surgery, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA

Conflicts of Interest

The authors declare no conflicts of interest.