Lung tissue was acquired from all COVID-19 associated autopsy cases of the University Medical Center of Augsburg, Germany, during the local first wave of the pandemic in 2020 (n = 19, Table 1). All patients had been tested positive for SARS-CoV-2. Post-mortem SARS-CoV-2 RNA was detectable in swabs of the upper and lower airways (Pharynx, Trachea, and Bronchus) at the time of autopsy in all but two cases via RT-qPCR (S-gene, N-gene, ORF1-gene). Respiratory insufficiency was the determining cause of death in 12 cases, while all but one of the remaining causes (septic shock) were COVID-associated, too. One patient had died of cardiac failure, showing only limited DAD patterns in the lungs.

Respecting the wish of their next of kin, a minimal or full-body autopsy was performed, resulting in a variable number of investigated lung lobes per patient. Out of each available lung lobe, three samples were taken, formalin-fixated and paraffin-embedded. This study was approved by the internal review board of the UKA (BKF No. 2020–18) and the ethics committee of the University of Munich (Project number 20–426, COVID-19 registry of the UKA). Informed consent was collected from the next of kin for all included patients.

On the basis of whole slide scans (Pannoramic scan II, 3D Histech, Budapest, Hungary) and by using a fully automated system (TMA Grand Master by 3D Histech, Budapest, Hungary), we prepared tissue microarrayed blocks containing 10 cores of each lung (core diameter: 1.5 mm), with a maximum of 85 cores per block.

TMA of COVID-lymph nodes (diameter 1.5 mm) were constructed manually, using a Beecher Manual Tissue Arrayer (Model MTA-1, Estigen OÜ, Tartu, Estonia).

An HE- and a multiplexed immunofluorescence-stain were performed on one slide each of a microarrayed paraffin block. For immunofluorescence staining, we used a kit with antibodies against CD20, CD68/CD168, CD11c, and MHC Class II (UltiMapper I/O APC-Kit, Ultivue, Cambridge, MA, USA) according to the manufacturer's protocol.

After staining, all slides from the TMA-blocks were scanned (Pannoramic Scan II scanner, 3D-Histech, Budapest, Hungary). Scans were obtained in 20x magnification, with automated calculation of the exposure time for each of the channels (DAPI, FITC, TRITC, Cy5, Cy7) of the multiplexed immunofluorescence stain. For further analyses of the virtual slides, we used the Multiplex Phenotyping Module by Visiopharm (Visiopharm, Hørsholm, Denmark).

We discarded all cores that had significant overlap with the neighboring cores (which rendered interpretation impossible) due to processing issues, got crumpled or were not stained appropriately. Areas with lots of cell debris and strong FITC-noise interfering with the analysis were excluded from the analysis (Figure 1). This reduced the core sample count to 71 and the number of patients included to 17.

The Multiplex Phenotyping Module by Visiopharm (Visiopharm, Hørsholm, Denmark), an AI-based software was trained to identify and count the different cells by surface antigen expression.

This approach offers the opportunity of automatically evaluating the cell count, not on a systemic, but a local level. It furthermore erases the need for complicated cell extraction and resulting artifacts, a problem dendritic cells are particularly prone to (17).

A pre-trained, deep-learning based Nuclei Detection APP by Visiopharm (detecting DAPI-signals) was trained on a representative selection of the slides (~189,000 iterations) and then used to label and count all nuclei of each lung core (Figure 2). These labels provide the basis for further analysis by the Multiplex phenotyping module. The Phenotyping Module automatically categorizes the cells according to their antigen expression.

Single-stain (CD80 and CD86) and double-stain (CD11c + CD83) were digitalized by a 3D Histech PANNORAMIC 1000 scanner. The resulting .mrxs-files were viewed with 3D Histech's own Case Viewer software. The double stain slides were viewed unaltered and counted manually. CD80 and CD86-stained slides were viewed using the “Gradient map visualization” plugin, which creates a heat map based on brown color signals in a digital slide and was developed for conventional IHC specifically (24). Cells surrounded by at least 50% intense positive signaling were counted as positive. For CD80, the gamma-value of all digital slides was set to 1.8 to suppress background and weak stain signals. Manual counts by LB were afterwards evaluated by a board-certified pathologist (BM). BM recounted a sample of roughly 10% of all manually counted slides. The count results were fit in a simple linear regression model [R² = 0.9019 (double stain); R² = 0.8228 (CD80 and CD86 count)]. CD11c stains of the lymph nodes were estimated by LS and BM.

The remaining 610 cores were assessed individually, whereas data of one lung lobe was compiled afterwards and statistically analyzed as a unit. To ensure comparability of lung lobes with a differing number of conserved cores, the cell count of the observed cell phenotypes in one lobe was interpreted in relation to the total number of nuclei in the lobe (for example “Macrophage Fraction” refers to the macrophage count relative to the total number of counted nuclei in the same lobe).

As cell counts showed no normal distribution, a Kruskal-Wallis test was performed on each of the mentioned cell counts to detect significant differences of the median for stages 0–2. The findings of stage 3 cores were not included for statistical testing due to the low sample size of only three remaining lobes. A Mann-Whitney-U test was used afterwards to specifically detect which two of the groups differed significantly. A p < 0.05 was considered significant.

We fit a simple linear regression model to correlate viral load and mDC maturation using GraphPad Prism. Data of viral load (Ct values of each lobe) were taken from the work by Hirschbühl et al. (23) and are available in the Supplementary Material of this study.

Out of 71 lung lobes from a total of 17 patients, nine were classified as non-DAD, 16 as stage 1, 43 as stage 2 and 3 as stage 3 DAD (for an overview of the selection process see Figure 1). Five patients showed the same stage in all analyzed lobes, two of which were full body autopsies. Two cases presented with lobes that showed more than two different stages.

In conventional IHC we found a rather constant CD11c⁺ cell count throughout groups 2 and 3 (with mDCs making up roughly 1/3 of all cells). Only two lymph nodes could be assigned to group 1. Both of these showed a higher CD11c count (around 2/3 of all cells). Both control lymph nodes showed roughly the same distribution as groups 2 and 3. Statistical analysis was not attempted due to the low case number.

The UltiMapper kits rely on antibodies marked with a DNA sequence unique for every antibody-type. Signal amplification is achieved by in-situ amplification, reducing the risk of background noise compared to Tyramide Signal Amplification-based methods (25). This allows for the simultaneous use of up to eight different antibodies. Analyzing the generated multidimensional dataset (considering the simultaneous use of four different antibodies can indicate 2⁴ = 16 different states for a single cell) makes AI-based analysis almost mandatory. Visiopharm software has already been used in peer-reviewed studies for analyzing conventional immunohistochemistry and multiplexed immunofluorescence (26, 27).

Depending on their maturation state, dendritic cells express a variety of surface antigens and receptors. Chemokines like CCL3 and CCL5 are produced by a wide variety of pulmonary cells, including epithelial cells, macrophages, neutrophils and fibroblasts (28–32). These chemokines direct circulating DCs expressing CCR1 and CCR5 toward the lung (17, 33), whereas a downregulation in CCR1 and CCR5 results in impaired migration toward the site of inflammation (34). Numerous studies have shown CCR1 and CCR5-expressing DCs to migrate toward a higher CCL5-concentration (28, 35, 36), in vitro as well as specifically in airway mucosa (37). DCs in lungs have been shown to express both of these receptors (17). Carrying a sentinel-like function, these local immature DCs take up and process antigens. It is only in combination with danger-signals like pathogen-associated molecular patterns (PAMP) that DCs start to undergo a maturation process. At the end of this process, the matured DC has migrated to the T-cell region of lymphoid organs and gained the ability to activate naïve T-cells (a process also known as “licensing”). DCs are the most potent initiators of T-cell response (18). Maturation is accompanied by a radical shift in surface antigen expression. As mature DCs are first and foremost professional antigen-presenting cells, receptors for antigen presentation like MHC class I and II are upregulated along with co-stimulating molecules like CD40, CD80, CD83, CD86 (38–40). Chemokine receptors CCR1 and CCR5 are downregulated, ending the stationing of the cell–a process. Sozzani et al. (41) described as “weighing the anchor”. CCR7 on the other hand is upregulated and aids in recruiting the cell toward lymphatic tissue (17, 42–44). Thus, an impaired maturation process hampers DC homing, antigen presentation and ultimately impedes an effective T-cell response.

Some immune evasion strategies of viral pathogens, including coronaviridae, aim at the maturation of DCs. MHC class I and II upregulation is impaired in MERS-CoV infected DCs (45). SARS-CoV, while able to infect monocyte-derived DCs (46), also hampered maturation of DCs (no significant upregulation of MHC class I, II, CD83, CD86) (47). Simultaneously, SARS-CoV induced the expression of CCR-1 and CCR-5 (48). For SARS-CoV-2, data so far are also pointing toward impaired maturation of mDCs (21, 22). It should be noted that while Law et al. reported a significant increase in the expression of CCR1 and CCR5, they interpreted this finding as facilitating migration from the site of infection toward lymph nodes. To our knowledge, this does not concur with current findings regarding these receptors (28, 35).

Interestingly, we found mDCs in stage 1 lung lobes also had failed to upregulate MHC class II. This is noteworthy as an inverse correlation between DAD stage and viral load has been described previously (23). Based on the data of this study, we attempted to fit a simple linear regression between the viral load and mDC maturation. A correlation could not be detected (R² = 0.003928, p = 0.6256).

To confirm our findings, we attempted conventional IHC staining on other DC maturation markers, namely CD80, CD86, and CD83. The results of the CD80 count were overall non-conclusive, although there was a detectable difference between non-DAD and stage 2 lobes. It is likely that the results were skewed by the overall high count of alveolar macrophages, most of which appeared to be CD80⁺. CD80 has been shown to be expressed by macrophages (49).

As the CD83 antibody proved to be rather unspecific, we conducted a duplex stain in combination with CD11c. The distribution of double positive cells as well as of CD86⁺ cells resembled that of MHC II⁺ mDCs, with no detectable increase of activated DC in stage 1. Even more so, stage 1 lobes showed a significantly lower count than stage 2 lobes in the double stain [and a near significant difference to non-DAD as well as stage 2 in CD83 (Figure 3)].

Considering the fairly low number of available lymph nodes, especially in group one (including two cases only), the results should be interpreted very cautiously. Assuming the higher CD11c⁺ count of the two cases of group 1 is representative and does not present outliers, one could argue that this indicates an increased DC homing in very early COVID-lungs (hence a physiological response to a pathogen). In some cases though, viral immune evasion might lead to an ineffective immune response (partially caused by impaired DC homing) and to development of DAD, represented by groups 2 and 3. It should be noted that group 2 also drained lung lobes in stage 1. This interpretation is, however, highly speculative.

The simultaneous induction of receptors tying the DCs to inflammatory foci and subsequent impaired maturation are possible explanations for our findings. While immature dendritic cells keep getting recruited, they ultimately fail to fulfil their primary goal: homing to lymph nodes and activating T-lymphocytes. This results in an observable accumulation of dendritic cells in the lungs and prevents an effective T-cell response.

One recent study by Onodi et al. reports a SARS-CoV-2 induced significant upregulation of maturation markers in (CD11c⁻) plasmacytoid predendritic cells specifically (50), a subtype of DCs that is similar to mDCs regarding some functions, but of lymphoid origin. Our study, by design, only reports on CD11c⁺ mDCs. Considering the different changes of both pDCs and mDCs to COVID-19 in cell count alone, it is reasonable to assume that the molecular response could also vary to a greater extend. Further research needs to be conducted to evaluate the similarities or differences in the molecular reaction of both DC types.

It is not unlikely that severe DAD might cause an increased release of chemokines compared to earlier stages, thus causing an intensified recruitment of DCs to foci of advanced lung damage. It has indeed been established that the cytokine storm in severely affected patients explicitly implies elevated levels of CCL3 amongst other cytokines (51). Conclusive data discussing a possible correlation between cytokine levels and DAD stage specifically have not yet been published.

Furthermore, we did not detect an increased number of macrophages in later stages. Monocytes are closely related to mDCs and are known to rely on a similar pattern chemotactic factors for trafficking toward inflammation (52). This renders an elevated chemokine level in later stages even less likely to be causal for the increased local mDC count.

There are no approved drugs to specifically induce DCs maturation. Recent studies are investigating the possible use of CCR5-inhibitors in COVID-19, primarily targeting monocyte trafficking (53, 54). Future works might uncover possible effects on mDCs.

There are some limitations to this study.

Post-mortem investigations are prone to autolysis, possibly affecting the results. However, as insightful as in vivo investigations might be, lung biopsies from COVID-19 patients with ARDS are not justifiable.

Although our study included only 17 cases and should thus be interpreted with caution, these cases represent roughly 77% of all COVID-19 deceased of the first wave at the UKA. Unfortunately, we were not able to acquire a sufficient number of lymph nodes for a conclusive analysis, which might have further elucidated our hypothesis.

Lastly, CD80, CD83, and CD86 show varying specificity for activated dendritic cells, impeding IHC analysis. More detailed histological single-cell characterization could be made possible by further development of multiplexed immunohistochemistry in the foreseeable future.