Mycobiome analysis in fungal infected formalin-fixed and paraffin-embedded tissues for identification of pathogenic fungi: a pilot study

Introduction

Fungal organisms are frequently observed in surgical pathological diagnosis. Recent developments in technology have enabled the identification of fungi using a variety of methods¹. These novel methods are based on analyzing fungal genomes or proteomes using fresh tissues^2,3. However, these methods are difficult to apply in formalin-fixed and paraffin-embedded (FFPE) tissues used in surgical pathology. Sanger sequencing and immunohistochemistry staining have been used in FFPE tissue, but most medical institutions only identify fungi based on morphological findings^4–9. The identification of fungi only by morphological findings can lead to inadequate treatment for patients due to misdiagnosis, which can often result in fatal consequences¹⁰. As different antifungal agents are preferentially used at the initial infection stage depending on the fungus, it is necessary to identify the exact fungi present through testing methods in addition to the morphological findings¹¹.

In order to more accurately identify fungi in FFPE tissues, it is essential to use genomic information. DNA markers that could be used to identify fungi include the internal transcribed spacer (ITS) region, small subunit (nrSSU-18S), large subunit (nrLSU-26S or 28S), elongation factor 1-alpha (EF1α), and the largest (RPB1) and second largest (RPB2) subunits of RNA polymerase^1,2. Among these markers, ITS could be relatively easily and effectively used for fungal identification, and the database of the ITS regions of fungi is available^12–14. In general, sequencing equipment is classified into first- (Sanger sequencing), second- (massively parallel sequencing), and third-generation (real-time and single molecule sequencing) equipment according to key analytical methods¹⁵. Third-generation sequencing technology is characterized by direct sequencing of nucleotides without PCR amplification. Oxford Nanopore Technologies (ONT) introduced several sequencing devices using nanopore sequencing technology, which measures the change in current that occurs when a nucleotide sequence passes through a narrow channel¹⁶. There is an advantage in identifying each DNA sequence without PCR amplification, and it is expected that the DNA can be effectively detected in spite of DNA degradation during FFPE tissue preparation and storage.

Extracting only fungal DNA from the fungal infection site of FFPE tissue is very difficult. Therefore, we decided to use the mycobiome analysis method. However, no studies have attempted to analyze the mycobiome in FFPE tissue. Therefore, this study was conducted as a pilot study on the development of a method for finding pathogenic fungi using mycobiome analysis in fungal infected FFPE tissues.

Methods

Results

A total of 2,526 DNA sequences were sequenced¹⁹. No DNA was sequenced in three cases. Of the remaining 46 cases, the sequenced DNA of 22 cases were not found in the ITS database. We were able to identify 342 fungal sequences in 24 (49.0%, 24/49) cases. The number of DNA sequences identified per case are summarized in Figure 1. The median number of the detected fungal DNA sequences per case was 3 (1Q: 1 and 3Q: 14.25). The clinicopathological information and the number of fungal species detected per case are summarized in Table 1. The remaining cases, in which fungal DNA was not sequenced, are listed in Table 2. Detailed information related to the BLAST program is summarized in Supplementary Table 1 (see Extended data)¹⁹. The relationship between the detected fungal DNA and the ITS region is summarized in Table 3 and visualized in Figure 2. Most (215, 62.87%) fungal DNA sequences contained the entire region of ITS1 or ITS2. The remaining 127 fungal DNA sequences were identified as fungi using a partial sequence of ITS1, ITS2, 5.8S, LSU or SSU.

Figure 1. Summary of the number of DNA sequences identified per case.

Figure 2. Visualization of nucleotide sequences used for fungal identification according to the degree of ITS1 and ITS2 sequence matching (the nucleotide sequences of the reference database were aligned with the starting base of the ITS2 region).

The mean FFPE block storage period of fungal and no fungal cases was 4.2 years and 4.8 years, respectively (p = 0.410). The DNA concentrations measured by NanoDrop after DNA extraction in the groups with and without fungal DNA detection were 71.2 ng/µl and 104.4 ng/µl, respectively, with no statistical significance (p=0.272). The fungal DNA detection rate was not statistically significant (p=0.376)

Representative microscopic images of some cases with inconsistent pathologic diagnosis and mycobiome analysis are summarized in Figure 3 (original images are provided as Underlying data)²⁰. Aspergillus and Candida species were detected by sequencing in case 1-01 (Figure 3A). Compared to the Aspergillus species commonly found in nasal cavities, thinner hyphae were observed, which may have been misleadingly similar to sulfur granules of the pathologically diagnosed Actinomyces. Aspergillus and Acremonium species were detected by sequencing in case 1-03 (Figure 3B). Contrary to case 1-01, it is thought to have been misdiagnosed as Mucor species because its hyphae appeared slightly wider than Aspergillus species. Case 1-08 (Figure 3C) was also misdiagnosed as Actinomyces, in what is thought to be a similar manner to case 1-01. Alternaria, Fusarium and Aspergillus species were detected with numerous sequencing reads. In case 2-08 (Figure 3D), yeast-form fungi were observed in alveolar macrophages using a Gomori methenamine-silver stain and Starmerella cellae was identified by sequencing. Starmerella cellae is a relatively recently identified ovoid to ellipsoidal fungus²¹. Case 2-09 (Figure 3E) is a fungus found in the pharynx, which shows morphological findings different from those of Candida, Aspergillus, and Mucor. In other words, yeast-form or short branching-type fungal nuclei were found, like in Pneumocystis jiroveci. Cladosporium coloradense was identified by sequencing. Case 3-04 (Figure 3F) is a yeast-form fungus found in subcutaneous tissue and Candida glabrata was identified by numerous sequencing reads.

Figure 3. Representative microscopic images with inconsistent pathological and molecular diagnoses (magnification x600).

A) case 1-01, B) case 1-03, C) case 1-08, D) case 2-08, E) case 2-09 and F) case 3-04.

All three cases diagnosed with Actinomyces were found to have fungi (cases 1-01, 1-02 and 1-08). Case 3-02 was had a pathological diagnosis of Blastomyces dermatitidis by sequencing analysis (summary of sequencing results: 575 nucleotide sequences of U18364.1 were identical without gap opening and 549 nucleotide sequences of EF592163.1 were identical without gap opening). One strand of fungal DNA was identified from the DNA extracted from the FFPE tissue, and the BLASTN analysis showed the highest probability match was Candida africana with a bit score of 468. In this case, the bit score was 154 based on the separate sequence of Blastomyces dermatitidis. Among three cases of pathologically diagnosed cryptococcosis, one case (case 3-10) was found to have a nucleotide sequence of Cryptococcus neoformans with a bit score of 296. In case 4-12, the nucleotide sequence of Candida africana with a bit score of 438 was suggested, with the possibility of Pneumocystis jiroveci with a bit score of 73.1. Four cases of pathologically unidentifiable fungi (cases 2-03, 2-08, 2-09 and 3-04) were found to most likely be Acremonium acutatum, Starmerella cellae, Cladosporium coloradense, and Candida glabrata, respectively.

Discussion

We performed mycobiome analysis on fungal infected FFPE tissues using nanopore sequencing. The detected fungal DNA occupied approximately one-third of the ITS1 entire region, the ITS2 entire region, and other regions. The advantages of nanopore sequencing compared with Sanger sequencing are as follows. First, unlike Sanger's method, which requires a lot of DNA, nanopore sequencing can be performed with a small amount of DNA (in theory, even with one strand). Second, the nanopore sequencing method can sequence DNA separately, even if the sample contains a variety of lengths of DNA²². Third, nanopore sequencing equipment (i.e., MinION) can be operated at a lower cost (about $1,000) than Sanger sequencing equipment. This low initial cost is a critical factor in the introduction of equipment in small pathology laboratories. Compared to second-generation sequencing equipment, nanopore sequencing has the advantage of sequencing damaged DNA because there is no PCR amplification process in the sequencing process itself. In this study, about one third of the fungal DNA was not an ITS1 entire match or an ITS2 entire match. Therefore, these fungi can be effectively detected using nanopore sequencing.

It is necessary to find pathogenic fungi among various fungi detected by mycobiome analysis. It may be assumed that dominant fungi are associated with the disease at the site of fungal infection. In cases 1-07 and 1–10, diagnosed as esophageal candidiasis, Candida species comprised 72% and 93% of the fungal DNA, respectively. In such cases, it would be clear to diagnose that the infection is caused by Candida species. In case 1-09, however, only 7% of the fungal DNA was Candida species. Similarly, Rhizopus species were detected in Mucor infections, but as only a small fraction of the fungal DNA. Considering these cases, it is considered desirable to include the process of detecting normal flora in adjacent uninfected sites and selecting pathogenic fungi associated with infection.

We performed multiplex PCR to increase fungal DNA concentration. However, 2,184 non-fungal DNA sequences accounted for 86.46% of the total DNA. Of the 2,184 DNA sequences not identified as fungi, only 382 (17.50%) were identified in the human genome (GRCh38). Because ribosomal DNA is present in all living organisms, including fungi, humans, and bacteria, it may be difficult to amplify only the fungal ITS region. Nevertheless, we were only able to detect fungal DNA in 49.0% of cases. This detection rate is not high enough to be applied in actual clinical situation and needs improvement. There was no difference in the storage period or DNA quality of the FFPE block between cases where fungal DNA was detected and cases that were not detected. We suggest the following about the causes of low fungal DNA detection rates. The first and most important is a low sequencing output. The third-generation sequencing technique is known to have less sequencing output than the second-generation sequencing equipment. Therefore, increasing the purity of the DNA to be sequenced is important for research. Since we used multiplex sequencing, we were forced to reduce the amount of DNA we analyzed per sample. We believe that the lack of sufficient DNA sequencing in each case was the main cause of the failure to detect fungal DNA. The second is the lack of fungal DNA evaluation methods. Unlike human cells, in which cell viability (closely related to DNA quality) could be predicted by evaluation of a haemotoxylin and eosin stained slide, how to evaluate the viability of fungi in a haemotoxylin and eosin stain is not known. Many studies have shown that inflammation and hypoxia are closely related, and hypoxia induced by inflammation caused by fungal infections may result in damage to fungal DNA and consequently affect the sequencing²³. Third, the aforementioned ribosomal DNA is present in almost all living things. We believe this could be overcome by using microdissection, which extracts DNA from fungi as much as possible. Fourth, there is a lack of high-quality databases. Even in the database of NCBI RefSeq Targeted Loci Project, there is only one species (murina) in the Pneumocystis genus. In other words, to diagnose Pneumocystis jiroveci, a separate database should be constructed. In addition, when we analyzed the GenBank file, there were 5,602 ITS1 full sequences and 7,552 ITS2 full sequences in 11,133 genera. Therefore, one well-known fungal ITS DB is not sufficient for identification of fungi, and several DBs must be synthesized to analyze the results. Fifth, it is known that tumor biopsy specimens of 50µm thickness can be used to sufficiently perform next-generation sequencing for diagnostic purposes²⁴. Our study was also performed with a 50µm thickness. However, if thicker tissues (e.g., 100µm thick) were used, the fungal detection rate could be increased.

Candida albicans is the most commonly isolated Candida species in the clinical setting. Of the 162 Candida species, 141 Candida africana (87.04%) and eight Candida albicans (4.94%) sequences were detected. These two Candida species are similar in ITS sequences. Comparing the ITS sequences (ITS1+5.8S+ITS2) of Candida africana (NR_138276.1, 447bp) with Candida albicans (NR_125332.1, 446bp), there is 99.192% identity. There is a 1 bp gap in ITS1 in Candida albicans. Two mismatched nucleotides are located in ITS2, where the DNA base "C" in Candida albicans is "T" in Candida africana. It is generally known that nanopore sequencing has a slightly higher error rate than second-generation NGS sequencing²⁵. DNA extracted from FFPE tissue is known to have a higher C:T conversion than DNA extracted from fresh frozen tissue²⁶. Because of these, it is difficult to use our method to precisely identify fungi, especially at the species level.

In summary, we have identified the possibility of finding pathogenic fungi through mycobiome analysis in fungal infected FFPE tissues. However, we have found problems to be solved in further studies, such as increasing sequencing output, increasing fungal DNA concentration, excluding normal flora, and expanding fungal databases. If we develop a method to characterize pathogenic fungi in FFPE tissues in a follow-up study, we think it will help patients to use appropriate antifungal agents.

Data availability