Cannabis microbiome sequencing reveals several mycotoxic fungi native to dispensary grade Cannabis flowers

The Center for Disease Control estimates 128,000 people in the U.S. are hospitalized annually due to food borne illnesses. This has created a demand for food safety testing targeting the detection of pathogenic mold and bacteria on agricultural products. This risk extends to medical Cannabis and is of particular concern with inhaled, vaporized and even concentrated Cannabis products . As a result, third party microbial testing has become a regulatory requirement in the medical and recreational Cannabis markets, yet knowledge of the Cannabis microbiome is limited. Here we describe the first next generation sequencing survey of the fungal communities found in dispensary based Cannabis flowers by ITS2 sequencing, and demonstrate the sensitive detection of several toxigenic Penicillium and Aspergillus species, including P. citrinum and P. paxilli, that were not detected by one or more culture-based methods currently in use for safety testing.


Introduction
Our knowledge of the natural microbiome of field-grown Cannabis in terms of rhizosphere bacteria, and endophytic fungi is limited to just a few focused studies [1][2][3] . Very little is known about the potential for bacterial and fungal contamination on medicinal Cannabis. Nevertheless, many states in the U.S. are now crafting regulations for detection of microbial contamination on Cannabis in the absence of any comprehensive survey of actual samples. A few of these regulations are inducing growers to "heat kill" or pasteurize Cannabis flowers to lower microbial content. While this seems a harmless suggestion, we must remain aware of how these drying techniques may create false negatives in culture-based safety tests used to monitor colony-forming units (CFU). Even though pasteurization may be effective at sterilizing some of the microbial content, it does not eliminate various pathogenic toxins or spores. Aspergillus spores and mycotoxins are known to resist pasteurization 4,5 . Similar thermal resistance has been reported for E. coli produced Shiga toxin 6 . While pasteurization may reduce CFU's used in petri-dish or plating based safety tests, it does not reduce the microbial toxins, spores or DNA encoding these toxins.
Monitoring for mycotoxic fungi in cannabis preparations has been recommended as part of routine safety testing by the Cannabis Safety Institute. A major driver for this recommendation has been numerous reported cases of serious or fatal pulmonary Aspergillosis associated with marijuana smoking in immunocompromised patients 7-9 . The major cannabinoids have been shown to be potent inhibitors of several cytochrome P450 enzymes at therapeutic concentrations, including 1A1, 1A2, 1B1 2B6, 2C19, 2D6, 3A4 and 3A5 10 . Some of these enzymes have been implicated in the metabolism of the fungal toxins aflatoxin and ochratoxin [11][12][13] . This raises questions about potential interactions and appropriate safety tolerances for mycotoxins in patients being treated with cannabinoid therapeutics. In addition, some Fusarium species that produce toxins have proven to be difficult to selectively culture with tailored media 14-16 . This is a common problem associated with culture-based systems as carbon sources are not exclusive to certain microbes and only 1% of microbial species are believed to be culturable 17 .
While the risks of mycotoxic fungal contamination have been well studied in the food markets, the presence of the fungal populations present on Cannabis flowers has never been surveyed with next generation sequencing techniques [18][19][20][21][22][23] . With the publication of the Cannabis genome 24,25 and many other pathogenic microbial genomes, quantitative PCR assays have been developed that can accurately quantify fungal DNA present in Cannabis samples 26 . Here, we analyze the yeast and mold species present in 10 real world, dispensary-derived Cannabis samples by quantitative PCR and sequencing, and demonstrate the presence of several mycotoxin producing fungal strains that are not detected by widely used culture-based assays.

Culture-based methods
The culture-based methods selected for testing here represent those currently in use by established medicinal Cannabis safety testing laboratories. 3.55ml of tryptic soy broth (TSB) was used to wet 250mg of homogenized flower in a whirlpack bag. TSB was aspirated from the reverse side of the 100μm mesh filter and placed into a Biolumix TM growth vial and spread onto a 3M Petri Film TM and a SimPlate TM (3M Petrifilm TM 3M Microbiology, St. Paul, MN, USA; SimPlates TM Biocontrol Systems, Bellevue, WA, USA; BioLumix TM Neogen, Lansing MI, USA) according to the respective manufacturers' recommendations. Biolumix TM vials were grown and monitored for 48 hours while Petri-films TM and SimPlates TM were grown for 5 days. Petri-films TM and SimPlates TM were colony counted manually by three independent observers. Samples were tested on total coliform, total entero, total aerobic, and total yeast and mold. Only total yeast and mold discrepancies were graduated to sequencing.

DNA purification
Plant DNA was extracted with SenSATIVAx according to manufacturers' instructions (Medicinal Genomics part #420001). DNA was eluted with 50μl ddH20.

Amendments from Version 1
The statement in the abstract saying that the findings "demonstrate the limitations in the culture-based regulations" was removed. We added a brief mention of the limited literature relating to Cannabis microbiomes to the introduction. Clarification as to the scope of the study including only analysis of fungal microbiomes was added. We revised paragraph 2 of the introduction to focus on the numerous literature reports of pulmonary aspergillosis associated with cannabis use instead of potential dangers of mycotoxin toxicity. We provided additional background on the overlap of cannabinoid and mycotoxin metabolism via cytochrome P450 system. Figure 2 was revised and the figure legend expanded. We removed spurious hits to non-fungal species and reported only high-confidence species detected with 10 reads or more. This resulted in a higher number of species reported for some samples and fewer for others. The first paragraph of the discussion section was expanded to describe the additional findings. We added a sentence to the end of the second paragraph of the discussion mentioning the two existing publications on endophytic fungi in Cannabis. We removed the comment on potential growth inhibition relating to terpenoids. With respect to potential paxilline contamination, we de-emphasized the concern based solely on the detection of P. paxilli, and stated instead that if the results were verified by tests indicating high levels of paxilline then it may be a cause for concern. The comments relating to the sensitivity of ELISA assays were deleted. Some clarification was added to the concluding paragraph to emphasize the need to ensure that all species of potential concern can be detected, and also the need for additional studies to characterize a broader diversity samples, including measurements of toxin levels where relevant.

REVISED
Total Yeast and Mold assay and ITS amplification A commercially available total yeast and mold qPCR assay (TYM-PathogINDICAtor, Medicinal Genomics, Woburn MA) was used to screen for fungal DNA in a background of host Cannabis DNA. The TYM qPCR assay targets the ribosomal DNA Internal Transcribed Spacer region 2 (ITS2) using modified primers described previously 27,28 . Fungal DNA amplified using these primers may also be subjected to next generation sequencing to identify the contributing yeast and mold species. ITS sequencing has been widely used to identify and enumerate fungal species present in a given sample 29 .
Tailed PCR cloning and sequencing DNA libraries were constructed with 250ng DNA using New England Biolabs (Ipswich, MA) NEBNext Quick ligation module (NEB # E6056S). End repair used 3μl of enzyme mix, 6.5μl of reagent mix, 55.5μl of DNA + ddH20. Reaction was incubated at 30°C for 20 minutes. After end repair, ligation was performed directly with 15μl of blunt end TA mix, 2.5μl of Illumina adaptor (10μM) and 1μl of ligation enhancer (assumed to be 20% PEG 6000). After 15 minute ligation at 25°C, 3μl of USER enzyme was added to digest the hairpin adaptors and prepare for PCR. The USER enzyme was tip-mixed and incubated at 37°C for 20 minutes. After USER digestion, 86.5μl of SenSATIVAx was added and mixed. The samples were placed on a magnet for 15 minutes until the beads cleared and the supernatant could be removed. Beads were washed twice with 150μl of 70% EtOH. Beads were left for 10 minute to air dry and then eluted in 25μl of 10mM Tris-HCl.
Library PCR 25μl 2X Q5 polymerase was added to 23μl of DNA with 1μl of i7 index primer (25μM) and 1μl universal primer (25μm). After an initial 95°C for 10s, the library was amplified for 15 cycles of 95°C 10s, 65°C 90s. Samples were purified by mixing 75μl of SenSATIVAx into the PCR reaction. The samples were placed on a magnet for 15 minutes until the beads cleared and the supernatant could be removed. Beads were washed twice with 150μl of 70% EtOH. Beads were left for 10 minute to air dry and then eluted in 25μl of 10mM Tris-HCl. Samples were prepared for sequencing on the MiSeq version 2 chemistry according to the manufacturers' instructions. 2×250bp reads were selected to obtain maximal ITS sequence information.

PaxP verification PCR
Primers described by Shirazi-zand et al. were utilized to amplify a segment of the 725bp PaxP gene from Penicillium paxilli. 25μl LongAmp (NEB) 4μl 10μM primer, 1μl DNA (14ng/μl), 20μl ddH20 to make a 50μl PCR reaction. Cycling conditions were slightly modified to accommodate a different polymerase. 95°C for 30s followed by 28 cycles of 95°C 15s, 55°C for 30s, 65°C 2.5 minutes. Samples were purified with 50μl of SenSATIVAx as described above. 1μl of purified PCR product was sized on Agilent HS 2000 chip. Nextera libraries and sequencing were performed according to instructions from Illumina using 2×75bp sequencing on a version 2 MiSeq.
Penicillium Citrinum verification PCR Citrinum forward GATTTTCCAAAATGCCGTCT and Citrinum reverse GCTCAAGCATTAATCTAGCTA primers were used with identical PCR conditions as above with the exception using 35 cycles of PCR. Samples were purified with 50μl of SenSATIVAx as described above. 1μl of purified PCR product was sized on Agilent HS 2000 chip. Nextera libraries and sequencing were performed according to instructions from Illumina using 2×75bp sequencing on a version 2 MiSeq. Reads were mapped to Genbank accession number LKUP01000000. Mappings were confirmed using BLAST to NCBI to ensure the strongest hits were to P. citrinum.

Analysis
Reads were demultiplexed and trimmed with Casava 1.8.2 and trim_ galore v0.4.1 (http://www.bioinformatics.babraham.ac.uk/projects/ trim_galore/). FLASH v1.2.11 30 was used to merge the reads using max_overlap 150. The reads were aligned to microbial references using MG-RAST v3.2 31 . Alignments and classifications were confirmed with a second software tool from One Codex (https://onecodex.com/) and critical pathways identified for further evaluation with PCR of toxin producing genes. Reads are deposited in NCBI under SRA accession: SRP065410. Nextera 2×75bp sequencing of the PaxP gene was mapped to accession number HM171111.1 with CLCbio Workstation V4 at 98% identity over 80% of the read. One Codex analysis was put into Public mode under the following public URLs:

Results
We purified DNA from Cannabis samples obtained from two different geographic regions (Amsterdam and Massachusetts) several years apart (2011 and 2015). The majority of samples purified and screened with ITS qPCR were negative for amplification signal implying reagents clean of fungal contamination. Six of the 17 dispensary-derived Cannabis samples tested positive for yeast and mold in the TYM qPCR assay. These results were compared with the results derived from three commercially available culture-based detection systems for each of the 17 samples (3M Petrifilm TM 3M Microbiology, St. Paul, MN, USA; SimPlates TM Biocontrol Systems, Bellevue, WA, USA; BioLumix TM Neogen, Lansing MI, USA; Figure 1). Of the 6 qPCR positive samples, two tested negative in all 3 culture-based assays and four tested negative in 1 or 2 of the culture-based assays ( Table 1). None of the qPCR negative samples tested positive in any of the culturebased assays. Each of the 6 discordant samples was subjected to ITS sequencing to precisely identify the collection of microbes present. Four additional samples from a different geographic origin (Amsterdam) were also subjected to ITS sequencing, for a total of 10 Cannabis samples.
Each discordant sample presented with an array of microbial species, as shown in Figure 2. No sample presented with a single dominant species, and each sample displayed multiple species of interest. Of particular concern were the identified DNA sequences from toxin producing species: Aspergillus versicolor 32-36 , Aspergillus terreus 37 , Penicillium citrinum 38-40 , Penicillium paxilli 41,42 .
We further analyzed the ITS sequence alignments using the whole genome shotgun based microbiome classification software known as One Codex 43 . Nine of the ten samples sequenced showed the presence of P. paxilli (Figure 3). To verify the accuracy of this ITS phylotyping, a gene involved in the paxilline toxin biosynthesis pathway of P. paxilli was amplified with PaxPss1 and PaxPss2 primers described by Saikia et al. 44 . The resulting 725bp amplicon (expected size) was sequenced to confirm the presence of the P. paxilli biosynthesis gene in the Cannabis sample KD8 (Figure 4). This was successfully repeated with primers designed to target genes in the citrinin pathway of P. citrinum. There were some discrepancies between the results derived from the two software platforms (One Codex and MG-RAST). The MG-RAST analysis, using merged, paired reads correlated better with the PCR results. While One Codex predicted and confirmed KD8 as having the highest P. paxilli content, the One Codex platform is optimized for whole genome shotgun data and may not be able to differentiate the 18S sequence differences (391/412 aligned bases) between these two species with a K-mer based approach.
With the confirmed presence of P. paxilli, we are curious to find out whether the toxin, paxilline, is present in the samples. Development of monoclonal antibodies to paxilline has recently been described 45 , but commercial ELISA assays with sensitivity under 50ppb do not appear to be available at this time. A >50ppb multiplexed ELISA assay is available from Randox Food Diagnostics (Crumlin, UK). Detection with LC-MS/MS has also been described 46,47 , however, and experiments are underway to determine whether paxilline can be identified in the background of cannabinoids and terpenes present in Cannabis samples.  Figure 1A. qPCR signal from TYM (red line) test run concurrently (multiplexed) with a plant internal control marker (green line). This marker targets a conserved region in the Cannabis genome and should show up in every assay (upper left). SimPlates count the number of discolored wells (purple to pink) as a proxy for CFU/gram. Only total aerobic show growth (upper right). Petrifilm only demonstrate colonies on total aerobic platings (lower left). Biolumix demonstrate no signal across all 4 tests (lower right). Figure 1B. Sample KD8 fails to culture any total yeast and mold yet demonstrates significant TYM qPCR signal. Sample was graduated to ITS based next generation sequencing. Figure 1C. Sample Liberty Haze was tested with 3 culture based methods and compared to qPCR. Sample was graduated to ITS based next generation sequencing. Table 1. Samples were cultured with 3 different techniques and compared to quantitative PCR (qPCR). Biolumix had the lowest sensitivity failing to pick up 4/17 samples detected with other culture-based platforms. qPCR identified 2 samples that were not picked up by any other method. Positive qPCR samples were sequenced to identify the contributing signal. Highlighted samples fail the 10,000 CFU/g cutoffs which equates to a Cq of 26 on the qPCR assay according to the manufacturers' instructions. (f) is fail or over 10,000 CFU/g. (p) is pass or under 10,000 CFU/g. The raw CFU numbers can be deduced by dividing the CFU number by the 1,000 fold dilution factor used in this study.

Samples
Total Yeast and Mold (10,000 CFU/g = fail)  Histograms are provided for each of the Cannabis samples that tested negative by at least one culture based method and positive using a qPCR-based total yeast and mold test. The number of reads corresponding to each detected fungal species is indicated to the right of each bar. Species detected with less than 10 reads are not included. The overall read counts per sample are more a reflection of sample normalization for sequencing than of the absolute fungal DNA levels.
Figure 3. One Codex classification of ITS reads. P. paxilli is the most frequently found contaminant in Cannabis flowers. P. citrinum is not in the One Codex database at this time. One Codex utilizes a fast k-mer based approach for whole genome shotgun classification and can be influenced by read trimming and database content. The reads provided to MG-RAST were trimmed and FLASH'd (paired end reads merged when overlapping) prior to classification. K-mer based approaches can significantly differ from longer word size methods and this underscores the importance of confirmatory PCR in microbiome analysis.  Quantitative PCR is agnostic to water activity and can be performed in hours instead of days. The specificity and sensitivity provides important information on samples that present risks invisible to culture based systems. The drawback to qPCR is the method's indifference to living or non-living DNA. While techniques exist to perform live-dead qPCR, the live status of the microbes is unrelated to toxin potentially produced while the microbes were alive. ELISA assays exist to screen for some toxins 51 . Current state-recommended ELISA's do not detect citrinin or paxilline, the toxins produced by P. citrinum and P. paxilli, respectively. The predominance of these Penicillium species in a majority of the samples tested is interesting. Several Penicillium species are known to be endophytes on various plant species, including P. citrinum 18 , and this raises the question of whether they may be common Cannabis endophytes. Indeed, P. citrinum and a species identified as P. copticola (a member of the citrinun section 51 ) have previously been identified as Cannabis endophytes, along with several Aspergillus species 2,3 .
Paxilline is a tremorgenic and ataxic potassium channel blocker and has been shown to attenuate the anti-seizure properties of cannabidiol in certain mouse models 52-54 . Paxilline is reported to have tremorgenic effects at nanomolar concentrations and is responsible for Ryegrass-staggers disease 55 . Cannabidiol is often used at micromolar concentrations for seizure reduction and contamination with paxilline, if confirmed, would be a cause for concern. Citrinin is a mycotoxin that disrupts Ca2+ efflux in the mitochondrial permeability transition pore (mPTP) 56-63 . Ryan et al. demonstrated that cannabidiol affects this pathway suggesting a similar potential cause for concern regarding CBD-citrinin interaction 64 . Considering the hydrophobicity of these mycotoxins and the growing interest in the use of extracted oils from CBD-rich Cannabis strains for treatment of drug resistant epilepsy 65-70 , more precise molecular screening of fungal toxins in these products might be warranted.
ITS amplification and sequencing offers a hypothesis-free testing approach that can be employed to identify a broad range of fungal species present in a given sample. Appropriate primer design can survey a broad spectrum of fungal genomes while affording rapid iteration of design. Quantitative PCR has also demonstrated single molecule sensitivity and linear dynamic range over 5 orders of magnitude offering a very sensitive approach for detection of microbial risks. Our survey of Cannabis flowers in this study was limited, however. Further studies are required to survey a broader range of samples, and to determine whether paxilline, citrinin, aflatoxin or ochratoxin can be detected at concentrations that represent a clinical risk in Cannabis samples or extracts derived from plants that test positive for the fungi known to produce those toxins.

Conclusions
Several toxigenic fungi were detected in dispensary-derived Cannabis samples using molecular amplification and sequencing techniques. These microbes were not detected using traditional culture-based platforms. These results suggest that culture based techniques borrowed from the food industry should be re-evaluated for Cannabis testing to ensure that they are capable of detecting the prevalent species detected by molecular methods with adequate sensitivity. We recommend that additional sequencing studies be performed to characterize the fungal and bacterial microbiomes of a more diverse selection of Cannabis samples. Such sampling should include dispensary-derived samples from both indoor and outdoor crops, as well as samples from police seizures from well-provenanced foreign sources, such as Mexico. Finally, further studies should be performed to measure toxin levels in strains that test positive for toxigenic species.

Competing interests
The authors are employees of Medicinal Genomics Corporation (MGC). MGC manufactures qPCR reagents utilized in this study.

Grant information
The author(s) declared that no grants were involved in supporting this work.

F1000Research
Open Peer Review The authors have adequately addressed most of my concerns with the exception of the issue of re-evaluating the standard methods used for establishing food safety thresholds as applied to cannabis.
The problem is with the use of the word "re-evaluate". The authors are suggesting that the negative results they obtained using cell culture-based platforms in samples that were positive using their more sensitive molecular amplification indicate a possible limitation of standard methods as applied to cannabis. This may or may not be true. The concern I have is whether their technique is overly sensitive based on current limits used in the food industry. It may be that use of molecular DNA amplification methods indicate the presence of several toxigenic species of fungi and other pathogens in food but at such low levels that standard culture-based methods would not detect them. In this case there would be no need to re-evaluate the use of standard methods that establish safety thresholds. In short, without some type of calibration between the author's technique and traditional methods there is no way of knowing whether their sensitivity is too high or traditional methods used in the food industry are somehow not capable of detecting pathogens in cannabis preparations and therefore need to be "re-evaluated".

I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
No competing interests were disclosed. The manuscript is well written and appropriate as an article in F1000 Research. The abstract states that their findings, "demonstrate the limitations in the culture-based regulations", but this conclusion does not follow from their data. In fact, their results show that their DNA based method is overly sensitive at detecting potential pathogens. Whether culture-based regulations are appropriate or not would have to be F1000Research detecting potential pathogens. Whether culture-based regulations are appropriate or not would have to be validated and tested directly, not simply inferred from the presence of microbial DNA. The authors should remove this cautionary sentence in the abstract and throughout the manuscript until it has been validated. This would make the manuscript more balanced and justified.
A better understanding of the microbiome and mycobiome in cannabis is an important endeavor in part, because recent work on the human microbiome has revealed that microbial constituents of the microbiome and fungi interact cooperatively and non cooperatively to influence human health. Recently studies focused on the human gut mycobiome have been performed using deep sequencing of the ITS1 region for identification of fungi in fecal samples from healthy individuals and the researchers identified 184 fungal species in total. Human oral and lung tissues testing indicate the presence of Aspergillus, , and in healthy individuals. An emerging theme from this new field of Cryptococcus, Fusarium Alternaria study indicates complex microbial communities distributed across the body that fundamentally contribute to the development, physiology and metabolic homeostasis of the macro-organism. The same is likely true in plants, like cannabis. Because of this beneficial dynamic interplay between microbes and hosts a complete absence of amplified DNA or RNA microbial markers would be unexpected. The question is, "What levels are safe?" and the answer to this question has not been established.
From a consumer safety perspective the rationale for microbial testing in food and pharmaceuticals is to prevent infection of highly toxigenic microbes that grow readily and are consumed in rather large quantities. For example, infects grain in as much as 30% of Sub-Saharan African maize Aspergillus flavus and the Aflatoxin regulatory limit for maize is 10 ppb because in this region maize is consumed in kilogram quantities on a daily basis. The risk of health hazard to individuals arising from microbial toxins or mycotoxins is largely proportional to the consumption amounts and frequency. Compared to other regulated foods or tobacco it is expected that the level of cannabis consumption would be minimal and would therefore present minimal risk, nevertheless, there is burgeoning interest in putting in place regulatory requirements for medical and recreational Cannabis in some markets.
In this present study the authors use state-of-the art technology to identify DNA-based markers associated with a variety of microbes and, as expected, some are pathogenic. The authors state that "Mycotoxin monitoring in Cannabis preparations is important since aflatoxin produced by Aspergillus species is a carcinogen.", but there is no actual data showing that the strain of they identify is Aspergillus toxigenic.
is a common fungus that is found in the human mycobiome in oral, lung, Aspergillus gastrointestinal tract. Detection of markers in minute quantities alone is not necessarily a Aspergillus health concern unless it is also coincident with live cells that can grow readily and secrete toxin that are above some threshold or there is an imbalance in homeostatic growth limiting factors. The authors call into question culture-based testing, which is the standard in Food safety and USP regulatory guidelines based on comparison to their highly sensitive DNA based detection. Their results using standard regulatory methods, for the most part, would not indicate a food safety problem. The authors should add this emphasis. The authors correctly identify a limitation of their study in that the qPCR based testing has an "indifference to living or non-living DNA" and because of this their PCR based approach may be unnecessarily sensitive. To date there are little to no guidelines for thresholds on many mycotoxins or bacterial toxins that have been established in the cannabis industry, so their findings help inform regulators as to which type of toxins might be relevant for further analysis.
The authors state, "Health compromised patients exposed to aflatoxin and clearance-inhibiting cannabinoids raise new questions in regards to the current safety tolerances to aflatoxin." but the authors present no data showing the presence of aflatoxin in any of their samples. So their caution is highly speculative and they should indicate that unless they provide data supporting their caution.

F1000Research
As mentioned above, dynamic homeostatic processes limit the growth of microbes and fungi in living organisms and the authors point out that "several studies have demonstrated plant phytochemicals and terpenoids like eugenol can inhibit the growth of fungi. It is possible the different water activity of the culture assay compared to the natural terpene rich flower environment is contributing to the false negative test results." but these phytochemicals may also prevent the growth of fungi and bacteria in the plant despite the presence of microbial DNA.
Furthermore they state that "While techniques exist to perform live-dead qPCR, the live status of the microbes is unrelated to toxin potentially produced while the microbes were alive. ELISA assays exist to screen for some toxins. Current state-recommended ELISA's do not detect citrinin or paxilline, the toxins produced by P. citrinum and P. paxilli, respectively. The predominance of these Penicillium species in a majority of the samples tested is interesting." Ideally the authors would test for these toxins in their most positive samples.
The authors state that, "Cannabidiol is often used at micromolar concentrations for seizure reduction implying sub-percentage contamination of paxilline could still be a concern" but this is highly speculative and the authors should de-emphasize the "concern" and state instead that if their results were verified by tests indicating high levels of paxilline then it may be cause for concern. The same is true for their concern about Citrinin and aflatoxin and the authors should state this.
The authors state, "While ELISA assays are easy point of use tests that can be used to detect fungal toxins, they can suffer from lack of sensitivity and cross reactivity. ITS amplification and sequencing offers hypothesis-free testing that can complement the lack of specificity in ELISA assays." ELISA and rapid diagnostic lateral flow tests are standard in the food safety industry for measuring toxins. I see no need to call into question protein based ELISA methods without even testing them in the first place. All diagnostic tests have sensitivity and selectivity limitations which is why they need to be tested and verified using other analytical methods.
The authors state that "Appropriate primer design can survey a broad spectrum of microbial genomes while affording rapid iteration of design. Quantitative PCR has also demonstrated single molecule sensitivity and linear dynamic range over 5 orders of magnitude offering a very robust approach for detection of microbial risks. This may be important for the detection of nanomolar potency mycotoxins". The ability to detect single copies of genes makes their system highly sensitive, but does not indicate level of mycotoxin. The authors should point out the limitations of their approach and discuss the possibility that it would likely generate a high degree of false positive results compared to culture-based standard methods.
The authors state, "These results demonstrate that culture based techniques superimposed from the food industry should be re-evaluated based on the known microbiome of actual Cannabis flowers in circulation at dispensaries." This statement appears to be too strong in light of their data. Without validation for the presence of toxins above a safety threshold there is no need to re-evaluate the standard methods in the food industry.

I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
No competing interests were disclosed. The manuscript is well written and appropriate as an article in F1000 Research. The abstract states that their findings, "demonstrate the limitations in the culture-based regulations", but this conclusion does not follow from their data. In fact, their results show that their DNA based method is overly sensitive at detecting potential pathogens. Whether culture-based regulations are appropriate or not would have to be validated and tested directly, not simply inferred from the presence of microbial DNA. The authors should remove this cautionary sentence in the abstract and throughout the manuscript until it has been validated. This would make the manuscript more balanced and justified.
Author response: The last sentence of the abstract was revised to read: Here we describe the first next generation sequencing survey of the fungal communities found in dispensary based Cannabis flowers by ITS2 sequencing, and demonstrate the sensitive detection of several toxigenic and species, including that were not detected by Penicillium Aspergillus P. citrinum and P. paxilli, one or more culture-based methods currently in use for safety testing.

A better understanding of the microbiome and mycobiome in cannabis is an important endeavor in part, because recent work on the human microbiome has revealed that microbial constituents of the microbiome and fungi interact cooperatively and non cooperatively to influence human health. Recently studies focused on the human gut mycobiome have been performed using deep sequencing of the ITS1 region for identification of fungi in fecal samples from healthy individuals and the researchers identified 184 fungal species in total. Human oral and lung tissues testing indicate the presence of Aspergillus, Cryptococcus, Fusarium, and Alternaria in healthy individuals.
An emerging theme from this new field of study indicates complex microbial communities distributed across the body that fundamentally contribute to the development, physiology and metabolic homeostasis of the macro-organism. The same is likely true in plants, like cannabis. Because of this beneficial dynamic interplay between microbes and hosts a complete absence of amplified DNA or RNA microbial markers would be unexpected. The question is, "What levels are safe?" and the answer to this question has not been established.
From a consumer safety perspective the rationale for microbial testing in food and pharmaceuticals is to prevent infection of highly toxigenic microbes that grow readily and are consumed in rather large quantities. For example, Aspergillus flavusinfects grain in as much as 30% of Sub-Saharan African maize and the Aflatoxin regulatory limit for maize is 10 ppb because in this region maize is consumed in kilogram quantities on a daily basis. The risk of health hazard to individuals arising from microbial toxins or mycotoxins is largely proportional to the consumption amounts and frequency. Compared to other regulated foods or tobacco it is expected that the level of cannabis consumption would be minimal and would therefore present minimal risk, nevertheless, there is burgeoning interest in putting in place regulatory requirements for medical and recreational Cannabis in some markets.
In this present study the authors use state-of-the art technology to identify DNA-based markers associated with a variety of microbes and, as expected, some are pathogenic. The authors state that "Mycotoxin monitoring in Cannabis preparations is important since aflatoxin produced by Aspergillus species is a carcinogen.", but there is no actual data showing that the strain of Aspergillus they identify is toxigenic. Aspergillus is a common fungus that is found in the human mycobiome in oral, lung, gastrointestinal tract. Detection of Aspergillus markers in minute quantities alone is not necessarily a health concern unless it is also coincident with live cells that F1000Research mycobiome in oral, lung, gastrointestinal tract. Detection of Aspergillus markers in minute quantities alone is not necessarily a health concern unless it is also coincident with live cells that can grow readily and secrete toxin that are above some threshold or there is an imbalance in homeostatic growth limiting factors. The authors call into question culture-based testing, which is the standard in Food safety and USP regulatory guidelines based on comparison to their highly sensitive DNA based detection. Their results using standard regulatory methods, for the most part, would not indicate a food safety problem. The authors should add this emphasis. The authors correctly identify a limitation of their study in that the qPCR based testing has an "indifference to living or non-living DNA" and because of this their PCR based approach may be unnecessarily sensitive. To date there are little to no guidelines for thresholds on many mycotoxins or bacterial toxins that have been established in the cannabis industry, so their findings help inform regulators as to which type of toxins might be relevant for further analysis.
The authors state, "Health compromised patients exposed to aflatoxin and clearance-inhibiting cannabinoids raise new questions in regards to the current safety tolerances to aflatoxin." but the authors present no data showing the presence of aflatoxin in any of their samples. So their caution is highly speculative and they should indicate that unless they provide data supporting their caution.
Author response: We thank the reviewer for the background and context provided to his concern.
The questions about what levels of toxins are safe or acceptable fall well beyond the scope of the present study. We concede, however, that some statements made in the introduction may have been too speculative. We have revised paragraph 2 of the introduction to focus on the numerous literature reports of pulmonary aspergillosis associated with cannabis use instead of potential mycotoxin toxicity. We have also provided more background on the overlap of cannabinoid and mycotoxin metabolism via cytochrome P450 system as follows. Monitoring for mycotoxic fungi in cannabis preparations has been recommended as part of routine safety testing by the Cannabis Safety Institute. A major driver for this recommendation has been numerous reported cases of serious or fatal pulmonary Aspergillosis associated with marijuana smoking in immunocompromised patients . The major cannabinoids have been shown to be potent inhibitors of several cytochrome P450 enzymes at therapeutic concentrations, including 1A1, 1A2, 1B1 2B6, 2C19, 2D6, 3A4 and 3A5 . Some of these enzymes have been implicated in the metabolism of the fungal toxins aflatoxin and ochratoxin . This raises questions about potential interactions and appropriate safety tolerances for mycotoxins in patients being treated with cannabinoid therapeutics. In addition, some species that produce toxins have proven to Fusarium be difficult to selectively culture with tailored media . This is a common problem associated with culture-based systems as carbon sources are not exclusive to certain microbes and only 1% of microbial species are believed to be culturable .
The first sentence of the next paragraph was edited slightly to follow more logically: While the risks of mycotoxic fungal contamination have been well studied in the food markets, the presence of the fungal populations present on flowers has never been surveyed with next generation Cannabis sequencing techniques . Author response: Upon reconsideration, we concluded that there is no way to predict how growth of the specific fungi detected might be affected by the terpenoids in Cannabis flowers. Those compounds may inhibit the growth of some fungi, while permitting the growth of others. In any case, the effect of terpenes would not relevant to the ability of the fungi to grow on culture media lacking those compounds. That sentence was deleted and the last sentence of paragraph 1 in the discussion was edited as follows: It is possible the different water activity of the culture assay compared to the natural flower environment is contributing to the false negative test results.
Furthermore they state that "While techniques exist to perform live-dead qPCR, the live status of the microbes is unrelated to toxin potentially produced while the microbes were alive. ELISA assays exist to screen for some toxins. Current state-recommended ELISA's do not detect citrinin or paxilline, the toxins produced by P. citrinum and P. paxilli, respectively. The predominance of these Penicillium species in a majority of the samples tested is interesting." Ideally the authors would test for these toxins in their most positive samples.
Author response: We stated transparently that this needed to be done as a future direction. We don't see it as an essential part of the current publication, which had the goal of simply cataloging the fungal content of dispensary-derived samples.
The authors state that, "Cannabidiol is often used at micromolar concentrations for seizure reduction implying sub-percentage contamination of paxilline could still be a concern" but this is highly speculative and the authors should de-emphasize the "concern" and state instead that if their results were verified by tests indicating high levels of paxilline then it may be cause for concern. The same is true for their concern about Citrinin and aflatoxin and the authors should state this. We agree. Paragraph 3 of the discussion was edited to address this criticism as follows: Paxilline is a tremorgenic and ataxic potassium channel blocker and has been shown to attenuate the anti-seizure properties of cannabidiol in certain mouse models . Paxilline is reported to have tremorgenic effects at nanomolar concentrations and is responsible for Ryegrass-staggers disease . Cannabidiol is often used at micromolar concentrations for seizure reduction and contamination with paxilline, if confirmed, would be a cause for concern. Citrinin is a mycotoxin that disrupts Ca2+ efflux in the mitochondrial permeability transition pore (mPTP) . Ryan et al. demonstrated that cannabidiol affects this pathway suggesting a similar potential cause for concern regarding CBD-citrinin interaction . Considering the hydrophobicity of these mycotoxins and the growing interest in the use of extracted oils from CBD-rich Cannabis strains for treatment of drug resistant epilepsy , more precise molecular screening of fungal toxins in these products might be warranted.

F1000Research
The authors state that "Appropriate primer design can survey a broad spectrum of microbial genomes while affording rapid iteration of design. Quantitative PCR has also demonstrated single molecule sensitivity and linear dynamic range over 5 orders of magnitude offering a very robust approach for detection of microbial risks. This may be important for the detection of nanomolar potency mycotoxins". The ability to detect single copies of genes makes their system highly sensitive, but does not indicate level of mycotoxin. The authors should point out the limitations of their approach and discuss the possibility that it would likely generate a high degree of false positive results compared to culture-based standard methods.
Author response: The comments about ELISA assays was deleted and the paragraph was edited to focus on detection of fungal species, not toxins, as follows. ITS amplification and sequencing offers a hypothesis-free testing approach that can be employed to identify a broad range of fungal species present in a given sample. Appropriate primer design can survey a broad spectrum of fungal genomes while affording rapid iteration of design. Quantitative PCR has also demonstrated single molecule sensitivity and linear dynamic range over 5 orders of magnitude offering a very sensitive approach for detection of microbial risks. Our survey of flowers in this study Cannabis was limited, however. Further studies are required to survey a broader range of samples, and to determine whether paxilline, citrinin, aflatoxin or ochratoxin can be detected at concentrations that represent a clinical risk in samples or extracts derived from plants that test positive for Cannabis the fungi known to produce those toxins.
The authors state, "These results demonstrate that culture based techniques superimposed from the food industry should be re-evaluated based on the known microbiome of actual Cannabis flowers in circulation at dispensaries." This statement appears to be too strong in light of their data. Without validation for the presence of toxins above a safety threshold there is no need to re-evaluate the standard methods in the food industry.
Author response: We respectfully disagree with this comment of the reviewer. The sentence was taken out of context. We were not trying to suggest that the standard methods in use in the food industry should be re-evaluated for all applications, only the use of those methods for medicinal testing. The Conclusions paragraph was modified to clarify as follows. Several toxigenic Cannabis fungi were detected in dispensary-derived samples using molecular amplification and Cannabis sequencing techniques. These microbes were not detected using traditional culture-based platforms. These results suggest that culture based techniques borrowed from the food industry should be re-evaluated for testing to ensure that they are capable of detecting the Cannabis prevalent species detected by molecular methods with adequate sensitivity. We recommend that additional sequencing studies be performed to characterize the fungal and bacterial microbiomes of a more diverse selection of samples. Such sampling should include Cannabis dispensary-derived samples from both indoor and outdoor crops, as well as samples from police seizures from well-provenanced foreign sources, such as Mexico. Finally, further studies should be performed to measure toxin levels in strains that test positive for toxigenic species. The foliar microbiome ( , phyllosphere, as opposed to rhizosphere) can be partitioned into two aka groups: live upon the leaf epidermis, and occupy intercellular spaces within the epiphytes endophytes leaf. Culture-based detection systems normally surface-sterilize plant samples, so they assume that cultured organisms are endophytes. NextGen and qPCR should identify both epiphytes and endophytes. Classic epiphytes identified in your study include and Kabatiella (Aureobasidium) microsticta Sarocladium .

(Acremonium) strictum
Phyllosphere organisms may be plant pathogens, and cause disease symptoms; diseased plants should never reach a dispensary. However, phyllosphere organisms may act as symbionts (good for the plant) or commensals (indifferent), and their asymptomatic presence is not easily detected. Nevertheless these cryptic organisms may cause disease in humans. The spores from phyllosphere fungi readily pass through waterpipes ( 2. Explain the methodology used to select three culture-based detection systems in this study. Are they the most widely-used systems? Are they the systems recommended by cannabis regulatory agencies? If the method was simply "convenience sampling," say so.
3. Methods used in the qPCR ITS assay should be described in the Methods section, not the Results section. Figure 2, "DNA sequencing of ITS2 amplicons from culture negative samples that are qPCR positive for total yeast and mold tests," deserves some comment. Some of the taxa are not yeasts or molds. They include angiosperms ), a protozoan ( ), (Zea mays, Pachysandra procumbens Sterkiella histriomuscorum and an "uncultured bacterium." Comment please. Where are the bacteria? Much of the concern over microbiology and food safety focuses on human enteric pathogens ( spp.) and opportunistic bacteria ( e.g., Escherichia coli, Salmonella e.g., Pseudomonas ). Previous culture-based studies have isolated these organisms from aeruginosa, Berkholderia cepacia ( Taylor 1982, Ungerleider 1982. The only bacterium you identify (in Figure 2