Exploring near-infrared spectroscopy and hyperspectral imaging as novel characterization methods for anaerobic gut fungi

Abstract Neocallimastigomycota are a phylum of anaerobic gut fungi (AGF) that inhabit the gastrointestinal tract of herbivores and play a pivotal role in plant matter degradation. Their identification and characterization with marker gene regions has long been hampered due to the high inter- and intraspecies length variability in the commonly used fungal marker gene region internal transcribed spacer (ITS). While recent research has improved methodology (i.e. switch to LSU D2 as marker region), molecular methods will always introduce bias through nucleic acid extraction or PCR amplification. Here, near-infrared spectroscopy (NIRS) and hyperspectral imaging (HSI) are introduced as two nucleic acid sequence-independent tools for the characterization and identification of AGF strains. We present a proof-of-concept for both, achieving an independent prediction accuracy of above 95% for models based on discriminant analysis trained with samples of three different genera. We further demonstrated the robustness of the NIRS model by testing it on cultures of different growth times. Overall, NIRS provides a simple, reliable, and nondestructive approach for AGF classification, independent of molecular approaches. The HSI method provides further advantages by requiring less biomass and adding spatial information, a valuable feature if this method is extended to mixed cultures or environmental samples in the future.


Introduction
Anaerobic gut fungi (AGF) are an early branching lineage of fungi (phylum Neocallimastigomycota) that inhabit the digestive tr act of herbivor es .T hese unique fungi ha v e e volv ed to thriv e in the o xygen-de pri ved milieu of their host's digestive system, where they act as crucial symbionts in the intricate process of plant biomass degradation and nutrient cycling (Gruninger et al. 2014, Hess et al. 2020 ).They possess a unique set of potent enzymes to breakdown lignocellulosic plant material, making them promising candidates for biotechnological applications, such as bioethanol pr oduction, gener ation of biomolecules, or biogas pr oduction from organic wastes (Dollhofer et al. 2015, Yildirim et al. 2017, Vinzelj et al. 2020, Liu et al. 2021 ).
AGF display a complex life cycle, switching between a nonmotile v egetativ e state and a mobile zoospore phase .T heir flagellated zoospor es attac h to plant material, wher e they encyst and form either a filamentous or a bulbous thallus that penetr ates and gr ows thr ough the plant substr ate.Repr oduction in AGF occurs pr esumabl y asexuall y, via formation of one or more spor angia that r elease ne w motile zoospor es upon matur ation, thus completing the life cycle.For species with monocentric, filamentous thalli, anucleate rhizoids are formed, where the nucleus remains in the cyst and only one sporangium is formed.For species with polycentric, filamentous thalli, nuclei migrate through the rhizoidal system, which may lead to the formation of m ultiple spor angia.Species with bulbous thallus de v elopment produce spherical holdfasts for the penetration of plant matter (Theodorou et al. 1996, Gruninger et al. 2014, Hanafy et al. 2022 ).
While having been described and isolated nearly 50 years a go (Or pin 1975 ), AGF ar e still poorl y understood.This is mainly due to challenges in their isolation and their complex growth r equir ements (Vinzelj et al. 2022, Meili et al. 2023 ).Further, genome-based studies on AGF are hampered by their large and re petiti ve genomes with high AT-content (up to 83%), impeding standard sequencing a ppr oac hes (Youssef et al. 2013, Edw ar ds et al. 2017, Meili et al. 2023 ).There are currently 22 described genera of AGF (Hanafy et al. 2022, Pratt et al. 2023 ) and cultureindependent techniques suggest the existence of e v en mor e than double this amount (Paul et al. 2018, Meili et al. 2023 ).The internal transcribed spacer 1 (ITS1) is a region commonly used to identify fungal species and is also fr equentl y used as a barcode for AGF (Sc hoc h et al. 2012, Edw ar ds et al. 2019, Hess 2020et al. 2020 ).Ho w e v er, this r egion exhibits significant length heterogeneity within AGF strains (Edw ar ds et al. 2008 ), with variation of up to 13% from clones within a single culture (Callaghan et al. 2015 ).This makes sequence alignment challenging and poses problems to ITS1 based phylogeny (Edw ar ds et al. 2017, 2019, Hanafy et al. 2020 ).While ITS2 is also a standard region for fungal taxonomy, it is less common than ITS1 for AGF.More recently, the usage of the 28S large ribosomal subunit (LSU), (D1/D2 domain) was br oadl y adopted by the AGF r esearc h comm unity for phylogenetic and taxonomic analyses .T his region contains negligible intr astr ain sequence div er gence and displays fe wer length heter ogeneities (Hanafy et al. 2020, Elshahed et al. 2022 ).LSU-based taxonomy has led to the construction of a well-r esolv ed phylogenetic tree, still with unr esolv ed taxonomic issues at the famil y le v el due to lack of genomic data (Hanafy et al. 2023 ).Overall, the challenges posed by the unique genomic features of AGF and struggles with commonly used fungal barcodes (Schoch et al. 2012, Edw ar ds et al. 2017 ) make alternative identification and characterization methods of high interest.
Apart from the genomic sequence of an organism, the relative abundances of biomolecules in its cells, whether they are used for cell-structur e, metabolism, or ener gy stor a ge, can pr ovide insights into the unique biochemical composition of an organism.Hence, the chemical configuration of cells can be used for discrimination of strains .T his is where analytical techniques, such as mass spectr ometry and infr ar ed spectr oscop y, expand our kno wledge of species identification via a more wholistic cellular c har acterization on top of standar d DN A-based a ppr oac hes (Ngo-Thi et al. 2003, Erukhimovitch et al. 2005, Krásný et al. 2013, Chalupová et al. 2014 ).
In recent decades, near-infrared spectroscopy (NIRS) has emerged as a po w erful tool to discriminate between different bacterial str ains, e v en down to the subspecies le v el (Cámar a- Martos et al. 2012, Quintelas et al. 2018, Tian et al. 2021 ).NIRS involv es the measur ement of light absor ption in the near-infr ar ed region of the electromagnetic spectrum (750-2500 nm).This technique exploits the overtone and combination vibrations of various molecular bonds, giving br oad absor ption bands.This results in a so-called "spectr al finger print" of eac h sample.Variations in k e y biomolecules , including proteins , lipids , and carbohydrates , can be detected with NIRS and conclusions about their r elativ e abundance and hence the composition of the sample can be drawn.Further, thr ough c hemometrics and m ultiv ariate data-anal yses, these variations can be used to correlate samples with certain spectra and thereby classify unknown samples to a r efer ence str ain.NIRS is further adv anta geous for biological samples as it is nondestructive and requires minimal sample preparation.This enables further downstr eam anal ysis of samples, suc h as nucleic acid extraction, mass spectrometry, or biochemical characterization (Burns and Ciurczak 2008, Pasquini 2018, Ozaki et al. 2021 ).
The implementation of NIR for identification of filamentous fungi is less established and fungal studies gener all y use visible light in combination with NIR or hyperspectr al ima ging (HSI) a ppr oac hes to c har acterize str ains (P etisco et al. 2008, Piekar czyk et al. 2019, Lu et al. 2020 ).Ov er all, the a pplication of NIRS on whole biomass of fungi poses a valuable alternative discrimination method, independent of standard molecular a ppr oac hes, complementary to the already common practice of MALDI-TOF MS (Hendrickx 2017, Gómez-Velásquez et al. 2021 ).
HSI-a technique that combines features of light microscopy and infr ar ed spectr oscopy (Khan et al. 2018, Wang et al. 2018 )can yield further insight into the spatial composition of samples.Ther eby, spectr al information on specific colonies of micr oor ganisms or distinct features of fungal thalli, with resolutions up to a few μm are feasible (Go w en et al. 2015, Soni et al. 2022 ).HSI is curr entl y used to answer complex r esearc h questions that r equir e spatial r esolution, suc h as detecting food contamination (Soni et al. 2022 ) or differentiating developmental stages of fungal growth on agar plates (Lu et al. 2020 ).For samples with little available biomass, mid-infr ar ed (MIR) light (2.5-25 μm) could pr ov e adv anta geous ov er NIR, as absor ption is gener all y m uc h str onger in this region.This is due to the lo w er energy of MIR light, which excites the more commonly occurring fundamental vibration modes within molecules .T hese fundamental excitations can be assigned to certain molecular bonds and hence, as compared to the combination and overtone bands in NIR, also functional gr oups ar e identified mor e easil y (P asquini 2018(P asquini , Be ć et al. 2022 ) ).
In this study, based on the intriguing insights that the NIRS and HSI methods can offer when inspecting biological samples, and to expand our discrimination efforts beyond the standard molecular a ppr oac hes for AGF, we had the three following aims: (i) NIR classification: whether NIRS is a valid technique to differentiate among AGF strains and further, if this is possible, which functional groups (and hence which biomolecules) are responsible for this discrimination; (ii) cultur e a ge anal ysis and model extension: if it is possible to detect changes in biomass composition of fungal cultures over time and whether fungal strains harvested at different cultivation times could still be differentiated by a NIRS model; (iii) HSI: lastly, w e w anted to test HSI as a tool for discrimination of AGF str ains and e v aluate the suitability to identify morphologicall y differ ent structur es within a str ain.

Str ain cultiv a tion and independent char acteriza tion
Thr ee pur e cultur es of AGF str ains wer e used for differ entiation in these experiments .T hey were chosen to serve as refer ence for differ ent gr owth c har acteristics of AGF: Anaerom yces mucronatus (polycentric, filamentous growth; sequence accession number: ON614226-ON614231), Caecomyces communis (monocentric, bulbous growth; sequence accession number: OP216660), and P ecoram yces ruminantium (monocentric, filamentous gr owth; str ain C1A sequence accession number: JN939127; Youssef et al. 2013 ).Eac h str ain w as identified b y micr oscopic inspection (ima ges displayed in appendix Supplementary Fig. 9 ) and sequencing of the D1/D2 region of the LSU using the GGNL1F and GGNL4R primer pair (Nagler et al. 2018 ).The absence of commonly co-occurring methanogens was confirmed by brightfield-and fluorescence microscopy (Co-Factor F 420 ) and PCR against the V4 region of 16S RNA with the 515f and 806r primer pair (Ca por aso et al. 2011 ).Further, no methane production was detected with gas composition measurement according to Wunderer et al. ( 2022 ) (GC-2010, Shimadzu).Defined media (omission of clarified rumen fluid, tryptone, and yeast extr act; Str obl et al. 2024 ) media, containing no antibiotics was used for AGF biomass generation.Briefly, 150 ml of salt solution 1 and 150 ml of salt solution 2 were combined and 2 g of xylan (from beech wood, Carl Roth), 2 ml of resazurin, 2 ml of hemin, and 700 ml of distilled w ater w ere added.The solution was heated up close to the boiling point and was then cooled under pure CO 2 flux.Then, 6 g of NaHCO 3 , 3 g of cellobiose (Carl Roth), 10 ml of trace element solution, and 1 g of l -cysteine HCl were added, and pH was adjusted to 6.9 with NaHCO 3 .A volume of 0.01 ml of vitamin solution per 1 ml of medium was added just before inoculation.Anoxic serum bottles containing 45 ml medium were inoculated with 5 ml of w ell-gro wing, 1-w eek-old fungal cultur es fr om the AGF cultur e collection at the Department of Micr obiology at Universität Innsbruck.

Sample preparation
30 pur e cultur es per str ain wer e gr own for 7 days in 50 ml defined medium (total n = 90) at 39 • C in anoxic serum bottles (henceforth, this group will be referred to as "1 w").Samples were then centrifuged (12 000 rcf for 4 min) and the supernatant was discarded.Subsequently, the fungal biomass was washed thrice with 30 ml deionized w ater, discar ding the supernatant after each centrifugation step.Samples were then lyophilized (VaCo 2, Zirbus, Bad Grund, Germany).The obtained fungal po wder w as homogenized with a spatula before being filled into quartz reflectance cuv ettes for measur ement and pac ked with a metal cylinder.As not all strains had produced enough biomass for downstream analysis an ov er all sample number of 78 was ac hie v ed [ P. ruminantium ( n = 25), C. communis ( n = 25), and A. mucronatus (n = 28)].For the cultur e a ge anal ysis and model extension (ii) additional biomass samples, harv ested at differ ent time points after inoculation wer e gener ated.Thr ee samples of each strain were taken at 72 h after inoculation, as preliminary experiments demonstrated the highest zoospore density for this timepoint (this group will be referred to as "72 h").For 72 h samples, larger bottles containing 100 ml of medium were used to generate sufficient biomass after shorter gr owth time.Additionall y old cultur es wer e taken [ P. ruminantium ( n = 6), C. communis ( n = 3), A. mucronatus ( n = 8)] that had been left at 39 • C for over 3 weeks, up to 3 months (this group will be r eferr ed to as "≥3 w").Samples of the 72 h and ≥3 w group were treated as described above prior to measurement.

Measurement
Samples wer e measur ed with the Büc hi NIRFlex N-500 (Büc hi Labortechnik A G , Flawil, Switzerland) using the Solids XL top piece.Each sample was measured in triplicate using reflectance mode and 32 scans per measur ement.Spectr a wer e obtained at a wav elength r ange of 4000-10 000 cm −1 and a r esolution of 4 cm −1 .The chitin standard (chitin from shrimp cells, Sigma Aldrich) was measur ed identicall y to the l yophilized samples.

Sample preparation
Thr ee biological r eplicates of pur e cultur es of eac h str ain wer e grown for 72 h in 50 ml defined medium (total n = 9).For the HSI a ppr oac h, a shorter cultiv ation period was used to pr e v ent formation of a ggr egates in fungal cultures, as this could lead to consider able ov erlay of mor phologicall y differ ent structur es during microscopy (see Supplementary Fig. 9 B).Twice, 2 ml of culture were withdr awn fr om eac h bottle via a syringe and centrifuged at 500 rcf for 5 min, discarding the supernatant.Centrifugation speed w as lo w er ed for this a ppr oac h, compar ed to NIRS sample pr epar ation, to pr eserv e the mor phological integrity of the str ains .T hen samples were washed with 1 ml of 1x PBS solution, centrifuged as before, and the supernatant was discarded.The harvested cells wer e then r esuspended in 250 μl of a 1:1 solution of 96% ethanol and 1x PBS solution.Two replicates of each suspended cell solution (each 10 μl) were put on CaF 2 -plates for imaging analysis and dried at room temperature .T he three biological replicates, together with two tec hnical r eplicates fr om eac h cultur e bottle and the duplicate measurement resulted in 12 plates per str ain.Befor e measur ement, CaF 2 -plates wer e dried by an ethanol series (50%, 80%, and 96%) and lastly, samples were dried at 50 • C for 30 min.The ethanol series had the additional effect of removing remaining salt from the ethanol/PBS washing step.

Measurement
Due to the low absorption intensities in the NIR-region, MIR was chosen as appropriate wavelength region for fungal specimen with only a few μm thickness (Pasquini 2018, Ozaki et al. 2021 ).Imaging was carried out using the PerkinElmer Spotlight 400 FT-IR Imaging System and the PerkinElmer Spectrum 400 FT -IR/FT -NIR Spectrometer (PerkinElmer Inc., Waltham, MA, USA).Transmission mode was employed for measurement.The wavelength range used was 1000-4000 cm −1 with a resolution of 4 cm −1 .A pixel size of 6.25 μm was chosen with 16 scans per pixel.For each plate, a minimum of three different cell structures with thickness, leading to sufficient absorption of MIR light, were selected, resulting in 156 total measurements [ P. ruminantium ( n = 59), C. communis ( n = 47), and A. mucronatus ( n = 48)].Additionally, for P. ruminantium 39 spectra of sporangia were collected, as this strain exhibited most c har acteristic, mor phological differ ences between hyphae and sporangia.

NIR
All spectr a wer e pr ocessed using The Unscr ambler X (Version 10.5, C AMO, Oslo, Norwa y).For NIR, the three technical replicate measur ements of eac h sample wer e r educed to giv e an av er a ge spectrum and then transformed from reflectance to absorbance data.Spectr a wer e then tr ansformed thr ough standard normal v ariate (SNV) to normalize the spectra and the second deri vati ve was performed using the Sa vitzky-Gola y method (Sa vitzky and Golay 1964 ) with polynomial order 2 and 11 smoothing points.For NIR classification analysis , wa velengths where no water absorption occurs were selected, as this rendered best results due to the absence of interfer ence fr om differ ent degr ees of r esidual moisture in the samples .Wa venumbers 4000-4988, 5452-6608, and 7136-10 000 cm −1 were chosen as optimal regions for differentiation of strains.Principal component analysis (PCA) was performed.The classification dataset (total sample number 78) was then r andoml y split into a calibr ation and v alidation set at the commonly used ratio of 70/30 (calibration set 54 samples; validation set 24 samples).A linear discriminant analysis (LDA) was performed with the calibration set, using the Mahalanobis method (Mahalanobis 1936 ) and se v en components .T he created LDA was used to predict the unknown samples of the independent validation set.This was carried out in triplicate to obtain reliable results fr om r andom sample splitting.
For the culture age analysis and model extension (ii) the dataset was enlarged to include spectra of strains at different time points post inoculation (72 h, 1 w, > 3 w).PCA was performed.The validity of LDA to discriminate samples by their strain at differ ent gr owth stages was tested by running an LDA (Mahalanobis method; Mahalanobis 1936 ; eight components) on the entire sample set.A splitting into calibr ation/v alidation set was not possible as before, as not sufficient samples fr om differ ent gr owth sta ges wer e av ailable to be r andoml y included in either set.

HSI
For the HSI a ppr oac h, ima ges wer e pr ocessed using SpectrumImage R1.9.0.0030 (Perkin-Elmer Inc.) and the respective spectra for eac h str ain wer e cr eated thr ough coaddition of a ppr opriate ima ge ar eas.Spectr a wer e tr ansformed fr om tr ansmittance to absorbance data and SNV was performed.Water and carbon dioxide absor ption r egions wer e r emov ed to exclude effects fr om moisture fluctuations in samples and atmospheric carbon dioxide on models .T his a ppr oac h r esulted in a wav enumber r egion of 1000-1772 cm −1 that was used further on.PCA and LDA analysis were run on obtained data.For LDA, a slightly lo w er sample split ratio than the pr e viousl y used 70/30 was employed, with 100/56 for calibration and validation set, respectively.As in the general NIR classification (i), LDA was calculated and used to predict the unknown samples of the independent validation set at three iterations.

NIR classification (i)
Fungal spectra sho w ed a pattern similar to pure chitin, which is an abundant molecule in the AGF biomass (Fig. 1 ) (Rezaeian et al. 2004b, Ga y 1991 ).T he first observed peak (4000 cm −1 ) is a combination of C-H str etc hing with C-C str etc hing.In the highlighted region A1, AGF peaks did not match the absorption peaks of the chitin spectrum.The peaks (4324 and 4270 cm −1 ) were assigned to C-H combinations of lipids.Region A2 encompassed three distinct peaks, where the outer two (4840 and 4592 cm −1 ) are associated with amides and the central peak (4800 cm −1 ) is associated with O-H functional groups.Region A3 contained absorption from C-H combinations and C-H first ov ertone str etc hing, whic h ho w e v er cannot be assigned to distinct biomolecules (Workman and Weyer 2007 ).Ne v ertheless, these r egions wer e r ele v ant for the discrimination of strains.Mean spectra of the pigmented fungal samples sho w ed a stronger absorption at high wavenumbers ( > 9000 cm −1 ) when compared to the white chitin reference.
The three tested strains were clearly separated in PCA analysis (Fig. 2 ).Components 2, 3, and 5 sho w ed the shar pest separ ation.While PC1 (41%) found the strongest difference among samples, it was not useful for separation of strains.PC4 (6%) only separated three C. communis samples and, therefore, was also not useful for str ain differ entiation, hence PC5 was used instead.The loadings plot emphasizes how the selected regions A1-3 (compare Fig. 1 ) are most important for the differentiation of strains in multivariate analysis .T he used components explained 52% of the o v er all variance.
For the classification of str ains, eac h r andom sample split resulted in a prediction accuracy of the LDA model, built on the calibration set of 100%.Each of the three independent validation sets was predicted with an accuracy of 96%.

Culture age analysis and model extension (ii)
The spectra of fungal strains sho w ed clear differences in their absor ption c har acteristics based on cultur e a ge (Fig. 3 ).The oldest cultures ( ≥3 w) exhibited the str ongest intr a gr oup heter ogeneity.In the lipid region (A1), the oldest cultures displayed the strongest absorption, while the youngest cultures did so in the C-H region (A3).In the protein and carbohaydrate region (A2), a pattern of incr easing absor ption of the amide bands (4840 and 4592 cm −1 ) compared to the O-H absorption (4800 cm −1 ) over time was observed.At 72 h the O-H peak almost matched the amide absorption, while in the ≥3 w samples the peak could bar el y be observed.The region > 9000 cm −1 sho w ed an absorption increase from 72 h, over 1-3-week-old cultures.
The three tested culture ages were clearly separated by PCA (Fig. 4 ).The ≥3 w samples again sho w ed the strongest intragroup heterogeneity.As in Fig. 2 , the loadings plot emphasizes the importance of the highlighted regions (compare Figs 1 and 3 ) for the differentiation of different culture ages .T he PC A explained 91% of the ov er all v ariance.
The pr ediction accur acy of the LDA model discriminating strains, calculated for the different culture ages, was 98%.

HSI (iii)
The gener al sc hematic pr ocedur e for gener ating the r espectiv e spectra with the HSI approach is portrayed in Fig. 5 .HSI helped to not only gain re presentati ve spectra for each strain, but also spectra of morphologically different structures , i.e .hyphae versus sporangia.The mean spectra sho w ed differences in their absorption c har acteristics in the MIR region, with the finger print r egion (1772-1000 cm −1 ) exhibiting strongest differences .T he morphological structures were discriminated by PCA using the fingerprint r egion (a ppendix Supplementary Fig. 8 ).As the spor angial spectra sho w ed a m uc h gr eater intr a gr oup heter ogeneity, onl y hyphal images and images containing hyphae as well as sporangia were used for discrimination through LDA.
A str ong absor ption peak fr om O-H str etc hing was seen in all av er a ged spectr a (3280 cm −1 ).As this could not onl y arise fr om c hitin and carbohydr ates, but also fr om r esidual moistur e in the samples, the entire region was excluded for strain differentiation.This included also the C-H str etc hing peaks from CH 2 (2960 and 2872 cm −1 ) and CH 3 (2928 and 2852 cm −1 ) groups, as well as CO 2 absor ption fr om atmospheric carbon dioxide (ca.2349 cm −1 ).This left the fingerprint region for robust classification to 1800-1000 cm −1 .Here, peaks of amide I (1644 cm −1 ), arising from C = O str etc hing, amide II (1544 cm −1 ), arising fr om C-N str etc hing and N-H in plane bending; and amide III (1244 cm −1 ), originating from a complex mixture of functional groups present in amides, as well as peaks from chitin and proteins, were identified (López-Lorente and Mizaikoff 2016 ).C-H bending was also observed (1452 and 1384 cm −1 ).C-O str etc hing peaks associated with ethers (1152 cm −1 ), and alcohols (1080 and 1030 cm −1 ) fr om c hitin and carbohydr ates wer e observ ed.Peaks with contributions from antisymmetric (1240 cm −1 ) and symmetric (1080 cm −1 ) str etc hing of phosphate groups were identified, which can be associated with DN A, RN A, and phospholipids .T he peak at 1240 cm 1 was not observed in the chitin spectrum, corroborating the assignment to phosphate groups.
The three strains were clearly separated by PC A (Fig. 7 ).T he influence of the r espectiv e peaks in the fingerprint region on discrimination can be observed in the loadings plot.The PCA explained 96% of the ov er all v ariance.
For classification, the random sample splitting resulted in an av er a ge pr ediction accur acy of 99% (SD ± 1%) with the LDA model.For the independent validation set an av er a ge pr ediction accur ay of 97% (SD ± 3%) was ac hie v ed.

NIR classification
NIR has been pr ov en a suitable method for c har acterization of AGF biomass composition and in addition a valid discrimination method for the tested AGF strains .T he gener al absor ption tr end of fungal cultures follo w ed the absorption characteristics of chitin, which together with molecularly similar glucans, are the most abundant cell wall components of fungi (Ruiz-Herr er a and Ortiz-Castellanos 2019 ).The differentiation of our fungal strains by the NIRS model w as, ho w e v er, mostl y based on differ ences in the relative abundance of other macromolecules found in fungal cells, namel y carbohydr ates , lipids , and proteins .For example , the observed C-H combination peaks (Fig. 1 , highlighted region A1) of the AGF samples did not match the chitin spectrum and could be   2009) found c har acteristic peaks fr om fatty acids in the NIR-spectra of Ascomycota strains at these positions, corr obor ating the assignment of these specific peaks to fungal lipids.Different compositions of membrane and stor a ge lipids of AGF strains led to differences in the spectral fingerprint of the samples in region A1, allowing for differentiation of strains with our NIRS model.Chitin and other biomolecules could, ho w e v er, still have an influence on this region.In the second highlighted region A2 (Fig. 1 ), the two observed amide peaks likely arose from pr otein-and c hitin absor ption, while the observ ed O-H peak likel y stemmed fr om carbohdydr ate-and c hitin absor ption (Workman and Weyer 2007 , Ishigaki et al. 2021 ).They can ther efor e be used to estimate the r elativ e abundance of said biomolecules in fungal biomass .Anaeromyces mucronatus displa yed the highest amide absorption, leading to the conclusion that this strain has the highest protein + chitin to carbohydrates + chitin ratio.Caecomyces communis displayed the strongest O-H absorption compared to the other peaks, suggesting a lo w er protein + chitin content of this str ain, with a r elativ el y higher carbohydr ate + c hitin content.P ecoramyces ruminantium also exhibited high O-H absorption.Differences in the composition of fungal biomass with higher or lo w er abundances of protein and carbohydrates further allo w ed the discrimination of strains by the NIRS model.The third highlighted region A3 (Fig. 1 ) contained absorption from overtones and combinations of C-H str etc hing.C-H bonds ar e pr esent in all macr omolecules constituting organic biomass, and therefore a broad absor ption band fr om man y ov erla pping peaks is expected.Due to this str ong ov erla p, assignment to distinct macr omolecules was not possible.Ho w e v er, differ ences in r elativ e abundance of certain molecules and hence differences in shapes of the absorption band can still be picked up by the model for correct differentiation of strains (Burns andCiurczak 2008 , Pasquini 2018 ).
Despite the importance of the highlighted regions, the remaining absor ption r egions wer e r ele v ant for fungal classification as well.Models using only the wavelengths in regions A1-3 performed worse compared to models using the entire spectrum without water bands (data not shown).An explanation for this could be differences in the absorption region of C-H second overtones and combinations (roughly 8250 cm −1 ), differences in absorption of pigments near the visible light region (above 9000 cm −1 ) or influence of the C-H + C-C combination (4000 cm −1 ) (Calderón et al. 2009 , Workman andWeyer 2007 ).
The classification of AGF strains through NIRS revealed reliable and accur ate r esults, with a high pr ediction accur acy of 96% for unknown AGF samples in the independent validation set.In comparison, Sc hoc h et al. ( 2012) performed a r ound-r obin test for identification of Fungi by standard DNA barcode regions such as ITS and LSU.For earl y div er ging linea ges, whic h contain the Neocallimastigomycota, pr obabilities of corr ect identification of 62% for ITS and 75% for LSU were reported.One, ho w ever, has to consider that in this study discrimination of AGF strains has been carried out on the genus le v el with mer el y thr ee out of the 22 described AGF genera.This limitation was mainly caused by the limited availability of pure cultures and with the three used being the only available pure cultures with our research consortium.Further, r ecentl y the de v elopment of v arious primer pairs specifically designed for detection and/or quantification of AGF strains (Kittelmann et al. 2012, Edw ar ds et al. 2017, Young et al. 2022 ), as compared to standard fungal barcoding primers used by Sc hoc h et al. ( 2012 ), hav e enhanced detection, identification, and discrimination of AGF.No current study, ho w e v er, has inv estigated the success rates of AGF identification with these state of the art molecular techniques in general laboratory settings.While the comparison of identification accuracies highlights the potential of the NIRS method in combination with LDA modeling, further compar ativ e studies ar e r equir ed.The corr obor ation of the NIRS to discriminate more strains at genus or even species level needs to be tested by including additional strains.
Besides identification accuracy, other aspects of standard nucleic acid-based techniques should be consider ed.Sc hoc h et al. ( 2012), for example, reported a PCR amplification success rate of only 65% for early diverging fungal lineages.As the NIRS method does not r equir e PCR amplification this is an additional beneficial feature of this approach for AGF assignment to reference spectra.Further, the proposed method does not require chemicals for nucleic acid extraction and sequencing (Rittenour et al. 2012 ), making it pr efer able considering the principles of green chemistry (DeVierno Kreuder et al. 2017 ), and eliminating potential biases brought in by nucleic acid extraction and PCR.
An additional benefit of NIRS is the nondestructive nature of the technique that allows for further downstr eam anal ysis of samples.In this study, further measur ements wer e conducted with the same samples to differ entiate str ains thr ough DART mass spectrometry and MALDI-TOF MS (unpublished results).
To the best of our knowledge, this is the first study to report the use of NIR for the identification of fungal biomass.As could be shown for the selected AGF, this method allows the simple, fast, and nondestructive distinction of fungal pur e cultur es.With the de v elop model, the thr ee selected cultur es can be discriminated and unknown cultures from these strains identified.The inclusion of more known strains as pure culture references could enable the de v elopment of a model for discriminating many more AGF strains.It could therefore be used as an alternative approach to molecular identification and c har acterization of str ains, a common practice in standard laboratory settings with MALDI-TOF MS (Hendrickx 2017, Gómez-Velásquez et al. 2021 ).It bears the additional benefit of being c hea per and requiring less chemicals than MALDI-TOF MS.Ho w e v er, NIRS identification r equir es fungal samples to be acquired in pure culture and with sufficient biomass to carry out measurements .T he HSI approach circumvents these limitations by adding spatial resolution to the data that allows for c har acterization on a small scale, with little biomass and the abil-    C. communis (red), and A. mucronatus (blue)] and c hitin r efer ence (gr een).The finger print r egion is highlighted as it is most r ele v ant for differ entiation of samples.Bottom: Individual spectr a of AGF samples in the zoomed-in finger print r egion ar e shown with the same color coding as abov e including peak assignment to functional gr oups.ity to separate fungi from particular plant matter found in growth media, or using mixed cultures for analyses (see HSI).

Culture age analysis and model extension
As the growth and maturation of fungal strains corresponds to the formation of different quantities of certain macromolecules, AGF biomass can exhibit different amounts of c hitin, pr otein, carbohydrates, or lipids depending on the culture age or growth phase (Phillips andGordon 1989 , Gay 1991 ).Since differentiation of strains in the NIRS model is based on the r elativ e abundances of these macr omolecules, cultur e a ge could hav e a significant effect on the robustness of the model.We therefore investigated whether identification of strains was still possible with strains harv ested after differ ent gr owth times.We tested cultur es gr own for 72 h, 1 week, and over 3 weeks, to reflect cultures in their most activ e gr owth phase, matur e-, and e v entuall y decaying cultur es, r espectiv el y.
The LDA of AGF samples fr om differ ent cultur e a ges r ender ed a good prediction accuracy of the model for discriminating the strain (98%).With the limited number of samples from each strain at different culture ages, a random splitting of the sample set into calibration and validation set was not possible.Despite the shortfall of external validation, the accurate model predictions sho w ed that discrimination of samples by strain was still possible when different culture ages are used.As only one additional component was employed compared to the LDA model used for aim (i), the chance of overfitting was low (Reyna et al. 2017 ).With a larger sample set, including more strains harvested at each selected time point, a mor e r obust model, encompassing various differ ent cultur e a ges could ther efor e be designed and v erified by external validation (Reyna et al. 2017, Pasquini 2018 ).
Figure 7. Multiv ariate anal ysis for differ entiation of AGF str ains with the HSI a ppr oac h.On the left a PCA scor es 3D plot of AGF str ains [ P. ruminantium (black), C. communis (red), and A. mucronatus (blue)] is depicted.On the right the loadings plot shows the influence of the finger print r egion (1772-1000 cm −1 ) on the discrimination of the r espectiv e str ains .T he PC A ordination explains for 96% of toal variance (PC1: 80%, PC2: 12%, and PC3: 4%).
Ov er all, 3-week-old samples showed the largest intragroup heterogeneity (Figs 3 B and 4 ).This is likely a result of the largest time interval in between individual samples of this group, with up to 2 months.Ne v ertheless, taking this into account, the > 3 w samples still clustered well together in multivariate analysis.
We further investigated the changes in biomass composition ov er time thr ough inter pr etation of NIR absorbance of the respectiv e cultur e a ges and found clear differences in the c har acteristics of the mean spectr a (Fig. 3 ).Especiall y in region A2, a clear shift of absorption over time was observed.Here, the intensities of the O-H band decreased, while the absorption of the amide bands incr eased, when moving fr om younger to older cultur es .T his indicates a decrease of carbohydrate + chitin and an increase of pr otein + c hitin r atios in the fungal biomass o ver time .T his result was also corr obor ated by pr e vious studies .Ga y ( 1991 ), for example, used biochemical methods to determine the protein and chitin content of AGF strains and found an increase in protein and mor e significantl y an incr ease in the pr otein to c hitin r atio o ver time .Phillips and Gordon ( 1989 ) used biochemical methods to determine the stor a ge carbohydr ate content of AGF strains and found the highest amounts of carbohydrates in fungal biomass during active growth from 18 to 75 h.Interpretation of the absorption of C-H functional groups in regions A1 and A3 was not as straightforw ar d, as changes over time could result from a multitude of different biomolecules, as described before (see NIR classification).
The increase of absorption with culture age in the region > 9000 cm −1 could be caused by an increase of pigment content, which absorbs in the visible light region and absorption shoulders r eac hing into the NIR range at high wavenumbers (Calderón et al. 2009, Muniz-Miranda et al. 2019 ).Dark coloration of fungal cultur es ov er time is commonl y observ ed in AGF cultur es and is belie v ed to be some kind of a stress response and may e v en correlate to the formation of pigmented aer o-toler ant r esting sta ges (Wubah et al. 1991 ).While the absorption in this region matched pigmentation patterns, it could also be stemming from commonly observed scattering effects at high wavenumbers in the NIR re-gion.These artifacts could lead to a ppar ent differ ences in spectr a, due to different particle sizes of the dried fungal biomass matrices (Burns andCiurczak 2008 , Xie andGuo 2020 ).Inter estingl y, despite clear differences in the absorption in this region, it had very little influence on the separation of samples for the LDA based on cultur e a ge (see loadings plot; Fig. 4 ).
In addition to culture age, the selected medium for fungal growth could have significant influence on the observed spectral patterns of each strain.A switch from cellobiose and xylan as used in these experiments to other C-sources could lead to differ ent r elative abundances of macromolecules in AGF biomass or changes in the speed of accumulation of macromolecules over time.For instance, when cultures in our experiment w ere gro wn on the same medium with the addition of yeast extract, a much faster build up of proteins was observed (data not shown).Due to these influences, a standardized medium as well as growing time is recommended when using NIR as a c har acterization and discrimination tool for AGF.Additional studies could shed light on the influences of media composition on AGF biomass build up over time.
While a proof of principle for identifying AGF str ains fr om differ ent cultur e a ges has been shown with this study, further inv estigations with larger sample sets are needed to r obustl y confirm the applicability of the NIRS method to classify AGF strains at differ ent cultur e a ges .T he additional use of HSI could help to shed light on the influence of certain morphological structures on the ov er all bioc hemical composition of fungal biomass.

HSI
The HSI a ppr oac h yielded accur ate and r eliable classification r esults, with a prediction accuracy of 97% in independent validation sets .T his was e v en higher than the classification accuracy of the NIRS model (see NIR classification), corr obor ating HSI as another sound alternative for AGF strain classification.With HSI, only minimal biomass is needed and a few hyphae can generate sufficient absorption data for spectroscopic classification of strains .T his not only enables specific detection and differentiation of strains not available in pure culture, it also overcomes one of the main limitations of the NIRS a ppr oac h, whic h r equir es lar ger amount of biomass.An additional benefit is the ability to harvest consecutiv el y limited amounts of biomass from the same culture bottle, facilitating the study of fungal biomass over time, as compared to the necessity of whole culture harvest for the NIRS method.
Due to the utilization of MIR compared to NIR, the assignment of peaks in HSI spectra was more straightforw ar d.With MIR, fundamental excitations can be correlated to distinct functional gr oups, compar ed to the difficult inter pr etation of combination bands in NIR.An example is the direct assignment of peaks to phosphate groups (e.g.1240 cm −1 ; see Fig. 6 ).Absorption from this functional group cannot be detected with NIR, as o vertones , which occur at roughly double the wavenumber as the fundamental excitation, would only be detectable in the MIR region.Howe v er, these ov ertones ar e not detected in MIR, as they exhibit m uc h lo w er absorption than fundamental excitations, leading to them being concealed.Combination bands of C-H str etc hing with phosphate groups could occur in the C-H region of the NIR spectrum (Fig. 2 , region C), as the sum of the individual wavenumbers (ca.2900 and 1240 cm −1 , r espectiv el y) adds up to just over 4000 cm −1 .Ho w e v er, these bands also contain mor e fr equentl y occurring C-H str etc hing plus C-H bending combinations (ca.2900 and ca.1400 cm −1 ), masking phosphate contribution (Burns and Ciurczak 2008, Pasquini 2018, Be ć et al. 2022 ).
Ov er all, individual fungal spectr a sho w ed clear differences in their spectral fingerprint between strains, while strong similarities wer e observ ed for samples fr om the same str ain.The most pr ominent peaks in the MIR spectra of AGF strains were the amide peaks (see Fig. 6 ) that are associated with chitin as well as with protein.Furthermore, ether or alcohol functional groups contributed to the C-O str etc hing that was associated with carbohydrates or chitin and observed in all AGF strains (Salman et al. 2010 , Prabu andNatarajan 2012 ).The same region of the spectra, ho w ever, might also be influenced by phospholipids, DNA, and RNA absorption (Parker and Quinn 2013 ).Interestingly, this region was very prominent for A. mucronatus samples, possibly overshadowing absorption at other wa velengths .T his could be a result of strong absor ption fr om membr ane lipids, RN A, or DN A molecules for this str ain.Anaerom yces mucronatus displays pol ycentric, filamentous gr owth, with man y highl y br anc hed and constricted hyphae (Breton et al. 1990 ).This ov erla p of m ultiple hyphae could r esult in strong absorption from phospholipids in this region.A contribution fr om mor e e v enl y distributed nuclei and associated DNA could also be an explanation for this strong absorption.Overall, the absorption from this region had a strong influence on the discrimination of strains in multivariate analysis (Fig. 7 ).
PCA of morphological structures of P. ruminantium specimen r e v ealed a clear separation of hyphae and sporangia (appendix Supplementary Fig. 8 ) with sporangia displaying higher heterogeneity than hyphae .T his stronger heterogeneity of sporangia spectra can be explained through different degrees of development of sporangia.Young sporangia could exhibit different composition of macromolecules compared to fully mature sporangia.Furthermor e, hyphae wer e scanned as networks with m ultiple o verla ying structures , hence alread y re presenting a complex image.
HSI allows for differentiation of morphological structures for P. ruminantium samples and r e v eals differ ences in their composition due to the r espectiv e absor ption intensities.Anaerom yces mucronatus was not investigated in this way, as the strain available in our lab sterilely reproduces through hyphal growth and therefore does not contain sporangia, leading to mor phologicall y ho-mogeneous thalli (Fliegerová et al. 2002 ).The bulbous thalli of C. communis with little to no hyphal structures (e.g.holdfasts), show little mor phological v ariation (especiall y in dried samples), which is why this strain was not in vestigated.T he approach for detecting mor phological differ ences in AGF cultur es could, ho w e v er, easil y be extended to other monocentric or nonsterile polycentric AGF strains.
Ov er all, HSI giv es us the adv anta ge of visual and spectral information on the same sample (Wang et al. 2018, Soni et al. 2022, Wu et al. 2022 ).It could ther efor e be used to addr ess similar r esearc h questions as FISH, while circumventing the need to design specific probes and protocols (Moter andGöbel 2000 , Baschien et al. 2001 ).Futur e r esearc h could focus on the identification of str ains gr owing in cocultur e, as individual AGF could be visuall y distinguished from one another and then classified by their r espectiv e spectral pattern.As such an experiment with competitive exclusion in artificial cocultures could be considered.Furthermore, AGF ar e often cultiv ated in media containing plant material (e.g.wheat str aw, rice str aw, and sor ghum; Stabel et al. 2021, Vinzelj et al. 2022 ), which makes classification with NIR more challenging, as the plant material remains in the sample after processing.With spatial resolution, the absorption from plant particles could be avoided, and through prior collection of r efer ence spectr a fr om plant material and fungi, HSI could also allow for the visual differentiation of the tw o b y the r espectiv e spectr al information (Lu et al. 2020, Soni et al. 2022 ).With the present study, we could, howe v er show a fundamental proof of concept.
In general, the HSI approach could potentially lead to classification of AGF from en vironmental samples , such as animal feces or rumen.Rezaeian et al. ( 2004a ) described the visual detection of AGF from digested plant material taken from sheep intestines.As such, rumen content or feces could be microscopically scanned for fungal thalli, of whic h r epr esentativ e spectr a could be recorded.The spectral information could then be used to assign found fungal thalli to specific species .T his could give information about the presence and frequency of occurrence of specific species in environmental samples, and hence enable a culture-independent method for AGF classification.Ho w ever, prior formation of novel models including man y, e v en yet uncultiv ated, AGF str ains and assessment of the matrix effects of different intestines and feces contents on AGF classification would be r equir ed for reliable applications.

Conclusion and outlook
The described NIRS method demonstrates a simple and accurate a ppr oac h for AGF classification.As a proof of principle, we could also demonstrate that using cultures of different age still enabled separ ation of str ains by a discriminant model.Ho w e v er, pur e cultures with sufficient biomass are required for the NIRS method.
The HSI a ppr oac h onl y r equir es minimal biomass and adds spatial resolution to the gathered data.This enables the classification of strains with little material available and could e v en allow for classification in medium containing particular matter or other impurities.Future studies could look into the possibilities of AGF classification harvested from more complex growth substrates, whic h could potentiall y e v en enable AGF classification from environmental samples, leading to a culture-independent classification a ppr oac h.For this , hurdles , such as matrix effects of different environmental samples and the inclusion of more strains, even uncultured ones, would need to be overcome.
In the present study, three strains were classified on the genus le v el.With the av ailability of mor e pur e AGF str ains, the curr ent a ppr oac hes could be tested on a larger set of genera and the ability to differentiate strains even at the species level could be e v aluated.This is especiall y inter esting for str ains with monocentric, filamentous growth, which currently cannot be readily distinguished by light microscopy.The inclusion of a larger set of AGF strains would also be essential to ac hie v e r eliable r esults for classification in environmental samples.We thus also advocate for the inclusion of spectroscopic (NIR/MIR) characterization of nov el str ains, to gener ate a compr ehensiv e database of so far isolated strains.

Figure 1 .
Figure 1.(A) Mean spectra of the respective fungal strains [ P. ruminantium (black), C. communis (red), A. mucronatus (blue), and a spectrum of pur e c hitin (gr een)].The spectr a of the individual fungal r eplicates ar e shown with the same color coding in panel (B).Important r egions, highlighted as A1-3 in Fig. 1 (A) are depicted enlarged: (A1) Lipid region.(A2) Protein and carbohydrate region.(A3) C-H stretching, first o vertone , and combination absorption region.

Figure 2 .
Figure 2. Multiv ariate anal ysis for differ entiation of AGF str ains by NIR.Second deriv ativ e data and absorption regions not containing w ater w ere used for m ultiv ariate anal ysis .On the left a PC A scor es 3D plot of AGF str ains [ P. ruminantium (blac k), C. communis (red), and A. mucronatus (blue)] is depicted.On the right the loadings plot with the r espectiv e components is shown.The gap in the graph results from the removed water bands.Highlighted regions A1-3 are shown as in Fig. 1 (PC2: 23%, PC3: 15%, and PC5: 4%).

Figure 3 .
Figure 3. (A) Mean spectra of the r espectiv e gr owth time (72 h: or ange, 1 w: c y an, and ≥3 w: dark green).The spectra of the individual fungal replicates are shown in panel (B) with the same color coding.Important regions are highlighted by labeled boxes (A1-3), and zoomed graphs of those regions are shown in panels A1-3 as in Fig. 1 .(A1) Lipid region.(A2) Protein and carbohydrate region.(A3) C-H absorption region.

Figure 4 .
Figure 4. Multiv ariate anal ysis for differ entiation of cultur e a ge with NIR.Second deriv ativ e data and absorption regions not containing water were used for m ultiv ariate anal ysis .On the left a PC A scor es 3D plot of AGF str ains harv ested at thr ee differ ent time points (72 h: or ange, 1 w: c y an, and ≥3 w: dark green) is depicted.On the right the loadings plot with the respective components is shown.The gap in the graph results from the removed water bands.Highlighted regions A1-3 are shown as in Fig. 3 (PC1:71%, PC2: 13%, and PC3: 7%).

Figure 5 .
Figure 5. (A) Brightfield image of P. ruminantium with its' r espectiv e absor ption ima ge of the highlighted r egion (B).(C) Coaddition of spectr a within the highlighted area of sporangia and hyphae yielded the respective average spectra.This graph depicts the mean spectra of all hyphae (red) and sporangia (blue).

Figure 6 .
Figure6.Top: mean HSI-spectra of fungal strains [ P. ruminantium (black), C. communis (red), and A. mucronatus (blue)] and c hitin r efer ence (gr een).The finger print r egion is highlighted as it is most r ele v ant for differ entiation of samples.Bottom: Individual spectr a of AGF samples in the zoomed-in finger print r egion ar e shown with the same color coding as abov e including peak assignment to functional gr oups.