Abstract
The emergence of Batrachochytrium salamandrivorans (Bsal) poses an imminent threat to caudate biodiversity worldwide, particularly through anthropogenic-mediated means such as the pet trade. Bsal is a fungal panzootic that has yet to reach the Americas, Africa, and Australia, presenting a significant biosecurity risk to naïve amphibian populations lacking the innate immune defenses necessary for combating invasive pathogens. We explored the capability of near-infrared spectroscopy (NIRS) coupled with predictive modeling as a rapid, non-invasive Bsal screening tool in live caudates. Using eastern newts (Notopthalmus viridescens) as a model species, NIR spectra were collected in tandem with dermal swabs used for confirmatory qPCR analysis. We identified that spectral profiles differed significantly by physical location (chin, cloaca, tail, and foot) as well as by Bsal pathogen status (control vs. exposed individuals; p < 0.05). The support vector machine algorithm achieved a mean classification accuracy of 80% and a sensitivity of 92% for discriminating Bsal-control (-) from Bsal-exposed (+) individuals. This approach offers a promising method for identifying Bsal-compromised populations, potentially aiding in early detection and mitigation efforts alongside existing techniques.
Similar content being viewed by others
Introduction
Chytridiomycosis is a skin-eating fungal disease that has proven fatal to hundreds of amphibian species worldwide1,2. The disease is caused by two fungal pathogens: Batrachochytrium dendrobatidis (Bd) and B. salamandrivorans (Bsal), whose primary hosts include anurans (i.e., frogs and toads) and caudates (i.e., salamanders and newts), respectively3. Amphibians rely on their permeable skin to mediate water uptake as well as to regulate respiratory processes. Thus, when chytrid fungus zoospores embed and replicate within the epidermal layers of an individual there can be devastating effects on the amphibian’s respiratory and osmoregulatory functions1,4. The chytrid panzootic instigated by Bd and Bsal pathogens stands out as one of the most catastrophic threats to global wildlife biodiversity, as chytridiomycosis has led to the decline of over 500 amphibian species and the extirpation of approximately 100 species2. Although Bd is persistent and causing widespread losses across the Americas, Europe, and Australia2, Bsal has not yet emerged in naturally occurring amphibian populations outside of Europe and Asia5. Martel et al.6 has traced the endemic range of Bsal back to Asia, similar to Bd’s origin, where it has been presumed as naturally occurring for approximately the last 30–55 million years.
As with any emerging infectious disease, Bsal presents itself as an imminent threat to naïve amphibian populations that have not coevolved with this pathogen historically. For example, drastic declines have been observed in fire salamander (Salamandra salamandra) populations due to the recent Bsal invasion into Europe with several reports of Bsal emergence specifically in Germany, Belgium, the Netherlands, and Spain6,7,8. Given that there is an ample swath of potential hosts for Bsal in North America, especially within the salamander biodiversity hotspots (e.g., Appalachia Mountains) of the southeastern United States9,10, proactive measures should be taken to avert this deadly fungus from entering the U.S. and mitigate its emergence within vulnerable amphibian populations. Disease mitigation can be done by monitoring and restricting the importation of chytrid-afflicted amphibians into novel geographic locations, especially by providing oversight and improving the efficacy of biosecurity measures11,12.
The first line of defense for mitigating pathogen spillover is through awareness of disease presence and prevalence, namely through disease detection technologies. Such technologies can be used to monitor animals as they cross country borders and state lines, usually through anthropogenic-mediated translocation (e.g., pet trade)11. The saying, “If you can’t measure it, you can’t manage it”, rings true for sick animals in the pet trade who appear healthy (i.e., no clinical signs of disease) and bypass biosecurity checkpoints with little questioning or diagnostic testing11. Currently, histological examinations and quantitative polymerase chain reaction (qPCR) are utilized to evaluate the etiology of lesions and to quantify fungal pathogen loads13. Although these histological and molecular tools are considered rapid and reliable for assessing chytrid pathology, they are primarily constrained to in-lab use, may take several hours or even days before results are available, and both require highly trained technicians to perform14. Considering the magnitude of the threat that Bsal presents as an emerging infectious disease, a diagnostic technology that is non-invasive, field compatible, and requires a fraction of the time to execute compared to traditional methodologies would help advance the detection of this devastating disease in real-time.
Near-infrared spectroscopy (NIRS) is one such tool that can be used, once validated with already established histological and molecular analyses, to rapidly evaluate the Bsal clinical status in living individuals. NIRS is a biophotonic technology that uses electromagnetic energy (i.e., light) in the near-infrared spectrum (700–2500 nm) to capture a suite of biochemical information inherent to the sample or live specimen under investigation15. A spectrometer is used to direct electromagnetic (EM) radiation, which penetrates several millimeters deep, into the epithelial tissue and measures the amount of EM radiation reflected, absorbed, or transmitted from the sample16,17. The resulting spectral signature relates to the biochemical profile of constituents within the tissue, as the pattern of observable peaks and troughs are a reflection of the bonds and functional groups representing compounds that vibrate and absorb energy at very specific wavelengths in the NIR spectrum15,18. Although spectroscopy can be used as a standalone technique to assess the presence and quantity of specific molecular structures present in a sample, the spectroscopic models as they pertain to use in a live-animal system often treat the spectral signatures as a holistic measure of the animal’s physiological profile. In the latter case, the spectral signatures can be categorized and correlated with a reference method (e.g., qPCR), and then machine learning methodologies are applied to generate prediction models. For example, NIRS has been used to answer a wide variety of questions within animal physiology, including the evaluation of demographic traits (e.g., biological sex, age, taxonomy), reproduction (e.g., gravidity, estrus, pregnancy), nutrition, and disease diagnosis19,20. NIRS has previously been applied in amphibians to evaluate biological sex, gravidity, taxonomy, and pheromone expression in a handful of species21,22,23,24,25,26,27,28. In the context of amphibian health, developing NIRS as a disease detection tool initially requires combining reference method (e.g., histological examination and/or qPCR) results with spectral data to establish diagnostic credibility of the NIRS-generated model. After spectroscopic models have been developed for the physiological trait of interest, subsequent spectral data can be readily assessed to provide disease diagnostics in real-time.
Amphibians have unique skin traits that make them ideal candidates for evaluating whether NIRS disease screening is possible in live animal systems. For example, their skin is a complex living tissue that serves as an important immune organ, linking physical, microbiological, chemical, and immunological blockades to pathogen insult29. Because amphibian skin is permeable and functions as a multi-faceted mucosal organ (e.g., for respiration, osmoregulation, reproduction, immune function)14,30, it serves as a biochemical gland that can be directly measured to evaluate aspects of the individual’s physiological state in real-time. Moreover, amphibians are easily and safely handled without anesthesia, making them excellent candidates for developing live animal screening techniques.
This study aimed to evaluate the performance of NIRS and chemometric analyses, in tandem with qPCR, for diagnosing chytrid fungus presence/absence in eastern newts (Notophthalmus viridescens) challenged with a uniform dosage of Bsal zoospores under laboratory conditions. We assessed whether NIRS could be used to discriminate between the spectral skin profiles of Bsal-challenged (+) vs. Bsal-sham (−) control individuals, as newts exposed to pathogen insult are known to respond differentially based on the composition and expression of their skin microbiome31. In addition, we evaluated how model performance for classifying spectra from healthy vs. infectious individuals was affected by disease intensity, as individuals undergoing varying levels of pathogen insult (e.g., mild vs. advanced clinical disease progression) are expected to vary with respect to overall physiological state compared to healthy individuals. We posited that spectral profiles would be most similar between Bsal-control (−) individuals and those experiencing low Bsal loads, while individuals exhibiting high Bsal loads should have distinct biochemical separation from Bsal-control (−) individuals. Therefore, we predicted that classification performance would increase as Bsal loads and associated skin damage due to chytridiomycosis increased.
Results
The spectra, principal component loadings, and scores
Spectral data across the NIR range (700–2500 nm) were categorized and averaged according to the physical location from which the spectra were acquired (chin, cloaca, tail, and foot) as well as by the infection status: Bsal-exposed (+) or Bsal-control (−) (Fig. 1A, B). The transformed spectral signatures varied by body region, especially in the third overtone of the NIR spectrum from 700–1200 nm. Additionally, major differences in spectral patterns occurred across the first and second overtones spanning 1200–1700 nm (Fig. 1A). Marked variability in the NIR spectra was also observed between Bsal (−) and Bsal (+) individuals in each of the four body regions (Fig. 1B). For example, Bsal (+) spectra in the foot, tail, and cloacal regions exhibited markedly lower transformed absorbance values compared to their respective Bsal (−) counterparts, particularly in the wavelength region spanning 1380–1400 nm. The principal component (PC) scores plot indicates distinct separation of spectra from the four regions examined, such that the chin, cloaca, and tail PC scores were predominantly constrained to the negative hemisphere of PC-1, while the foot PC scores were confined to the positive hemisphere of PC-1 (Fig. 1C). Additionally, the spectral scores for the chin and cloaca as well as tail and cloaca regions were more tightly clustered compared to the chin and tail regions. PERMANOVA revealed significant differences (p < 0.001) between the four spectral scanning regions with respect to mean values of PC-1 and PC-2, with the adonis pairwise comparison test indicating significant differences (p < 0.001) between all possible region pairs except the cloaca and chin (F1,1930 = 0.7, p > 0.05; see Supplementary Materials Table S1 for all pairwise comparison test statistics). Misclassifications occurred between regions with more similar biochemical profiles; chin spectra were more likely to be misclassified as cloaca spectra (6.6%; 64/966) rather than tail (2.7%; 26/966) or foot (0.4%; 4/966) spectra (Table 1). Biochemical differences inherent to each body region are shown as the dominant peaks found in the positive or negative direction of the PCA loadings (Fig. 1D), where the first two PCs accounted for 93% of the total database variance, with PC-1 and PC-2 accounting for 87% and 6% of that variance, respectively. As prediction models successfully discriminated between the four regions with an accuracy >90% (Table 1), we treated the spectra from each region as separate datasets to be analyzed independently at first and then as a combined, four-region model.
A Transformed reflectance for all NIR (700–1900 nm) spectra, and B Transformed absorbance NIR spectra (900–1500 nm) were categorized and averaged according to region of acquisition. C Principal component scores plot from transformed spectra (700–2500 nm) indicated by body region. D PC loadings highlighting the dominant peaks in the scores plot: PC-1 (87%) and PC-2 (6%) accounted for 93% of total database variance.
Fungal loads and mortality throughout the study for PCR vs. NIRS + PCR groups
Fungal loads were quantified via qPCR, which varied by individual and time point of sample collection with an average of 80,170 ± 187,506 (mean ± SD) Bsal genomic copies/μL throughout the study period (Fig. 2). As expected, fungal load increased steadily throughout the study duration, such that significant (p < 0.05) increases in mean zoospore genetic copies occurred between days 6 (9.2 ± 13.1 Bsal genomic copies/μL) and 12 (1653.2 ± 2563), days 12 and 18 (4148 ± 4907), days 18 and 24 (47,718.7 ± 71,220.5), days 24 and 30 (110,382.7 ± 219,004.0), as well as days 30 and 36 (192,038.9 ± 300,784.6). Although N. viridescens are highly susceptible and most individuals succumb to Bsal-induced mortality, the relatively small 5 × 103 inoculation dose administered in the present study is comparable to the median infectious dose in this species (i.e., 3000 zoospores/10 mL H2O is the concentration required to infect 50% of individuals in a population); roughly 10% of exposed individuals appear healthy with no clinical disease signs and no pathogens detected31,32. Thus, it is possible for some individuals to experience high infection loads while others remain healthy, resulting in a wide range of clinical states (and respective pathogen loads) among individuals throughout this study.
Minima and maxima are indicated by whiskers, lower and upper interquartile ranges by boxes, medians by black horizontal lines, and means by white diamonds. Day 42 post exposure also included mortality swabs. Only individuals that had a positive qPCR detection are shown.
When comparing mortality data between the Bsal-exposed (+) qPCR-only (i.e., did not receive additional handling or electromagnetic radiation in the form of NIR spectra acquisition) and NIRS approaches tested, we observed 57% (4/7) mortality of individuals in the qPCR-only group compared to 23% (7/30) mortality in the NIRS group on day 36 post-Bsal exposure. By study termination on day 42, 100% (7/7) of the qPCR-only group succumbed to mortality vs. 60% (18/30) mortality observed in the NIRS group. Additionally, the diagnostic approach was a significant predictor (p < 0.001) when evaluating mean fungal loads between the qPCR-only and NIRS groups, such that NIRS-diagnosed individuals (82,037.1 ± 200,076.2 copies/μL) yielded lower mean pathogen replication rates compared to the qPCR-only group (84,286.1 ± 118,723.7 copies/μL) throughout the study period, although there is overlap between the confidence intervals (Fig. 2). This finding suggests either that near-infrared electromagnetic radiation directed at individuals could have an anti-fungal effect or that additional handling of animals with nitrile gloves for even short periods of time might affect pathogen load33.
Experiment 1: discriminating spectra of Bsal-exposed (+) vs. Bsal-control (−) individuals
For Experiment 1, the PC scores plot visualizing Bsal-control (−) vs. Bsal-exposed (+) spectra showed poor spectral separation when PC-1 and PC-2 were solely applied for each of the four regions (Supplementary Materials Fig. S1). However, the scores plot for the foot region utilizing more than two PCs enabled us to visualize some separation occurring between individuals undergoing different pathogen status designations (Fig. S2). For example, when PC-2 (19% database variation explained) and PC-3 (16%) were plotted against PC-1 (46%) and PC-4 (7%), spectral separation between Bsal (−) and Bsal (+) groups was observed, such that the Bsal-control (−) spectra tended to exhibit greater biochemical variation, as indicated by a wider spread of scores across all hemispheres of the examined PCs compared to Bsal-exposed (+) spectra. Although a large amount of spectral overlap occurred between the two groups, it is important to recognize that PCA is an unsupervised dimensionality reduction algorithm that describes and visualizes the database variation, specifically through patterns that emerge without predefined class labels. The spectral patterns generated by PCA provide only a cursory glance into how the prediction models may parse out variation in the structure of the datasets examined.
When model performances were aggregated according to each region of the animal that was scanned, mean classification accuracy varied significantly (F4,145 = 4.7, p < 0.01), such that the foot region yielded the highest overall accuracy (75.0 ± 3.9%) compared to the cloaca (67.9 ± 8.3%) region but did not differ significantly (p > 0.05) from the chin (72.1 ± 6.1%), tail (70.6 ± 6.7%), or combined (70.9 ± 6.8%) region datasets. When evaluating models of the foot region dataset for classifying Bsal-exposed (+) vs. Bsal-control (−) spectra, significant differences in mean classification accuracy were observed across the six algorithms tested (F5,24 = 18.4, p < 0.001), for which the accuracy ranged from 67.5 ± 2.1% (K-nearest neighbor; KNN hereafter) to 77.4 ± 1.6% (support vector machine; SVM). While the SVM algorithm performed significantly (p < 0.05) better than KNN with respect to mean classification accuracy, the SVM did not differ (p > 0.05) in its performance compared to the generalized linear model (GLM; 75.5 ± 2.1%), linear discriminant analysis (LDA; 77.1 ± 2.3%), partial least squares (PLS; 76.0 ± 2.1%), or random forest (RF; 76.6 ± 1.5%) algorithms (Fig. 3). Overall, the SVM algorithm applied to the combined four-region dataset accurately classified 92.0% of Bsal-exposed (+) spectra and 67.6% of Bsal-control (−) spectra. The confusion matrix for the SVM yielding the highest sensitivity (i.e., true-positive classification rate) is provided in Fig. 4.
Significant differences between models are denoted by different letters (p < 0.05), as indicated by the ANOVA results using Tukey HSD. Minima and maxima are indicated by whiskers, lower and upper interquartile ranges by boxes, medians by black horizontal lines, and means by gold diamonds. GLM generalized linear model with ElasticNet regression, KNN K-nearest neighbor, LDA linear discriminant analysis, PLS partial least squares, RF random forest, SVM support vector machine.
Confusion matrix and performance metrics for the top-performing model (support vector machine) on the combined, four-region test dataset (N = 3864 spectra, n = 966 spectra per region).
Experiment 2: discriminating spectra of individuals with varying levels of Bsal infection vs. Bsal (−) controls
For Experiment 2, the spectral profiles as shown through principal component scores plots appeared very similar between the Bsal-low (+) and Bsal-high (+++) groups (shown for foot region; Fig. 5). For the foot region, when PC-1 was plotted against PC-2 and PC-3, we observed a pattern of tight clustering between the Bsal (+) groups, whereas the scores plots for spectra from Bsal (−) animals showed less overlap and spanned a greater portion of both PC-1 and PC-2 compared to those of the Bsal-exposed (+) animals. Additionally, greater spectral overlap tended to occur between control (−) and mildly infected (+) individuals (F1,244 = 2.3, p > 0.05) than between the spectra belonging to control (−) and highly infected (+++) individuals (F1,217 = 11.1, p < 0.005; see Table S2 for all other pairwise adonis comparisons).
The principal component scores plots visualize spectral differences for the transformed NIR spectra (700–2500 nm) between the three groups examined across PC-1 (56% of database variance explained), PC-2 (15%), and PC-3 (13%). The five subplots are visualized using identical upper and lower ranges of PC-1 and PC-2 but are faceted by varying ranges of PC-3.
Experiment 2a: Bsal-low (+) vs. Bsal-control (−)
When classifying spectra from Bsal-control (−) and Bsal (+) animals characterized as having a low fungal load, mean classification accuracy varied significantly among the four physical regions (F4,145 = 5.4, p < 0.001; Table 2). Spectra from the chin (62.0 ± 1.6%) and tail (61.7 ± 3.7%) yielded significantly (p < 0.05) higher performance accuracies compared to those from the cloaca (55.7 ± 4.9%) but did not differ significantly (p > 0.05) from the foot (59.9 ± 3.4%) or combined region (57.6 ± 3.7%) datasets. For the combined region dataset, classification models using the SVM (62.5 ± 1.8%) algorithm yielded higher (p < 0.05) classification accuracies compared to those with GLM (55.7 ± 1.5%), KNN (52.4 ± 2.2%), LDA (57.0 ± 2.2%), and PLS (56.8 ± 1.8%) algorithms (see Table 2 for all other pairwise comparisons).
Experiment 2b: Bsal-moderate (++) vs. Bsal-control (−)
When discriminating spectra from Bsal-control (−) and Bsal-exposed (++) animals characterized as having a moderate fungal load, there was a significant effect of physical region on classification performance (F4,145 = 7.0, p < 0.001), such that the tail region (65.1 ± 6.3%) yielded higher mean classification rates than the foot region (57.9 ± 6.4%). Similar to the results found in Experiment 2a, there was no significant difference (p > 0.05) between the six algorithms tested for any of the four regions examined except for the combined region dataset (F5,24 = 6.0, p < 0.05). Within the combined region dataset, the LDA (60.7 ± 1.7%), PLS (61.0 ± 1.6%), RF (60.6 ± 3.3%), and SVM (62.2 ± 1.4%) algorithms classified spectra at significantly (p < 0.05) higher rates than the KNN (55.2 ± 3.4%) algorithm (Table 2).
Experiment 2c: Bsal-high (+++) vs. Bsal-control (−)
Spectra from Bsal -control (−) and Bsal-exposed (+++) animals, characterized as having a high fungal load, exhibited a significant effect of physical region on classification performance (F4,145 = 4.1, p < 0.01), such that the tail region (63.1 ± 8.4%) yielded higher mean classification rates than the chin region (56.8 ± 5.2%; see Table 2 for all pairwise comparisons). There was a difference between the six algorithms tested with respect to mean classification performance for the cloaca (F5,24 = 3.9, p < 0.05), tail (F5,24 = 3.3, p < 0.05), and combined region (F5,24 = 7.3, p < 0.05) datasets, but not for the chin (F5,24 = 2.6, p > 0.05) and foot (F5,24 = 1.8, p > 0.05) regions. For the tail region dataset, the RF (67.5 ± 2.5%) and SVM (66.2 ± 8.7%) algorithms yielded significantly higher (p < 0.05) classification accuracies compared to the KNN (51.7 ± 7.7%) algorithm (see Table 2 for all pairwise comparisons). Among the three experimental spectral datasets of Bsal-exposed tested against Bsal-control [−] spectra (low [+], medium [++], and high [+++]), there was no significant difference (p > 0.05) with respect to model performance discriminating spectra between Bsal-control (−) and Bsal-afflicted (+) individuals.
Discussion
Chytridiomycosis stands out as one of the most destructive threats to biodiversity among other salient wildlife pathogens (e.g., white-nose syndrome in bats and West Nile virus in birds) to date, with roughly 6.5% of described amphibians (>500 species) suffering declines or extirpations due to chytrid pathogen exposure2. Despite the widespread application of spectroscopic technologies for evaluating and characterizing many clinical disease states in plants34,35, domesticated animals36,37,38,39, and people40,41,42, there have been limited reports, if any, on the effectiveness of NIRS-based disease detection in amphibian species. This work highlights NIRS as a practical diagnostic tool for assessing deadly fungal pathogen occurrence in a susceptible amphibian species. In Experiment 1, we compared spectra from Bsal-control (−) and Bsal-exposed (+) individuals and found that Bsal infection status in N. viridescens individuals could be classified with a promisingly high level of accuracy (80%), with a sensitivity >90% and a specificity >65%, the former associated with classification rate of spectra from Bsal-exposed (+) individuals and the latter with Bsal-control (−) individuals. From these results, we conclude that with relatively minimal sampling effort, chytrid-afflicted populations may be reliably identified, enabling mitigative strategies to be enacted in a timely manner. The support vector machine and random forest algorithms yielded the highest performance outcomes from each of the four body regions tested, with the foot region providing the highest mean classification accuracy and the cloacal region the lowest. One reason for the SVM and RF outperforming the other algorithms tested could be due to their adaptability in handling complex multivariate datasets. Whereas some models such as LDA and GLM (i.e., linear models) are limited in their ability to identify relationships in non-linear data structures, it is purported that SVM and RF algorithms are better suited for modeling non-linear data in hyper-dimensional spaces43. As such, the “curse of dimensionality” necessitates predictive algorithms to navigate the complex hyper-dimensional space in search of meaningful relationships that may exist between the predictor variable space and the response variable of interest44. However, algorithms perform differentially under varying data science scenarios as the mathematical approach (e.g., internal procedures for feature selection, dimensionality reduction, regularization, hyperparameter tuning) used to achieve classification or regression tasks varies markedly between algorithms45. Therefore, the use of a multi-algorithm approach, as demonstrated here, has been recommended to identify candidate models when assessing spectroscopic datasets for the first time22.
In Experiment 2, we compared model performance for distinguishing Bsal-control (−) spectra from Bsal-exposed (+) spectra of varying loads (low, medium, and high gene replication rates) and observed no differences in mean classification accuracy among the datasets tested. Overall, the tail region tended to yield the highest classification accuracies for distinguishing Bsal-exposed (+) and Bsal-control (−) individuals. All algorithms performed comparably, except for K-nearest neighbors, which yielded the lowest classification accuracies in several instances (e.g., all combined region datasets in Experiment 2). In Experiments 2a and 2c of this study comparing Bsal-control (−) spectra vs. Bsal-low (+) and Bsal-high (+++) spectra, we posited that the spectral profiles belonging to Bsal-control (−) vs. highly infectious Bsal (+++) individuals would exhibit the greatest spectral separation due to these groups undergoing starkly different physiological states (as opposed to weakly positive individuals and their healthy counterparts). PERMANOVA revealed greater spectral similarities among the Bsal-low (+) and Bsal-control (−) groups, whereas Bsal-low (+) and high (+++) as well as high (+++) and control (−) groups differed significantly with respect to principal component scores. Overall, spectral differences among control (−) and pathogen-exposed (+) groups indicated a biochemically divergent response between individuals undergoing pathogen insult compared to those who were not. Predictive outcomes for classifying disease states in Experiment 1 were possibly more conclusive than in Experiment 2 due to differences in sample size. Regardless, our results suggest that NIRS is capable of detecting the large-scale biochemical changes in the skin that are expected to arise in disease-afflicted newts compared to healthy newts. Importantly, the NIRS prediction models were capable of detecting individuals that were inoculated with Bsal but were not necessarily exhibiting clinical symptoms. For example, roughly 30% (n = 147/483 spectra) of the global database was comprised of Bsal (+) spectra characterized as having low fungal loads (i.e., spectra categorized as belonging to quartile 1 in Experiment 2a), so even spectra associated with very low pathogen replication rates, as indicated by qPCR, were distinguishable from Bsal-control (−) spectral samples, with the majority of models achieving >70% mean classification accuracy.
The NIRS diagnostic method proposed in this study may be applied to rapidly assess the disease status of individuals, even for pathogen-exposed individuals showing no clinical symptoms, after prediction models have been calibrated for targeted amphibian species and pathogens (e.g., Bd, Bsal, Ranavirus) of interest. We suggest that NIRS could serve as a preliminary and rapid screening tool when validated with qPCR analysis and histological examination of vulnerable amphibian populations in a wide variety of contexts. For example, handheld NIRS instruments, which are portable and easy-to-operate, may have prediction models installed for detection of pathogens hosted by species that are commonly transported as part of the pet trade. Such a technology could be used to help mitigate the spread of potentially harmful pathogens between captive and wild populations, as well as between state lines and country borders. Given that the southeastern United States is a biodiversity hotspot for native salamander populations, which are particularly vulnerable to Bsal invasion10, timely action is needed to develop rapid disease screening technologies if we are to limit the spread of this harmful pathogen12. Organizations (e.g., academic institutions, US Fish and Wildlife Services, Customs and Border Protection) interested in implementing NIRS technologies for disease surveillance applications may do so by acquiring the necessary hardware and open-source software resources (R code provided) detailed in the Methods and Supplementary Materials sections. In accordance with statute 18 USC 42 of the Lacey Act46, the goal of which is to limit importation of amphibians deemed injurious to nationwide biodiversity and ecosystem services within the United States, rapid diagnostic tools such as NIRS could be tested in a wide range of amphibian species47, particularly taxa that are heavily traded (e.g., genus Rana)48 or those designated as carrier or amplification species10.
Previous work with N. viridescens experiencing varying clinical disease states after exposure to Bsal found marked differences in the abundance of fungal and bacterial communities that comprise the skin microbiome of this species31. It is of importance to note that skin microbiome composition (e.g., antimicrobial peptides and proteins, fungi, bacteria) has been linked to crucial Bd and Bsal-inhibitory properties in several amphibians31,49,50,51,52, and may be a physiological trait that can be identified and characterized via NIRS. Pathological outcomes, such as necrotic skin ulcerations, hemorrhages, hyperplasia, hyperkeratosis, electrolyte imbalance, and dehydration4,5,53,54 may also be contributing to the spectral variation present between healthy (−) vs. Bsal-exposed (+) individuals. Altogether, we posit that one, or a combination of these factors, provided sufficient variation among spectral profiles for generating the informative prediction models shown in this study. More specifically, we demonstrate the efficacy of a broad-based model for Bsal detection in N. viridescens, whereby the instrument user can scan newts of varying infection load and detect Bsal pathogen exposure from any one of the four body regions examined with a fairly high level of accuracy. Our results indicate that the combined, four-region model performed comparably to the single-region models ( ~ 80%; Table 2 and Figs. 3 and 4) in mature N. viridescens, a species found in aquatic environments. It is also possible that the water band between 1300 and 1600 nm is contributing to the signal used to produce prediction models. Until recently, the general consensus asserted was that water, with its broad absorption bands, overlapped and negatively impacted signal quality of the analyte(s) under investigation55. However, the emerging field of Aquaphotomics demonstrates it is also possible that water may serve as a signal amplifier, as shifts in complex bio-aqueous systems, coined the water molecular network, respond in step according to perturbations occurring both in vivo and in the surrounding environment56. The changing water signal itself (observed at the water absorption bands near 1190 and 1450 nm) is possibly reflective of the pathogenic disease, as Bsal alters the osmotic balance in the skin of exposed individuals4 and likely causes shifts in water structure and composition (e.g., free vs. bound water molecules, dimers, water solvation shells) that can potentially be used to differentiate between healthy and disease-afflicted individuals56. Indeed, more than 500 water absorbance bands have been identified in association with relevant biomolecules within the near-infrared spectrum, and there are now several studies that have employed this framework to better understand the impact of the water molecular network on various disease states36,37,57. Given the aquatic nature of the live specimen under investigation, an aquaphotomics framework would be appropriate to apply and is therefore a valuable consideration for future studies56,58.
Radiation across various regions of the electromagnetic spectrum has previously been shown to have therapeutic effects, including the use of ultraviolet radiation as a fungicide (Candida sp.) on impression materials used in dentistry, and infrared radiation as a mitigative strategy for several diseases such as Parkinson’s disease and cancer59,60. As electromagnetic radiation applied to the dermal surface of amphibians undergoing NIR spectra acquisition may have fungicidal effects, we compared the mean fungal zoospore loads of individuals that underwent NIRS data collection and were subsequently swabbed vs. individuals that were swabbed only. The diagnostic method was a strong predictor of Bsal load progression, such that the NIRS-treated + qPCR group had significantly lower fungal loads compared to the qPCR-only group throughout the study duration, which may be a consequence of electromagnetic radiation dampening fungal propagation. In addition, previous work found that handling amphibians with nitrile gloves actively kills chytrid zoospores33; thus, we need to consider handling of the animals as another factor impacting our results. In conjunction with having lower Bsal pathogen loads on average, the NIRS group succumbed to mortality more slowly and at a lower overall magnitude than the PCR-only group (60% vs. 100%, respectively, on day 42). These findings warrant further investigation of electromagnetic radiation in the infrared spectrum as a potential treatment for chytrid-afflicted individuals.
NIRS technologies, coupled with predictive modeling, offer several advantages over traditional disease screening tools, such as minimal sample processing, rapid analysis of a high throughput of samples, and the automation of physiological assessments. Technologies that can be automated help to eliminate inter-observer bias and make evaluations of multi-factorial complex traits, like disease, more reliable and straightforward to apply. Additionally, the automated nature of machine learning coupled with NIRS may serve to increase the capacity of such tools for usage by a larger number of practitioners, as the training time necessary to operate the instrumentation is small relative to the molecular techniques traditionally applied for disease diagnosis. It is the relative ease of use and high throughput screening associated with NIRS that has led many major industries to employ this technique for quality control and assurance purposes for a diverse range of products across the medical, agricultural, and engineering sectors over the last few decades42,61,62. Moreover, spectroscopic instrumentation has seen vast improvements in recent years with respect to form factor, portability (e.g., <50 grams for handheld spectrometers)63, and integrated software (e.g., applications that can be operated via mobile devices) which may further enhance the utility of this tool by a wider userbase. Although spectroscopic based studies involving wildlife have emerged only within the last couple decades, their development and deployment have been applied with great versatility such that NIRS has served as an effective monitoring tool for informing wildlife management and conservation-based initiatives for dozens of threatened species19,20.
In summary, the results from this study provide support for NIRS as a diagnostic tool for evaluating Bsal pathogen presence/absence in N. viridescens, a newt species that is widely distributed across the United States and highly susceptible to Bsal. As it stands, over 50% of salamander species (426/759, according to the IUCN Red List) face threats of extinction64. If Bsal is introduced to North America, it is predicted that up to 140 salamander species could experience population declines10. Thus, the implementation of rapid screening techniques for invasive pathogens is needed if we are to ensure the sustainability of North American amphibian biodiversity. Given that NIRS screening technologies can be highly portable, cost-effective, and easy to operate, it could serve as a useful Bsal surveillance strategy in the future and with additional testing could be applicable to a wider range of species.
Methods
Study population
A study population of eastern newts (Notophthalmus viridescens, IUCN conservation status: Least Concern), also known as the red-spotted newt, was collected (N = 50; Tennessee Wildlife Resources Agency Scientific Collection permit #1504) in a pond located near Knoxville, TN, USA. N. viridescens is a ubiquitous amphibian species present across eastern North America and serves as an appropriate study species for assessing chytridiomycosis pathogenesis due to its high susceptibility to Batrachochytrium salamandrivorans (Bsal)4. Only sexually mature (specific ages unknown) N. viridescens (n = 22 females [2.69 ± 0.67 g] and 28 males [2.49 ± 0.56 g]) were selected for sampling in this study, which were captured in early February 2023 via seine net during the pre-spring breeding season. Sex was determined based on the presence or absence of enlarged nuptial pads and swollen cloacae exhibited by males. N. viridescens were singly housed in sterilized 2000-cm3 containers containing semi-circular PVC hides and filled with 300 mL of dechlorinated, aged water that was changed every 3 days. Newts were fed a standardized dry weight (2% of the individual’s body weight) of bloodworms (Chironomus plumosus) 24 h preceding all water changes.
Prior to the Bsal challenge, all newts underwent a heat treatment of 30 °C lasting 10 days in Conviron® growth chambers (Winnipeg, Canada) to clear potential infection burdens (e.g., Batrachochytrium dendrobatidis, Bd, known to be present in North American amphibian populations), following Chatfield et al.65. To ensure that N. viridescens were clear of Bd and Bsal, newts were swabbed and tested for coinfections via qPCR14,31 before and after the heat treatment following the swabbing method described by Carter et al.31, where 10 passes across both the abdomen and the bottom of each foot with a rayon-tipped swab were made. Three individuals tested positive for Bd and were excluded from the experiment and one newt was found deceased (cause of death unknown) before the study began. Following heat treatment, newts were maintained at 14 °C in order for them to acclimatize and restore their natural biochemical profiles. The remaining 46 individuals (n = 22 females and 24 males) were selected at random in an 80:20 split design to receive either the fungal challenge (n = 37) or a sham water (n = 9) treatment. We have complied with all relevant ethical regulations for animal use. All animal procedures were approved and adhered to the animal care and use guidelines set forth by the University of Tennessee Institutional Animal Care and Use Committee under Protocol #2632-0321.
Spectral data acquisition
NIR reflectance spectra (N = 4332) were collected with an ASD FieldSpec® 3 Indico®Pro (Malvern Panalytical, ASD Analytical Spectral Devices Inc., Boulder, CO, USA). Spectra were collected in triplicate using a small diameter (1.5 mm) contact probe from the chin, cloacal, basal tail, and foot region (N = 1083 spectra per region; Fig. 6) from 46 N. viridescens over 3 months (February–April 2023). To ensure disease transmission avoidance between the Bsal (−) and Bsal (+) groups, Bsal (−) individuals were handled and swabbed before sampling the Bsal (+) group. NIR spectra were always collected prior to being swabbed. To avoid biochemical interference as well as cross-contamination between individuals, the detector end of the contact probe was sanitized with 70% ethanol in between individuals. A baseline of reflectance was established by capturing a white (Spectralon®) reference between each animal. All spectra were collected across the wavelength range of 350–2500 nm (resolution of 1.4 nm for the 350–1000 nm range and 2 nm for the 1000–2500 nm range), where each spectral replicate was comprised of 50 scans, then averaged into a single spectrum (integration time approximately 34 ms). Regions within the visible-light spectrum were omitted so that only wavelengths within the near infrared range (700–2500 nm) would be included as variables in the predictive modeling analysis described below.
Spectra in the near-infrared region (700–2500 nm) were collected using a small diameter (1.5 mm) fiberoptic reflectance contact probe from four regions: the A chin, B cloaca, C base of tail, and D foot (N = 966 spectra analyzed per region) prior to dermal swab collection from the ventral side and foot region.
Pathogen exposure and experimental design
The 37 newts randomly selected for chytrid exposure were inoculated with a uniform dosage of 5 × 103 Bsal zoospores/10 mL autoclaved dechlorinated water, and the remaining 9 newts serving as Bsal (−) controls were inoculated with autoclaved dechlorinated water only. This dose concentration was selected based on a previous multidose-response experiment, where several concentrations and temperatures were tested to evaluate N. viridescens susceptibility under varying environmental conditions31. Since the primary goal of this study was to determine if small levels of chytrid pathogen loads could be detected using NIRS, we selected the relatively smaller magnitude 5 × 103 dose (vs. 5 × 104–6), which enabled us to collect the maximal number of spectra across a feasible study duration, while capturing the full breadth of Bsal disease progression (i.e., individuals exhibiting low loads with no clinical symptoms to advanced stages of clinical disease). Two separate experiments were developed to evaluate the diagnostic capability of NIRS for assessing Bsal-disease status, as well as to determine if the animal’s level of infection impacted NIRS model performance. Experiment 1 compared spectra from Bsal-control (−) vs. Bsal-exposed (+) individuals, in which spectra were colleted from each of the four body regions (chin, tail, foot, cloaca), where individuals were defined as Bsal-exposed (+) if they were challenged with the 5 × 103 dose of Bsal zoospores, regardless of whether the qPCR method had a positive detection reading for Bsal or not; this is often the case for low exposure dosages, where positive chytrid detection often does not occur until at least 1-week post-infection10. Individuals were categorized as Bsal-control (−) if they received the control water treatment or if the spectra were collected prior to the disease challenge. Experiment 2 compared the model performance for discriminating Bsal-controls (−) vs. three varying levels of Bsal (+) infection, including: Bsal-low (+) quartiles [1,2] vs. Bsal-control (−) (Experiment 2a); Bsal-moderate (++) quartiles [2,3] vs. Bsal-control (−) (Experiment 2b); and Bsal-high (+++) quartiles [3,4] vs. Bsal-control (−) (Experiment 2c; see Table 2).
Following the Bsal pathogen challenge day, all animals had their spectra collected and were swabbed in 6-day intervals up through 42 days post-infection (DPI) or until mortality (see Fig. 7 for experimental timeline). The experiment was carried out until a mortality rate of 60% was attained in the Bsal-exposed (+) treatment group. Mortality events were recorded as animals reached terminal endpoints (i.e., excessive lethargy and loss of righting reflex), which necessitated humane euthanasia through dermal application of Orajel™66. Of the 37 animals exposed to Bsal zoospores, 30 had swabs and spectra collected, while seven were randomly selected to serve as Bsal (+) controls that were not evaluated using the NIRS diagnostic approach but were swabbed for genetic analyses (i.e., qPCR-only). This was to account for the effects of additional handling with nitrile gloves known to actively function as a fungicide33, in addition to potential therapeutic effects of electromagnetic radiation emitted from the NIRS spectroradiometer during spectral data collection. Bsal isolate (AMFP13/1)5 from a morbid S. salamandra in the Netherlands was cultured and enumerated at the University of Tennessee-Knoxville for lab-based experimental inoculations.
Dermal swabs and spectra were collected at three pre-exposure (blue) and seven post-exposure (red) time points. DPI = days post-infection. The image of N. viridescens was accessed from the Public Domain (source: Wikimedia Commons).
Statistics and reproducibility
To evaluate whether the qPCR-only (N = 38 swabs) and NIRS (N = 164 swabs) groups differed with respect to fungal loads throughout the study period, a generalized additive model for location, scale, and shape (GAMLSS) was carried out in the “GAMLSS” package67 with diagnostic approach (qPCR vs. NIRS + qPCR) and days post-infection as fixed effects (two, separate models), newt sample ID as a random effect, and reverse Gumbel used as the data distribution family. Mean zoospore equivalents (i.e., Bsal load replication rates) were log-transformed for statistical modeling due to a high level of skew. Model fit was evaluated via Q-Q Plots and checking the normality of residuals. The alpha value was set to 0.05 for significance testing.
Predictive models were built using a multi-model benchmark approach22 in the software, R68 with the ‘benchmark’ function within the “mlr” package69. Spectral data preprocessing included smoothing (Savitsky-Golay, first order polynomial with a window size of 11) and first derivative (Savitsky-Golay, first order polynomial with 3 symmetric kernel points) functions with the “prospectr” package70, then feature selection was carried out with the “Boruta” package71. This significantly reduced the total number of predictor variables in the dataset (range = 36–340 variables were selected and used to calibrate models, depending on the dataset for each respective region; Table 2). Six machine learning algorithms were applied for calibrating and testing models, including a generalized linear model with ElasticNet regression (GLM), linear discriminant analysis (LDA), K-nearest neighbors (KNN), partial least squares (PLS), random forest (RF), and support vector machine (SVM). Parameters and hyperparameters along with their tuning constraints (i.e., upper and lower range) for each of the algorithms applied are specified and briefly described in the Supplementary Materials. Spectral datasets were first partitioned into blocks (spectral replicates = 3), then a K-fold nested cross-validation (k = 5 folds) procedure was applied for each algorithm for calibrating and testing models, whereby the spectral dataset is split into five folds with the first four folds used to calibrate the model and the remaining fold treated as test data. The five total data folds are sequentially rotated so that each fold is represented once in the test dataset while remaining independent from the training dataset. All reported classification accuracy values in this study reflect model performances drawn solely from the test data folds. This approach ensures that the reported metrics reflect the model’s true generalizability. Calibration and internal validation results might overestimate the model’s performance because they are derived from data used during the model selection and tuning process. As a result, the model may become overfitted to the specifics of the training data, leading to an overly optimistic assessment of its performance. By focusing on the outer validation folds, we provide a more accurate estimate of how the model will perform on unseen data. The five accuracies generated for each model were aggregated and compared using an ANOVA, and a Tukey HSD test was utilized to evaluate significant (alpha = 0.05) differences between the mean classification accuracies of all algorithms and body regions investigated. Assumptions for ANOVA were met according to Shapiro-Wilk, Bartlett’s, and Durbin-Watson tests. Principal component scores plots were generated to visualize spectral separation between Bsal (+) vs. Bsal (−) groups using JMP 14.072.
First, we produced models to classify the four regions from which spectra were collected (i.e., chin, cloaca, tail, and foot) in order to determine if the spectral profiles differed by body region. Additionally, a PERMANOVA was carried out using the “vegan” package73 in R to assess if principal component scores differed between the four spectral acquisition locations. As regions differed significantly (p < 0.001), models for classifying the spectra of Bsal-exposed (+) vs. Bsal-control (−) individuals were built independently for each of the four spectral acquisition regions as well as in a combined dataset that included all four body areas, hereafter referred to as ‘combined region’. The combined region dataset was examined to evaluate model performance from a holistic viewpoint representing the body’s overall response to pathogen insult. As Bsal (−) control spectra were found to be distinguishable between the three pre-challenge time points, only spectra collected following the heat treatment were included in the analysis (i.e., post-acclimatization and pre-challenge spectra were included, yet pre-acclimatization spectra were excluded). All prediction models were built with balanced spectral datasets. As the Bsal (+) spectral database (N = 600 per region) was larger than the Bsal (−) database (N = 483 per region), 117 spectra from the Bsal (+) database were selected at random and removed from the combined four-region database to ensure datasets were balanced. In order to evaluate how disease intensity affected model performance, Bsal (+) spectra were partitioned into three separate datasets, based on whether they had low, moderate, or high fungal replication rates (as indicated by qPCR). PERMANOVA was applied to evaluate differences in spectral scores among the Bsal-control (−), Bsal-low (+) and Bsal-high (+++) pathogen load groups, then post-hoc pairwise comparisons were evaluated using the adonis function with Bonferroni corrected p-values.
Spectra belonging to Bsal-exposed (+) individuals had their Bsal fungal loads quantified via qPCR, were categorized according to quartile, and then prediction models for discriminating Bsal-control (−) from Bsal-exposed (+) were generated based on whether the Bsal load was low, moderate, or high. Bsal loads were categorized as low if they were within quartiles 1 and 2 (range = 0.4–3719.6 zoospore equivalents), moderate if they were within quartiles 2 and 3 (range = 74.8–67,801.9 zoospore equivalents), and high if they were within quartiles 3 and 4 (range = 3771.8–1,176,733.8 zoospore equivalents). Therefore, only spectra belonging to individuals with positive Bsal detections, as per qPCR, were included in Experiment 2 (i.e., Bsal-exposed spectra were excluded if their corresponding fungal loads were below 0.4 zoospore genome equivalents or if they were undetectable via qPCR). All datasets were analyzed as binary classification problems; the number of spectra evaluated in Experiments 1 and 2a–c are shown in Table 3. The code and source data (spectral dataset) used to develop predictive models (and underlying figures) for classifying Bsal pathogen presence/absence in this study are provided as Supplementary Code 1 and Supplementary Data 1, respectively.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
All data required to reproduce study results can be found within the manuscript and the attached Supplementary Materials.
Code availability
All code required to reproduce study results can be found within the manuscript and the attached Supplementary Materials.
References
Voyles, J. et al. Pathogenesis of chytridiomycosis, a cause of catastrophic amphibian declines. Science 326, 582–585 (2009).
Scheele, B. C. et al. Amphibian fungal panzootic causes catastrophic and ongoing loss of biodiversity. Science 363, 1459–1463 (2019).
Lips, K. R. Overview of chytrid emergence and impacts on amphibians. Philos. Trans. R. Soc. B Biol. Sci. 371, 20150465 (2016).
Sheley, W. C. et al. Electrolyte imbalances and dehydration play a key role in Batrachochytrium salamandrivorans chytridiomycosis. Front. Vet. Sci. 9, 1055153 (2023).
Martel, A. et al. Batrachochytrium salamandrivorans sp. nov. causes lethal chytridiomycosis in amphibians. Proc. Natl Acad. Sci. USA 110, 15325–15329 (2013).
Martel, A. et al. Recent introduction of a chytrid fungus endangers Western Palearctic salamanders. Science 346, 630–631 (2014).
Stegen, G. et al. Drivers of salamander extirpation mediated by Batrachochytrium salamandrivorans. Nature 544, 353–356 (2017).
Lötters, S. et al. The amphibian pathogen Batrachochytrium salamandrivorans in the hotspot of its European invasive range: past-present-future. Salamandra 56, 173–188 (2019).
Barrett, K., Nibbelink, N. P. & Maerz, J. C. Identifying priority species and conservation opportunities under future climate scenarios: amphibians in a biodiversity hotspot. J. Fish. Wildl. Manag. 5, 282–297 (2014).
Gray, M. J. et al. Broad host susceptibility of North American amphibian species to Batrachochytrium salamandrivorans suggests high invasion potential and biodiversity risk. Nat. Commun. 14, 3270 (2023).
Fu, M. & Waldman, B. Novel chytrid pathogen variants and the global amphibian pet trade. Conserv. Biol. 36, 1–9 (2022).
Gray, M. J. et al. Batrachochytrium salamandrivorans: the North American response and a call for action. PLoS Pathog. 11, e1005251 (2015).
Thomas, V. et al. Recommendations on diagnostic tools for Batrachochytrium salamandrivorans. Transbound. Emerg. Dis. 65, e478–e488 (2018).
Boyle, D. G., Boyle, D. B., Olsen, V., Morgan, J. A. T. & Hyatt, A. D. Rapid quantitative detection of chytridiomycosis (Batrachochytrium dendrobatidis) in amphibian samples using real-time Taqman PCR assay. Dis. Aquat. Organ. 60, 141–148 (2004).
Pasquini, C. Near infrared spectroscopy: fundamentals, practical aspects and analytical applications. J. Braz. Chem. Soc. 14, 198–219 (2003).
Padalkar, M. V. & Pleshko, N. Wavelength-dependent penetration depth of near infrared radiation into cartilage. Analyst 140, 2093–2100 (2015).
Beć, K. B., Grabska, J. & Huck, C. W. Near-infrared spectroscopy in bio-applications. Molecules 25, 2948 (2020).
Stuart, B. H. Infrared Spectroscopy: Fundamentals and Applications (John Wiley & Sons, 2004).
Morgan, L. R., Marsh, K. J., Tolleson, D. R. & Youngentob, K. N. The application of NIRS to determine animal physiological traits for wildlife management and conservation. Remote Sens. 13, 3699 (2021).
Vance, C. K., Tolleson, D. R., Kinoshita, K., Rodriguez, J. & Foley, W. J. Near infrared spectroscopy in wildlife and biodiversity. J. Infrared Spectrosc. 24, 1–25 (2016).
Chen, L.-D. et al. Near-infrared spectroscopy (NIRS) as a method for biological sex discrimination in the endangered houston toad (Anaxyrus houstonensis). Methods Protoc. 5, 4 (2021).
Chen, L.-D., Caprio, M. A., Chen, D. M., Kouba, A. J. & Kouba, C. K. Enhancing predictive performance for spectroscopic studies in wildlife science through a multi-model approach: A case study for species classification of live amphibians. PLoS Comput. Biol. 20, e1011876 (2024).
Torralvo, K., Durgante, F., Pasquini, C. & Magnusson, W. E. Near infrared spectroscopy for the identification of live anurans: towards rapid and automated identification of species in the field. J. Near Infrared Spectrosc. 31, 80–88 (2023).
Torralvo, K., Magnusson, W. E. & Durgante, F. Effectiveness of Fourier transform near-infrared spectroscopy spectra for species identification of anurans fixed in formaldehyde and conserved in alcohol: a new tool for integrative taxonomy. J. Zool. Syst. Evol. Res. 59, 442–458 (2021).
Vance, C. K., Kouba, A. J. & Willard, S. T. Near infrared spectroscopy applications in amphibian ecology and conservation: gender and species identification. NIR N. 25, 10–15 (2014).
Vance, C. K. et al. Near infrared reflectance spectroscopy studies of Chinese giant salamanders in aquaculture production. NIR N. 26, 4–7 (2015).
Helser, T. E., Benson, I. M. & Barnett, B. K. Proceedings of the research workshop on the rapid estimation of fish age using Fourier Transform Near Infrared Spectroscopy (FT-NIRS). AFSC Processed Report 53. (NOAA Institutional Repository, 2019).
Chen, L.-D. et al. Evaluation of reproductive status using near infrared spectroscopy in an endangered anuran. Reprod. Fertil. Dev. 34, 236–237 (2022).
Varga, J. F. A., Bui-Marinos, M. P. & Katzenback, B. A. Frog skin innate immune defences: sensing and surviving pathogens. Front. Immunol. 10, 3128 (2019).
Zouboulis, C. C. The skin as an endocrine organ. Dermatoendocrinol 1, 250–252 (2009).
Carter, E. D. et al. Winter is coming- temperature affects immune defenses and susceptibility to Batrachochytrium salamandrivorans. PLoS Pathog. 17, 1–20 (2021).
Malagon, D. A. et al. Host density and habitat structure influence host contact rates and Batrachochytrium salamandrivorans transmission. Sci. Rep. 10, 5584 (2020).
Thomas, V. et al. Instant killing of pathogenic chytrid fungi by disposable nitrile gloves prevents disease transmission between amphibians. PLoS ONE 15, 1–16 (2020).
Farber, C., Mahnke, M., Sanchez, L. & Kurouski, D. Advanced spectroscopic techniques for plant disease diagnostics. A review. Trends Anal. Chem. 118, 43–49 (2019).
Wang, Y.-H., Li, J.-J. & Su, W.-H. An integrated multi-model fusion system for automatically diagnosing the severity of wheat fusarium head blight. Agriculture 13, 1381 (2023).
Santos-Rivera, M. et al. Profiling Mannheimia haemolytica infection in dairy calves using near infrared spectroscopy (NIRS) and multivariate analysis (MVA). Sci. Rep. 11, 1392 (2021).
Meilina, H., Kuroki, S., Jinendra, B. M., Ikuta, K. & Tsenkova, R. Double threshold method for mastitis diagnosis based on NIR spectra of raw milk and chemometrics. Biosyst. Eng. 104, 243–249 (2009).
Tolleson, D. R. et al. Fecal NIRS: Detection of tick infestations in cattle and horses. Vet. Parasitol. 144, 146–152 (2007).
Cugmas, B. et al. Detection of canine skin and subcutaneous tumors by visible and near-infrared diffuse reflectance spectroscopy. J. Biomed. Opt. 20, 037003 (2015).
Kondepati, V. R., Heise, H. M. & Backhaus, J. Recent applications of near-infrared spectroscopy in cancer diagnosis and therapy. Anal. Bioanal. Chem. 390, 125–139 (2008).
Sakudo, A., Baba, K. & Ikuta, K. Discrimination of influenza virus-infected nasal fluids by Vis-NIR spectroscopy. Clin. Chim. Acta 414, 130–134 (2012).
Sakudo, A. Near-infrared spectroscopy for medical applications: current status and future perspectives. Clin. Chim. Acta 455, 181–188 (2016).
Pichler, M. & Hartig, F. Machine learning and deep learning—a review for ecologists. Methods Ecol. Evol. 14, 994–1016 (2023).
Altman, N. & Krzywinski, M. The curse(s) of dimensionality. Nat. Methods 15, 399–400 (2018).
Kuhn, M. & Johnson, K. Applied Predictive Modeling (Springer, 2013).
U.S. Fish and Wildlife Service. Injurious wildlife species; listing salamanders due to risk of salamander chytrid fungus. Fed. Regist. 81, 1534–1556 (2016).
Grear, D. A., Mosher, B. A., Richgels, K. L. D. & Grant, E. H. C. Evaluation of regulatory action and surveillance as preventive risk-mitigation to an emerging global amphibian pathogen Batrachochytrium salamandrivorans (Bsal). Biol. Conserv. 260, 109222 (2021).
Connelly, P. J., Ross, N., Stringham, O. C. & Eskew, E. A. United States amphibian imports pose a disease risk to salamanders despite Lacey Act regulations. Commun. Earth Environ. 4, 351 (2023).
Woodhams, D. C. et al. Resistance to chytridiomycosis varies among amphibian species and is correlated with skin peptide defenses. Anim. Conserv. 10, 409–417 (2007).
Rollins-Smith, L. A., Reinert, L. K., O’leary, C. J., Houston, L. E. & Woodhams, D. C. Antimicrobial peptide defenses in amphibian skin. Integr. Comp. Biol. 45, 137–142 (2005).
Kearns, P. J. et al. Fight fungi with fungi: antifungal properties of the amphibian mycobiome. Front. Microbiol. 8, 2494 (2017).
Bates, K. A. et al. Microbiome function predicts amphibian chytridiomycosis disease dynamics. Microbiome 10, 44 (2022).
Sabino-Pinto, J. et al. First detection of the emerging fungal pathogen Batrachochytrium salamandrivorans in Germany. Amphib. Reptil. 36, 411–416 (2015).
Allain, S. Emerging infectious disease threats to European herpetofauna. Herpetol. J. 29, 189–206 (2019).
Chen, D., Hu, B., Shao, X. & Su, Q. Removal of major interference sources in aqueous near-infrared spectroscopy techniques. Anal. Bioanal. Chem. 379, 143–148 (2004).
Muncan, J. & Tsenkova, R. Aquaphotomics- from innovative knowledge to integrative platform in science and technology. Molecules 24, 2742 (2019).
Jinendra, B. et al. Near infrared spectroscopy and aquaphotomics: novel approach for rapid in vivo diagnosis of virus infected soybean. Biochem. Biophys. Res. Commun. 397, 685–690 (2010).
Van De Kraats, E. B., Munćan, J. S. & Tsenkova, R. N. Aquaphotomics origin, concept, applications and future perspectives. Int. J. Hist. Chem. 3, 13–28 (2019).
Tsai, S.-R. & Hamblin, M. R. Biological effects and medical applications of infrared radiation. J. Photochem. Photobiol. B 170, 197–207 (2017).
Ishida, H., Nahara, Y., Tamamoto, M. & Hamada, T. The fungicidal effect of ultraviolet light on materials impression. J. Prosthet. Dent. 65, 532–535 (1991).
Cozzolino, D. Infrared spectroscopy as a versatile analytical tool for the quantitative determination of antioxidants in agricultural products, foods and plants. Antioxidants 4, 482–497 (2015).
Silva, W. K., Chicoma, D. L. & Giudici, R. In-situ real-time monitoring of particle size, polymer, and monomer contents in emulsion polymerization of methyl methacrylate by near infrared spectroscopy. Polym. Eng. Sci. 51, 2024–2034 (2011).
Beć, K. B., Grabska, J., Siesler, H. W. & Huck, C. W. Handheld near-infrared spectrometers: where are we heading?. NIR N. 31, 28–35 (2020).
IUCN. The IUCN Red List of Threatened Species. http://www.iucnredlist.org (2020).
Chatfield, M. & Richards-Zawacki, C. Elevated temperature as a treatment for Batrachochytrium dendrobatidis infection in captive frogs. Dis. Aquat. Organ. 94, 235–238 (2011).
Cecala, K. & Price, S. J. A comparison of the effectiveness of recommended doses of MS222 (tricaine methanesulfonate) and Orajel® (benzocaine) for amphibian anesthesia. Herpetol. Rev. 38, 63 (2007).
Stasinopoulos, D. M. & Rigby, R. A. Generalized additive models for location scale and shape (GAMLSS) in R. J. Stat. Softw. 23, 1–46 (2007).
R Core Team. R: A language and environment for statistical computing. https://www.r-project.org/ (2022).
Bischl, B. et al. Mlr: Machine learning in R. J. Mach. Learn. Res. 17, 1–5 (2016).
Stevens A. & Ramirez-Lopez L. An introduction to the prospectr package. R package Vignette R package version 0.2.8. (2025).
Kursa, M. B. & Rudnicki, W. R. Feature selection with the boruta package. J. Stat. Softw. 36, 1–13 (2010).
JMP®, Version 14. SAS Institute Inc., Cary, NC, USA (1989-2023).
Oksanen, J. et al. vegan: Community Ecology Package. R package version 2.6-4. (2022).
Acknowledgements
This project was supported in part by the U.S. Department of Agriculture, Agricultural Research Service, Biophotonics project #6066-31000-015-00D under National Institute of Food and Agriculture Hatch project accession numbers W3173 and W4173, two Institute of Museum and Library Services National Leadership Grant on Amphibian Sustainable Collections Management #MG-30-17-0052-17 and MG-251614-OMS-22, the Association of Zoos and Aquariums and Disney Conservation Grant Fund grants CGF16-1396 and CGF-19-1618, and the U.S. Forest Service International Programs co-operative agreement grant #18-DG-11132762-248 on Coupled Forest-Human Ecosystems: Connecting Forest Management with Conservation Initiatives. This project was also partially funded by the U.S. National Science Foundation (DEB EEID grant 1814520) and the U.S. Department of Agriculture NIFA Hatch project #1012932. Special thanks to Rajeev Kumar for providing technical assistance on this project.
Author information
Authors and Affiliations
Contributions
This study was conceptualized by C.K.K., L.-D.C., and A.J.K. The experimental design and NIR spectral collection were conducted by L.-D.C., E.D.C., M.P.U., C.T.M., M.J.G., and D.L.M. The molecular qPCR analyses were performed by E.D.C., M.P.U., and C.T.M. Predictive modeling, formal statistical analyses, and figure/table creation were conducted by L.-D.C. and D.M.C. The original draft was prepared by L.-D.C., E.D.C., and D.M.C., with all remaining authors having contributed to the review and editing of this article. This work was supported with funds acquired by C.K.K., A.J.K., M.J.G., and D.L.M. who also provided supervision and support.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Communications Biology thanks Minjie FU and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Dr Hannes Schuler and Dr Ophelia Bu.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Chen, LD., Carter, E.D., Urban, M.P. et al. Near-infrared spectroscopy as a diagnostic screening tool for lethal chytrid fungus in eastern newts. Commun Biol 8, 625 (2025). https://doi.org/10.1038/s42003-025-08025-8
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s42003-025-08025-8
