A two‐photon fluorescence micro‐spectroscopy method for chemotaxonomic characterization of fossil resins

Abstract

Premise

Amber, a fossilized chemical defensive exudate, is important for understanding ancient plant–insect/pathogen interactions, yet its botanical origins, especially for Cretaceous amber, remain enigmatic. Here, we develop and apply a fluorescence micro‐spectroscopy approach within the context of a larger study on the provenance of amber.

Methods

Emission spectra were collected from 21 amber, copal, and Defaunation resin samples at six excitation wavelengths. The emission spectra at all excitation wavelengths were deconvolved using simultaneous fitting, with each emission spectrum modeled as a linear combination of four Gaussian components.

Results

Differences between Gaussian means separated the samples into five groups. Cross‐referencing these groupings with prior studies on their botanical origins indicate that they corresponded broadly to ambers from well‐defined conifer families, including the Araucariaceae (group 1 associated with Wollemia and Araucaria, groups 2 and 3 associated with Agathis) and Cupressaceae (group 4), and Baltic amber (group 5) with unresolved taxonomic affinity.

Discussion

Fluorescence micro‐spectroscopy is a useful approach for characterizing fossil resin in a reproducible and nondestructive manner. Variations in the fluorescence properties of resins do not correspond to variations in age or maturity but rather to plant origin. Therefore, this approach provides an additional line of evidence for inferring the botanical origins of amber.

Amber, or ancient tree resin, is a rare example of a fossilized chemical defensive exudate, making it a valuable repository for information on ancient plant–insect and plant–pathogen interactions (Labandeira, 2014; Seyfullah et al., 2018; McCoy et al., 2021; Delclòs et al., 2023). However, amber is typically preserved separately from other remains of the tree that produced it, so identifying its botanical origin is key for placing amber within the larger paleoecological context of ancient plant–insect and plant–pathogen interactions. For this reason, interpreting the botanical sources of amber is a major area of inquiry within amber paleobiology (e.g., Yu et al., 2023; Pańczak et al., 2023, 2024).

There are two basic approaches to determining the botanical origin of an amber deposit: the paleobotanical approach and the chemical approach (Langenheim, 2003). The paleobotanical approach involves determining which tree has the most abundant remains in or in association with the amber. For example, a study of a modern resiniferous forest in Madagascar by Peñalver et al. (2018) found that the majority of the plant remains caught in non‐natural sticky traps often came from the resin‐producing trees themselves (Hymenaea verrucosa Gaertn., Fabaceae). However, some amber assemblages, especially the Cretaceous ones, contain few to no plant bioinclusions (Delclòs et al., 2023), while others have such an abundant and diverse set of plant bioinclusions that it is unclear which correspond to the amber‐producing tree(s) (e.g., Baltic amber; see Sadowski et al., 2017).

The chemical approach involves characterizing the composition of amber using any of a variety of analytical techniques—many of which are destructive, although some need only a very small sample—then comparing it to the chemical composition of amber (older than 2.58 Ma), copal (2.58 Ma–1760 CE, including Pleistocene and Holocene copal), or Defaunation resin (resin produced after the commencement of the Industrial Revolution) with a known botanical origin (Langenheim, 2003; Solórzano‐Kraemer et al., 2020; Pańczak et al., 2023, 2024); these comparisons are complicated by the chemical alteration that occurs as resin matures into amber. Note that in this paper, we are using the terms amber, copal, and Defaunation resin as defined by Solórzano‐Kraemer et al. (2020); other terminologies have also been developed and are reviewed by Solórzano‐Kraemer et al. (2020). For example, the presence of Asiatic acid in Miocene amber from Indonesia indicated that it originated from the dipterocarpaceous genus Shorea Roxb. ex C.F. Gaertn. (Simoneit et al., 2020). Similarly, the abundance of chamaecydin, ferruginol, 7α‐p‐cymenylferruginol, and taxodione in one type of resinite from the Eocene Geiseltal fossil site in Germany links it to the cupressaceous genus Taxodium Rich. (Simoneit et al., 2021). However, amber is extremely chemically complex, and most chemical techniques cannot elucidate all details of its chemistry.

Because each approach has inherent limitations, new analytical strategies continue to be developed to improve taxonomic resolution and preserve sample integrity. One underutilized approach in this area is fluorescence micro‐spectroscopy. Fluorescence spectroscopy has previously been applied to amber, but not in the context of determining its botanical origins, and only as an average over volume rather than with microscopic resolution (e.g., Islas et al., 2001; Bellani et al., 2005). Amber exhibits intrinsic fluorescence due to polycyclic and oxygenated chromophores embedded in its resin matrix, yet this property has rarely been explored systematically for taxonomic classification. Fluorescence emission spectra offer a chemically sensitive, nondestructive, and spatially resolved signal that complements existing analytical techniques (Lakowicz, 2006; Stoneman and Raicu, 2023). Two‐photon fluorescence microscopy enables high‐resolution imaging and deep optical sectioning with reduced photodamage (Denk et al., 1990; Biener et al., 2014). When combined with spectral deconvolution methods, such as Gaussian fitting, it provides a robust framework for extracting reproducible chemical signatures from amber, copal, and Defaunation resin.

Here, we present a new method using two‐photon fluorescence micro‐spectroscopy (Raicu et al., 2009; Tahoun et al., 2024; Stoneman et al., 2024a) to determine the botanical origins of amber and apply it within the context of a larger study to develop a reproducible protocol for chemotaxonomic classification. In addition, we evaluate the new protocol to determine whether the fluorescence emission properties of these samples—captured across multiple excitation wavelengths and modeled as linear mixtures of Gaussian components—can be used to characterize resin composition and discriminate between resins of different botanical origin through the following steps.

• 1.We assess whether the fluorescence emission properties vary between resin samples; if all samples fluoresce identically, this will not be a useful tool for characterization.
• 2.We assess whether age and geological maturation influence the fluorescence properties significantly. Typically, the source of Holocene resin and copal samples are known to the species level, whereas the origin of amber samples is debated at even the family level (Langenheim, 2003). Therefore, being able to directly compare Holocene samples with well‐constrained botanical origins to ancient samples with disputed or unknown origins is critical.
• 3.Finally, we carry out a preliminary qualitative assessment on the current set of samples to confirm that variations in fluorescence properties correspond to variations in likely botanical origins.

By developing a detailed, reproducible protocol for characterizing and quantifying the fluorescence properties of resin samples of a wide range of geological ages and degrees of maturation, and by carrying out these initial steps for assessing the utility of this new method for characterizing resins and inferring taxonomic origins, this study provides a solid foundation for future studies to more quantitatively evaluate competing hypotheses about the botanical origins of major amber deposits.

METHODS

Materials

We analyzed 21 samples of conifer‐derived amber, copal, and Defaunation resin from various assemblages: three samples of Cretaceous amber from Lebanon, seven samples of Cretaceous amber from Spain, three samples of Cretaceous amber from France, four samples of Eocene Baltic amber, one sample of Pleistocene Agathis australis (D. Don) Loudon copal from New Zealand, and three samples of Holocene resin (Wollemia nobilis W.G. Jones, K.D. Hill & J.M. Allen, Araucaria columnaris (G. Forst.) Hook., and Araucaria cunninghamii Mudie) from The Huntington in San Marino, California, USA (Figures 1, 2; Table 1). The Cretaceous amber samples and the sample of Agathis australis copal were previously analyzed by McCoy et al. (2021) using solid‐phase microextraction gas chromatography–mass spectrometry (SPME GC‐MS); photography was carried out on these samples prior to being powdered for SPME GC‐MS analysis. The Baltic amber samples, which contain unidentified plant inclusions, come from the Paleobotany Collection of the Field Museum of Natural History in Chicago, Illinois, USA. It is thought that Baltic amber was produced by only one species of tree (Wolfe et al., 2009; Weitschat and Wichard, 2010; Dolezych et al., 2011), so these samples will allow us to determine how much variation exists in the fluorescence properties of amber from one tree species. Baltic amber was found originally in several deposits across Europe (Weitschat and Wichard, 2010). The chemical composition of Baltic amber from all these deposits is similar enough that they are all recognizably the same type of amber (often termed succinite), but their isotopic composition can differ (Wolfe et al., 2009, 2016; Mänd et al., 2018). The fluorescence spectra for three of the Baltic amber samples (FMNH PE 93421, FMNH PE 93422, and FMNH PE 93423) have previously been published (Stoneman et al., 2024a) but were subjected to additional analysis for the current study. The fourth sample of Baltic amber (FMNH PP 62126) is analyzed here for the first time.

Figure 1.

Samples included in this study; locality details are provided in Table 1, and the sample numbers agree with those listed there. Scale bars = 5 mm. (A) Sample 1, Holocene Wollemia nobilis resin. (B) Sample 2, Holocene Araucaria cunninghamii resin. (C) Sample 3, Holocene Araucaria columnaris resin. (D) Sample 4, Pleistocene copal, 41,370 ± 2590 years (Solórzano‐Kraemer et al., 2020). (E) Sample 5, Baltic amber, Eocene. (F) Sample 6, Baltic amber, Eocene. (G) Sample 7, Baltic amber, Eocene. (H) Sample 8, Baltic amber, Eocene. (I) Sample 9, French amber, Santonian. (J) Sample 10, Spanish amber, Cenomanian. (K) Sample 11, French amber, Cenomanian. (L) Sample 12, French amber, Albian. (M) Sample 13, Spanish amber, Albian. (N) Sample 14, Spanish amber, Albian. (O) Sample 15, Spanish amber, Albian. (P) Sample 16, Spanish amber, Albian. (Q) Sample 17, Spanish amber, Albian. (R) Sample 18, Spanish amber, Albian–Cenomanian. (S) Sample 19, Lebanese amber, Barremian. (T) Sample 20, Lebanese amber, Barremian. (U) Sample 21, Lebanese amber, Barremian.

Figure 2.

Selected photos of sampled trees and outcrops (arrows); locality details are provided in Table 1, and the sample numbers agree with those listed there. (A) A Holocene Wollemia nobilis tree (Araucariaceae) (left), with sampled resin (right) (Sample 1). (B) Pleistocene copal outcrop (Sample 4). (C) Cenomanian amber outcrop (Sample 10). (D) Cenomanian amber outcrop (Sample 11). (E) Albian amber outcrop (Sample 12). (F) Albian amber outcrop (Sample 13). (G) Barremian amber outcrop (Samples 20 and 21).

Table 1.

List of samples, showing sample details, the wavelengths (nm) corresponding to the emission maximum for the four Gaussian components (λ₁–λ₄), and the assigned group.

				Wavelength corresponding to emission maximum (nm)
Sample	Amber type	Locality	Age	λ1	λ2	λ3	λ4	Group
1a	Resin	Wollemia nobilis, The Huntington, San Marino, California, USA; accession no. 95020	Holocene	475.58	516.00	566.72	647.68	NA
1b	Resin	Wollemia nobilis, The Huntington, San Marino, California, USA; accession no. 95020	Holocene	460.48	503.30	548.00	650.70	1
2	Resin	Araucaria cunninghamii, The Huntington, San Marino, California, USA; accession no. 614	Holocene	456.85	497.80	543.57	609.00	1
3a	Resin	Araucaria columnaris, The Huntington, San Marino, California, USA; accession no. 37224	Holocene	495.80	573.37	635.83	666.36	NA
3b	Resin	Araucaria columnaris, The Huntington, San Marino, California, USA; accession no. 37224	Holocene	454.99	494.84	543.33	611.33	1
4	Copal	Agathis australis, Baylys Beach, near Dargaville, North Island, New Zealand	Pleistocene	444.83	492.84	537.10	608.90	2
5	Baltic	FMNH PE 93421, no specific locality information known	Eocene	469.47	511.82	547.41	636.18	5
6	Baltic	FMNH PE 93422, no specific locality information known	Eocene	469.75	511.07	544.26	601.13	5
7	Baltic	FMNH PE 93423, no specific locality information known	Eocene	468.04	508.67	545.22	614.32	5
8	Baltic	FMNH PP 62126, no specific locality information known	Eocene	471.65	511.40	546.75	609.01	5
9	French	Piolenc, Vancluse, France	Santonian	450.91	506.58	559.22	611.09	4
10	Spanish	Fuensanta (Nava), Asturias, Spain	Cenomanian	441.11	491.75	537.48	612.35	2
11	French	Les Renardières, Rochefort, France	Cenomanian	444.82	502.30	555.03	621.40	1
12	French	Cadeuil, Rochefort, France	Albian	451.58	505.46	555.83	597.57	1
13	Spanish	Arroyo de la Pascueta (Rubielos de Mora), Teruel, Spain	Albian	448.41	499.15	549.53	646.49	1
14	Spanish	Salinillas de Buradón, Álava, Spain	Albian	432.27	488.13	542.06	649.64	3
15	Spanish	San Just, Teruel, Spain	Albian	449.36	502.70	550.85	639.77	1
16	Spanish	El Soplao, Cantabria, Spain	Albian	446.66	492.33	535.52	604.26	2
17	Spanish	Cabañas de Virtus, Burgos, Spain	Albian	436.58	486.77	541.97	644.70	3
18	Spanish	Reocín, Cantabria, Spain	Albian–Cenomanian	443.07	495.26	542.56	612.55	1
19	Lebanese	Hammana, Lebanon	Barremian	441.46	491.29	538.14	624.12	2
20	Lebanese	Jezzine Top, Lebanon	Barremian	438.84	498.97	548.69	607.97	1
21	Lebanese	Jezzine Base, Lebanon	Barremian	454.15	501.24	543.72	603.75	1

Note: NA = not applicable (used for samples that were not placed in a group).

Two‐photon fluorescence micro‐spectroscopy

The protocols for imaging the amber samples follow those developed by Stoneman et al. (2024a), as briefly explained below. For this study, we developed and optimized a new data analysis process to allow for spectral fingerprinting of amber. This data analysis process is described subsequently.

Two‐photon microscope with spectral resolution and line‐scan excitation

The imaging instrument used here is the two‐photon microscope with spectral resolution in the University of Wisconsin–Milwaukee Optical Micro‐Spectroscopy Imaging Facility, which was previously described in detail (Biener et al., 2014). The excitation source of the micro‐spectroscope is a femtosecond‐pulsed ultrafast laser (Mai Tai, Spectra Physics, Milpitas, California, USA) tunable between 690 and 1040 nm. Excitation was performed at six discrete wavelengths across this range. The excitation beam is shaped into a line at the sample plane using a cylindrical mirror placed in the excitation path before the galvanometric scanners. True line‐scan excitation allows for the simultaneous excitation of hundreds of sample voxels, similar to operating hundreds of individual imaging instruments concurrently. This approach significantly boosts signal levels, enabling a considerable reduction in excitation power, which is critical for imaging organically preserved fossil material that may be susceptible to photobleaching and laser damage (Stoneman et al., 2024a). Incorporating this line‐scan technology enhances sensitivity, which is particularly beneficial for samples with low expression levels.

Samples were mounted on an inverted microscope (Nikon Eclipse Ti‐U, Nikon Instruments, Melville, New York, USA) equipped with a 10× infinity‐corrected objective lens with a numerical aperture (NA) of 0.3. Fluorescence emission from the sample was then dispersed through a transmission grating onto an electron‐multiplying CCD camera (iXon X3, Andor Technologies, Belfast, Northern Ireland). The resulting fluorescence image featured spatial information along one axis and spectral information along the other, providing full spectral data for each voxel. The time to collect the full spectral output from a voxel was the laser dwell time in that region. By acquiring frames at different excitation light spots in the sample, the microscope reconstructed four‐dimensional fluorescence images, with three dimensions in the spatial dimension and one spectral dimension, with only a single laser pass through each respective voxel.

Spectral acquisition parameters

Each sample was imaged at six different excitation wavelengths: 760, 800, 840, 880, 920, and 960 nm. For each sample, three fields of view (FOVs) were selected, and each FOV was scanned once at each of the six excitation wavelengths. Each excitation scan produced a micro‐spectroscopic image stack comprising two spatial dimensions and one spectral dimension, with emission recorded across channels separated by a minimum of 1 nm. Emission spectra were captured over a range of 430 to 650 nm. A fixed camera offset bias was subtracted from each image pixel prior to analysis. For each of the six excitation wavelengths, the fluorescence emission spectra from the three FOVs were averaged, resulting in six distinct average emission spectra per sample (Figure 3A).

Figure 3.

Line charts showing example spectra, Gaussian components, and fitting curve for sample 10 (Spanish amber, Fuensanta [Nava], Asturias, Spain, Cenomanian). Such plots are provided for all samples in Appendix S1. (A) Typical amber emission spectra. (B) Fitting curve based on four Gaussian components whose maxima were constrained at the same position across all the excitation wavelengths.

Gaussian deconvolution of average emission spectra

Fluorescent molecules are typically characterized by an emission spectrum that can be approximately modeled as one to two Gaussian functions. Because the total number of fluorescent species in each of our amber samples was not known a priori, we devised a systematic process to identify the optimal number of Gaussian‐like contributions needed to fit each sample, which was facilitated by the fact that different excitation wavelengths should excite to a different extent for each fluorescent species in the samples. To facilitate this process, we fitted each average fluorescence emission spectrum with a linear combination of N Gaussian components using the following equation:

$$I_{\mathit{model}}({\lambda }_{\mathit{ex}},{\lambda }_i)=\sum _{k=1}^N{A}_{{\lambda }_{\mathit{ex}},k}\text{exp}\left[-\frac{{({\lambda }_k-{\lambda }_i)}^2}{2{\sigma }_k^2}\right]$$ (1)

where ${\lambda }_i$ is the emission wavelength channel, $A_{{\lambda }_{\mathit{ex}},k}$ the amplitude, ${\lambda }_k$ the center wavelength (mean), and ${\sigma }_k$ the standard deviation for each Gaussian component k. Fitting was performed independently for each excitation wavelength, but with the constraint that the Gaussian means (${\lambda }_k$) and widths (${\sigma }_k$) remained fixed across all six excitation wavelengths for a given sample. Only the amplitudes, $A_{{\lambda }_{\mathit{ex}},k}$, were allowed to vary between excitation wavelengths. The optimal parameters were determined by minimizing the residual between the modeled and measured fluorescence spectra across all excitation wavelengths. Residuals were calculated as the root mean square error (RMSE) between the measured and modeled intensity values for each emission channel as follows:

$$R=\sqrt{{\sum }_{{\lambda }_{ex}}{\sum }_i{(I_{measured}({\lambda }_{ex},{\lambda }_i)-I_{model}({\lambda }_{ex},{\lambda }_i))}^2}$$ (2)

where ${\lambda }_{\mathit{ex}}$ is the set of excitation wavelengths (760, 800, 840, 880, 920, and 960 nm) and ${\lambda }_i$ the emission channels. Curve fitting was performed using the Solver add‐in in Microsoft Excel (Microsoft, Redmond, Washington, USA), employing a nonlinear least‐squares optimization routine. The final number of Gaussian components (N) considered to uniquely characterize all samples investigated without overfitting or redundancies for each individual sample was four.

RESULTS

Qualitative characterization of the effects of amberization on fluorescence

Before applying spectral deconvolution and quantitative modeling, we first examined the general fluorescence emission characteristics of the samples to assess variability across the resin‐to‐amber age/maturation spectrum. The fluorescence emission properties of Defaunation resin samples differed from those of amber/copal in two notable ways—lower overall intensity and greater internal heterogeneity. Both of these differences in fluorescence properties can be attributed to chemical differences between Defaunation resin and amber/copal.

First, the fluorescence intensity of the Defaunation resin samples was consistently lower than the fluorescence intensity for the amber and copal samples. The particular molecular species responsible for the fluorescence of amber have not been fully characterized. However, those fluorescent species that have been identified in amber (e.g., aromatic compounds such as benzo[a]anthracene in late Albian to early Cenomanian Burmese “chameleon” amber [Li et al., 2024] and perylene in blue Miocene Dominican amber [Bellani et al., 2005]) are typically produced as resin matures into amber (through a complex and not yet fully understood process called “amberization”) and are not present in large amounts in resin. Therefore, the lower fluorescence intensity in the Defaunation resin samples is likely a factor of their young age and limited geological maturation. Notably, perylene can also be produced through wood‐degrading fungal activity (Marynowski et al., 2013), and fungal activity may play an as‐yet‐unexplored role in amber fluorescence.

Second, two of the Defaunation resin samples (from Wollemia nobilis and Araucaria columnaris) were more internally variable than the other Defaunation resin sample (A. cunninghamii), copal sample, and amber samples. For these two Defaunation resin samples, different regions of interest had noticeably different fluorescence emission spectra, so these samples were further separated into two subsamples to encompass this variation (samples 1a and 1b of W. nobilis and samples 3a and 3b of A. columnaris in Table 1). In each of these samples, one subsample (subsamples 1a and 3a) differed from amber primarily in the shape of the fluorescence emission curve produced through excitation at 760 nm, which was strongly left‐skewed. The other subsample (subsamples 1b and 3b) looked roughly similar to amber, although with lower fluorescence intensity, as was discussed above.

There are a few reasons why resin might be more heterogeneous than amber (Seyfullah et al., 2018). Resin, as a complex mixture of volatile, semi‐volatile, and non‐volatile compounds that commonly trap organic debris, is inherently heterogeneous (Pańczak et al., 2023, 2024). In addition, the amberization process, which involves the oxidation of organic material, loss of volatile compounds, and cross‐linking and polymerization of other compounds (Clifford and Hatcher, 1995; Ragazzi et al., 2003; Tonidandel et al., 2008; Pańczak et al., 2023, 2024), may occur heterogeneously throughout a piece of resin, as the original chemical composition is one of many factors that can influence amberization (Tonidandel et al., 2008).

We suggest that the regions of varying fluorescence properties in the Wollemia nobilis and Araucaria columnaris resin samples represent regions that have undergone different degrees of amberization. There are no patterns to this variation that might indicate what aspect of amberization most strongly influences fluorescence. For example, if the periphery fluoresced differently than the center, it might suggest that oxidation controls fluorescence. Therefore, to compare them to the more fully amberized samples (copal and amber), we used only the regions within samples 1 and 3 that were more consistent with typical amber fluorescence patterns (samples 1b and 3b in Table 1).

Our preliminary observations underscore the potential of fluorescence data to reveal structural and chemical heterogeneity across stages of resin fossilization. In other words, resin fluorescence does vary between samples, and the age and maturation of resin samples does not strongly influence their fluorescence properties in general. To rigorously evaluate these patterns and to determine whether specific spectral features can be used to chemotaxonomically classify fossil resins, we applied next a quantitative fluorescence analysis using spectral deconvolution and Gaussian modeling, followed by a systematic, qualitative grouping of samples based on component emission parameters.

Quantitative fluorescence characterization of resin samples

Spectral deconvolution and component determination

To extract meaningful chemical information from the fluorescence emission profiles of the fossilized resin samples, we applied a linear mixture model (LMM), which contained a pre‐selected number of Gaussian functions with adjustable wavelengths associated with their emission maximum, widths, and amplitudes. Each average emission spectrum was unmixed into a small number of elementary spectral components using the LMM, enabling us to isolate contributions from distinct fluorescent species. Those components are hypothesized to correspond to individual fluorophores or dominant spectral motifs within the complex amber matrix.

We devised a systematic process for identifying the minimum number of such irreducible spectral components needed to accurately model the fluorescence spectra across all excitation wavelengths while minimizing the number of fitting parameters. In this regard, exciting each sample at multiple wavelengths was important. Each fluorescent compound may exhibit a unique dependence of its extinction coefficient on excitation wavelength, resulting in excitation‐specific differences in composite emission spectrum. These wavelength‐dependent excitation efficiencies alter the relative contributions of each fluorophore to the observed emission spectra, thereby providing additional constraints that help reduce the number of free parameters required to model the spectra accurately.

We began by using two Gaussian components to visually capture the region at and around the main peak of an individual emission spectrum for each excitation wavelength. A constant value was then added to the sum of the two Gaussian functions used to fit the experimental spectra to also capture some of the short‐ and long‐wavelength wings of the spectra. This protocol did not fully capture the non‐constant behavior at the two wings of each spectrum. Therefore, we looked for commonalities between the positions of all the Gaussian functions for all six emission spectra and roughly identified four to five groups of peak positions for each set of six emission spectra for all 21 samples investigated. Thus, four Gaussian functions and a bias parameter were added together to model the emission spectra, and the six spectra were simultaneously fit, as described in the Methods section. Specifically, we independently adjusted the amplitude of each Gaussian component, because each excitation wavelength excited each fluorescent species to a different extent. The peak positions and widths of the Gaussian components were also treated as fitting parameters but were constrained to remain constant across all six excitation spectra for a given sample, thereby representing shared spectral features independent of excitation wavelength. Finally, we removed the bias value and refined the simultaneous fit for each sample by using the peak positions and widths obtained in the previous fitting attempt as initial guesses. Through this iterative procedure, we found that four Gaussian components were sufficient to reproducibly fit the emission spectra of each sample across all excitation wavelengths and for all samples investigated. Fewer than four Gaussian components failed to resolve important spectral features. Representative examples of these fitted emission spectra and their deconvoluted components are shown in Figure 3B.

Importantly, while an individual fluorescent species may often exhibit an emission spectrum that closely resembles a single Gaussian distribution, it is also common for more complex fluorophores to display additional shoulders or sub‐peaks—due, for example, to conformational isomers, vibrational transitions, or subspecies—resulting in a broader or asymmetric spectrum (Donaldson, 2020; Adhikari et al., 2021). These features can often be decomposed into two overlapping Gaussian components. Therefore, the four Gaussian components used in our model may represent approximately two to four chemically distinct fluorescent species in each sample.

Once the four‐Gaussian model was established as the minimal configuration capable of capturing the emission spectra, we proceeded to extract quantitative fitting parameters using this model framework. For each sample, the emission spectra collected at all six excitation wavelengths were fitted simultaneously using a constrained model. The amplitudes of the four Gaussian components were allowed to vary independently for each excitation wavelength, while the center wavelengths and widths of each component were held fixed across excitations. This allowed the capture of excitation‐dependent intensity changes while preserving a consistent spectral fingerprint for each sample. Optimal parameter values were obtained by minimizing the RMSE between modeled and measured intensity values across all emission channels and excitation wavelengths (Eq. 2).

The resulting center wavelengths of the four Gaussian components—representing the characteristic emission peaks of each modeled fluorophore—were extracted for each sample and served as the key parameters for classification. These center wavelengths exhibited systematic differences across resin types and were used as the primary features for grouping the samples into chemotaxonomic clusters, as described below. Details of the four Gaussian components, along with the average fluorescence emission spectrum for each sample, are given in Appendix S1.

Grouping samples based on fluorescence properties

The wavelength at maximum emission of the four Gaussian components varied among the 21 samples and was the most useful parameter for distinguishing subgroups. Each sample was characterized by four emission maxima corresponding to its fitted spectral components. These peak positions reflect the intrinsic fluorescence features of the underlying chemical constituents and were held fixed across excitation wavelengths, providing a consistent spectral fingerprint for each sample.

We visually examined trends in ${\lambda }_k$ values across samples and found that variation in the parameters extracted for Gaussian components 1, 2, and 3 (i.e., ${\lambda }_1$, ${\lambda }_2$, and ${\lambda }_3$) provided strong discriminatory power. To visualize these spectral differences, we plotted the extracted values of ${\lambda }_1$, ${\lambda }_2$, and ${\lambda }_3$ for each individual sample as coordinates along the x‐, y‐, and z‐axes of a three‐dimensional scatter plot (Figure 4), where each point corresponds to a single sample, positioned according to its component peak wavelengths. Visually, the samples formed distinct clusters within this spectral parameter space, revealing natural groupings based on emission peak positions. No mathematical test was used to assess these clusters; in future analyses, groups will be developed based on modern samples of known botanical origins and defined taxonomically. To highlight these clusters, we assigned colors to the five groups identified, facilitating interpretation of their separation in the plot. Based on clustering patterns in the ${\lambda }_k$ space, we separated the amber samples into five groups (Table 1, Figure 4). In most cases, members within a group shared at least three closely aligned Gaussian center wavelengths, suggesting shared chemical features or precursor compounds. The fourth component (${\lambda }_4$) did not strongly influence group assignments, although it remained broadly consistent within groups.

Figure 4.

Three‐dimensional scatter plot of wavelengths at maximum emission of Gaussian components 1–3 (i.e., ${\lambda }_1$, ${\lambda }_2$, and ${\lambda }_3$) for all 21 samples included in this study. Key: Group 1, blue circles; Group 2, red squares; Group 3, pink diamonds; Group 4, black star; Group 5, green triangles.

Notably, the samples did not group by age or geographic origin (Table 1), suggesting that fluorescence‐based groupings may reflect differences in botanical origin. This grouping strategy is entirely data‐driven, relying only on intrinsic spectral properties and without prior assumptions about taxonomy, geographic origin, or geological age. Once these spectral groups were defined, we examined external metadata (geochemical or paleobotanical evidence) to suggest which botanical origins were represented in each group (see Discussion).

Group 1 (Table 1, blue circles in Figure 4) consists of three samples of Holocene araucariaceous resin produced by Wollemia nobilis, Araucaria cunninghamii, and A. columnaris (samples 1b, 2, 3b). In addition, it includes one sample of Cenomanian French amber (sample 11), one sample of Albian French amber (sample 12), two samples of Albian Spanish amber (samples 13, 15), one sample of Albian–Cenomanian Spanish amber (sample 18), and two samples of Barremian Lebanese amber (samples 20, 21). Group 2 (Table 1, red squares in Figure 4) is made up of the Pleistocene Agathis australis (Araucariaceae) copal from Baylys Beach (sample 4). In addition, it contains one sample of Cenomanian Spanish amber (sample 10), one sample of Albian Spanish amber (sample 16), and one sample of Barremian Lebanese amber (sample 19). Group 3 (Table 1, pink diamonds in Figure 4) comprises two samples of Albian Spanish amber (samples 14, 17). Group 4 contains only one sample (Table 1, black star in Figure 4): the French amber from the Santonian of Piolenc (sample 9). Group 5 includes four samples of Baltic amber without precise locality information (samples 5, 6, 7, 8; green triangles in Figure 4).

DISCUSSION

Chemical and botanical insights of fluorescence‐derived groups

The only samples with clear, traceable botanical origins are the araucariaceous resin and copal. The araucariaceous resin samples (from the genera Wollemia W.G. Jones, K.D. Hill & J.M. Allen and Araucaria Juss.) are found in Group 1, and the copal sample (genus Agathis Salisb.) is in Group 2. The samples of Cretaceous amber from Lebanon, France, and Spain are distributed across Groups 1–4; their botanical origins are not definitively known, but various previously published analyses, reviewed in the following paragraphs, provide some information.

Little outcrop‐specific research has been done on the botanical origin of Lebanese amber; however, Lebanese amber is most commonly attributed to the Araucariaceae or Cheirolepidiaceae (Seyfullah et al., 2018; Pańczak et al., 2024). For Cretaceous amber from Lebanon, which would include all samples in this study, a botanical origin from the Araucariaceae is more strongly supported. Pyrolysis gas chromatography–mass spectrometry (GC‐MS) revealed it contains compounds with a labdane moiety, as well as abietane and pimarane, a composition very similar to modern araucariaceous resin; Fourier transform infrared spectroscopy (FTIR), one of the most standard and effective analyses for determining the botanical origins of amber (e.g., Seyfullah et al., 2015), has also demonstrated that Cretaceous Lebanese amber yields a spectrum very similar to that of modern araucariaceous resin (Nohra et al., 2014; Pańczak et al., 2024). However, an origin from the Cheirolepidiaceae cannot be ruled out (Nohra et al., 2014). Plant remains are found in association with Lebanese amber, but these have been interpreted as either araucariaceous or cheirolepidiaceous depending on the author (Azar et al., 2010). Therefore, the Lebanese amber samples (sample 19 in Group 2 and samples 20 and 21 in Group 1) are likely araucariaceous, although we cannot rule out a cheirolepidiaceous origin.

In general, based on infrared spectroscopy, solid‐state ¹³C nuclear magnetic resonance (¹³C NMR) spectroscopy, and thermochemolysis GC‐MS, the source of French amber shows a taxonomic shift between the Lower and Upper Cretaceous. Lower Cretaceous French ambers (which include samples 11 and 12 in Group 1) are typically attributed to the Araucariaceae or the Cheirolepidiaceae due to abundant associated araucariaceous and cheirolepidiaceous plant remains in the amber‐bearing rock and cheirolepidiaceous remains and agathic acid (a biomarker of the Araucariaceae) in the amber itself (Perrichot, 2003; Nohra et al., 2015). In contrast, Upper Cretaceous French amber (including sample 9 in Group 4) is generally considered to be produced by the Cupressaceae; notably, amber from Piolenc, the same locality that produced our sample 9, contains callitrisic acid and cuparene, which are biomarkers of the Cupressaceae (Nohra et al., 2015; Lenen et al., 2023). There is a distinct lack of fossil plant bioinclusions in the Piolenc amber, but the few specimens identified are araucariaceous (Nohra et al., 2015). However, amber is often produced in mixed forests, and amber, especially Cretaceous amber, can be associated with plant remains from multiple families (e.g., Delclòs et al., 2007, 2023). Moreover, although plant remains in amber can be a valuable tool for determining the amber's origin, this likely only works well for amber with abundant plant bioinclusions, and therefore it may not be effective for Cretaceous ambers with minimal plant bioinclusions (Peñalver et al., 2018), such as the Piolenc assemblage. Overall evidence suggests that Piolenc amber was produced by the Cupressaceae (Nohra et al., 2015; Lenen et al., 2023). Nevertheless, it is possible that the araucariaceous inclusions originated from the tree that produced Piolenc amber.

Cretaceous amber from Spain is primarily attributed to either the Araucariaceae or the Cheirolepidiaceae, based on its chemical composition, plant inclusions in the amber, associated plant fossils in the amber‐bearing rock, and extent of Araucariaceae‐dominated forests and Cheirolepidiaceae‐dominated forests in the Cretaceous of Spain. These attributions vary both geographically and stratigraphically; therefore, assessing the most likely botanical source of each sample requires consideration both of its specific locality and age. For most samples, this is discussed in more detail below; however, the localities that produced sample 10 (in Group 2) and sample 17 (in Group 3) have not been studied with respect to their origins, so the likely botanical source of these samples remains unknown (Delclòs et al., 2007).

Albian Spanish amber from Teruel (which includes the localities that produced samples 13 and 15 in Group 1) contains high levels of amberene and labdane resin acids and an absence of ferruginol, as well as abundant associated araucariaceous fossils in the amber‐bearing rock, which is consistent with an origin from the Araucariaceae (Menor‐Salván et al., 2016; Álvarez‐Parra et al., 2021; Barrón et al., 2023). Sample 16 from Group 2 and sample 18 in Group 1 come from Cantabria (El Soplao and Reocín, respectively). Early GC‐MS analyses of Cantabrian amber suggested an origin from the Cheirolepidiaceae or Cupressaceae due to the presence of biomarkers such as ferruginol (Menor‐Salván et al., 2010). For El Soplao amber specifically, more recent analyses showed it had low levels of amberene and was produced in Cheirolepidiaceae‐dominated forests, supporting the original cheirolepidiaceous interpretation (Menor‐Salván et al., 2016). However, Kvaček et al. (2018) disputed such an origin for El Soplao amber, because they identified compression fossils as araucariaceous in the amber‐bearing strata. There are also associated cheirolepidiaceous remains, including one bioinclusion in the amber (Delclòs et al., 2023). Based on geochemistry and paleobotany, both families are equally likely as the amber‐producing trees. In fact, both families may have contributed to this amber assemblage. For Reocín amber specifically, a more recent study identified high amounts of amberene and indicated that it was likely produced in an Araucariaceae‐dominated forest, suggesting a likely araucariaceous origin (Menor‐Salván et al., 2016).

One of the Spanish amber samples in Group 3 (sample 14) comes from the Salinillas de Buradón locality. Although the botanical origin of the amber from this locality has not been directly studied, more extensive analyses have been carried out on the Albian Peñacerrada I and II localities, which are at the same stratigraphic level as our locality and are situated very close to it, only 10–11 km away (Peñalver et al., 2025). GC‐MS analysis of Albian amber from Peñacerrada II (Álava) demonstrated that it contains large amounts of kaurenes as well as potential derivatives of agathic acid, supporting a botanical origin from the Araucariaceae, possibly from the genus Agathis (Chaler and Grimalt, 2005). In addition, amber from Peñacerrada I and II lacks the common biomarkers of the Podocarpaceae and Cupressaceae (Alonso et al., 2000; Chaler and Grimalt, 2005). Menor‐Salván et al. (2016) carried out GC‐MS analysis on Albian amber from Peñacerrada I and II and determined that it contains high quantities of amberene and other norlabdane diterpenoids and low amounts of abietanes, which is consistent with an araucariaceous origin. Moreover, Menor‐Salván et al. (2016) argued that the Cretaceous forests in the Álava region were dominated by the Araucariaceae as the major resin‐producing trees. Finally, a paleobotanical investigation of the Peñacerrada I locality revealed araucariaceous remains as bioinclusions in the amber, which further supports the Araucariaceae as the source (Kvaček et al., 2018). Thus, we suggest an araucariaceous origin for our sample 14, and possibly an origin from the genus Agathis, due to the provenance of the Peñacerrada I and II samples from the same stratigraphic level; however, future studies on the Salinillas de Buradón outcrop specifically may change this interpretation.

Therefore, for the majority of the samples in Group 1, an araucariaceous origin is most strongly supported. However, a cheirolepidiaceous origin cannot be excluded in all cases. This suggests that Group 1 includes amber originating from both families, but more likely from the Araucariaceae. Group 2 includes one sample (Pleistocene copal) of definite araucariaceous origin. The other amber samples are of araucariaceous or possibly of cheirolepidiaceous origin. Therefore, Group 2 most likely represents amber of the Araucariaceae; however, it may instead represent samples from both families. Although the two Spanish amber samples in Group 3 come from poorly known localities, studies of nearby localities from the same or similar stratigraphic horizon suggest that the Group 3 samples correspond to an origin from the Araucariaceae. Finally, a cupressaceous origin is strongly supported for the one sample in Group 4.

Group 5 is composed entirely of Baltic amber samples. The fluorescence properties of these samples are variable, which implies that the amber produced by one single tree species does not fluoresce identically, possibly due to variations in chemical composition related to different sites of deposition. Although the botanical origin of Baltic amber is still debated (Wolfe et al., 2009), the Cupressaceae have never been suggested as a likely origin and the Cheirolepidiaceae were long extinct by the Eocene (Smith et al., 2024), when Baltic amber was being deposited (Weitschat and Wichard, 2010; Barreda et al., 2012; Yu et al., 2023). Anderson et al. (1992) chemically categorized amber into different classes based on their structural characteristics using pyrolysis GC‐MS and grouped Baltic amber with araucariaceous ambers into their class I. A later, more detailed analysis (based on FTIR, NMR, and GC‐MS) also noted compositional similarities between Baltic amber and Agathis resin (Czechowski et al., 1996). However, other chemical differences, such as an absence of succinic acid in araucariaceous resin, as well as the lack of araucariaceous bioinclusions in Baltic amber, make this family unlikely (Sadowski et al., 2017, 2022; Pańczak et al., 2023).

Instead, the two most likely sources for Baltic amber are the Sciadopityaceae and Pinaceae. The Sciadopityaceae have been suggested based on chemical analyses and the recent identification of sciadopityaceous bioinclusions in Baltic amber (Wolfe et al., 2009; Sadowski et al., 2016). The Pinaceae have also been suggested based on paleobotanical evidence; the most abundant plant bioinclusions in Baltic amber, and plant remains otherwise associated with Baltic amber, pertain to the Pinaceae (Weitschat and Wichard, 2010; Dolezych et al., 2011). Therefore, the origin of the Baltic amber in Group 5, while still unclear, is likely a conifer family different than those that produced the ambers of Groups 1–4. Additional studies that include a wider range of modern Defaunation resin samples, particularly pinaceous or sciadopityaceous resin, will be necessary to assess which familial origin for Baltic amber is best supported by its fluorescence properties.

Broader interpretation of spectral groupings

Variation in the fluorescence properties of araucariaceous amber

Our data cluster into five distinct groups, three of which (Groups 1, 2, 3) comprise samples of likely araucariaceous origin and those of definite araucariaceous origin. This provides strong evidence that araucariaceous amber, copal, and Defaunation resin encompass multiple groups with distinct fluorescence properties. The fluorescence emission spectrum of an amber sample is a combination of the fluorescence emitted by all the fluorescent molecules in the amber (Islas et al., 2001; Matuszewska and Czaja, 2002; Bellani et al., 2005; Al‐Tamimi et al., 2020; Stoneman et al., 2024a), and distinct fluorescence properties should therefore correspond to distinct chemical compositions. Menor‐Salván et al. (2016) identified multiple chemically distinct groups of araucariaceous amber (their Types 1 and 3, which differ in their relative proportions of various terpenoids). Only three localities provided samples for both this study and for that of Menor‐Salván et al. (2016): the Albian San Just locality in Teruel, which produced sample 15 in our study (assigned to Group 1, based on fluorescence properties) and sample 5 in Menor‐Salván et al. (2016) (assigned to Type 1, based on terpenoid composition); the Albian El Soplao locality in Cantabria, which produced sample 16 in this study (assigned to Group 1) and sample 3 in Menor‐Salván et al. (2016) (assigned to Type 1); and the Albian–Cenomanian Reocín locality in Cantabria, which produced sample 18 in our study (assigned to Group 2) and sample 11 in Menor‐Salván et al. (2016) (assigned to Type 2).

These associations support our preliminary hypothesis that our Group 1 corresponds to Type 1 in Menor‐Salván et al. (2016) and that Group 2 corresponds to Type 2 in Menor‐Salván et al. (2016); however, additional samples analyzed using the techniques from both studies would be necessary to confirm this. Based on our current findings, however, our results agree with Menor‐Salván et al. (2016) that Type 1/Group 1 was produced by the Araucariaceae. They interpret their Type 2 amber to be produced by the Cheirolepidiaceae, whereas we suggest that an araucariaceous origin is more likely for our Group 2. However, as both families likely contributed to the amber assemblage at the Albian El Soplao locality, it is also possible that the samples from the two studies analyzed from this locality may have originated from both families.

Within Groups 1, 2, and 3, there are some samples whose origins are known to the genus or species level, and the distribution of these samples suggests that these three groups may correspond to taxa in the Araucariaceae. The resin samples produced by Wollemia nobilis, Araucaria cunninghamii, and A. columnaris all fall into Group 1, while the copal sample produced by Agathis australis places in Group 2. The amber sample loosely associated with the genus Agathis (sample 14 in Table 1) is in Group 3. Therefore, it is likely that Group 1 samples were produced by Wollemia and Araucaria, or their close, now‐extinct relatives. Groups 2 and 3 seem to more closely correspond to the Agathis, or its close, now‐extinct relatives. Groups 2 and 3 may belong together as subgroups of a larger Agathis‐associated group. These results are similar to the results of a previous FTIR study of araucariaceous resin that found that Araucaria and Wollemia resins were most similar to each other in terms of FTIR spectra, and that Agathis resin was both distinct from and more variable than the resins of Araucaria and Wollemia (Seyfullah et al., 2015).

Distinguishing between araucariaceous and cheirolepidiaceous resins

Cheirolepidiaceous resin cannot easily be chemically characterized, because the family is extinct (Smith et al., 2024), but some posit that it was chemically very similar to araucariaceous resin, making the amber from these two families difficult to distinguish through geochemical analysis (Pańczak et al., 2023). Paleobotanical analysis is also rarely diagnostic, due to factors such as a dearth of plant bioinclusions in the amber, a mixture of araucariaceous and cheirolepidiaceous remains in the same strata that bear the amber, and the difficulty of distinguishing between the plant remains of the two families (Perrichot, 2003; Azar et al., 2010; Escapa and Leslie, 2017; Kvaček et al., 2018; Delclòs et al., 2023). Therefore, an approach to characterization that will clearly distinguish araucariaceous from cheirolepidiaceous amber is still needed.

There is no clear cluster of samples in this study with similar fluorescence properties and cheirolepidiaceaous origins. This is because our dataset lacks samples unequivocally produced by this family. Alternatively, fluorescence spectroscopy, like other analytical methods, is not able to clearly distinguish between araucariaceous and cheirolepidiaceous amber.

CONCLUSIONS

Our results demonstrate that different samples of amber have distinct fluorescence properties, showing that two‐photon fluorescence micro‐spectroscopy is a new, promising way to categorize and differentiate amber based on its fluorescence properties, instead of relying only on chemistry. Critically, two‐photon fluorescence micro‐spectroscopy is nondestructive and can be carried out on very small samples just a few millimeters in length. Importantly, our results suggest that the fluorescence properties of amber may provide information about its botanical origin that is not strongly influenced by taphonomy or geological age. Most notably, our results suggest that the two‐photon fluorescence micro‐spectroscopy method that we present here can identify the origin of amber more precisely than simply to the family level. Further research on amber from known plants will confirm whether our method can rigorously characterize the fluorescence properties of resin produced by different taxa.

AUTHOR CONTRIBUTIONS

V.E.M., M.R.S., and V.R. conceptualized the project. X.D., E.P., and C.T.G. carried out fieldwork to collect samples. M.R.S. and V.R. performed the methods development work, fluorescence spectroscopy analysis, and Gaussian deconvolution and preliminary evaluation of the fluorescence data. V.E.M., K.M.M.B., X.D., E.P., and C.T.G. interpreted the fluorescence data within a paleobiological context. V.E.M., M.R.S., and V.R. wrote the first draft of the manuscript. All authors edited, revised, and approved the manuscript.

CONFLICT OF INTEREST STATEMENT

Carole T. Gee is an associate editor of Applications in Plant Sciences, but took no part in the peer review and decision‐making processes for this paper.

Supporting information

Appendix S1: Details of the fluorescence emission spectra and Gaussian deconvolution for all samples.

DATA AVAILABILITY STATEMENT

Datasets are available at https://doi.org/10.6084/m9.figshare.27221637 (Stoneman et al., 2024b).