Identifying biochemical phenotypic differences between cryptic species

Abstract

Molecular genetic methods can distinguish divergent evolutionary lineages in what previously appeared to be single species, but it is not always clear what functional differences exist between such cryptic species. We used a metabolomic approach to profile biochemical phenotype (metabotype) differences between two putative cryptic species of the earthworm Lumbricus rubellus. There were no straightforward metabolite biomarkers of lineage, i.e. no metabolites that were always at higher concentration in one lineage. Multivariate methods, however, identified a small number of metabolites that together helped distinguish the lineages, including uncommon metabolites such as Nε-trimethyllysine, which is not usually found at high concentrations. This approach could be useful for characterizing functional trait differences, especially as it is applicable to essentially any species group, irrespective of its genome sequencing status.

1. Introduction

Molecular genetic methods have been so successful in revealing hidden diversity that it is no longer surprising to find that what we previously thought of as a single species is, in fact, two (or more) cryptic species. However, the ecological implications of distinguishing such cryptic species are not necessarily straightforward. Mechanisms for species isolation are sometimes identifiable, e.g. segregation based on non-visual mating signals [1], but in many other cases the drivers of cryptic speciation are yet to be resolved, and the resulting functional differences between species remain to be clarified [2].

One approach to understanding functional diversity is to sequence the genomes of the different cryptic species. However, knowing a genome sequence is not equivalent to understanding all of its ecological consequences. Firstly, many genes will not be annotated; secondly, even for known genes, it may not be possible to infer organism-level effects from genetic differences alone. As a result, we also need to measure phenotypic differences. Untargeted metabolic profiling (metabolomics) offers one of the most direct measures of cellular phenotype: it integrates the effects of regulation at different biological levels—transcriptional, translational and post-translational [3]. Metabolomics has been frequently used as a chemotaxonomic tool for classifying different groups, most usually of plants, but previous studies have generally focused on well-defined populations or cultivars; often, samples from different geographical locations are analysed, with no attempt to distinguish between genetic and environmental effects [4,5]. There have been few attempts to analyse the effects of genetic variation on metabolome profiles in an environmental context [5–7].

Since morphological taxonomy relies on visible traits, it has been suggested that cryptic species may be more frequent when non-visual cues dominate intra-species interactions and hence selection for morphological traits is limited [8]. The soil is one such case, and analyses for springtails [9], oribatids [10] and molluscs [11,12] have all revealed a high degree of crypsis. Earthworms have high potential for local population isolation, and many named earthworm species are probably species complexes [13]. Lumbricus rubellus is present as potential cryptic species in the UK, as it is found as two lineages with around 14% sequence divergence in the COII gene; as it is important to anchor molecular observations to classic taxonomic descriptions [14], and this has not yet been done for L. rubellus, we follow previous studies' precedent in calling these lineages A and B [15]. Little is yet known about any phenotypic differentiation of the lineages, although there is some variation in glandular tumescences between them [16]. We previously found unexpected diversity in secondary metabolites between two closely related earthworm species, Eisenia fetida and Eisenia andrei [17], and so we wondered if metabolic differences could also be useful in the current case. We used NMR-based metabolomics to identify biochemical differences between L. rubellus lineages from multiple field populations, spanning a wide range of environmental factors.

2. Material and methods

(a). Earthworm collection and genotyping

We collected L. rubellus earthworms on two independent series of field trips, which we designate as ‘set 1’ and ‘set 2’ worms. Both sets of worms contained individuals from both lineages A and B. They were collected by digging and hand-sorting from October to November 2010 (set 1) and November 2006 to April 2008 (set 2). The earthworms were snap-frozen on-site (set 1), or transported back to the laboratory and depurated on filter paper for 48 h (set 2). Genotyping was carried out by sequencing the COII gene for all individual worms, as previously described [18].

(b). Sample preparation

The extraction procedure for the set 1 worms is described in detail elsewhere [19]. Briefly, the worms were ground under liquid nitrogen and the frozen powder extracted directly into water/acetonitrile/methanol (1 : 2 : 2 ratio v/v/v). A high-concentration metabolite was then removed using solid phase extraction. For the set 2 worms, we ground the worms under liquid nitrogen using a pestle and mortar, and then extracted the tissue into 60% acetonitrile, before drying the samples under reduced pressure. All samples were reconstituted in NMR buffer immediately before analysis. The NMR buffer contained phosphate buffer (pH 7.0) plus approximately 1 mM of an internal chemical shift reference (sodium 4,4-dimethyl-4-silapentane-1-sulfonate-²H₆ for the set 1 worms, and sodium trimethylsilyl-²H₄-propionate for the set 2 worms) and was made up in ²H₂O. Earthworm metabolite extracts are prone to further enzymatic alterations when redissolved in an aqueous buffer, and so we either used a brief heat-treatment step as previously described [19], which allowed samples to be run under automation (set 1 worms), or kept the samples on ice until acquisition (set 2 worms).

(c). Metabolite analysis

We obtained metabolite profiles by proton nuclear magnetic resonance (NMR) spectroscopy, as described in detail by Beckonert et al. [20]: briefly, we acquired the spectra at 600 MHz using an Avance DRX600 spectrometer (Bruker, Rheinstetten, Germany) with an inverse-geometry 5 mm probe with samples held at 300 K. We processed the spectra in iNMR 3 (Nucleomatica, Molfetta, Italy): the spectra were zero-filled by a factor of 1.5, and an exponential apodization function equivalent to 0.5 Hz line-broadening was applied, followed by Fourier transformation. The spectra were corrected for phase and a first-order polynomial baseline correction made. We data-reduced the spectra by integrating into 104 bins manually selected to contain, as far as possible, resonances from individual metabolites. The data were then normalized to the total signal intensity. Statistical analysis was carried out using Simca-P+ 13.2 (Umetrics, Crewe, UK) and JMP 9.0.0 (SAS, Marlow, UK).

3. Results

We collected adult L. rubellus from different field populations (the set 1 worms), chosen to cover a range of geographical and soil characteristics (table 1), and then acquired untargeted metabolite profile data. By far the largest source of metabolic variation was which site the worms were collected from. Unsupervised clustering methods clearly showed discrimination between individual sites, and the ratio of between-site to within-site variance was high across all individual bins (electronic supplementary material, figure S1).

Table 1.

Sites and site properties.

set	site name	sampling date	latitude	longitude	soil pH	soil moisture (%)	soil organic carbon (%)	land use	lineage A worms found	lineage B worms found
1	SCTc	11/10/2010	53.640668	−0.567680	6.79	17.8	6.04	grassland	26	0
1	SCTp	19/10/2010	53.595921	−0.594973	6.43	20.7	9.14	grassland	14	0
1	SND	20/10/2010	53.075826	−4.033613	6.04	49	13.9	grassland	14	11
1	CWMc	21/10/2010	52.360926	−3.747023	5.65	38.8	18.4	grassland	1	13
1	CWMp	21/10/2010	52.360089	−3.743421	5.62	28.9	12.7	mine site	3	14
1	DRA	11/10/2010	52.193504	−1.762976	6.54	33.8	17.1	grassland	0	21
1	AMTp	11/10/2010	51.536457	−2.623850	6.3	38.2	24.5	grassland	22	0
1	PTBc	12/10/2010	51.548574	−3.671153	6.26	44.3	18.1	grassland	16	3
1	PTBp	12/10/2010	51.555836	−3.744873	5.97	25.6	15.1	grassland	21	0
1	AMCc	11/10/2010	51.533153	−2.667012	5.86	30.6	14.2	grassland	1	1
1	AMCp	11/10/2010	51.513495	−2.668987	6.5	25.5	7.94	grassland	17	0
1	DGCc	12/10/2010	50.544322	−4.222484	5.49	36.1	17.2	grassland	5	15
1	DGCp	12/10/2010	50.538145	−4.222097	5.3	45.4	29.7	mine site	7	8
1	SHPc	11/10/2010	51.330935	−2.763240	5.38	28.6	11.7	grassland	1	15
1	SHPp	11/10/2010	51.313283	−2.793163	6.33	38.9	33	mine site	0	23
1	PDW	11/10/2010	51.126701	−1.638419	6.75	31.5	12.9	grassland	8	1
1	AHT	10/11/2010	51.154994	−0.860450	4.93	28	13.2	woodland	6	8
2	CL01	20/11/2006	51.696182	−3.888506	n.a.	n.a.	n.a.	grassland	23	0
2	CL02	20/11/2006	51.708333	−3.865556	n.a.	n.a.	n.a.	woodland	6	9
2	CL11	01/12/2006	51.731944	−3.831944	n.a.	n.a.	n.a.	woodland	10	0
2	CL12	29/04/2008	51.698889	−3.886667	n.a.	n.a.	n.a.	grassland	9	0
2	DP03	23/05/2007	51.437833	−3.218722	n.a.	n.a.	n.a.	woodland	10	5
2	DP04	23/05/2007	51.443861	−3.239361	n.a.	n.a.	n.a.	farm site	10	0
2	DP06	31/05/2007	51.493167	−3.203889	n.a.	n.a.	n.a.	woodland	10	0
2	DP07	31/05/2007	51.452583	−3.305306	n.a.	n.a.	n.a.	woodland	10	0
2	RU01	27/03/2007	51.573333	−3.183611	n.a.	n.a.	n.a.	woodland	10	0
2	RU02	24/04/2007	51.574444	−3.184444	n.a.	n.a.	n.a.	mine site	10	0
2	RU03	28/03/2007	51.571944	−3.181111	n.a.	n.a.	n.a.	woodland	9	10
2	RU05	03/04/2007	51.563333	−3.178889	n.a.	n.a.	n.a.	grassland	10	0
2	RU06	04/04/2007	51.560000	−3.187222	n.a.	n.a.	n.a.	grassland	20	0
2	RU07	04/04/2007	51.567778	−3.180000	n.a.	n.a.	n.a.	grassland	10	0
2	RU08	25/04/2007	51.568889	−3.178611	n.a.	n.a.	n.a.	woodland	10	0

The lineages were unevenly distributed across sites: some had only lineage A worms, some only B and some both (table 1). We used orthogonal partial least-squares discriminant analysis (OPLS-DA) to test for robust metabolic lineage differences despite the site variation. We fitted a model with one predictive and two orthogonal components (based on leave-one-site-out cross-validation) and validated it by permutation analysis. The two lineages were clearly different; the loadings indicated a number of bins for further investigation, in particular two at 3.10 and 7.69 ppm (figure 1a,b). We assigned the first of these as the betaine compound laminine (Nε-trimethyllysine, NCBI PubChem ID 159659), which has recently been identified in earthworms [21]. The other is not yet assigned but is from an aromatic metabolite structurally similar to aromatic compounds such as p-hydroxybenzoate (electronic supplementary material, figure S2).

Figure 1.

The two earthworm lineages have robust metabolic differences. (a) Cross-validated scores for OPLS-DA analysis of lineages A and B; individual data points are shown on top of boxplots with distribution shape. (b) OPLS-DA loadings; points also identified as significant by univariate analysis (table 2) marked in red. Laminine and unassigned aromatic compound labelled on plot. (c) Set 1 worms and (d) set 2 worms: laminine versus aromatic compound, both shown as log₁₀-transformed intensities. Lineage A, black; lineage B, red. Points are connected to the group centroids. (Online version in colour.)

We then used univariate analysis as a complementary approach to identify metabolites related to lineage: we selected only sites with both A and B worms present (with a minimum of four worms per lineage per site), reducing the dataset to five sites and 89 individual worms, and evaluated linear models against the categorical variables site and lineage, plus an interaction term, for all bins. No metabolites were always higher in one lineage in all worms from all sites. However, five metabolites were significantly related to lineage (p < 0.01), even when site was included as a factor. Four of these retained significant associations with lineage when considering the potential confounding factors of soil pH, soil moisture, soil organic carbon and land use (table 2). These included the two metabolites identified earlier from the OPLS-DA, laminine and the unassigned aromatic metabolite. These two metabolites alone were sufficient to discriminate the two lineages to a reasonable extent (figure 1c; area under ROC curve = 0.84).

Table 2.

Linear model statistics for four metabolite bins related to lineage; two-parameter models evaluated against lineage and each potential confounder in turn, i.e. metabolite concentration = f(lineage, confounder). For each model, the top line of the cell gives ; the middle line gives the variance ratio (F) and associated p-value (in parentheses) for the effect of lineage; and the bottom line gives F- and p-values for the effect of the potential confounder.

potential confounder	δ 2.22 ppm (unassigned)	δ 2.89 ppm (unassigned)	δ 3.10 ppm (laminine)	δ 7.70 ppm (unassigned aromatic metabolite)
site	0.34	0.42	0.59	0.24
	19.4 (p < 0.0001)	11.3 (p = 0.0012)	7.65 (p = 0.007)	16.7 (p = 0.0001)
	9.47 (p < 0.0001)	14.9 (p < 0.0001)	26.5 (p < 0.0001)	2.3 (p = 0.065)
soil pH	0.07	0.17	0.42	0.19
	8.47 (p = 0.0046)	10.0 (p = 0.0022)	7.17 (p = 0.0089)	19.5 (p < 0.0001)
	0.03 (p = 0.86)	13.3 (p = 0.0004)	49.6 (p < 0.0001)	0.59 (p = 0.44)
soil organic carbon (%)	0.08	0.06	0.09	0.20
	9.19 (p = 0.003)	5.81 (p = 0.018)	10.6 (p = 0.0016)	21.5 (p < 0.0001)
	0.53 (p = 0.53)	2.00 (p = 0.16)	0.30 (p = 0.59)	1.9 (p = 0.17)
soil moisture (%)	0.14	0.05	0.16	0.19
	8.28 (p = 0.0051)	5.97 (p = 0.017)	9.99 (p = 0.0022)	22.6 (p < 0.0001)
	6.28 (p = 0.014)	1.32 (p = 0.25)	6.69 (p = 0.011)	1.27 (p = 0.26)
land use	0.12	0.07	0.25	0.22
	12.0 (p = 0.0008)	6.03 (p = 0.016)	6.26 (p = 0.014)	18.1 (p < 0.0001)
	2.89 (p = 0.060)	1.84 (p = 0.16)	9.53 (p = 0.0002)	2.49 (p = 0.089)

We validated this separation of the two lineages using an entirely independent set of field samples (set 2 worms). Comparing the same two metabolites demonstrated similar differences between the lineages (figure 1d).

4. Discussion

There have been a surprising number of earthworm metabolomic studies to date [22], but little evidence so far of possible intra-species metabolic variation. A previous observation of possible bimodal distributions of metabolites in L. rubellus was not based on enough individuals to confirm the existence of sub-populations [23]. Metabolomics is often used to study the relationship of genotype to phenotype [24]. However, are metabolite differences necessarily functional? A metabolic change represents an actual biochemical difference, whereas changes at the genome level do not always result in phenotypic changes. However, changes in a biochemical function do not necessarily imply a change in ecological function. It is likely, though, that the metabolic differences are physiologically relevant, as NMR detects relatively high-concentration metabolites only. For instance, laminine is an intermediate in carnitine biosynthesis, but it is unlikely that the metabolic changes between the lineages are caused by differences in carnitine metabolism: usually this metabolite is not detectable by NMR in tissue extracts and so its occurrence in earthworms in relatively high concentrations probably has a different biochemical explanation. Earthworms contain a remarkable variety of betaine metabolites, and laminine appears to be produced as part of this general biochemical ability to synthesize betaines [21]. It is not yet known why earthworms make this range of betaines, but it is plausible that they contribute to earthworms' ability to survive in habitats of variable moisture, and/or osmotic stress [25]. Most probably the difference between the lineages is a consequence of mutations in the genes coding for the enzymes that synthesize betaines in earthworms; we do not yet know if there are ecophysiological lineage differences because of this.

The metabolic separation between the lineages is only partial, i.e. the metabolomic data could not be used as an alternative for classifying them with complete specificity (figure 1). Furthermore, it is not just a matter of baseline noise in the data: these metabolite concentrations can also be affected by other environmental factors, which can have larger effects than lineage, e.g. as shown by the variance ratios for lineage + confounder for laminine (table 2). The association with lineage is despite the fact that other factors also affect the metabolite levels.

A key benefit of metabolomic approaches is that samples can be compared directly without any need for prior knowledge of gene/protein sequence, nor homology-based classification of identities [17]. This is clearly true for primary metabolites: many common high-concentration metabolites are detected routinely by NMR. Here, though, the untargeted approach has identified lineage-related biochemical differences in ‘unusual’ compounds. Hence, metabolomic analysis can identify genuinely novel phenotypic differences between otherwise-identical cryptic species, even though taken from many sub-populations from varied environments. This approach could therefore become an important new molecular tool for functional ecologists, allowing generation of testable hypotheses about cryptic species.

Funding statement

This study was supported by the UK Natural Environment Research Council, grant no. NE/H009973/1, and by the Leverhulme Trust, grant no. F/00407/AI.