Brood-site deceptive A. maculatum is a rhizomatous perennial woodland herb (2n = 4x = 56) that is widespread throughout Western and Central Europe, including the British Isles, and reaches as far south as Italy, Northern Spain, and the Balkans (Boyce, 2006; Espíndola et al., 2010). It is thermogenically active, exhibits a sapromyiophilous pollination strategy, and emits a strong dung-like scent for attracting moth fly pollinators during the evening on the first day of anthesis (Kite et al., 1998; Marotz-Clausen et al., 2018). The inflorescence of A. maculatum consists of a spadix (fleshy spike) and a spathe (bract), is protogynous, and the anthesis lasts <2 days (Lack and Diaz, 1991; Marotz-Clausen et al., 2018). The spathe, which completely encloses the spadix during floral development, partially opens during anthesis to reveal the sterile appendix of the apical part of the spadix. This appendix produces and releases the scent for pollinator attraction, e.g., Lack and Diaz (1991) and Scheven (1994). At the base of the spadix, female (fertile and sterile) flowers are situated lowest, followed upwards by male flowers and staminodes (sterile male flowers). All flowers remain enveloped by the spathe during anthesis, forming a chamber that is closed by the staminodes throughout the female stage to prevent trapped insects from leaving. Pollinators are attracted in the evening on the first day of anthesis, during the female stage, slip and fall into the floral chamber, and are trapped overnight (Lack and Diaz, 1991; Gibernau et al., 2004; Espíndola et al., 2011). On the next morning, during the male stage, they are dusted with pollen, before being released at around noon when the staminodes and spathe wither (Lack and Diaz, 1991; Espíndola and Alvarez, 2011). After pollination in spring, red berry-like fruits develop as an infructescence until summer (Lack and Diaz, 1991).

During springtime in 2017–2019, we collected scents from randomly chosen A. maculatum individuals of six populations located north of the Alps (n = 106; Northwestern Austria: JOS, Josefiau; Central/Southern Germany: BUR, Burg Hohenstein; HOH, Hohendilching; MUR, Murnau; NEC, Horb am Neckar; Northern Switzerland: RÜM, Rümikon) and five populations from south of the Alps (n = 127; Northern Italy: DAO, Daone; LIM, Limone-Piemonte; MAH, Santa Maria Hoé; MON, Montese; UDI, Udine) (Figure 1 and Supplementary Table 1). We kept a minimum distance of 1 m between sampled individuals to avoid sampling potential clones, as A. maculatum can also propagate vegetatively by fragmenting rhizomes (Lack and Diaz, 1991). In summer, we harvested fruits from all individuals surveyed for scent. At most sites, we recorded scent and the fruit set of 15 individuals, except for each of the largest population per region (JOS and DAO; n = 70 each), and a northern population (HOH; n = 7) where only a few individuals had flowered at the time of scent sampling (Figure 1 and Supplementary Table 1).

Scent sampling took place on the first day of anthesis during the female stage between 6.00 p.m. and 7:30 p.m. which is the period of maximum scent emission (Marotz-Clausen et al., 2018), employing a non-invasive dynamic headspace technique. We enclosed each inflorescence in situ using a plastic oven bag (c. 30 cm × 12 cm; Toppits®, Melitta, Germany) and immediately collected scent for 5 min at 200 ml min⁻¹ on adsorbent tubes (inner diameter: 2 mm) filled with a mixture of Tenax-TA (mesh 60–80) and Carbotrap B (mesh 20–40; 1.5 mg each; both Supelco, Germany), using a battery-operated vacuum pump (rotary vane pump G12/01 EB, Gardner Denver Austria GmbH, Vienna, Austria; Marotz-Clausen et al., 2018). In the same way, we collected scent samples from leaves and ambient air as negative controls in each population.

The dynamic headspace samples were analysed by thermal desorption-gas chromatography/mass spectrometry (TD-GC/MS; Marotz-Clausen et al., 2018), and obtained data were handled using GCMSolution v.4.41 (Shimadzu Corporation, Kyoto, Japan) (see Supplementary Methods 1 for details). Compounds were chemically identified by comparison of Kováts' retention indices (KRIs), based on commercially available n-alkanes (C₇-C₂₀; Sigma Aldrich, Vienna, Austria), and mass spectra to data available in the libraries of Adams (Adams, 2007), FFNSC 2, Wiley9, NIST11, and ESSENTIAL OILS (available in MassFinder 3, Hochmuth Scientific Consulting, Hamburg, Germany). We established our own library of mass-spectral and KRIs for semi-automatic analysis (Supplementary Methods 1). Whenever possible, compounds were verified by comparison with authentic reference compounds available in the collection of the Plant Ecology Lab of Salzburg University or with chemically synthesised reference compounds (Supplementary Methods 2). In total, 233 scent samples yielded a sufficiently informative chromatogram and were included in the analyses (Figure 1). Ultimately, a compound was only considered if it occurred in more than three scent samples and did not occur in leaf and air controls.

Percentage fruit set, i.e., number of fruits by total number of flowers per individual × 100, was determined as a measure of female reproductive success. For selection analyses, we further estimated relative fruit set, i.e., number of fruits per individual divided by mean number of fruits per given population, (e.g., Gross et al., 2016), for the most extensively sampled populations JOS and DAO. In one southern population (MON), a shallow landslide destroyed all plants, with the exception of one; hence, this population was excluded from fruit set analyses.

The total absolute amount of scent was highly variable among the 233 sampled individuals of A. maculatum (range of 1–2,052 ng inflorescence⁻¹ h⁻¹; Table 1). When taken together, northern plants released a 3-fold lower amount of scent than those from the South, along with differences among populations within regions (permANOVA: region: pseudo-F_{(1, 222)} = 25.7, population nested within the region: pseudo-F_{(9, 222)} = 5.36, both P < 0.001). For three of the five southern populations (MAH, MON, LIM; Figure 1), we estimated a median scent amount of c. 200 ng inflorescence⁻¹ h⁻¹, while DAO and UDI showed 1.5-fold higher and 5-fold lower amounts, respectively (Table 1 and Supplementary Table 1). For three of the six northern populations (MUR, NEC, RÜM; Figure 1), median estimates ranged between 40 and 81 ng inflorescence⁻¹ h⁻¹, while amounts in the remaining populations were manifold higher (JOS and HOH) or lower (BUR) (Table 1 and Supplementary Table 1).

Across all scent samples, we detected a total of 289 floral volatiles (285 north vs. 277 south), of which 92 could be chemically identified (Table 1 and Supplementary Table 2). A median of 102 compounds per individual was recorded (Figure 2), and the number of compounds was independent of the region (permANOVA: pseudo-F_{(1, 222)} = 1.98, P = 0.16) but varied among populations within regions (pseudo-F_{(9, 222)} = 4.57, P = 0.001). At the population level, between 186 (BUR) and 271 (JOS) compounds were recorded in the North, and between 188 (MON) and 257 (DAO) in the South (Figure 2). The two most extensively sampled northern (JOS) vs. southern (DAO) populations covered 96 vs. 94% of their respective regional diversity (Figure 2) and together 99% (287/289) of the total number of compounds (Table 1 and Supplementary Table 2). The five most frequent compounds found in more than 99% of the samples were the nitrogen-bearing compound indole, the monoterpenoids 3,7-dimethyloct-1-ene and β-citronellene, the sesquiterpenoid β-caryophyllene, and the unidentified UNK1492 (Table 1).

The absolute amounts of single compounds significantly differed between regions (permANOVA: pseudo-F_{(1, 222)} = 22.53, P < 0.001), between JOS and DAO only (pseudo-F_{(1, 109)} = 9.95, P < 0.001), and among populations within regions (pseudo-F_{(9, 222)} = 6.44, P < 0.001). However, differences were more pronounced between regions than among populations within regions (north vs. south OOB error: 10.3%; among populations within north OOB error: 27.3%; within south OOB error: 25.2%). Only a few abundant compounds dominated the scent bouquet of A. maculatum, including indole, β-citronellene, the unknown UNK1415, and 3,7-dimethyloct-2-ene (all abundant in both regions), p-cresol (most abundant only north), and 2-heptanone (only south, Table 1).

We also detected differences in the relative amounts of scent compounds between regions (permANOVA: pseudo-F_{(1, 222)} = 30.18, P < 0.001), between JOS and DAO only (pseudo-F_{(1, 109)} = 22.81, P < 0.001), and among populations within regions (pseudo-F_{(9, 222)} = 4.90, P < 0.001; Figure 2). Again, these differences were more pronounced at the inter-regional than within-region levels (north vs. south OOB error: 9%; among populations within north OOB error: 35.8%; among populations within south OOB error: 24.4%; see also Supplementary Figure 2).

Across all populations, variation in absolute or relative amounts of scent could not be explained by their geographic distances (Mantel's Rho = 0.108, P = 0.25 and Rho = −0.154, P = 0.85, respectively).

Among the 25 compounds each that were most responsible for regional differences in the absolute and relative datasets in the randomForest analyses, 20 were common to both datasets (Supplementary Table 3). These 20 compounds included 2-heptanone, 2-heptanol, and α- and β-citronellene, all of which were more abundant (in relative and absolute amounts) south of the Alps, and 1-pentadecanol, the unknown UNK1503, p-cresol, and indole, which occurred in higher amounts north of the Alps (Table 1 and Supplementary Tables 2, 3). Many of these compounds, and some non-overlapping ones (absolute: α-copaene, β-caryophyllene; relative: UNK1409, bicyclogermacrene), explained most of the variation in scent among all samples (for relative data see Figure 3; for absolute data see Supplementary Table 4).

There was also a considerably high variation in scent within populations, most prominently in the most extensively sampled northern (JOS) and southern (DAO) populations, which harboured almost all of the absolute and relative scent variation of their respective regions (for relative data see Figure 3).

Among the 233 individuals surveyed for inflorescence scent, 113 set fruit in summer. Percentages of fruit set were significantly higher north of the Alps (42 ± 41% mean ± SD, 0–100% Min–Max) than south of the Alps [26 ± 33% mean ± SD, 0–100% Min–Max; Figure 4; region: F_{(1, 209)} = 10.11, P = 0.002] and differed significantly among populations within regions [population nested within region: F_{(8, 209)} = 2.23, P = 0.03].

In the most extensively sampled northern (JOS) and southern (DAO) populations, we tested 19 and three compounds for phenotypic selection, respectively, as they correlated with relative fruit set in the elastic net and Boruta analyses (see section Materials and Methods; Supplementary Methods 3). Among those 22 compounds, seven showed signals of linear phenotypic selection (two of which as an interaction), all in the north, and two for non-linear (quadratic) phenotypic selection, all in the south (Figure 5). Seven of the overall nine compounds that were under phenotypic selection correlated positively with relative fruit set (linear: 2-heptanol, 2-nonanol, α-terpinene, UNK681, and UNK1496 together with UNK1503; non-linear: sabinene), while two correlated negatively (linear: UNK960; non-linear: 4-terpinenol; Figure 5 and Supplementary Table 5).

Among the 25 compounds that most strongly contributed to the absolute (and relative) differences in scent between the regions, only four were under selection (north: 2-heptanol, 2-nonanol, UNK681, UNK1503), but not others (e.g., 2-heptanone, α- and β-citronellene; Supplementary Table 3). Differences in absolute and relative scent traits between the northern JOS and the southern DAO populations remained significant, regardless of performing permANOVA separately on the nine compounds that correlated with relative fruit set in the elastic net/Boruta analyses and were under selection [absolute vs. relative datasets: pseudo-F_{(1, 109)} = 18.4 vs. 30.8, both P < 0.001], or on the 93 compounds that were not under selection and did not correlate with fruit set [absolute vs. relative datasets: pseudo-F_{(1, 109)} = 9.8 vs. 24.5, both P < 0.001; see Supplementary Figure 3, Supplementary Methods 3].

With 289 floral volatiles recorded, the inflorescence scent diversity of A. maculatum is extraordinarily high and not matched by any other plant species to the best of our knowledge. In fact, we are not aware of any species from which more than 200 floral compounds are reported, a number that a single A. maculatum individual can reach by three quarters (max. = 152 VOCs; Figure 2). This difference in the number of scent compounds between A. maculatum and other species cannot just be explained by differences in techniques used for scent analyses, given that scents of a high number of species were analysed using a similar approach as we did (dynamic headspace and thermal desorption of samples; Gottsberger et al., 2013; Borchsenius et al., 2016; Lukas et al., 2019). Species closest to the high number of VOCs in A. maculatum include the sapromyiophilous Sauromatum guttatum (Araceae, with altogether 196 different VOCs; Skubatz et al., 1996; Hadacek and Weber, 2002) and Aristolochia gigantea (Aristolochiaceae, 168 VOCs; Martin et al., 2017), as well as the insect-pollinated and rewarding Geonoma macrostachys (Arecaceae, 176 VOCs; Borchsenius et al., 2016) and Echinopsis ancistrophora (Cactaceae, 145 VOCs; Schlumpberger and Raguso, 2008). Other species for which c. 100 VOCs are described likewise include insect-pollinated and rewarding species [e.g., Philodendron bipinnatifidum (Araceae), Gottsberger et al., 2013; Pyrus communis (Rosaceae), Lukas et al., 2019], but also the sexually deceptive orchid Ophrys sphegodes (Ayasse et al., 2000). Thus, high numbers of compounds are found across a wide range of plant families and are apparently not restricted to a specific pollination system.

One explanation for the high diversity of scent compounds in A. maculatum is that this species likely imitates the various breeding substrates of its moth fly pollinators, all potentially differently scented. The two main pollinators, P. phalaenoides and P. grisescens, breed in a variety of different substrates such as rotting manure from cattle and horse, fungi (P. grisescens), waste pits, mud-flats, plant litter in drainages (P. phalaenoides), and the hygropetric zones of river banks and ponds (Satchell, 1947; Ježek, 1990; Sigsgaard et al., 2020). Arum maculatum emits compounds described from several substrates, such as cattle and horse manure (e.g., indole, p-cresol, skatole), fungi (e.g., 1-octen-3-ol, (E)-2-octen-1-ol, 3-octanone), and general degrading and fermenting plant or animal material (e.g., 2,3-heptanedione, acetoin, butanoic acid) (Dormont et al., 2010; Jürgens et al., 2013). Highly specialised deceptive plant systems frequently rely on only a few volatiles to attract pollinators; they seem to imitate a more specific model, thus releasing less complex scent blends (e.g., Wee et al., 2018).

The number of volatiles detected across the 233 samples (11 populations) of A. maculatum (289 VOCs) is five to 10 times higher than previously reported for this species (18–61, and 143 VOCs in total; Scheven, 1994; Diaz and Kite, 2002; Chartier et al., 2013; Marotz-Clausen et al., 2018; Szenteczki et al., 2021; and references therein). This discrepancy cannot be explained by differences in sample size, as a similar number of individuals were surveyed in those previous studies (n = 222 in total, representing 23 populations). Interestingly, we found a similar number of compounds in some individuals (up to 152 VOCs; median of 102; Figure 2) as overall detected previously (Diaz and Kite, 2002; Chartier et al., 2013; Marotz-Clausen et al., 2018; Szenteczki et al., 2021; and references therein). With the exception of two studies (Marotz-Clausen et al., 2018; Szenteczki et al., 2021), each sharing one of our sampled populations (JOS and MON, respectively), all previous studies sampled scents in other populations across Europe (Kite, 1995; Diaz and Kite, 2002; Chartier et al., 2013). Thus, some of the differences in the number of A. maculatum compounds detected across studies might reflect population-specific scent characteristics (see Figure 2). However, more importantly we believe that the discrepancy in the number of compounds recorded largely reflects differences in methodology between the present and previous studies. For example, these are: higher sensitivity of modern GC/MS systems, usage of less selective adsorbent agents [Carbotrap/Tenax-TA vs. polydimethylsiloxane/ divinylbenzene (Chartier et al., 2013) vs. polydimethylsiloxane (Szenteczki et al., 2021)]; in situ vs. ex situ samplings (Scheven, 1994; Marotz-Clausen et al., 2018), and including all vs. only compounds above a specific threshold in relative amounts (Chartier et al., 2013; Szenteczki et al., 2021). Among the 92 compounds chemically identified in this study, more than half (50) were previously unknown to be released by A. maculatum. Some of these newly described compounds for A. maculatum are known from other species of Araceae (e.g., α-cubebene, β-phellandrene, γ-terpinene, Sauromatum guttatum, Hadacek and Weber, 2002), or other plant families (e.g., methyl anthranilate, isobutyl butyrate, citronellal, Knudsen et al., 2006; El-Sayed, 2019). To the best of our knowledge, this study is, however, the first to identify p-cresyl butyrate as a floral scent compound.

In the region south of the Alps, the qualitative, absolute, and relative differences in scent among populations of A. maculatum may be related to the pollinator assemblages that are in this region more diverse in terms of abundance, species composition, and sex ratio (Espíndola et al., 2011; Laina et al., unpublished data). In the region north of the Alps, females of P. phalaenoides are the principal pollinators in all studied populations (Espíndola et al., 2011; Laina et al., unpublished data), even though other Psychoda spp. also occur in this region (Chartier et al., 2013; Laina et al., unpublished data). Hence, the scent variation we observed among northern populations is not reflected by variations in pollinator spectra in this region. In this study, all variations in scent were more pronounced between regions than among populations within each region. This strong regional component of scent variation in A. maculatum across the Alps thus accords with strong differences in pollinator spectra (Espíndola et al., 2011; Laina et al., unpublished data) and coincides with a genetic (AFLP) subdivision of A. maculatum across this geographic barrier (Espíndola and Alvarez, 2011). Differing pollinator availability has been linked to different climatic factors (Espíndola et al., 2011), which might also influence scent variation (e.g., Farré-Armengol et al., 2014). However, a preliminary transplant experiment shows that A. maculatum individuals originating from north or south of the Alps keep their population-typic scent after transplantation (Gfrerer et al., unpublished data). This suggests that abiotic factors do not directly influence scent emissions (see also Szenteczki et al., 2021). However, we cannot exclude that they might exert differential selection pressures, thus influencing evolutionary processes that may lead to differences in scent emission between the regions. Previous studies in A. maculatum also found population effects in scent composition, e.g., Chartier et al. (2013) and Szenteczki et al. (2021). Nonetheless, our study is the first to demonstrate such population differentiation in scent across the Alps. Intraspecific variation in floral scent among populations and regions has also been reported for other plant species (e.g., Dötterl et al., 2005; Chapurlat et al., 2019; and Schlumpberger and Raguso, 2008), including sapromyiophilous species (e.g., Chen et al., 2017). In some of those, this variation, as shown in this study, could be linked to pollinator assemblages and/or genetic patterns (e.g., Chapurlat et al., 2019), but not in others (Dötterl et al., 2005; Schlumpberger and Raguso, 2008).

The two most extensively sampled northern (JOS) and southern (DAO) populations differed in absolute and relative amounts of scent, regardless of whether the analyses were conducted on all compounds, on only those that correlated with relative fruit set and were under selection, or on those that did not correlate with fruit set (Material and methods, Supplementary Figure 3). Thus, this regional difference in scent could be caused by different selection regimes, as well as other reasons, such as phenotypic plasticity (but see Szenteczki et al., 2021) or genetic drift (Herrera et al., 2006; Majetic et al., 2009). In support of differential selection, we detected population-specific signatures of phenotypic selection on scent in JOS and DAO, possibly due to different olfactory preferences of those Psychoda species that dominate the pollinator spectra of A. maculatum in the northern (female P. phalaenoides) vs. southern (and P. grisescens) regions (Espíndola et al., 2011; Chartier et al., 2013; Szenteczki et al., 2021; Laina et al., unpublished data).

For the five compounds under phenotypic selection that we were able to chemically identify, i.e., 2-nonanol, 2-heptanol, sabinene, 4-terpinenol and α-terpinene, information on their attractiveness to pollinators of A. maculatum is lacking. However, the aliphatic compounds 2-heptanol and 2-nonanol are known, either together or alone, as attractants for bees (Meliponini, Pianaro et al., 2009) and kleptoparasitic flies (Heiduk et al., 2016). They are also known as (sex-)pheromones of female Diptera (Cecidomyiidae, Censier et al., 2014) and female non-Diptera (Trichoptera, Löfstedt et al., 1994). The monoterpenoids sabinene, α-terpinene, and 4-terpinenol are defence substances of some insects (Coleoptera, e.g., Wheeler et al., 2002; Lepidoptera, Ômura et al., 2006) but are used by others (e.g., Lepidoptera, Baur et al., 1993) as oviposition stimulants. The latter two volatiles are also pheromones of fruit flies (Fletcher et al., 1992). In summary, these five compounds, found to be under phenotypic selection, elicit responses in insects other than moth flies. Furthermore, they are known from the floral scent of other sapromyiophilous species (e.g., Hadacek and Weber, 2002; Johnson and Jürgens, 2010), and some of them (α-terpinene and 4-terpinenol) are also known from cattle dung (Dormont et al., 2010; Sládeček et al., 2021), i.e., one of the oviposition substrates of moth flies. Further research is required to establish whether these five compounds, which are all widespread floral scent compounds (Knudsen et al., 2006; El-Sayed, 2019), are attractive to the pollinators of A. maculatum.

Several of the compounds most responsible for regional differences in inflorescence scent, e.g., 2-heptanone, 3,7-dimethyloct-1-ene, UNK966 (Supplementary Table 3), did not show signals of phenotypic selection (see Figure 5). Thus, the different selection regimes cannot explain several of the most obvious differences in scent between A. maculatum from north and south of the Alps (see also Supplementary Figure 2). However, some other compounds that also differed in absolute amounts between regions (2-heptanol, 2-nonanol, UNK681, sabinene; Supplementary Tables 3, 5) were under phenotypic selection, either in northern JOS or southern DAO (Figure 5), and some of the differences between regions could, therefore, be due to differential selection.

Somewhat unexpectedly, we did not find phenotypic selection for the most abundant compounds in the scent of A. maculatum, e.g., indole, β-citronellene, unknown UNK1415, with the exception of 2-heptanol (Figure 5). Even more surprisingly, we also did not find phenotypic selection for those compounds known to attract P. phalaenoides, i.e., indole, 2-heptanone, p-cresol, and α-humulene (Scheven, 1994; Kite et al., 1998), occurring both north and south of the Alps, and also in JOS and DAO (Espíndola et al., 2011). This contrasts with most other studies, where main compounds and/or pollinator attractants showed signals of phenotypic selection (but see Chapurlat et al., 2019). In Penstemon digitalis (Plantaginaceae), one of the main compounds, linalool, was under phenotypic selection and attractive to bumblebees in the laboratory but not in field bioassays (Parachnowitsch et al., 2012; Burdon et al., 2020). Possible explanations for not finding phenotypic selection on the main compounds of A. maculatum include the following: (1) their released amounts are high enough to achieve maximum pollinator attractiveness (see also Chapurlat et al., 2019); (2) there are opposing selection pressures on these compounds by different pollinators or herbivores, resulting in zero ‘net' selection (e.g., Knauer and Schiestl, 2017; Chapurlat et al., 2019); and (3) their relationship with flower visitors is non-linear and non-quadratic (e.g., Galen et al., 2011). Although our multivariate models detected non-linear phenotypic selection by including quadratic terms, such quadratic analyses cannot uncover all potential non-linear relationships (e.g., Stinchcombe et al., 2008). Hence, we cannot exclude the possibility that such abundant and/or attractive compounds are still under phenotypic selection, which in turn calls for future statistical developments that allow testing for any kind of non-linear multivariate relationships.

Deceptive plant species might experience stronger selection than rewarding ones (Sletvold and Ågren, 2014). However, by comparison with rewarding species (Parachnowitsch et al., 2012; Gross et al., 2016; Gervasi and Schiestl, 2017; Chapurlat et al., 2019), we found that deceptive A. maculatum does not release a higher number of volatiles with signatures of phenotypic selection (7 vs. 3–42%), but these volatiles appear to be under slightly stronger positive linear phenotypic selection (−0.3 to 0.5 vs. −0.3 to 0.4, Min to Max; Figure 5; Parachnowitsch et al., 2012; Gross et al., 2016; Chapurlat et al., 2019) and stronger non-linear phenotypic selection (−0.9 to 9 vs. −0.5 to −0.3, Min to Max; Figure 5; Gervasi and Schiestl, 2017). Future studies on other deceptive plant species that also attract specific pollinators by chemical cues, but have lower levels of fruit set than A. maculatum (such as many orchids, e.g., Tremblay et al., 2005), might reveal even stronger signatures of phenotypic selection.